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, 2013, 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 #ifdef __KERNEL__
  40 # include <linux/module.h>
  41 # include <linux/pagemap.h>
  42 # include <linux/miscdevice.h>
  43 # include <linux/init.h>
  44 # include <linux/utsname.h>
  45 #else
  46 # include <liblustre.h>
  47 #endif
  48
  49 #include <lustre_acl.h>
  50 #include <lustre_ioctl.h>
  51 #include <obd_class.h>
  52 #include <lustre_lmv.h>
  53 #include <lustre_fid.h>
  54 #include <lprocfs_status.h>
  55 #include <lustre_param.h>
  56 #include <lustre_log.h>
  57 #include <cl_object.h>
  58 #include <lclient.h>
  59
  60 #include "mdc_internal.h"
  61
  62 #define REQUEST_MINOR 244
  63
  64 struct mdc_renew_capa_args {
  65         struct obd_capa        *ra_oc;
  66         renew_capa_cb_t         ra_cb;
  67 };
  68
  69 static int mdc_cleanup(struct obd_device *obd);
  70
  71 int mdc_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
  72                     const struct req_msg_field *field, struct obd_capa **oc)
  73 {
  74         struct lustre_capa *capa;
  75         struct obd_capa *c;
  76         ENTRY;
  77
  78         /* swabbed already in mdc_enqueue */
  79         capa = req_capsule_server_get(&req->rq_pill, field);
  80         if (capa == NULL)
  81                 RETURN(-EPROTO);
  82
  83         c = alloc_capa(CAPA_SITE_CLIENT);
  84         if (IS_ERR(c)) {
  85                 CDEBUG(D_INFO, "alloc capa failed!\n");
  86                 RETURN(PTR_ERR(c));
  87         } else {
  88                 c->c_capa = *capa;
  89                 *oc = c;
  90                 RETURN(0);
  91         }
  92 }
  93
  94 static inline int mdc_queue_wait(struct ptlrpc_request *req)
  95 {
  96         struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
  97         int rc;
  98
  99         /* obd_get_request_slot() ensures that this client has no more
 100          * than cl_max_rpcs_in_flight RPCs simultaneously inf light
 101          * against an MDT. */
 102         rc = obd_get_request_slot(cli);
 103         if (rc != 0)
 104                 return rc;
 105
 106         rc = ptlrpc_queue_wait(req);
 107         obd_put_request_slot(cli);
 108
 109         return rc;
 110 }
 111
 112 /* Helper that implements most of mdc_getstatus and signal_completed_replay. */
 113 /* XXX this should become mdc_get_info("key"), sending MDS_GET_INFO RPC */
 114 static int send_getstatus(struct obd_import *imp, struct lu_fid *rootfid,
 115                           struct obd_capa **pc, int level, int msg_flags)
 116 {
 117         struct ptlrpc_request *req;
 118         struct mdt_body       *body;
 119         int                    rc;
 120         ENTRY;
 121
 122         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_GETSTATUS,
 123                                         LUSTRE_MDS_VERSION, MDS_GETSTATUS);
 124         if (req == NULL)
 125                 RETURN(-ENOMEM);
 126
 127         mdc_pack_body(req, NULL, NULL, 0, 0, -1, 0);
 128         lustre_msg_add_flags(req->rq_reqmsg, msg_flags);
 129         req->rq_send_state = level;
 130
 131         ptlrpc_request_set_replen(req);
 132
 133         rc = ptlrpc_queue_wait(req);
 134         if (rc)
 135                 GOTO(out, rc);
 136
 137         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 138         if (body == NULL)
 139                 GOTO(out, rc = -EPROTO);
 140
 141         if (body->mbo_valid & OBD_MD_FLMDSCAPA) {
 142                 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA1, pc);
 143                 if (rc)
 144                         GOTO(out, rc);
 145         }
 146
 147         *rootfid = body->mbo_fid1;
 148         CDEBUG(D_NET, "root fid="DFID", last_committed="LPU64"\n",
 149                PFID(rootfid), lustre_msg_get_last_committed(req->rq_repmsg));
 150         EXIT;
 151 out:
 152         ptlrpc_req_finished(req);
 153         return rc;
 154 }
 155
 156 /* This should be mdc_get_info("rootfid") */
 157 int mdc_getstatus(struct obd_export *exp, struct lu_fid *rootfid,
 158                   struct obd_capa **pc)
 159 {
 160         return send_getstatus(class_exp2cliimp(exp), rootfid, pc,
 161                               LUSTRE_IMP_FULL, 0);
 162 }
 163
 164 /*
