drivers/staging/lustre/lustre/mdc/mdc_request.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Intel Corporation.
  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 #define DEBUG_SUBSYSTEM S_MDC
  38
  39 # include <linux/module.h>
  40 # include <linux/pagemap.h>
  41 # include <linux/miscdevice.h>
  42 # include <linux/init.h>
  43 # include <linux/utsname.h>
  44
  45 #include "../include/lustre_acl.h"
  46 #include "../include/obd_class.h"
  47 #include "../include/lustre_fid.h"
  48 #include "../include/lprocfs_status.h"
  49 #include "../include/lustre_param.h"
  50 #include "../include/lustre_log.h"
  51
  52 #include "mdc_internal.h"
  53
  54 #define REQUEST_MINOR 244
  55
  56 struct mdc_renew_capa_args {
  57         struct obd_capa *ra_oc;
  58         renew_capa_cb_t  ra_cb;
  59 };
  60
  61 static int mdc_cleanup(struct obd_device *obd);
  62
  63 int mdc_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
  64                     const struct req_msg_field *field, struct obd_capa **oc)
  65 {
  66         struct lustre_capa *capa;
  67         struct obd_capa *c;
  68
  69         /* swabbed already in mdc_enqueue */
  70         capa = req_capsule_server_get(&req->rq_pill, field);
  71         if (capa == NULL)
  72                 return -EPROTO;
  73
  74         c = alloc_capa(CAPA_SITE_CLIENT);
  75         if (IS_ERR(c)) {
  76                 CDEBUG(D_INFO, "alloc capa failed!\n");
  77                 return PTR_ERR(c);
  78         } else {
  79                 c->c_capa = *capa;
  80                 *oc = c;
  81                 return 0;
  82         }
  83 }
  84
  85 static inline int mdc_queue_wait(struct ptlrpc_request *req)
  86 {
  87         struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
  88         int rc;
  89
  90         /* mdc_enter_request() ensures that this client has no more
  91          * than cl_max_rpcs_in_flight RPCs simultaneously inf light
  92          * against an MDT. */
  93         rc = mdc_enter_request(cli);
  94         if (rc != 0)
  95                 return rc;
  96
  97         rc = ptlrpc_queue_wait(req);
  98         mdc_exit_request(cli);
  99
 100         return rc;
 101 }
 102
 103 /* Helper that implements most of mdc_getstatus and signal_completed_replay. */
 104 /* XXX this should become mdc_get_info("key"), sending MDS_GET_INFO RPC */
 105 static int send_getstatus(struct obd_import *imp, struct lu_fid *rootfid,
 106                           struct obd_capa **pc, int level, int msg_flags)
 107 {
 108         struct ptlrpc_request *req;
 109         struct mdt_body       *body;
 110         int                 rc;
 111
 112         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_GETSTATUS,
 113                                         LUSTRE_MDS_VERSION, MDS_GETSTATUS);
 114         if (req == NULL)
 115                 return -ENOMEM;
 116
 117         mdc_pack_body(req, NULL, NULL, 0, 0, -1, 0);
 118         lustre_msg_add_flags(req->rq_reqmsg, msg_flags);
 119         req->rq_send_state = level;
 120
 121         ptlrpc_request_set_replen(req);
 122
 123         rc = ptlrpc_queue_wait(req);
 124         if (rc)
 125                 GOTO(out, rc);
 126
 127         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 128         if (body == NULL)
 129                 GOTO(out, rc = -EPROTO);
 130
 131         if (body->valid & OBD_MD_FLMDSCAPA) {
 132                 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA1, pc);
 133                 if (rc)
 134                         GOTO(out, rc);
 135         }
 136
 137         *rootfid = body->fid1;
 138         CDEBUG(D_NET,
 139                "root fid="DFID", last_committed=%llu\n",
 140                PFID(rootfid),
 141                lustre_msg_get_last_committed(req->rq_repmsg));
 142 out:
 143         ptlrpc_req_finished(req);
 144         return rc;
 145 }
 146
 147 /* This should be mdc_get_info("rootfid") */
 148 int mdc_getstatus(struct obd_export *exp, struct lu_fid *rootfid,
 149                   struct obd_capa **pc)
 150 {
 151         return send_getstatus(class_exp2cliimp(exp), rootfid, pc,
 152                               LUSTRE_IMP_FULL, 0);
 153 }
 154
 155 /*
 156  * This function now is known to always saying that it will receive 4 buffers
 157  * from server. Even for cases when acl_size and md_size is zero, RPC header
 158  * will contain 4 fields and RPC itself will contain zero size fields. This is
 159  * because mdt_getattr*() _always_ returns 4 fields, but if acl is not needed
 160  * and thus zero, it shrinks it, making zero size. The same story about
 161  * md_size. And this is course of problem when client waits for smaller number
 162  * of fields. This issue will be fixed later when client gets aware of RPC
 163  * layouts.  --umka
 164  */
 165 static int mdc_getattr_common(struct obd_export *exp,
 166                               struct ptlrpc_request *req)
 167 {
 168         struct req_capsule *pill = &req->rq_pill;
 169         struct mdt_body    *body;
 170         void           *eadata;
 171         int              rc;
 172
 173         /* Request message already built. */
 174         rc = ptlrpc_queue_wait(req);
 175         if (rc != 0)
 176                 return rc;
 177
 178         /* sanity check for the reply */
 179         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 180         if (body == NULL)
 181                 return -EPROTO;
 182
 183         CDEBUG(D_NET, "mode: %o\n", body->mode);
 184
 185         if (body->eadatasize != 0) {
 186                 mdc_update_max_ea_from_body(exp, body);
 187
 188                 eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
 189                                                       body->eadatasize);
 190                 if (eadata == NULL)
 191                         return -EPROTO;
 192         }
 193
 194         if (body->valid & OBD_MD_FLRMTPERM) {
 195                 struct mdt_remote_perm *perm;
 196
 197                 LASSERT(client_is_remote(exp));
 198                 perm = req_capsule_server_swab_get(pill, &RMF_ACL,
 199                                                 lustre_swab_mdt_remote_perm);
 200                 if (perm == NULL)
 201                         return -EPROTO;
 202         }
 203
 204         if (body->valid & OBD_MD_FLMDSCAPA) {
 205                 struct lustre_capa *capa;
 206                 capa = req_capsule_server_get(pill, &RMF_CAPA1);
 207                 if (capa == NULL)
 208                         return -EPROTO;
 209         }
 210
 211         return 0;
 212 }
 213
 214 int mdc_getattr(struct obd_export *exp, struct md_op_data *op_data,
 215                 struct ptlrpc_request **request)
 216 {
 217         struct ptlrpc_request *req;
 218         int                 rc;
 219
 220         /* Single MDS without an LMV case */
 221         if (op_data->op_flags & MF_GET_MDT_IDX) {
 222                 op_data->op_mds = 0;
 223                 return 0;
 224         }
 225         *request = NULL;
 226         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
 227         if (req == NULL)
 228                 return -ENOMEM;
 229
 230         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 231
 232         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
 233         if (rc) {
 234                 ptlrpc_request_free(req);
 235                 return rc;
 236         }
 237
 238         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
 239                       op_data->op_valid, op_data->op_mode, -1, 0);
 240
 241         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 242                              op_data->op_mode);
 243         if (op_data->op_valid & OBD_MD_FLRMTPERM) {
 244                 LASSERT(client_is_remote(exp));
 245                 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
 246                                      sizeof(struct mdt_remote_perm));
 247         }
 248         ptlrpc_request_set_replen(req);
 249
 250         rc = mdc_getattr_common(exp, req);
 251         if (rc)
 252                 ptlrpc_req_finished(req);
 253         else
 254                 *request = req;
 255         return rc;
 256 }
 257
 258 int mdc_getattr_name(struct obd_export *exp, struct md_op_data *op_data,
 259                      struct ptlrpc_request **request)
 260 {
 261         struct ptlrpc_request *req;
 262         int                 rc;
 263
 264         *request = NULL;
 265         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 266                                    &RQF_MDS_GETATTR_NAME);
 267         if (req == NULL)
 268                 return -ENOMEM;
 269
 270         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 271         req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 272                              op_data->op_namelen + 1);
 273
 274         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR_NAME);
 275         if (rc) {
 276                 ptlrpc_request_free(req);
 277                 return rc;
 278         }
 279
 280         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
 281                       op_data->op_valid, op_data->op_mode,
 282                       op_data->op_suppgids[0], 0);
 283
 284         if (op_data->op_name) {
 285                 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 286                 LASSERT(strnlen(op_data->op_name, op_data->op_namelen) ==
 287                                 op_data->op_namelen);
 288                 memcpy(name, op_data->op_name, op_data->op_namelen);
 289         }
 290
 291         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 292                              op_data->op_mode);
 293         ptlrpc_request_set_replen(req);
 294
 295         rc = mdc_getattr_common(exp, req);
 296         if (rc)
 297                 ptlrpc_req_finished(req);
 298         else
 299                 *request = req;
 300         return rc;
 301 }
 302
 303 static int mdc_is_subdir(struct obd_export *exp,
 304                          const struct lu_fid *pfid,
 305                          const struct lu_fid *cfid,
 306                          struct ptlrpc_request **request)
 307 {
 308         struct ptlrpc_request  *req;
 309         int                  rc;
 310
 311         *request = NULL;
 312         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
 313                                         &RQF_MDS_IS_SUBDIR, LUSTRE_MDS_VERSION,
 314                                         MDS_IS_SUBDIR);
 315         if (req == NULL)
 316                 return -ENOMEM;
 317
 318         mdc_is_subdir_pack(req, pfid, cfid, 0);
 319         ptlrpc_request_set_replen(req);
 320
 321         rc = ptlrpc_queue_wait(req);
 322         if (rc && rc != -EREMOTE)
 323                 ptlrpc_req_finished(req);
 324         else
 325                 *request = req;
 326         return rc;
 327 }
 328
 329 static int mdc_xattr_common(struct obd_export *exp,const struct req_format *fmt,
 330                             const struct lu_fid *fid,
 331                             struct obd_capa *oc, int opcode, obd_valid valid,
 332                             const char *xattr_name, const char *input,
 333                             int input_size, int output_size, int flags,
 334                             __u32 suppgid, struct ptlrpc_request **request)
 335 {
 336         struct ptlrpc_request *req;
 337         int   xattr_namelen = 0;
 338         char *tmp;
 339         int   rc;
 340
 341         *request = NULL;
 342         req = ptlrpc_request_alloc(class_exp2cliimp(exp), fmt);
 343         if (req == NULL)
 344                 return -ENOMEM;
 345
 346         mdc_set_capa_size(req, &RMF_CAPA1, oc);
 347         if (xattr_name) {
 348                 xattr_namelen = strlen(xattr_name) + 1;
 349                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 350                                      xattr_namelen);
 351         }
 352         if (input_size) {
 353                 LASSERT(input);
 354                 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
 355                                      input_size);
 356         }
 357
 358         /* Flush local XATTR locks to get rid of a possible cancel RPC */
 359         if (opcode == MDS_REINT && fid_is_sane(fid) &&
 360             exp->exp_connect_data.ocd_ibits_known & MDS_INODELOCK_XATTR) {
 361                 LIST_HEAD(cancels);
 362                 int count;
 363
 364                 /* Without that packing would fail */
 365                 if (input_size == 0)
 366                         req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
 367                                              RCL_CLIENT, 0);
 368
 369                 count = mdc_resource_get_unused(exp, fid,
 370                                                 &cancels, LCK_EX,
 371                                                 MDS_INODELOCK_XATTR);
 372
 373                 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
 374                 if (rc) {
 375                         ptlrpc_request_free(req);
 376                         return rc;
 377                 }
 378         } else {
 379                 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode);
 380                 if (rc) {
 381                         ptlrpc_request_free(req);
 382                         return rc;
 383                 }
 384         }
 385
 386         if (opcode == MDS_REINT) {
 387                 struct mdt_rec_setxattr *rec;
 388
 389                 CLASSERT(sizeof(struct mdt_rec_setxattr) ==
 390                          sizeof(struct mdt_rec_reint));
 391                 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
 392                 rec->sx_opcode = REINT_SETXATTR;
 393                 rec->sx_fsuid  = from_kuid(&init_user_ns, current_fsuid());
 394                 rec->sx_fsgid  = from_kgid(&init_user_ns, current_fsgid());
 395                 rec->sx_cap    = cfs_curproc_cap_pack();
 396                 rec->sx_suppgid1 = suppgid;
 397                 rec->sx_suppgid2 = -1;
 398                 rec->sx_fid    = *fid;
 399                 rec->sx_valid  = valid | OBD_MD_FLCTIME;
 400                 rec->sx_time   = get_seconds();
 401                 rec->sx_size   = output_size;
 402                 rec->sx_flags  = flags;
 403
 404                 mdc_pack_capa(req, &RMF_CAPA1, oc);
 405         } else {
 406                 mdc_pack_body(req, fid, oc, valid, output_size, suppgid, flags);
 407         }
 408
 409         if (xattr_name) {
 410                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 411                 memcpy(tmp, xattr_name, xattr_namelen);
 412         }
 413         if (input_size) {
 414                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
 415                 memcpy(tmp, input, input_size);
 416         }
 417
 418         if (req_capsule_has_field(&req->rq_pill, &RMF_EADATA, RCL_SERVER))
 419                 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
 420                                      RCL_SERVER, output_size);
 421         ptlrpc_request_set_replen(req);
 422
 423         /* make rpc */
 424         if (opcode == MDS_REINT)
 425                 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 426
 427         rc = ptlrpc_queue_wait(req);
 428
 429         if (opcode == MDS_REINT)
 430                 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 431
 432         if (rc)
 433                 ptlrpc_req_finished(req);
 434         else
 435                 *request = req;
 436         return rc;
 437 }
 438
 439 int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
 440                  struct obd_capa *oc, obd_valid valid, const char *xattr_name,
 441                  const char *input, int input_size, int output_size,
 442                  int flags, __u32 suppgid, struct ptlrpc_request **request)
 443 {
 444         return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
 445                                 fid, oc, MDS_REINT, valid, xattr_name,
 446                                 input, input_size, output_size, flags,
 447                                 suppgid, request);
 448 }
 449
 450 int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
 451                  struct obd_capa *oc, obd_valid valid, const char *xattr_name,
 452                  const char *input, int input_size, int output_size,
 453                  int flags, struct ptlrpc_request **request)
 454 {
 455         return mdc_xattr_common(exp, &RQF_MDS_GETXATTR,
 456                                 fid, oc, MDS_GETXATTR, valid, xattr_name,
 457                                 input, input_size, output_size, flags,
 458                                 -1, request);
 459 }
 460
 461 #ifdef CONFIG_FS_POSIX_ACL
 462 static int mdc_unpack_acl(struct ptlrpc_request *req, struct lustre_md *md)
 463 {
 464         struct req_capsule     *pill = &req->rq_pill;
 465         struct mdt_body *body = md->body;
 466         struct posix_acl       *acl;
 467         void               *buf;
 468         int                  rc;
 469
 470         if (!body->aclsize)
 471                 return 0;
 472
 473         buf = req_capsule_server_sized_get(pill, &RMF_ACL, body->aclsize);
 474
 475         if (!buf)
 476                 return -EPROTO;
 477
 478         acl = posix_acl_from_xattr(&init_user_ns, buf, body->aclsize);
 479         if (IS_ERR(acl)) {
 480                 rc = PTR_ERR(acl);
 481                 CERROR("convert xattr to acl: %d\n", rc);
 482                 return rc;
 483         }
 484
 485         rc = posix_acl_valid(acl);
 486         if (rc) {
 487                 CERROR("validate acl: %d\n", rc);
 488                 posix_acl_release(acl);
 489                 return rc;
 490         }
 491
 492         md->posix_acl = acl;
 493         return 0;
 494 }
 495 #else
 496 #define mdc_unpack_acl(req, md) 0
 497 #endif
 498
 499 int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
 500                       struct obd_export *dt_exp, struct obd_export *md_exp,
 501                       struct lustre_md *md)
 502 {
 503         struct req_capsule *pill = &req->rq_pill;
 504         int rc;
 505
 506         LASSERT(md);
 507         memset(md, 0, sizeof(*md));
 508
 509         md->body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 510         LASSERT(md->body != NULL);
 511
 512         if (md->body->valid & OBD_MD_FLEASIZE) {
 513                 int lmmsize;
 514                 struct lov_mds_md *lmm;
 515
 516                 if (!S_ISREG(md->body->mode)) {
 517                         CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, should be a "
 518                                "regular file, but is not\n");
 519                         GOTO(out, rc = -EPROTO);
 520                 }
 521
 522                 if (md->body->eadatasize == 0) {
 523                         CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, "
 524                                "but eadatasize 0\n");
 525                         GOTO(out, rc = -EPROTO);
 526                 }
 527                 lmmsize = md->body->eadatasize;
 528                 lmm = req_capsule_server_sized_get(pill, &RMF_MDT_MD, lmmsize);
 529                 if (!lmm)
 530                         GOTO(out, rc = -EPROTO);
 531
 532                 rc = obd_unpackmd(dt_exp, &md->lsm, lmm, lmmsize);
 533                 if (rc < 0)
 534                         GOTO(out, rc);
 535
 536                 if (rc < sizeof(*md->lsm)) {
 537                         CDEBUG(D_INFO, "lsm size too small: "
 538                                "rc < sizeof (*md->lsm) (%d < %d)\n",
 539                                rc, (int)sizeof(*md->lsm));
 540                         GOTO(out, rc = -EPROTO);
 541                 }
 542
 543         } else if (md->body->valid & OBD_MD_FLDIREA) {
 544                 int lmvsize;
 545                 struct lov_mds_md *lmv;
 546
 547                 if(!S_ISDIR(md->body->mode)) {
 548                         CDEBUG(D_INFO, "OBD_MD_FLDIREA set, should be a "
 549                                "directory, but is not\n");
 550                         GOTO(out, rc = -EPROTO);
 551                 }
 552
 553                 if (md->body->eadatasize == 0) {
 554                         CDEBUG(D_INFO, "OBD_MD_FLDIREA is set, "
 555                                "but eadatasize 0\n");
 556                         return -EPROTO;
 557                 }
 558                 if (md->body->valid & OBD_MD_MEA) {
 559                         lmvsize = md->body->eadatasize;
 560                         lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
 561                                                            lmvsize);
 562                         if (!lmv)
 563                                 GOTO(out, rc = -EPROTO);
 564
 565                         rc = obd_unpackmd(md_exp, (void *)&md->mea, lmv,
 566                                           lmvsize);
 567                         if (rc < 0)
 568                                 GOTO(out, rc);
 569
 570                         if (rc < sizeof(*md->mea)) {
 571                                 CDEBUG(D_INFO, "size too small:  "
 572                                        "rc < sizeof(*md->mea) (%d < %d)\n",
 573                                         rc, (int)sizeof(*md->mea));
 574                                 GOTO(out, rc = -EPROTO);
 575                         }
 576                 }
 577         }
 578         rc = 0;
 579
 580         if (md->body->valid & OBD_MD_FLRMTPERM) {
 581                 /* remote permission */
 582                 LASSERT(client_is_remote(exp));
 583                 md->remote_perm = req_capsule_server_swab_get(pill, &RMF_ACL,
 584                                                 lustre_swab_mdt_remote_perm);
 585                 if (!md->remote_perm)
 586                         GOTO(out, rc = -EPROTO);
 587         }
 588         else if (md->body->valid & OBD_MD_FLACL) {
 589                 /* for ACL, it's possible that FLACL is set but aclsize is zero.
 590                  * only when aclsize != 0 there's an actual segment for ACL
 591                  * in reply buffer.
 592                  */
 593                 if (md->body->aclsize) {
 594                         rc = mdc_unpack_acl(req, md);
 595                         if (rc)
 596                                 GOTO(out, rc);
 597 #ifdef CONFIG_FS_POSIX_ACL
 598                 } else {
 599                         md->posix_acl = NULL;
 600 #endif
 601                 }
 602         }
 603         if (md->body->valid & OBD_MD_FLMDSCAPA) {
 604                 struct obd_capa *oc = NULL;
 605
 606                 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA1, &oc);
 607                 if (rc)
 608                         GOTO(out, rc);
 609                 md->mds_capa = oc;
 610         }
 611
 612         if (md->body->valid & OBD_MD_FLOSSCAPA) {
 613                 struct obd_capa *oc = NULL;
 614
 615                 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA2, &oc);
 616                 if (rc)
 617                         GOTO(out, rc);
 618                 md->oss_capa = oc;
 619         }
 620
 621 out:
 622         if (rc) {
 623                 if (md->oss_capa) {
 624                         capa_put(md->oss_capa);
 625                         md->oss_capa = NULL;
 626                 }
 627                 if (md->mds_capa) {
 628                         capa_put(md->mds_capa);
 629                         md->mds_capa = NULL;
 630                 }
 631 #ifdef CONFIG_FS_POSIX_ACL
 632                 posix_acl_release(md->posix_acl);
 633 #endif
 634                 if (md->lsm)
 635                         obd_free_memmd(dt_exp, &md->lsm);
 636         }
 637         return rc;
 638 }
 639
 640 int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
 641 {
 642         return 0;
 643 }
 644
 645 /**
 646  * Handles both OPEN and SETATTR RPCs for OPEN-CLOSE and SETATTR-DONE_WRITING
 647  * RPC chains.
 648  */
 649 void mdc_replay_open(struct ptlrpc_request *req)
 650 {
 651         struct md_open_data *mod = req->rq_cb_data;
 652         struct ptlrpc_request *close_req;
 653         struct obd_client_handle *och;
 654         struct lustre_handle old;
 655         struct mdt_body *body;
 656
 657         if (mod == NULL) {
 658                 DEBUG_REQ(D_ERROR, req,
 659                           "Can't properly replay without open data.");
 660                 return;
 661         }
 662
 663         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 664         LASSERT(body != NULL);
 665
 666         och = mod->mod_och;
 667         if (och != NULL) {
 668                 struct lustre_handle *file_fh;
 669
 670                 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
 671
 672                 file_fh = &och->och_fh;
 673                 CDEBUG(D_HA, "updating handle from %#llx to %#llx\n",
 674                        file_fh->cookie, body->handle.cookie);
 675                 old = *file_fh;
 676                 *file_fh = body->handle;
 677         }
 678         close_req = mod->mod_close_req;
 679         if (close_req != NULL) {
 680                 __u32 opc = lustre_msg_get_opc(close_req->rq_reqmsg);
 681                 struct mdt_ioepoch *epoch;
 682
 683                 LASSERT(opc == MDS_CLOSE || opc == MDS_DONE_WRITING);
 684                 epoch = req_capsule_client_get(&close_req->rq_pill,
 685                                                &RMF_MDT_EPOCH);
 686                 LASSERT(epoch);
 687
 688                 if (och != NULL)
 689                         LASSERT(!memcmp(&old, &epoch->handle, sizeof(old)));
 690                 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
 691                 epoch->handle = body->handle;
 692         }
 693 }
 694
 695 void mdc_commit_open(struct ptlrpc_request *req)
 696 {
 697         struct md_open_data *mod = req->rq_cb_data;
 698         if (mod == NULL)
 699                 return;
 700
 701         /**
 702          * No need to touch md_open_data::mod_och, it holds a reference on
 703          * \var mod and will zero references to each other, \var mod will be
 704          * freed after that when md_open_data::mod_och will put the reference.
 705          */
 706
 707         /**
 708          * Do not let open request to disappear as it still may be needed
 709          * for close rpc to happen (it may happen on evict only, otherwise
 710          * ptlrpc_request::rq_replay does not let mdc_commit_open() to be
 711          * called), just mark this rpc as committed to distinguish these 2
 712          * cases, see mdc_close() for details. The open request reference will
 713          * be put along with freeing \var mod.
 714          */
 715         ptlrpc_request_addref(req);
 716         spin_lock(&req->rq_lock);
 717         req->rq_committed = 1;
 718         spin_unlock(&req->rq_lock);
 719         req->rq_cb_data = NULL;
 720         obd_mod_put(mod);
 721 }
 722
 723 int mdc_set_open_replay_data(struct obd_export *exp,
 724                              struct obd_client_handle *och,
 725                              struct lookup_intent *it)
 726 {
 727         struct md_open_data   *mod;
 728         struct mdt_rec_create *rec;
 729         struct mdt_body       *body;
 730         struct ptlrpc_request *open_req = it->d.lustre.it_data;
 731         struct obd_import     *imp = open_req->rq_import;
 732
 733         if (!open_req->rq_replay)
 734                 return 0;
 735
 736         rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
 737         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
 738         LASSERT(rec != NULL);
 739         /* Incoming message in my byte order (it's been swabbed). */
 740         /* Outgoing messages always in my byte order. */
 741         LASSERT(body != NULL);
 742
 743         /* Only if the import is replayable, we set replay_open data */
 744         if (och && imp->imp_replayable) {
 745                 mod = obd_mod_alloc();
 746                 if (mod == NULL) {
 747                         DEBUG_REQ(D_ERROR, open_req,
 748                                   "Can't allocate md_open_data");
 749                         return 0;
 750                 }
 751
 752                 /**
 753                  * Take a reference on \var mod, to be freed on mdc_close().