 165  * This function now is known to always saying that it will receive 4 buffers
 166  * from server. Even for cases when acl_size and md_size is zero, RPC header
 167  * will contain 4 fields and RPC itself will contain zero size fields. This is
 168  * because mdt_getattr*() _always_ returns 4 fields, but if acl is not needed
 169  * and thus zero, it shrinks it, making zero size. The same story about
 170  * md_size. And this is course of problem when client waits for smaller number
 171  * of fields. This issue will be fixed later when client gets aware of RPC
 172  * layouts.  --umka
 173  */
 174 static int mdc_getattr_common(struct obd_export *exp,
 175                               struct ptlrpc_request *req)
 176 {
 177         struct req_capsule *pill = &req->rq_pill;
 178         struct mdt_body    *body;
 179         void               *eadata;
 180         int                 rc;
 181         ENTRY;
 182
 183         /* Request message already built. */
 184         rc = ptlrpc_queue_wait(req);
 185         if (rc != 0)
 186                 RETURN(rc);
 187
 188         /* sanity check for the reply */
 189         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 190         if (body == NULL)
 191                 RETURN(-EPROTO);
 192
 193         CDEBUG(D_NET, "mode: %o\n", body->mbo_mode);
 194
 195         mdc_update_max_ea_from_body(exp, body);
 196         if (body->mbo_eadatasize != 0) {
 197                 eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
 198                                                       body->mbo_eadatasize);
 199                 if (eadata == NULL)
 200                         RETURN(-EPROTO);
 201         }
 202
 203         if (body->mbo_valid & OBD_MD_FLRMTPERM) {
 204                 struct mdt_remote_perm *perm;
 205
 206                 LASSERT(client_is_remote(exp));
 207                 perm = req_capsule_server_swab_get(pill, &RMF_ACL,
 208                                                 lustre_swab_mdt_remote_perm);
 209                 if (perm == NULL)
 210                         RETURN(-EPROTO);
 211         }
 212
 213         if (body->mbo_valid & OBD_MD_FLMDSCAPA) {
 214                 struct lustre_capa *capa;
 215                 capa = req_capsule_server_get(pill, &RMF_CAPA1);
 216                 if (capa == NULL)
 217                         RETURN(-EPROTO);
 218         }
 219
 220         RETURN(0);
 221 }
 222
 223 int mdc_getattr(struct obd_export *exp, struct md_op_data *op_data,
 224                 struct ptlrpc_request **request)
 225 {
 226         struct ptlrpc_request *req;
 227         int                    rc;
 228         ENTRY;
 229
 230         /* Single MDS without an LMV case */
 231         if (op_data->op_flags & MF_GET_MDT_IDX) {
 232                 op_data->op_mds = 0;
 233                 RETURN(0);
 234         }
 235         *request = NULL;
 236         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
 237         if (req == NULL)
 238                 RETURN(-ENOMEM);
 239
 240         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 241
 242         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
 243         if (rc) {
 244                 ptlrpc_request_free(req);
 245                 RETURN(rc);
 246         }
 247
 248         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
 249                       op_data->op_valid, op_data->op_mode, -1, 0);
 250
 251         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 252                              op_data->op_mode);
 253         if (op_data->op_valid & OBD_MD_FLRMTPERM) {
 254                 LASSERT(client_is_remote(exp));
 255                 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
 256                                      sizeof(struct mdt_remote_perm));
 257         }
 258         ptlrpc_request_set_replen(req);
 259
 260         rc = mdc_getattr_common(exp, req);
 261         if (rc)
 262                 ptlrpc_req_finished(req);
 263         else
 264                 *request = req;
 265         RETURN(rc);
 266 }
 267
 268 int mdc_getattr_name(struct obd_export *exp, struct md_op_data *op_data,
 269                      struct ptlrpc_request **request)
 270 {
 271         struct ptlrpc_request *req;
 272         int                    rc;
 273         ENTRY;
 274
 275         *request = NULL;