 754                  * It protects \var mod from being freed on eviction (commit
 755                  * callback is called despite rq_replay flag).
 756                  * Another reference for \var och.
 757                  */
 758                 obd_mod_get(mod);
 759                 obd_mod_get(mod);
 760
 761                 spin_lock(&open_req->rq_lock);
 762                 och->och_mod = mod;
 763                 mod->mod_och = och;
 764                 mod->mod_is_create = it_disposition(it, DISP_OPEN_CREATE) ||
 765                                      it_disposition(it, DISP_OPEN_STRIPE);
 766                 mod->mod_open_req = open_req;
 767                 open_req->rq_cb_data = mod;
 768                 open_req->rq_commit_cb = mdc_commit_open;
 769                 spin_unlock(&open_req->rq_lock);
 770         }
 771
 772         rec->cr_fid2 = body->fid1;
 773         rec->cr_ioepoch = body->ioepoch;
 774         rec->cr_old_handle.cookie = body->handle.cookie;
 775         open_req->rq_replay_cb = mdc_replay_open;
 776         if (!fid_is_sane(&body->fid1)) {
 777                 DEBUG_REQ(D_ERROR, open_req, "Saving replay request with "
 778                           "insane fid");
 779                 LBUG();
 780         }
 781
 782         DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
 783         return 0;
 784 }
 785
 786 static void mdc_free_open(struct md_open_data *mod)
 787 {
 788         int committed = 0;
 789
 790         if (mod->mod_is_create == 0 &&
 791             imp_connect_disp_stripe(mod->mod_open_req->rq_import))
 792                 committed = 1;
 793
 794         LASSERT(mod->mod_open_req->rq_replay == 0);
 795
 796         DEBUG_REQ(D_RPCTRACE, mod->mod_open_req, "free open request\n");
 797
 798         ptlrpc_request_committed(mod->mod_open_req, committed);
 799         if (mod->mod_close_req)
 800                 ptlrpc_request_committed(mod->mod_close_req, committed);
 801 }
 802
 803 int mdc_clear_open_replay_data(struct obd_export *exp,
 804                                struct obd_client_handle *och)
 805 {
 806         struct md_open_data *mod = och->och_mod;
 807
 808         /**
 809          * It is possible to not have \var mod in a case of eviction between
 810          * lookup and ll_file_open().
 811          **/
 812         if (mod == NULL)
 813                 return 0;
 814
 815         LASSERT(mod != LP_POISON);
 816         LASSERT(mod->mod_open_req != NULL);
 817         mdc_free_open(mod);
 818
 819         mod->mod_och = NULL;
 820         och->och_mod = NULL;
 821         obd_mod_put(mod);
 822
 823         return 0;
 824 }
 825
 826 /* Prepares the request for the replay by the given reply */
 827 static void mdc_close_handle_reply(struct ptlrpc_request *req,
 828                                    struct md_op_data *op_data, int rc) {
 829         struct mdt_body  *repbody;
 830         struct mdt_ioepoch *epoch;
 831
 832         if (req && rc == -EAGAIN) {
 833                 repbody = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 834                 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
 835
 836                 epoch->flags |= MF_SOM_AU;
 837                 if (repbody->valid & OBD_MD_FLGETATTRLOCK)
 838                         op_data->op_flags |= MF_GETATTR_LOCK;
 839         }
 840 }
 841
 842 int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
 843               struct md_open_data *mod, struct ptlrpc_request **request)
 844 {
 845         struct obd_device     *obd = class_exp2obd(exp);
 846         struct ptlrpc_request *req;
 847         struct req_format     *req_fmt;
 848         int                    rc;
 849         int                    saved_rc = 0;
 850
 851
 852         req_fmt = &RQF_MDS_CLOSE;
 853         if (op_data->op_bias & MDS_HSM_RELEASE) {
 854                 req_fmt = &RQF_MDS_RELEASE_CLOSE;
 855
 856                 /* allocate a FID for volatile file */
 857                 rc = mdc_fid_alloc(exp, &op_data->op_fid2, op_data);
 858                 if (rc < 0) {
 859                         CERROR("%s: "DFID" failed to allocate FID: %d\n",
 860                                obd->obd_name, PFID(&op_data->op_fid1), rc);
 861                         /* save the errcode and proceed to close */
 862                         saved_rc = rc;
 863                 }
 864         }
 865
 866         *request = NULL;
 867         req = ptlrpc_request_alloc(class_exp2cliimp(exp), req_fmt);
 868         if (req == NULL)
 869                 return -ENOMEM;
 870
 871         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 872
 873         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
 874         if (rc) {
 875                 ptlrpc_request_free(req);
 876                 return rc;
 877         }
 878
 879         /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
 880          * portal whose threads are not taking any DLM locks and are therefore
 881          * always progressing */
 882         req->rq_request_portal = MDS_READPAGE_PORTAL;
 883         ptlrpc_at_set_req_timeout(req);
 884
 885         /* Ensure that this close's handle is fixed up during replay. */
 886         if (likely(mod != NULL)) {
 887                 LASSERTF(mod->mod_open_req != NULL &&
 888                          mod->mod_open_req->rq_type != LI_POISON,
 889                          "POISONED open %p!\n", mod->mod_open_req);
 890
 891                 mod->mod_close_req = req;
 892
 893                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
 894                 /* We no longer want to preserve this open for replay even
 895                  * though the open was committed. b=3632, b=3633 */
 896                 spin_lock(&mod->mod_open_req->rq_lock);
 897                 mod->mod_open_req->rq_replay = 0;
 898                 spin_unlock(&mod->mod_open_req->rq_lock);
 899         } else {
 900                  CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
 901         }
 902
 903         mdc_close_pack(req, op_data);
 904
 905         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 906                              obd->u.cli.cl_default_mds_easize);
 907         req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
 908                              obd->u.cli.cl_default_mds_cookiesize);
 909
 910         ptlrpc_request_set_replen(req);
 911
 912         mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 913         rc = ptlrpc_queue_wait(req);
 914         mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 915
 916         if (req->rq_repmsg == NULL) {
 917                 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
 918                        req->rq_status);
 919                 if (rc == 0)
 920                         rc = req->rq_status ?: -EIO;
 921         } else if (rc == 0 || rc == -EAGAIN) {
 922                 struct mdt_body *body;
 923
 924                 rc = lustre_msg_get_status(req->rq_repmsg);
 925                 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
 926                         DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
 927                                   "= %d", rc);
 928                         if (rc > 0)
 929                                 rc = -rc;
 930                 }
 931                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 932                 if (body == NULL)
 933                         rc = -EPROTO;
 934         } else if (rc == -ESTALE) {
 935                 /**
 936                  * it can be allowed error after 3633 if open was committed and
 937                  * server failed before close was sent. Let's check if mod
 938                  * exists and return no error in that case
 939                  */
 940                 if (mod) {
 941                         DEBUG_REQ(D_HA, req, "Reset ESTALE = %d", rc);
 942                         LASSERT(mod->mod_open_req != NULL);
 943                         if (mod->mod_open_req->rq_committed)
 944                                 rc = 0;
 945                 }
 946         }
 947
 948         if (mod) {
 949                 if (rc != 0)
 950                         mod->mod_close_req = NULL;
 951                 /* Since now, mod is accessed through open_req only,
 952                  * thus close req does not keep a reference on mod anymore. */
 953                 obd_mod_put(mod);
 954         }
 955         *request = req;
 956         mdc_close_handle_reply(req, op_data, rc);
 957         return rc < 0 ? rc : saved_rc;
 958 }
 959
 960 int mdc_done_writing(struct obd_export *exp, struct md_op_data *op_data,
 961                      struct md_open_data *mod)
 962 {
 963         struct obd_device     *obd = class_exp2obd(exp);
 964         struct ptlrpc_request *req;
 965         int                 rc;
 966
 967         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 968                                    &RQF_MDS_DONE_WRITING);
 969         if (req == NULL)
 970                 return -ENOMEM;
 971
 972         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 973         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_DONE_WRITING);
 974         if (rc) {
 975                 ptlrpc_request_free(req);
 976                 return rc;
 977         }
 978
 979         if (mod != NULL) {
 980                 LASSERTF(mod->mod_open_req != NULL &&
 981                          mod->mod_open_req->rq_type != LI_POISON,
 982                          "POISONED setattr %p!\n", mod->mod_open_req);
 983
 984                 mod->mod_close_req = req;
 985                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched setattr");
 986                 /* We no longer want to preserve this setattr for replay even
 987                  * though the open was committed. b=3632, b=3633 */
 988                 spin_lock(&mod->mod_open_req->rq_lock);
 989                 mod->mod_open_req->rq_replay = 0;
 990                 spin_unlock(&mod->mod_open_req->rq_lock);
 991         }
 992
 993         mdc_close_pack(req, op_data);
 994         ptlrpc_request_set_replen(req);
 995
 996         mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 997         rc = ptlrpc_queue_wait(req);
 998         mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 999
1000         if (rc == -ESTALE) {
1001                 /**
1002                  * it can be allowed error after 3633 if open or setattr were
1003                  * committed and server failed before close was sent.
1004                  * Let's check if mod exists and return no error in that case
1005                  */
1006                 if (mod) {
1007                         LASSERT(mod->mod_open_req != NULL);
1008                         if (mod->mod_open_req->rq_committed)
1009                                 rc = 0;
1010                 }
1011         }
1012
1013         if (mod) {
1014                 if (rc != 0)
1015                         mod->mod_close_req = NULL;
1016                 LASSERT(mod->mod_open_req != NULL);
1017                 mdc_free_open(mod);
1018
1019                 /* Since now, mod is accessed through setattr req only,
1020                  * thus DW req does not keep a reference on mod anymore. */
1021                 obd_mod_put(mod);
1022         }
1023
1024         mdc_close_handle_reply(req, op_data, rc);
1025         ptlrpc_req_finished(req);
1026         return rc;
1027 }
1028
1029
1030 int mdc_readpage(struct obd_export *exp, struct md_op_data *op_data,
1031                  struct page **pages, struct ptlrpc_request **request)
1032 {
1033         struct ptlrpc_request   *req;
1034         struct ptlrpc_bulk_desc *desc;
1035         int                   i;
1036         wait_queue_head_t             waitq;
1037         int                   resends = 0;
1038         struct l_wait_info       lwi;
1039         int                   rc;
1040
1041         *request = NULL;
1042         init_waitqueue_head(&waitq);
1043
1044 restart_bulk:
1045         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
1046         if (req == NULL)
1047                 return -ENOMEM;
1048
1049         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1050
1051         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
1052         if (rc) {
1053                 ptlrpc_request_free(req);
1054                 return rc;
1055         }
1056
1057         req->rq_request_portal = MDS_READPAGE_PORTAL;
1058         ptlrpc_at_set_req_timeout(req);
1059
1060         desc = ptlrpc_prep_bulk_imp(req, op_data->op_npages, 1, BULK_PUT_SINK,
1061                                     MDS_BULK_PORTAL);
1062         if (desc == NULL) {
1063                 ptlrpc_request_free(req);
1064                 return -ENOMEM;
1065         }
1066
1067         /* NB req now owns desc and will free it when it gets freed */
1068         for (i = 0; i < op_data->op_npages; i++)
1069                 ptlrpc_prep_bulk_page_pin(desc, pages[i], 0, PAGE_CACHE_SIZE);
1070
1071         mdc_readdir_pack(req, op_data->op_offset,
1072                          PAGE_CACHE_SIZE * op_data->op_npages,
1073                          &op_data->op_fid1, op_data->op_capa1);
1074
1075         ptlrpc_request_set_replen(req);
1076         rc = ptlrpc_queue_wait(req);
1077         if (rc) {
1078                 ptlrpc_req_finished(req);
1079                 if (rc != -ETIMEDOUT)
1080                         return rc;
1081
1082                 resends++;
1083                 if (!client_should_resend(resends, &exp->exp_obd->u.cli)) {
1084                         CERROR("too many resend retries, returning error\n");
1085                         return -EIO;
1086                 }
1087                 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(resends), NULL, NULL, NULL);
1088                 l_wait_event(waitq, 0, &lwi);
1089
1090                 goto restart_bulk;
1091         }
1092
1093         rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
1094                                           req->rq_bulk->bd_nob_transferred);
1095         if (rc < 0) {
1096                 ptlrpc_req_finished(req);
1097                 return rc;
1098         }
1099
1100         if (req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK) {
1101                 CERROR("Unexpected # bytes transferred: %d (%ld expected)\n",
1102                         req->rq_bulk->bd_nob_transferred,
1103                         PAGE_CACHE_SIZE * op_data->op_npages);
1104                 ptlrpc_req_finished(req);
1105                 return -EPROTO;
1106         }
1107
1108         *request = req;
1109         return 0;
1110 }
1111
1112 static int mdc_statfs(const struct lu_env *env,
1113                       struct obd_export *exp, struct obd_statfs *osfs,
1114                       __u64 max_age, __u32 flags)
1115 {
1116         struct obd_device     *obd = class_exp2obd(exp);
1117         struct ptlrpc_request *req;
1118         struct obd_statfs     *msfs;
1119         struct obd_import     *imp = NULL;
1120         int                 rc;
1121
1122         /*
1123          * Since the request might also come from lprocfs, so we need
1124          * sync this with client_disconnect_export Bug15684
1125          */
1126         down_read(&obd->u.cli.cl_sem);
1127         if (obd->u.cli.cl_import)
1128                 imp = class_import_get(obd->u.cli.cl_import);
1129         up_read(&obd->u.cli.cl_sem);
1130         if (!imp)
1131                 return -ENODEV;
1132
1133         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_STATFS,
1134                                         LUSTRE_MDS_VERSION, MDS_STATFS);
1135         if (req == NULL)
1136                 GOTO(output, rc = -ENOMEM);
1137
1138         ptlrpc_request_set_replen(req);
1139
1140         if (flags & OBD_STATFS_NODELAY) {
1141                 /* procfs requests not want stay in wait for avoid deadlock */
1142                 req->rq_no_resend = 1;
1143                 req->rq_no_delay = 1;
1144         }
1145
1146         rc = ptlrpc_queue_wait(req);
1147         if (rc) {
1148                 /* check connection error first */
1149                 if (imp->imp_connect_error)
1150                         rc = imp->imp_connect_error;
1151                 GOTO(out, rc);
1152         }
1153
1154         msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1155         if (msfs == NULL)
1156                 GOTO(out, rc = -EPROTO);
1157
1158         *osfs = *msfs;
1159 out:
1160         ptlrpc_req_finished(req);
1161 output:
1162         class_import_put(imp);
1163         return rc;
1164 }
1165
1166 static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
1167 {
1168         __u32 keylen, vallen;
1169         void *key;
1170         int rc;
1171
1172         if (gf->gf_pathlen > PATH_MAX)
1173                 return -ENAMETOOLONG;
1174         if (gf->gf_pathlen < 2)
1175                 return -EOVERFLOW;
1176
1177         /* Key is KEY_FID2PATH + getinfo_fid2path description */
1178         keylen = cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf);
1179         OBD_ALLOC(key, keylen);
1180         if (key == NULL)
1181                 return -ENOMEM;
1182         memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
1183         memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf));
1184
1185         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
1186                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
1187
1188         if (!fid_is_sane(&gf->gf_fid))
1189                 GOTO(out, rc = -EINVAL);
1190
1191         /* Val is struct getinfo_fid2path result plus path */
1192         vallen = sizeof(*gf) + gf->gf_pathlen;
1193
1194         rc = obd_get_info(NULL, exp, keylen, key, &vallen, gf, NULL);
1195         if (rc != 0 && rc != -EREMOTE)
1196                 GOTO(out, rc);
1197
1198         if (vallen <= sizeof(*gf))
1199                 GOTO(out, rc = -EPROTO);
1200         else if (vallen > sizeof(*gf) + gf->gf_pathlen)
1201                 GOTO(out, rc = -EOVERFLOW);
1202
1203         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n%s\n",
1204                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno, gf->gf_path);
1205
1206 out:
1207         OBD_FREE(key, keylen);
1208         return rc;
1209 }
1210
1211 static int mdc_ioc_hsm_progress(struct obd_export *exp,
1212                                 struct hsm_progress_kernel *hpk)
1213 {
1214         struct obd_import               *imp = class_exp2cliimp(exp);
1215         struct hsm_progress_kernel      *req_hpk;
1216         struct ptlrpc_request           *req;
1217         int                              rc;
1218
1219         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_PROGRESS,
1220                                         LUSTRE_MDS_VERSION, MDS_HSM_PROGRESS);
1221         if (req == NULL)
1222                 GOTO(out, rc = -ENOMEM);
1223
1224         mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, 0, 0);
1225
1226         /* Copy hsm_progress struct */
1227         req_hpk = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_PROGRESS);
1228         if (req_hpk == NULL)
1229                 GOTO(out, rc = -EPROTO);
1230
1231         *req_hpk = *hpk;
1232         req_hpk->hpk_errval = lustre_errno_hton(hpk->hpk_errval);
1233
1234         ptlrpc_request_set_replen(req);
1235
1236         rc = mdc_queue_wait(req);
1237         GOTO(out, rc);
1238 out:
1239         ptlrpc_req_finished(req);
1240         return rc;
1241 }
1242
1243 static int mdc_ioc_hsm_ct_register(struct obd_import *imp, __u32 archives)
1244 {
1245         __u32                   *archive_mask;
1246         struct ptlrpc_request   *req;
1247         int                      rc;
1248
1249         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_REGISTER,
1250                                         LUSTRE_MDS_VERSION,
1251                                         MDS_HSM_CT_REGISTER);
1252         if (req == NULL)
1253                 GOTO(out, rc = -ENOMEM);
1254
1255         mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, 0, 0);
1256
1257         /* Copy hsm_progress struct */
1258         archive_mask = req_capsule_client_get(&req->rq_pill,
1259                                               &RMF_MDS_HSM_ARCHIVE);
1260         if (archive_mask == NULL)
1261                 GOTO(out, rc = -EPROTO);
1262
1263         *archive_mask = archives;
1264
1265         ptlrpc_request_set_replen(req);
1266
1267         rc = mdc_queue_wait(req);
1268         GOTO(out, rc);
1269 out:
1270         ptlrpc_req_finished(req);
1271         return rc;
1272 }
1273
1274 static int mdc_ioc_hsm_current_action(struct obd_export *exp,
1275                                       struct md_op_data *op_data)
1276 {
1277         struct hsm_current_action       *hca = op_data->op_data;
1278         struct hsm_current_action       *req_hca;
1279         struct ptlrpc_request           *req;
1280         int                              rc;
1281
1282         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1283                                    &RQF_MDS_HSM_ACTION);
1284         if (req == NULL)
1285                 return -ENOMEM;
1286
1287         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1288
1289         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_ACTION);
1290         if (rc) {
1291                 ptlrpc_request_free(req);
1292                 return rc;
1293         }
1294
1295         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
1296                       OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
1297
1298         ptlrpc_request_set_replen(req);
1299
1300         rc = mdc_queue_wait(req);
1301         if (rc)
1302                 GOTO(out, rc);
1303
1304         req_hca = req_capsule_server_get(&req->rq_pill,
1305                                          &RMF_MDS_HSM_CURRENT_ACTION);
1306         if (req_hca == NULL)
1307                 GOTO(out, rc = -EPROTO);
1308
1309         *hca = *req_hca;
1310
1311 out:
1312         ptlrpc_req_finished(req);
1313         return rc;
1314 }
1315
1316 static int mdc_ioc_hsm_ct_unregister(struct obd_import *imp)
1317 {
1318         struct ptlrpc_request   *req;
1319         int                      rc;
1320
1321         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_UNREGISTER,
1322                                         LUSTRE_MDS_VERSION,
1323                                         MDS_HSM_CT_UNREGISTER);
1324         if (req == NULL)
1325                 GOTO(out, rc = -ENOMEM);
1326
1327         mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, 0, 0);
1328
1329         ptlrpc_request_set_replen(req);
1330
1331         rc = mdc_queue_wait(req);
1332         GOTO(out, rc);
1333 out:
1334         ptlrpc_req_finished(req);
1335         return rc;
1336 }
1337
1338 static int mdc_ioc_hsm_state_get(struct obd_export *exp,
1339                                  struct md_op_data *op_data)
1340 {
1341         struct hsm_user_state   *hus = op_data->op_data;
1342         struct hsm_user_state   *req_hus;
1343         struct ptlrpc_request   *req;
1344         int                      rc;
1345
1346         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1347                                    &RQF_MDS_HSM_STATE_GET);
1348         if (req == NULL)
1349                 return -ENOMEM;
1350
1351         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1352
1353         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_GET);
1354         if (rc != 0) {
1355                 ptlrpc_request_free(req);
1356                 return rc;
1357         }
1358
1359         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
1360                       OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
1361
1362         ptlrpc_request_set_replen(req);
1363
1364         rc = mdc_queue_wait(req);
1365         if (rc)
1366                 GOTO(out, rc);
1367