 276         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 277                                    &RQF_MDS_GETATTR_NAME);
 278         if (req == NULL)
 279                 RETURN(-ENOMEM);
 280
 281         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 282         req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 283                              op_data->op_namelen + 1);
 284
 285         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR_NAME);
 286         if (rc) {
 287                 ptlrpc_request_free(req);
 288                 RETURN(rc);
 289         }
 290
 291         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
 292                       op_data->op_valid, op_data->op_mode,
 293                       op_data->op_suppgids[0], 0);
 294
 295         if (op_data->op_name) {
 296                 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 297                 LASSERT(strnlen(op_data->op_name, op_data->op_namelen) ==
 298                                 op_data->op_namelen);
 299                 memcpy(name, op_data->op_name, op_data->op_namelen);
 300         }
 301
 302         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 303                              op_data->op_mode);
 304         ptlrpc_request_set_replen(req);
 305
 306         rc = mdc_getattr_common(exp, req);
 307         if (rc)
 308                 ptlrpc_req_finished(req);
 309         else
 310                 *request = req;
 311         RETURN(rc);
 312 }
 313
 314 static int mdc_xattr_common(struct obd_export *exp,const struct req_format *fmt,
 315                             const struct lu_fid *fid,
 316                             struct obd_capa *oc, int opcode, obd_valid valid,
 317                             const char *xattr_name, const char *input,
 318                             int input_size, int output_size, int flags,
 319                             __u32 suppgid, struct ptlrpc_request **request)
 320 {
 321         struct ptlrpc_request *req;
 322         int   xattr_namelen = 0;
 323         char *tmp;
 324         int   rc;
 325         ENTRY;
 326
 327         *request = NULL;
 328         req = ptlrpc_request_alloc(class_exp2cliimp(exp), fmt);
 329         if (req == NULL)
 330                 RETURN(-ENOMEM);
 331
 332         mdc_set_capa_size(req, &RMF_CAPA1, oc);
 333         if (xattr_name) {
 334                 xattr_namelen = strlen(xattr_name) + 1;
 335                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 336                                      xattr_namelen);
 337         }
 338         if (input_size) {
 339                 LASSERT(input);
 340                 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
 341                                      input_size);
 342         }
 343
 344         /* Flush local XATTR locks to get rid of a possible cancel RPC */
 345         if (opcode == MDS_REINT && fid_is_sane(fid) &&
 346             exp->exp_connect_data.ocd_ibits_known & MDS_INODELOCK_XATTR) {
 347                 struct list_head cancels = LIST_HEAD_INIT(cancels);
 348                 int count;
 349
 350                 /* Without that packing would fail */
 351                 if (input_size == 0)
 352                         req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
 353                                              RCL_CLIENT, 0);
 354
 355                 count = mdc_resource_get_unused(exp, fid,
 356                                                 &cancels, LCK_EX,
 357                                                 MDS_INODELOCK_XATTR);
 358
 359                 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
 360                 if (rc) {
 361                         ptlrpc_request_free(req);
 362                         RETURN(rc);
 363                 }
 364         } else {
 365                 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode);
 366                 if (rc) {
 367                         ptlrpc_request_free(req);
 368                         RETURN(rc);
 369                 }
 370         }
 371
 372         if (opcode == MDS_REINT) {
 373                 struct mdt_rec_setxattr *rec;
 374
 375                 CLASSERT(sizeof(struct mdt_rec_setxattr) ==
 376                          sizeof(struct mdt_rec_reint));