1368         req_hus = req_capsule_server_get(&req->rq_pill, &RMF_HSM_USER_STATE);
1369         if (req_hus == NULL)
1370                 GOTO(out, rc = -EPROTO);
1371
1372         *hus = *req_hus;
1373
1374 out:
1375         ptlrpc_req_finished(req);
1376         return rc;
1377 }
1378
1379 static int mdc_ioc_hsm_state_set(struct obd_export *exp,
1380                                  struct md_op_data *op_data)
1381 {
1382         struct hsm_state_set    *hss = op_data->op_data;
1383         struct hsm_state_set    *req_hss;
1384         struct ptlrpc_request   *req;
1385         int                      rc;
1386
1387         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1388                                    &RQF_MDS_HSM_STATE_SET);
1389         if (req == NULL)
1390                 return -ENOMEM;
1391
1392         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1393
1394         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_SET);
1395         if (rc) {
1396                 ptlrpc_request_free(req);
1397                 return rc;
1398         }
1399
1400         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
1401                       OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
1402
1403         /* Copy states */
1404         req_hss = req_capsule_client_get(&req->rq_pill, &RMF_HSM_STATE_SET);
1405         if (req_hss == NULL)
1406                 GOTO(out, rc = -EPROTO);
1407         *req_hss = *hss;
1408
1409         ptlrpc_request_set_replen(req);
1410
1411         rc = mdc_queue_wait(req);
1412         GOTO(out, rc);
1413
1414 out:
1415         ptlrpc_req_finished(req);
1416         return rc;
1417 }
1418
1419 static int mdc_ioc_hsm_request(struct obd_export *exp,
1420                                struct hsm_user_request *hur)
1421 {
1422         struct obd_import       *imp = class_exp2cliimp(exp);
1423         struct ptlrpc_request   *req;
1424         struct hsm_request      *req_hr;
1425         struct hsm_user_item    *req_hui;
1426         char                    *req_opaque;
1427         int                      rc;
1428
1429         req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_REQUEST);
1430         if (req == NULL)
1431                 GOTO(out, rc = -ENOMEM);
1432
1433         req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM, RCL_CLIENT,
1434                              hur->hur_request.hr_itemcount
1435                              * sizeof(struct hsm_user_item));
1436         req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA, RCL_CLIENT,
1437                              hur->hur_request.hr_data_len);
1438
1439         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_REQUEST);
1440         if (rc) {
1441                 ptlrpc_request_free(req);
1442                 return rc;
1443         }
1444
1445         mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, 0, 0);
1446
1447         /* Copy hsm_request struct */
1448         req_hr = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_REQUEST);
1449         if (req_hr == NULL)
1450                 GOTO(out, rc = -EPROTO);
1451         *req_hr = hur->hur_request;
1452
1453         /* Copy hsm_user_item structs */
1454         req_hui = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM);
1455         if (req_hui == NULL)
1456                 GOTO(out, rc = -EPROTO);
1457         memcpy(req_hui, hur->hur_user_item,
1458                hur->hur_request.hr_itemcount * sizeof(struct hsm_user_item));
1459
1460         /* Copy opaque field */
1461         req_opaque = req_capsule_client_get(&req->rq_pill, &RMF_GENERIC_DATA);
1462         if (req_opaque == NULL)
1463                 GOTO(out, rc = -EPROTO);
1464         memcpy(req_opaque, hur_data(hur), hur->hur_request.hr_data_len);
1465
1466         ptlrpc_request_set_replen(req);
1467
1468         rc = mdc_queue_wait(req);
1469         GOTO(out, rc);
1470
1471 out:
1472         ptlrpc_req_finished(req);
1473         return rc;
1474 }
1475
1476 static struct kuc_hdr *changelog_kuc_hdr(char *buf, int len, int flags)
1477 {
1478         struct kuc_hdr *lh = (struct kuc_hdr *)buf;
1479
1480         LASSERT(len <= KUC_CHANGELOG_MSG_MAXSIZE);
1481
1482         lh->kuc_magic = KUC_MAGIC;
1483         lh->kuc_transport = KUC_TRANSPORT_CHANGELOG;
1484         lh->kuc_flags = flags;
1485         lh->kuc_msgtype = CL_RECORD;
1486         lh->kuc_msglen = len;
1487         return lh;
1488 }
1489
1490 #define D_CHANGELOG 0
1491
1492 struct changelog_show {
1493         __u64           cs_startrec;
1494         __u32           cs_flags;
1495         struct file     *cs_fp;
1496         char            *cs_buf;
1497         struct obd_device *cs_obd;
1498 };
1499
1500 static int changelog_kkuc_cb(const struct lu_env *env, struct llog_handle *llh,
1501                              struct llog_rec_hdr *hdr, void *data)
1502 {
1503         struct changelog_show *cs = data;
1504         struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
1505         struct kuc_hdr *lh;
1506         int len, rc;
1507
1508         if (rec->cr_hdr.lrh_type != CHANGELOG_REC) {
1509                 rc = -EINVAL;
1510                 CERROR("%s: not a changelog rec %x/%d: rc = %d\n",
1511                        cs->cs_obd->obd_name, rec->cr_hdr.lrh_type,
1512                        rec->cr.cr_type, rc);
1513                 return rc;
1514         }
1515
1516         if (rec->cr.cr_index < cs->cs_startrec) {
1517                 /* Skip entries earlier than what we are interested in */
1518                 CDEBUG(D_CHANGELOG, "rec=%llu start=%llu\n",
1519                        rec->cr.cr_index, cs->cs_startrec);
1520                 return 0;
1521         }
1522
1523         CDEBUG(D_CHANGELOG, "%llu %02d%-5s %llu 0x%x t="DFID" p="DFID
1524                 " %.*s\n", rec->cr.cr_index, rec->cr.cr_type,
1525                 changelog_type2str(rec->cr.cr_type), rec->cr.cr_time,
1526                 rec->cr.cr_flags & CLF_FLAGMASK,
1527                 PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid),
1528                 rec->cr.cr_namelen, changelog_rec_name(&rec->cr));
1529
1530         len = sizeof(*lh) + changelog_rec_size(&rec->cr) + rec->cr.cr_namelen;
1531
1532         /* Set up the message */
1533         lh = changelog_kuc_hdr(cs->cs_buf, len, cs->cs_flags);
1534         memcpy(lh + 1, &rec->cr, len - sizeof(*lh));
1535
1536         rc = libcfs_kkuc_msg_put(cs->cs_fp, lh);
1537         CDEBUG(D_CHANGELOG, "kucmsg fp %p len %d rc %d\n", cs->cs_fp, len,rc);
1538
1539         return rc;
1540 }
1541
1542 static int mdc_changelog_send_thread(void *csdata)
1543 {
1544         struct changelog_show *cs = csdata;
1545         struct llog_ctxt *ctxt = NULL;
1546         struct llog_handle *llh = NULL;
1547         struct kuc_hdr *kuch;
1548         int rc;
1549
1550         CDEBUG(D_CHANGELOG, "changelog to fp=%p start %llu\n",
1551                cs->cs_fp, cs->cs_startrec);
1552
1553         OBD_ALLOC(cs->cs_buf, KUC_CHANGELOG_MSG_MAXSIZE);
1554         if (cs->cs_buf == NULL)
1555                 GOTO(out, rc = -ENOMEM);
1556
1557         /* Set up the remote catalog handle */
1558         ctxt = llog_get_context(cs->cs_obd, LLOG_CHANGELOG_REPL_CTXT);
1559         if (ctxt == NULL)
1560                 GOTO(out, rc = -ENOENT);
1561         rc = llog_open(NULL, ctxt, &llh, NULL, CHANGELOG_CATALOG,
1562                        LLOG_OPEN_EXISTS);
1563         if (rc) {
1564                 CERROR("%s: fail to open changelog catalog: rc = %d\n",
1565                        cs->cs_obd->obd_name, rc);
1566                 GOTO(out, rc);
1567         }
1568         rc = llog_init_handle(NULL, llh, LLOG_F_IS_CAT, NULL);
1569         if (rc) {
1570                 CERROR("llog_init_handle failed %d\n", rc);
1571                 GOTO(out, rc);
1572         }
1573
1574         rc = llog_cat_process(NULL, llh, changelog_kkuc_cb, cs, 0, 0);
1575
1576         /* Send EOF no matter what our result */
1577         kuch = changelog_kuc_hdr(cs->cs_buf, sizeof(*kuch), cs->cs_flags);
1578         if (kuch) {
1579                 kuch->kuc_msgtype = CL_EOF;
1580                 libcfs_kkuc_msg_put(cs->cs_fp, kuch);
1581         }
1582
1583 out:
1584         fput(cs->cs_fp);
1585         if (llh)
1586                 llog_cat_close(NULL, llh);
1587         if (ctxt)
1588                 llog_ctxt_put(ctxt);
1589         if (cs->cs_buf)
1590                 OBD_FREE(cs->cs_buf, KUC_CHANGELOG_MSG_MAXSIZE);
1591         OBD_FREE_PTR(cs);
1592         return rc;
1593 }
1594
1595 static int mdc_ioc_changelog_send(struct obd_device *obd,
1596                                   struct ioc_changelog *icc)
1597 {
1598         struct changelog_show *cs;
1599         int rc;
1600
1601         /* Freed in mdc_changelog_send_thread */
1602         OBD_ALLOC_PTR(cs);
1603         if (!cs)
1604                 return -ENOMEM;
1605
1606         cs->cs_obd = obd;
1607         cs->cs_startrec = icc->icc_recno;
1608         /* matching fput in mdc_changelog_send_thread */
1609         cs->cs_fp = fget(icc->icc_id);
1610         cs->cs_flags = icc->icc_flags;
1611
1612         /*
1613          * New thread because we should return to user app before
1614          * writing into our pipe
1615          */
1616         rc = PTR_ERR(kthread_run(mdc_changelog_send_thread, cs,
1617                                  "mdc_clg_send_thread"));
1618         if (!IS_ERR_VALUE(rc)) {
1619                 CDEBUG(D_CHANGELOG, "start changelog thread\n");
1620                 return 0;
1621         }
1622
1623         CERROR("Failed to start changelog thread: %d\n", rc);
1624         OBD_FREE_PTR(cs);
1625         return rc;
1626 }
1627
1628 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
1629                                 struct lustre_kernelcomm *lk);
1630
1631 static int mdc_quotacheck(struct obd_device *unused, struct obd_export *exp,
1632                           struct obd_quotactl *oqctl)
1633 {
1634         struct client_obd       *cli = &exp->exp_obd->u.cli;
1635         struct ptlrpc_request   *req;
1636         struct obd_quotactl     *body;
1637         int                   rc;
1638
1639         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
1640                                         &RQF_MDS_QUOTACHECK, LUSTRE_MDS_VERSION,
1641                                         MDS_QUOTACHECK);
1642         if (req == NULL)
1643                 return -ENOMEM;
1644
1645         body = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1646         *body = *oqctl;
1647
1648         ptlrpc_request_set_replen(req);
1649
1650         /* the next poll will find -ENODATA, that means quotacheck is
1651          * going on */
1652         cli->cl_qchk_stat = -ENODATA;
1653         rc = ptlrpc_queue_wait(req);
1654         if (rc)
1655                 cli->cl_qchk_stat = rc;
1656         ptlrpc_req_finished(req);
1657         return rc;
1658 }
1659
1660 static int mdc_quota_poll_check(struct obd_export *exp,
1661                                 struct if_quotacheck *qchk)
1662 {
1663         struct client_obd *cli = &exp->exp_obd->u.cli;
1664         int rc;
1665
1666         qchk->obd_uuid = cli->cl_target_uuid;
1667         memcpy(qchk->obd_type, LUSTRE_MDS_NAME, strlen(LUSTRE_MDS_NAME));
1668
1669         rc = cli->cl_qchk_stat;
1670         /* the client is not the previous one */
1671         if (rc == CL_NOT_QUOTACHECKED)
1672                 rc = -EINTR;
1673         return rc;
1674 }
1675
1676 static int mdc_quotactl(struct obd_device *unused, struct obd_export *exp,
1677                         struct obd_quotactl *oqctl)
1678 {
1679         struct ptlrpc_request   *req;
1680         struct obd_quotactl     *oqc;
1681         int                   rc;
1682
1683         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
1684                                         &RQF_MDS_QUOTACTL, LUSTRE_MDS_VERSION,
1685                                         MDS_QUOTACTL);
1686         if (req == NULL)
1687                 return -ENOMEM;
1688
1689         oqc = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1690         *oqc = *oqctl;
1691
1692         ptlrpc_request_set_replen(req);
1693         ptlrpc_at_set_req_timeout(req);
1694         req->rq_no_resend = 1;
1695
1696         rc = ptlrpc_queue_wait(req);
1697         if (rc)
1698                 CERROR("ptlrpc_queue_wait failed, rc: %d\n", rc);
1699
1700         if (req->rq_repmsg) {
1701                 oqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1702                 if (oqc) {
1703                         *oqctl = *oqc;
1704                 } else if (!rc) {
1705                         CERROR ("Can't unpack obd_quotactl\n");
1706                         rc = -EPROTO;
1707                 }
1708         } else if (!rc) {
1709                 CERROR("Can't unpack obd_quotactl\n");
1710                 rc = -EPROTO;
1711         }
1712         ptlrpc_req_finished(req);
1713
1714         return rc;
1715 }
1716
1717 static int mdc_ioc_swap_layouts(struct obd_export *exp,
1718                                 struct md_op_data *op_data)
1719 {
1720         LIST_HEAD(cancels);
1721         struct ptlrpc_request   *req;
1722         int                      rc, count;
1723         struct mdc_swap_layouts *msl, *payload;
1724
1725         msl = op_data->op_data;
1726
1727         /* When the MDT will get the MDS_SWAP_LAYOUTS RPC the
1728          * first thing it will do is to cancel the 2 layout
1729          * locks hold by this client.