 377                 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
 378                 rec->sx_opcode = REINT_SETXATTR;
 379                 rec->sx_fsuid  = from_kuid(&init_user_ns, current_fsuid());
 380                 rec->sx_fsgid  = from_kgid(&init_user_ns, current_fsgid());
 381                 rec->sx_cap    = cfs_curproc_cap_pack();
 382                 rec->sx_suppgid1 = suppgid;
 383                 rec->sx_suppgid2 = -1;
 384                 rec->sx_fid    = *fid;
 385                 rec->sx_valid  = valid | OBD_MD_FLCTIME;
 386                 rec->sx_time   = cfs_time_current_sec();
 387                 rec->sx_size   = output_size;
 388                 rec->sx_flags  = flags;
 389
 390                 mdc_pack_capa(req, &RMF_CAPA1, oc);
 391         } else {
 392                 mdc_pack_body(req, fid, oc, valid, output_size, suppgid, flags);
 393         }
 394
 395         if (xattr_name) {
 396                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 397                 memcpy(tmp, xattr_name, xattr_namelen);
 398         }
 399         if (input_size) {
 400                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
 401                 memcpy(tmp, input, input_size);
 402         }
 403
 404         if (req_capsule_has_field(&req->rq_pill, &RMF_EADATA, RCL_SERVER))
 405                 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
 406                                      RCL_SERVER, output_size);
 407         ptlrpc_request_set_replen(req);
 408
 409         /* make rpc */
 410         if (opcode == MDS_REINT)
 411                 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 412
 413         rc = ptlrpc_queue_wait(req);
 414
 415         if (opcode == MDS_REINT)
 416                 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 417
 418         if (rc)
 419                 ptlrpc_req_finished(req);
 420         else
 421                 *request = req;
 422         RETURN(rc);
 423 }
 424
 425 int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
 426                  struct obd_capa *oc, obd_valid valid, const char *xattr_name,
 427                  const char *input, int input_size, int output_size,
 428                  int flags, __u32 suppgid, struct ptlrpc_request **request)
 429 {
 430         return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
 431                                 fid, oc, MDS_REINT, valid, xattr_name,
 432                                 input, input_size, output_size, flags,
 433                                 suppgid, request);
 434 }
 435
 436 int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
 437                  struct obd_capa *oc, obd_valid valid, const char *xattr_name,
 438                  const char *input, int input_size, int output_size,
 439                  int flags, struct ptlrpc_request **request)
 440 {
 441         return mdc_xattr_common(exp, &RQF_MDS_GETXATTR,
 442                                 fid, oc, MDS_GETXATTR, valid, xattr_name,
 443                                 input, input_size, output_size, flags,
 444                                 -1, request);
 445 }
 446
 447 #ifdef CONFIG_FS_POSIX_ACL
 448 static int mdc_unpack_acl(struct ptlrpc_request *req, struct lustre_md *md)
 449 {
 450         struct req_capsule     *pill = &req->rq_pill;
 451         struct mdt_body        *body = md->body;
 452         struct posix_acl       *acl;
 453         void                   *buf;
 454         int                     rc;
 455         ENTRY;
 456
 457         if (!body->mbo_aclsize)
 458                 RETURN(0);
 459
 460         buf = req_capsule_server_sized_get(pill, &RMF_ACL, body->mbo_aclsize);
 461
 462         if (!buf)
 463                 RETURN(-EPROTO);
 464
 465         acl = posix_acl_from_xattr(&init_user_ns, buf, body->mbo_aclsize);
 466         if (IS_ERR(acl)) {
 467                 rc = PTR_ERR(acl);
 468                 CERROR("convert xattr to acl: %d\n", rc);
 469                 RETURN(rc);
 470         }
 471
 472         rc = posix_acl_valid(acl);
 473         if (rc) {
 474                 CERROR("validate acl: %d\n", rc);
 475                 posix_acl_release(acl);
 476                 RETURN(rc);
 477         }
 478
 479         md->posix_acl = acl;
 480         RETURN(0);
 481 }
 482 #else
 483 #define mdc_unpack_acl(req, md) 0
 484 #endif
 485
 486 int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
 487                       struct obd_export *dt_exp, struct obd_export *md_exp,
 488                       struct lustre_md *md)
 489 {
 490         struct req_capsule *pill = &req->rq_pill;
 491         int rc;
 492         ENTRY;
 493
 494         LASSERT(md);
 495         memset(md, 0, sizeof(*md));
 496
 497         md->body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 498         LASSERT(md->body != NULL);
 499
 500         if (md->body->mbo_valid & OBD_MD_FLEASIZE) {
 501                 int lmmsize;
 502                 struct lov_mds_md *lmm;
 503
 504                 if (!S_ISREG(md->body->mbo_mode)) {
 505                         CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, should be a "
 506                                "regular file, but is not\n");
 507                         GOTO(out, rc = -EPROTO);
 508                 }
 509
 510                 if (md->body->mbo_eadatasize == 0) {
 511                         CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, "
 512                                "but eadatasize 0\n");
 513                         GOTO(out, rc = -EPROTO);
 514                 }
 515
 516                 lmmsize = md->body->mbo_eadatasize;
 517                 lmm = req_capsule_server_sized_get(pill, &RMF_MDT_MD, lmmsize);
 518                 if (!lmm)
 519                         GOTO(out, rc = -EPROTO);
 520
 521                 rc = obd_unpackmd(dt_exp, &md->lsm, lmm, lmmsize);
 522                 if (rc < 0)
 523                         GOTO(out, rc);
 524
 525                 if (rc < sizeof(*md->lsm)) {
 526                         CDEBUG(D_INFO, "lsm size too small: "
 527                                "rc < sizeof (*md->lsm) (%d < %d)\n",
 528                                rc, (int)sizeof(*md->lsm));
 529                         GOTO(out, rc = -EPROTO);
 530                 }
 531
 532         } else if (md->body->mbo_valid & OBD_MD_FLDIREA) {
 533                 int lmvsize;
 534                 struct lov_mds_md *lmv;
 535
 536                 if (!S_ISDIR(md->body->mbo_mode)) {
 537                         CDEBUG(D_INFO, "OBD_MD_FLDIREA set, should be a "
 538                                "directory, but is not\n");
 539                         GOTO(out, rc = -EPROTO);
 540                 }
 541
 542                 if (md->body->mbo_eadatasize == 0) {
 543                         CDEBUG(D_INFO, "OBD_MD_FLDIREA is set, "
 544                                "but eadatasize 0\n");
 545                         RETURN(-EPROTO);
 546                 }
 547
 548                 if (md->body->mbo_valid & OBD_MD_MEA) {
 549                         lmvsize = md->body->mbo_eadatasize;
 550                         lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
 551                                                            lmvsize);
 552                         if (!lmv)
 553                                 GOTO(out, rc = -EPROTO);
 554
 555                         rc = obd_unpackmd(md_exp, (void *)&md->lmv, lmv,
 556                                           lmvsize);
 557                         if (rc < 0)
 558                                 GOTO(out, rc);
 559
 560                         if (rc < sizeof(*md->lmv)) {
 561                                 CDEBUG(D_INFO, "size too small:  "
 562                                        "rc < sizeof(*md->lmv) (%d < %d)\n",
 563                                         rc, (int)sizeof(*md->lmv));
 564                                 GOTO(out, rc = -EPROTO);
 565                         }
 566                 }
 567         }
 568         rc = 0;
 569
 570         if (md->body->mbo_valid & OBD_MD_FLRMTPERM) {
 571                 /* remote permission */
 572                 LASSERT(client_is_remote(exp));
 573                 md->remote_perm = req_capsule_server_swab_get(pill, &RMF_ACL,
 574                                                 lustre_swab_mdt_remote_perm);
 575                 if (!md->remote_perm)
 576                         GOTO(out, rc = -EPROTO);
 577         } else if (md->body->mbo_valid & OBD_MD_FLACL) {
 578                 /* for ACL, it's possible that FLACL is set but aclsize is zero.