1730          * So the client must cancel its layout locks on the 2 fids
1731          * with the request RPC to avoid extra RPC round trips
1732          */
1733         count = mdc_resource_get_unused(exp, &op_data->op_fid1, &cancels,
1734                                         LCK_CR, MDS_INODELOCK_LAYOUT);
1735         count += mdc_resource_get_unused(exp, &op_data->op_fid2, &cancels,
1736                                          LCK_CR, MDS_INODELOCK_LAYOUT);
1737
1738         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1739                                    &RQF_MDS_SWAP_LAYOUTS);
1740         if (req == NULL) {
1741                 ldlm_lock_list_put(&cancels, l_bl_ast, count);
1742                 return -ENOMEM;
1743         }
1744
1745         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1746         mdc_set_capa_size(req, &RMF_CAPA2, op_data->op_capa2);
1747
1748         rc = mdc_prep_elc_req(exp, req, MDS_SWAP_LAYOUTS, &cancels, count);
1749         if (rc) {
1750                 ptlrpc_request_free(req);
1751                 return rc;
1752         }
1753
1754         mdc_swap_layouts_pack(req, op_data);
1755
1756         payload = req_capsule_client_get(&req->rq_pill, &RMF_SWAP_LAYOUTS);
1757         LASSERT(payload);
1758
1759         *payload = *msl;
1760
1761         ptlrpc_request_set_replen(req);
1762
1763         rc = ptlrpc_queue_wait(req);
1764         if (rc)
1765                 GOTO(out, rc);
1766
1767 out:
1768         ptlrpc_req_finished(req);
1769         return rc;
1770 }
1771
1772 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1773                          void *karg, void *uarg)
1774 {
1775         struct obd_device *obd = exp->exp_obd;
1776         struct obd_ioctl_data *data = karg;
1777         struct obd_import *imp = obd->u.cli.cl_import;
1778         struct llog_ctxt *ctxt;
1779         int rc;
1780
1781         if (!try_module_get(THIS_MODULE)) {
1782                 CERROR("Can't get module. Is it alive?");
1783                 return -EINVAL;
1784         }
1785         switch (cmd) {
1786         case OBD_IOC_CHANGELOG_SEND:
1787                 rc = mdc_ioc_changelog_send(obd, karg);
1788                 GOTO(out, rc);
1789         case OBD_IOC_CHANGELOG_CLEAR: {
1790                 struct ioc_changelog *icc = karg;
1791                 struct changelog_setinfo cs =
1792                         {.cs_recno = icc->icc_recno, .cs_id = icc->icc_id};
1793                 rc = obd_set_info_async(NULL, exp, strlen(KEY_CHANGELOG_CLEAR),
1794                                         KEY_CHANGELOG_CLEAR, sizeof(cs), &cs,
1795                                         NULL);
1796                 GOTO(out, rc);
1797         }
1798         case OBD_IOC_FID2PATH:
1799                 rc = mdc_ioc_fid2path(exp, karg);
1800                 GOTO(out, rc);
1801         case LL_IOC_HSM_CT_START:
1802                 rc = mdc_ioc_hsm_ct_start(exp, karg);
1803                 /* ignore if it was already registered on this MDS. */
1804                 if (rc == -EEXIST)
1805                         rc = 0;
1806                 GOTO(out, rc);
1807         case LL_IOC_HSM_PROGRESS:
1808                 rc = mdc_ioc_hsm_progress(exp, karg);
1809                 GOTO(out, rc);
1810         case LL_IOC_HSM_STATE_GET:
1811                 rc = mdc_ioc_hsm_state_get(exp, karg);
1812                 GOTO(out, rc);
1813         case LL_IOC_HSM_STATE_SET:
1814                 rc = mdc_ioc_hsm_state_set(exp, karg);
1815                 GOTO(out, rc);
1816         case LL_IOC_HSM_ACTION:
1817                 rc = mdc_ioc_hsm_current_action(exp, karg);
1818                 GOTO(out, rc);
1819         case LL_IOC_HSM_REQUEST:
1820                 rc = mdc_ioc_hsm_request(exp, karg);
1821                 GOTO(out, rc);
1822         case OBD_IOC_CLIENT_RECOVER:
1823                 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1, 0);
1824                 if (rc < 0)
1825                         GOTO(out, rc);
1826                 GOTO(out, rc = 0);
1827         case IOC_OSC_SET_ACTIVE:
1828                 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
1829                 GOTO(out, rc);
1830         case OBD_IOC_PARSE: {
1831                 ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
1832                 rc = class_config_parse_llog(NULL, ctxt, data->ioc_inlbuf1,
1833                                              NULL);
1834                 llog_ctxt_put(ctxt);
1835                 GOTO(out, rc);
1836         }
1837         case OBD_IOC_LLOG_INFO:
1838         case OBD_IOC_LLOG_PRINT: {
1839                 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
1840                 rc = llog_ioctl(NULL, ctxt, cmd, data);
1841                 llog_ctxt_put(ctxt);
1842                 GOTO(out, rc);
1843         }
1844         case OBD_IOC_POLL_QUOTACHECK:
1845                 rc = mdc_quota_poll_check(exp, (struct if_quotacheck *)karg);
1846                 GOTO(out, rc);
1847         case OBD_IOC_PING_TARGET:
1848                 rc = ptlrpc_obd_ping(obd);
1849                 GOTO(out, rc);
1850         /*
1851          * Normally IOC_OBD_STATFS, OBD_IOC_QUOTACTL iocontrol are handled by
1852          * LMV instead of MDC. But when the cluster is upgraded from 1.8,
1853          * there'd be no LMV layer thus we might be called here. Eventually
1854          * this code should be removed.
1855          * bz20731, LU-592.
1856          */
1857         case IOC_OBD_STATFS: {
1858                 struct obd_statfs stat_buf = {0};
1859
1860                 if (*((__u32 *) data->ioc_inlbuf2) != 0)
1861                         GOTO(out, rc = -ENODEV);
1862
1863                 /* copy UUID */
1864                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd),
1865                                      min((int) data->ioc_plen2,
1866                                          (int) sizeof(struct obd_uuid))))
1867                         GOTO(out, rc = -EFAULT);
1868
1869                 rc = mdc_statfs(NULL, obd->obd_self_export, &stat_buf,
1870                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1871                                 0);
1872                 if (rc != 0)
1873                         GOTO(out, rc);
1874
1875                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1876                                      min((int) data->ioc_plen1,
1877                                          (int) sizeof(stat_buf))))
1878                         GOTO(out, rc = -EFAULT);
1879
1880                 GOTO(out, rc = 0);
1881         }
1882         case OBD_IOC_QUOTACTL: {
1883                 struct if_quotactl *qctl = karg;
1884                 struct obd_quotactl *oqctl;
1885
1886                 OBD_ALLOC_PTR(oqctl);
1887                 if (oqctl == NULL)
1888                         GOTO(out, rc = -ENOMEM);
1889
1890                 QCTL_COPY(oqctl, qctl);
1891                 rc = obd_quotactl(exp, oqctl);
1892                 if (rc == 0) {
1893                         QCTL_COPY(qctl, oqctl);
1894                         qctl->qc_valid = QC_MDTIDX;
1895                         qctl->obd_uuid = obd->u.cli.cl_target_uuid;
1896                 }
1897
1898                 OBD_FREE_PTR(oqctl);
1899                 GOTO(out, rc);
1900         }
1901         case LL_IOC_GET_CONNECT_FLAGS:
1902                 if (copy_to_user(uarg, exp_connect_flags_ptr(exp),
1903                                  sizeof(*exp_connect_flags_ptr(exp))))
1904                         GOTO(out, rc = -EFAULT);
1905
1906                 GOTO(out, rc = 0);
1907         case LL_IOC_LOV_SWAP_LAYOUTS:
1908                 rc = mdc_ioc_swap_layouts(exp, karg);
1909                 GOTO(out, rc);
1910         default:
1911                 CERROR("unrecognised ioctl: cmd = %#x\n", cmd);
1912                 GOTO(out, rc = -ENOTTY);
1913         }
1914 out:
1915         module_put(THIS_MODULE);
1916
1917         return rc;
1918 }
1919
1920 int mdc_get_info_rpc(struct obd_export *exp,
1921                      obd_count keylen, void *key,
1922                      int vallen, void *val)
1923 {
1924         struct obd_import      *imp = class_exp2cliimp(exp);
1925         struct ptlrpc_request  *req;
1926         char               *tmp;
1927         int                  rc = -EINVAL;
1928
1929         req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
1930         if (req == NULL)
1931                 return -ENOMEM;
1932
1933         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
1934                              RCL_CLIENT, keylen);
1935         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
1936                              RCL_CLIENT, sizeof(__u32));
1937
1938         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
1939         if (rc) {
1940                 ptlrpc_request_free(req);
1941                 return rc;
1942         }
1943
1944         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
1945         memcpy(tmp, key, keylen);
1946         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
1947         memcpy(tmp, &vallen, sizeof(__u32));
1948
1949         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
1950                              RCL_SERVER, vallen);
1951         ptlrpc_request_set_replen(req);
1952
1953         rc = ptlrpc_queue_wait(req);
1954         /* -EREMOTE means the get_info result is partial, and it needs to
1955          * continue on another MDT, see fid2path part in lmv_iocontrol */
1956         if (rc == 0 || rc == -EREMOTE) {
1957                 tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
1958                 memcpy(val, tmp, vallen);
1959                 if (ptlrpc_rep_need_swab(req)) {
1960                         if (KEY_IS(KEY_FID2PATH))
1961                                 lustre_swab_fid2path(val);
1962                 }
1963         }
1964         ptlrpc_req_finished(req);
1965
1966         return rc;
1967 }
1968
1969 static void lustre_swab_hai(struct hsm_action_item *h)
1970 {
1971         __swab32s(&h->hai_len);
1972         __swab32s(&h->hai_action);
1973         lustre_swab_lu_fid(&h->hai_fid);
1974         lustre_swab_lu_fid(&h->hai_dfid);
1975         __swab64s(&h->hai_cookie);
1976         __swab64s(&h->hai_extent.offset);
1977         __swab64s(&h->hai_extent.length);
1978         __swab64s(&h->hai_gid);
1979 }
1980
1981 static void lustre_swab_hal(struct hsm_action_list *h)
1982 {
1983         struct hsm_action_item  *hai;
1984         int                      i;
1985
1986         __swab32s(&h->hal_version);
1987         __swab32s(&h->hal_count);
1988         __swab32s(&h->hal_archive_id);
1989         __swab64s(&h->hal_flags);
1990         hai = hai_zero(h);
1991         for (i = 0; i < h->hal_count; i++, hai = hai_next(hai))
1992                 lustre_swab_hai(hai);
1993 }
1994
1995 static void lustre_swab_kuch(struct kuc_hdr *l)
1996 {
1997         __swab16s(&l->kuc_magic);
1998         /* __u8 l->kuc_transport */
1999         __swab16s(&l->kuc_msgtype);
2000         __swab16s(&l->kuc_msglen);
2001 }
2002
2003 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2004                                 struct lustre_kernelcomm *lk)
2005 {
2006         struct obd_import  *imp = class_exp2cliimp(exp);
2007         __u32               archive = lk->lk_data;
2008         int                 rc = 0;
2009
2010         if (lk->lk_group != KUC_GRP_HSM) {
2011                 CERROR("Bad copytool group %d\n", lk->lk_group);
2012                 return -EINVAL;
2013         }
2014
2015         CDEBUG(D_HSM, "CT start r%d w%d u%d g%d f%#x\n", lk->lk_rfd, lk->lk_wfd,
2016                lk->lk_uid, lk->lk_group, lk->lk_flags);
2017
2018         if (lk->lk_flags & LK_FLG_STOP) {
2019                 /* Unregister with the coordinator */
2020                 rc = mdc_ioc_hsm_ct_unregister(imp);
2021         } else {
2022                 rc = mdc_ioc_hsm_ct_register(imp, archive);
2023         }
2024
2025         return rc;
2026 }
2027
2028 /**
2029  * Send a message to any listening copytools
2030  * @param val KUC message (kuc_hdr + hsm_action_list)
2031  * @param len total length of message
2032  */
2033 static int mdc_hsm_copytool_send(int len, void *val)
2034 {
2035         struct kuc_hdr          *lh = (struct kuc_hdr *)val;
2036         struct hsm_action_list  *hal = (struct hsm_action_list *)(lh + 1);
2037         int                      rc;
2038
2039         if (len < sizeof(*lh) + sizeof(*hal)) {
2040                 CERROR("Short HSM message %d < %d\n", len,
2041                        (int) (sizeof(*lh) + sizeof(*hal)));
2042                 return -EPROTO;
2043         }
2044         if (lh->kuc_magic == __swab16(KUC_MAGIC)) {
2045                 lustre_swab_kuch(lh);
2046                 lustre_swab_hal(hal);
2047         } else if (lh->kuc_magic != KUC_MAGIC) {
2048                 CERROR("Bad magic %x!=%x\n", lh->kuc_magic, KUC_MAGIC);
2049                 return -EPROTO;
2050         }
2051
2052         CDEBUG(D_HSM, " Received message mg=%x t=%d m=%d l=%d actions=%d "
2053                "on %s\n",
2054                lh->kuc_magic, lh->kuc_transport, lh->kuc_msgtype,
2055                lh->kuc_msglen, hal->hal_count, hal->hal_fsname);
2056
2057         /* Broadcast to HSM listeners */
2058         rc = libcfs_kkuc_group_put(KUC_GRP_HSM, lh);
2059
2060         return rc;
2061 }
2062
2063 /**
2064  * callback function passed to kuc for re-registering each HSM copytool
2065  * running on MDC, after MDT shutdown/recovery.
2066  * @param data archive id served by the copytool
2067  * @param cb_arg callback argument (obd_import)
2068  */
2069 static int mdc_hsm_ct_reregister(__u32 data, void *cb_arg)
2070 {
2071         struct obd_import       *imp = (struct obd_import *)cb_arg;
2072         __u32                    archive = data;
2073         int                      rc;
2074
2075         CDEBUG(D_HA, "recover copytool registration to MDT (archive=%#x)\n",
2076                archive);
2077         rc = mdc_ioc_hsm_ct_register(imp, archive);
2078
2079         /* ignore error if the copytool is already registered */
2080         return ((rc != 0) && (rc != -EEXIST)) ? rc : 0;
2081 }
2082
2083 /**
2084  * Re-establish all kuc contexts with MDT
2085  * after MDT shutdown/recovery.
2086  */
2087 static int mdc_kuc_reregister(struct obd_import *imp)
2088 {
2089         /* re-register HSM agents */
2090         return libcfs_kkuc_group_foreach(KUC_GRP_HSM, mdc_hsm_ct_reregister,
2091                                          (void *)imp);
2092 }
2093
2094 int mdc_set_info_async(const struct lu_env *env,
2095                        struct obd_export *exp,
2096                        obd_count keylen, void *key,
2097                        obd_count vallen, void *val,
2098                        struct ptlrpc_request_set *set)
2099 {
2100         struct obd_import       *imp = class_exp2cliimp(exp);
2101         int                      rc;
2102
2103         if (KEY_IS(KEY_READ_ONLY)) {
2104                 if (vallen != sizeof(int))
2105                         return -EINVAL;
2106
2107                 spin_lock(&imp->imp_lock);
2108                 if (*((int *)val)) {
2109                         imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
2110                         imp->imp_connect_data.ocd_connect_flags |=
2111                                                         OBD_CONNECT_RDONLY;
2112                 } else {
2113                         imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
2114                         imp->imp_connect_data.ocd_connect_flags &=
2115                                                         ~OBD_CONNECT_RDONLY;
2116                 }
2117                 spin_unlock(&imp->imp_lock);
2118
2119                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2120                                        keylen, key, vallen, val, set);
2121                 return rc;
2122         }
2123         if (KEY_IS(KEY_SPTLRPC_CONF)) {
2124                 sptlrpc_conf_client_adapt(exp->exp_obd);
2125                 return 0;
2126         }
2127         if (KEY_IS(KEY_FLUSH_CTX)) {
2128                 sptlrpc_import_flush_my_ctx(imp);
2129                 return 0;
2130         }
2131         if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2132                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2133                                        keylen, key, vallen, val, set);
2134                 return rc;
2135         }
2136         if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) {
2137                 rc = mdc_hsm_copytool_send(vallen, val);
2138                 return rc;
2139         }
2140
2141         CERROR("Unknown key %s\n", (char *)key);
2142         return -EINVAL;
2143 }
2144
2145 int mdc_get_info(const struct lu_env *env, struct obd_export *exp,
2146                  __u32 keylen, void *key, __u32 *vallen, void *val,
2147                  struct lov_stripe_md *lsm)
2148 {
2149         int rc = -EINVAL;
2150
2151         if (KEY_IS(KEY_MAX_EASIZE)) {
2152                 int mdsize, *max_easize;
2153
2154                 if (*vallen != sizeof(int))
2155                         return -EINVAL;
2156                 mdsize = *(int *)val;
2157                 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
2158                         exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
2159                 max_easize = val;
2160                 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
2161                 return 0;
2162         } else if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2163                 int *default_easize;
2164
2165                 if (*vallen != sizeof(int))
2166                         return -EINVAL;
2167                 default_easize = val;
2168                 *default_easize = exp->exp_obd->u.cli.cl_default_mds_easize;
2169                 return 0;
2170         } else if (KEY_IS(KEY_MAX_COOKIESIZE)) {
2171                 int mdsize, *max_cookiesize;
2172
2173                 if (*vallen != sizeof(int))
2174                         return -EINVAL;
2175                 mdsize = *(int *)val;
2176                 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_cookiesize)
2177                         exp->exp_obd->u.cli.cl_max_mds_cookiesize = mdsize;
2178                 max_cookiesize = val;
2179                 *max_cookiesize = exp->exp_obd->u.cli.cl_max_mds_cookiesize;
2180                 return 0;
2181         } else if (KEY_IS(KEY_DEFAULT_COOKIESIZE)) {
2182                 int *default_cookiesize;
2183
2184                 if (*vallen != sizeof(int))
2185                         return -EINVAL;
2186                 default_cookiesize = val;
2187                 *default_cookiesize =
2188                         exp->exp_obd->u.cli.cl_default_mds_cookiesize;
2189                 return 0;
2190         } else if (KEY_IS(KEY_CONN_DATA)) {
2191                 struct obd_import *imp = class_exp2cliimp(exp);
2192                 struct obd_connect_data *data = val;
2193
2194                 if (*vallen != sizeof(*data))
2195                         return -EINVAL;
2196
2197                 *data = imp->imp_connect_data;
2198                 return 0;
2199         } else if (KEY_IS(KEY_TGT_COUNT)) {
2200                 *((int *)val) = 1;
2201                 return 0;
2202         }
2203
2204         rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
2205
2206         return rc;
2207 }
2208
2209 static int mdc_pin(struct obd_export *exp, const struct lu_fid *fid,
2210                    struct obd_capa *oc, struct obd_client_handle *handle,
2211                    int flags)
2212 {
2213         struct ptlrpc_request *req;
2214         struct mdt_body       *body;
2215         int                 rc;
2216
2217         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_PIN);
2218         if (req == NULL)
2219                 return -ENOMEM;
2220
2221         mdc_set_capa_size(req, &RMF_CAPA1, oc);
2222
2223         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_PIN);
2224         if (rc) {
2225                 ptlrpc_request_free(req);
2226                 return rc;
2227         }
2228
2229         mdc_pack_body(req, fid, oc, 0, 0, -1, flags);
2230
2231         ptlrpc_request_set_replen(req);
2232
2233         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
2234         rc = ptlrpc_queue_wait(req);
2235         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
2236         if (rc) {
2237                 CERROR("Pin failed: %d\n", rc);
2238                 GOTO(err_out, rc);
2239         }
2240
2241         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2242         if (body == NULL)
2243                 GOTO(err_out, rc = -EPROTO);
2244
2245         handle->och_fh = body->handle;
2246         handle->och_magic = OBD_CLIENT_HANDLE_MAGIC;
2247
2248         handle->och_mod = obd_mod_alloc();
2249         if (handle->och_mod == NULL) {
2250                 DEBUG_REQ(D_ERROR, req, "can't allocate md_open_data");
2251                 GOTO(err_out, rc = -ENOMEM);
2252         }
2253         handle->och_mod->mod_open_req = req; /* will be dropped by unpin */
2254
2255         return 0;
2256
2257 err_out:
2258         ptlrpc_req_finished(req);
2259         return rc;
2260 }
2261
2262 static int mdc_unpin(struct obd_export *exp, struct obd_client_handle *handle,
2263                      int flag)
2264 {
2265         struct ptlrpc_request *req;
2266         struct mdt_body       *body;
2267         int                 rc;
2268
2269         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_UNPIN,
2270                                         LUSTRE_MDS_VERSION, MDS_UNPIN);
2271         if (req == NULL)
2272                 return -ENOMEM;
2273
2274         body = req_capsule_client_get(&req->rq_pill, &RMF_MDT_BODY);
2275         body->handle = handle->och_fh;
2276         body->flags = flag;
2277
2278         ptlrpc_request_set_replen(req);
2279
2280         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
2281         rc = ptlrpc_queue_wait(req);
2282         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
2283
2284         if (rc != 0)
2285                 CERROR("Unpin failed: %d\n", rc);
2286
2287         ptlrpc_req_finished(req);
2288         ptlrpc_req_finished(handle->och_mod->mod_open_req);
2289
2290         obd_mod_put(handle->och_mod);
2291         return rc;
2292 }
2293
2294 int mdc_sync(struct obd_export *exp, const struct lu_fid *fid,
2295              struct obd_capa *oc, struct ptlrpc_request **request)
2296 {
2297         struct ptlrpc_request *req;
2298         int                 rc;
2299
2300         *request = NULL;
2301         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
2302         if (req == NULL)
2303                 return -ENOMEM;
2304
2305         mdc_set_capa_size(req, &RMF_CAPA1, oc);
2306
2307         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
2308         if (rc) {
2309                 ptlrpc_request_free(req);
2310                 return rc;
2311         }
2312
2313         mdc_pack_body(req, fid, oc, 0, 0, -1, 0);
2314
2315         ptlrpc_request_set_replen(req);
2316
2317         rc = ptlrpc_queue_wait(req);
2318         if (rc)
2319                 ptlrpc_req_finished(req);
2320         else
2321                 *request = req;
2322         return rc;
2323 }
2324
2325 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
2326                             enum obd_import_event event)
2327 {
2328         int rc = 0;
2329
2330         LASSERT(imp->imp_obd == obd);
2331
2332         switch (event) {
2333         case IMP_EVENT_DISCON: {
2334 #if 0
2335                 /* XXX Pass event up to OBDs stack. used only for FLD now */
2336                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DISCON, NULL);
2337 #endif
2338                 break;
2339         }
2340         case IMP_EVENT_INACTIVE: {
2341                 struct client_obd *cli = &obd->u.cli;
2342                 /*
2343                  * Flush current sequence to make client obtain new one
2344                  * from server in case of disconnect/reconnect.
2345                  */
2346                 if (cli->cl_seq != NULL)
2347                         seq_client_flush(cli->cl_seq);
2348
2349                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
2350                 break;
2351         }
2352         case IMP_EVENT_INVALIDATE: {
2353                 struct ldlm_namespace *ns = obd->obd_namespace;
2354
2355                 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2356
2357                 break;
2358         }
2359         case IMP_EVENT_ACTIVE:
2360                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
2361                 /* redo the kuc registration after reconnecting */
2362                 if (rc == 0)
2363                         rc = mdc_kuc_reregister(imp);
2364                 break;
2365         case IMP_EVENT_OCD:
2366                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
2367                 break;
2368         case IMP_EVENT_DEACTIVATE:
2369         case IMP_EVENT_ACTIVATE:
2370                 break;
2371         default:
2372                 CERROR("Unknown import event %x\n", event);
2373                 LBUG();
2374         }
2375         return rc;
2376 }
2377
2378 int mdc_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
2379                   struct md_op_data *op_data)
2380 {
2381         struct client_obd *cli = &exp->exp_obd->u.cli;
2382         struct lu_client_seq *seq = cli->cl_seq;
2383
2384         return seq_client_alloc_fid(NULL, seq, fid);
2385 }
2386
2387 struct obd_uuid *mdc_get_uuid(struct obd_export *exp) {
2388         struct client_obd *cli = &exp->exp_obd->u.cli;
2389         return &cli->cl_target_uuid;
2390 }
2391
2392 /**
2393  * Determine whether the lock can be canceled before replaying it during
2394  * recovery, non zero value will be return if the lock can be canceled,
2395  * or zero returned for not
2396  */
2397 static int mdc_cancel_for_recovery(struct ldlm_lock *lock)
2398 {
2399         if (lock->l_resource->lr_type != LDLM_IBITS)
2400                 return 0;
2401
2402         /* FIXME: if we ever get into a situation where there are too many
2403          * opened files with open locks on a single node, then we really
2404          * should replay these open locks to reget it */
2405         if (lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_OPEN)
2406                 return 0;
2407
2408         return 1;
2409 }
2410
2411 static int mdc_resource_inode_free(struct ldlm_resource *res)
2412 {
2413         if (res->lr_lvb_inode)
2414                 res->lr_lvb_inode = NULL;
2415
2416         return 0;
2417 }
2418
2419 struct ldlm_valblock_ops inode_lvbo = {
2420         .lvbo_free = mdc_resource_inode_free,
2421 };
2422
2423 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
2424 {
2425         struct client_obd *cli = &obd->u.cli;
2426         struct lprocfs_static_vars lvars = { NULL };
2427         int rc;
2428
2429         OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
2430         if (!cli->cl_rpc_lock)
2431                 return -ENOMEM;
2432         mdc_init_rpc_lock(cli->cl_rpc_lock);
2433
2434         ptlrpcd_addref();
2435
2436         OBD_ALLOC(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
2437         if (!cli->cl_close_lock)
2438                 GOTO(err_rpc_lock, rc = -ENOMEM);
2439         mdc_init_rpc_lock(cli->cl_close_lock);
2440
2441         rc = client_obd_setup(obd, cfg);
2442         if (rc)
2443                 GOTO(err_close_lock, rc);
2444         lprocfs_mdc_init_vars(&lvars);
2445         lprocfs_obd_setup(obd, lvars.obd_vars);
2446         sptlrpc_lprocfs_cliobd_attach(obd);
2447         ptlrpc_lprocfs_register_obd(obd);
2448
2449         ns_register_cancel(obd->obd_namespace, mdc_cancel_for_recovery);
2450
2451         obd->obd_namespace->ns_lvbo = &inode_lvbo;
2452
2453         rc = obd_llog_init(obd, &obd->obd_olg, obd, NULL);
2454         if (rc) {
2455                 mdc_cleanup(obd);
2456                 CERROR("failed to setup llogging subsystems\n");
2457         }
2458
2459         return rc;
2460
2461 err_close_lock:
2462         OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
2463 err_rpc_lock:
2464         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
2465         ptlrpcd_decref();
2466         return rc;
2467 }
2468
2469 /* Initialize the default and maximum LOV EA and cookie sizes.  This allows
2470  * us to make MDS RPCs with large enough reply buffers to hold a default
2471  * sized EA and cookie without having to calculate this (via a call into the
2472  * LOV + OSCs) each time we make an RPC.  The maximum size is also tracked
2473  * but not used to avoid wastefully vmalloc()'ing large reply buffers when
2474  * a large number of stripes is possible.  If a larger reply buffer is
2475  * required it will be reallocated in the ptlrpc layer due to overflow.
2476  */
2477 static int mdc_init_ea_size(struct obd_export *exp, int easize,
2478                             int def_easize, int cookiesize, int def_cookiesize)
2479 {
2480         struct obd_device *obd = exp->exp_obd;
2481         struct client_obd *cli = &obd->u.cli;
2482
2483         if (cli->cl_max_mds_easize < easize)
2484                 cli->cl_max_mds_easize = easize;
2485
2486         if (cli->cl_default_mds_easize < def_easize)
2487                 cli->cl_default_mds_easize = def_easize;
2488
2489         if (cli->cl_max_mds_cookiesize < cookiesize)
2490                 cli->cl_max_mds_cookiesize = cookiesize;
2491
2492         if (cli->cl_default_mds_cookiesize < def_cookiesize)
2493                 cli->cl_default_mds_cookiesize = def_cookiesize;
2494
2495         return 0;
2496 }
2497
2498 static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2499 {
2500         int rc = 0;
2501
2502         switch (stage) {
2503         case OBD_CLEANUP_EARLY:
2504                 break;
2505         case OBD_CLEANUP_EXPORTS:
2506                 /* Failsafe, ok if racy */
2507                 if (obd->obd_type->typ_refcnt <= 1)
2508                         libcfs_kkuc_group_rem(0, KUC_GRP_HSM);
2509
2510                 obd_cleanup_client_import(obd);
2511                 ptlrpc_lprocfs_unregister_obd(obd);
2512                 lprocfs_obd_cleanup(obd);
2513
2514                 rc = obd_llog_finish(obd, 0);
2515                 if (rc != 0)
2516                         CERROR("failed to cleanup llogging subsystems\n");
2517                 break;
2518         }
2519         return rc;
2520 }
2521
2522 static int mdc_cleanup(struct obd_device *obd)
2523 {
2524         struct client_obd *cli = &obd->u.cli;
2525
2526         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
2527         OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
2528
2529         ptlrpcd_decref();
2530
2531         return client_obd_cleanup(obd);
2532 }
2533
2534
2535 static int mdc_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
2536                          struct obd_device *tgt, int *index)
2537 {
2538         struct llog_ctxt        *ctxt;
2539         int                      rc;
2540
2541         LASSERT(olg == &obd->obd_olg);
2542
2543         rc = llog_setup(NULL, obd, olg, LLOG_CHANGELOG_REPL_CTXT, tgt,
2544                         &llog_client_ops);
2545         if (rc)
2546                 return rc;
2547
2548         ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
2549         llog_initiator_connect(ctxt);
2550         llog_ctxt_put(ctxt);
2551
2552         return 0;
2553 }
2554
2555 static int mdc_llog_finish(struct obd_device *obd, int count)
2556 {
2557         struct llog_ctxt *ctxt;
2558
2559         ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
2560         if (ctxt)
2561                 llog_cleanup(NULL, ctxt);
2562
2563         return 0;
2564 }
2565
2566 static int mdc_process_config(struct obd_device *obd, obd_count len, void *buf)
2567 {
2568         struct lustre_cfg *lcfg = buf;
2569         struct lprocfs_static_vars lvars = { NULL };
2570         int rc = 0;
2571
2572         lprocfs_mdc_init_vars(&lvars);
2573         switch (lcfg->lcfg_command) {
2574         default:
2575                 rc = class_process_proc_param(PARAM_MDC, lvars.obd_vars,
2576                                               lcfg, obd);
2577                 if (rc > 0)
2578                         rc = 0;
2579                 break;
2580         }
2581         return(rc);
2582 }
2583
2584
2585 /* get remote permission for current user on fid */
2586 int mdc_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
2587                         struct obd_capa *oc, __u32 suppgid,
2588                         struct ptlrpc_request **request)
2589 {
2590         struct ptlrpc_request  *req;
2591         int                 rc;
2592
2593         LASSERT(client_is_remote(exp));
2594
2595         *request = NULL;
2596         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
2597         if (req == NULL)
2598                 return -ENOMEM;
2599
2600         mdc_set_capa_size(req, &RMF_CAPA1, oc);
2601
2602         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
2603         if (rc) {
2604                 ptlrpc_request_free(req);
2605                 return rc;
2606         }
2607
2608         mdc_pack_body(req, fid, oc, OBD_MD_FLRMTPERM, 0, suppgid, 0);
2609
2610         req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
2611                              sizeof(struct mdt_remote_perm));
2612
2613         ptlrpc_request_set_replen(req);
2614
2615         rc = ptlrpc_queue_wait(req);
2616         if (rc)
2617                 ptlrpc_req_finished(req);
2618         else
2619                 *request = req;
2620         return rc;
2621 }
2622
2623 static int mdc_interpret_renew_capa(const struct lu_env *env,
2624                                     struct ptlrpc_request *req, void *args,
2625                                     int status)
2626 {
2627         struct mdc_renew_capa_args *ra = args;
2628         struct mdt_body *body = NULL;
2629         struct lustre_capa *capa;
2630
2631         if (status)
2632                 GOTO(out, capa = ERR_PTR(status));
2633
2634         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2635         if (body == NULL)
2636                 GOTO(out, capa = ERR_PTR(-EFAULT));
2637
2638         if ((body->valid & OBD_MD_FLOSSCAPA) == 0)
2639                 GOTO(out, capa = ERR_PTR(-ENOENT));
2640
2641         capa = req_capsule_server_get(&req->rq_pill, &RMF_CAPA2);
2642         if (!capa)
2643                 GOTO(out, capa = ERR_PTR(-EFAULT));
2644 out:
2645         ra->ra_cb(ra->ra_oc, capa);
2646         return 0;
2647 }
2648
2649 static int mdc_renew_capa(struct obd_export *exp, struct obd_capa *oc,
2650                           renew_capa_cb_t cb)
2651 {
2652         struct ptlrpc_request *req;
2653         struct mdc_renew_capa_args *ra;
2654
2655         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_GETATTR,
2656                                         LUSTRE_MDS_VERSION, MDS_GETATTR);
2657         if (req == NULL)
2658                 return -ENOMEM;
2659
2660         /* NB, OBD_MD_FLOSSCAPA is set here, but it doesn't necessarily mean the
2661          * capa to renew is oss capa.
2662          */
2663         mdc_pack_body(req, &oc->c_capa.lc_fid, oc, OBD_MD_FLOSSCAPA, 0, -1, 0);
2664         ptlrpc_request_set_replen(req);
2665
2666         CLASSERT(sizeof(*ra) <= sizeof(req->rq_async_args));
2667         ra = ptlrpc_req_async_args(req);
2668         ra->ra_oc = oc;
2669         ra->ra_cb = cb;
2670         req->rq_interpret_reply = mdc_interpret_renew_capa;
2671         ptlrpcd_add_req(req, PDL_POLICY_LOCAL, -1);
2672         return 0;
2673 }
2674
2675 struct obd_ops mdc_obd_ops = {
2676         .o_owner            = THIS_MODULE,
2677         .o_setup            = mdc_setup,
2678         .o_precleanup       = mdc_precleanup,
2679         .o_cleanup        = mdc_cleanup,
2680         .o_add_conn      = client_import_add_conn,
2681         .o_del_conn      = client_import_del_conn,
2682         .o_connect          = client_connect_import,
2683         .o_disconnect       = client_disconnect_export,
2684         .o_iocontrol    = mdc_iocontrol,
2685         .o_set_info_async   = mdc_set_info_async,
2686         .o_statfs          = mdc_statfs,
2687         .o_pin        = mdc_pin,
2688         .o_unpin            = mdc_unpin,
2689         .o_fid_init         = client_fid_init,
2690         .o_fid_fini         = client_fid_fini,
2691         .o_fid_alloc    = mdc_fid_alloc,
2692         .o_import_event     = mdc_import_event,
2693         .o_llog_init    = mdc_llog_init,
2694         .o_llog_finish      = mdc_llog_finish,
2695         .o_get_info      = mdc_get_info,
2696         .o_process_config   = mdc_process_config,
2697         .o_get_uuid      = mdc_get_uuid,
2698         .o_quotactl      = mdc_quotactl,
2699         .o_quotacheck       = mdc_quotacheck
2700 };
2701
2702 struct md_ops mdc_md_ops = {
2703         .m_getstatus    = mdc_getstatus,
2704         .m_null_inode       = mdc_null_inode,
2705         .m_find_cbdata      = mdc_find_cbdata,
2706         .m_close            = mdc_close,
2707         .m_create          = mdc_create,
2708         .m_done_writing     = mdc_done_writing,
2709         .m_enqueue        = mdc_enqueue,
2710         .m_getattr        = mdc_getattr,
2711         .m_getattr_name     = mdc_getattr_name,
2712         .m_intent_lock      = mdc_intent_lock,
2713         .m_link      = mdc_link,
2714         .m_is_subdir    = mdc_is_subdir,
2715         .m_rename          = mdc_rename,
2716         .m_setattr        = mdc_setattr,
2717         .m_setxattr      = mdc_setxattr,
2718         .m_getxattr      = mdc_getxattr,
2719         .m_sync      = mdc_sync,
2720         .m_readpage      = mdc_readpage,
2721         .m_unlink          = mdc_unlink,
2722         .m_cancel_unused    = mdc_cancel_unused,
2723         .m_init_ea_size     = mdc_init_ea_size,
2724         .m_set_lock_data    = mdc_set_lock_data,
2725         .m_lock_match       = mdc_lock_match,
2726         .m_get_lustre_md    = mdc_get_lustre_md,
2727         .m_free_lustre_md   = mdc_free_lustre_md,
2728         .m_set_open_replay_data = mdc_set_open_replay_data,
2729         .m_clear_open_replay_data = mdc_clear_open_replay_data,
2730         .m_renew_capa       = mdc_renew_capa,
2731         .m_unpack_capa      = mdc_unpack_capa,
2732         .m_get_remote_perm  = mdc_get_remote_perm,
2733         .m_intent_getattr_async = mdc_intent_getattr_async,
2734         .m_revalidate_lock      = mdc_revalidate_lock
2735 };
2736
2737 int __init mdc_init(void)
2738 {
2739         int rc;
2740         struct lprocfs_static_vars lvars = { NULL };
2741         lprocfs_mdc_init_vars(&lvars);
2742
2743         rc = class_register_type(&mdc_obd_ops, &mdc_md_ops, lvars.module_vars,
2744                                  LUSTRE_MDC_NAME, NULL);
2745         return rc;
2746 }
2747
2748 static void /*__exit*/ mdc_exit(void)
2749 {
2750         class_unregister_type(LUSTRE_MDC_NAME);
2751 }
2752
2753 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2754 MODULE_DESCRIPTION("Lustre Metadata Client");
2755 MODULE_LICENSE("GPL");
2756
2757 module_init(mdc_init);
2758 module_exit(mdc_exit);