 579                  * only when aclsize != 0 there's an actual segment for ACL
 580                  * in reply buffer.
 581                  */
 582                 if (md->body->mbo_aclsize) {
 583                         rc = mdc_unpack_acl(req, md);
 584                         if (rc)
 585                                 GOTO(out, rc);
 586 #ifdef CONFIG_FS_POSIX_ACL
 587                 } else {
 588                         md->posix_acl = NULL;
 589 #endif
 590                 }
 591         }
 592         if (md->body->mbo_valid & OBD_MD_FLMDSCAPA) {
 593                 struct obd_capa *oc = NULL;
 594
 595                 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA1, &oc);
 596                 if (rc)
 597                         GOTO(out, rc);
 598                 md->mds_capa = oc;
 599         }
 600
 601         if (md->body->mbo_valid & OBD_MD_FLOSSCAPA) {
 602                 struct obd_capa *oc = NULL;
 603
 604                 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA2, &oc);
 605                 if (rc)
 606                         GOTO(out, rc);
 607                 md->oss_capa = oc;
 608         }
 609
 610         EXIT;
 611 out:
 612         if (rc) {
 613                 if (md->oss_capa) {
 614                         capa_put(md->oss_capa);
 615                         md->oss_capa = NULL;
 616                 }
 617                 if (md->mds_capa) {
 618                         capa_put(md->mds_capa);
 619                         md->mds_capa = NULL;
 620                 }
 621 #ifdef CONFIG_FS_POSIX_ACL
 622                 posix_acl_release(md->posix_acl);
 623 #endif
 624                 if (md->lsm)
 625                         obd_free_memmd(dt_exp, &md->lsm);
 626         }
 627         return rc;
 628 }
 629
 630 int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
 631 {
 632         ENTRY;
 633         RETURN(0);
 634 }
 635
 636 /**
 637  * Handles both OPEN and SETATTR RPCs for OPEN-CLOSE and SETATTR-DONE_WRITING
 638  * RPC chains.
 639  */
 640 void mdc_replay_open(struct ptlrpc_request *req)
 641 {
 642         struct md_open_data *mod = req->rq_cb_data;
 643         struct ptlrpc_request *close_req;
 644         struct obd_client_handle *och;
 645         struct lustre_handle old;
 646         struct mdt_body *body;
 647         ENTRY;
 648
 649         if (mod == NULL) {
 650                 DEBUG_REQ(D_ERROR, req,
 651                           "Can't properly replay without open data.");
 652                 EXIT;
 653                 return;
 654         }
 655
 656         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 657         LASSERT(body != NULL);
 658
 659         och = mod->mod_och;
 660         if (och != NULL) {
 661                 struct lustre_handle *file_fh;
 662
 663                 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
 664
 665                 file_fh = &och->och_fh;
 666                 CDEBUG(D_HA, "updating handle from "LPX64" to "LPX64"\n",
 667                        file_fh->cookie, body->mbo_handle.cookie);
 668                 old = *file_fh;
 669                 *file_fh = body->mbo_handle;
 670         }
 671         close_req = mod->mod_close_req;
 672         if (close_req != NULL) {
 673                 __u32 opc = lustre_msg_get_opc(close_req->rq_reqmsg);
 674                 struct mdt_ioepoch *epoch;
 675
 676                 LASSERT(opc == MDS_CLOSE || opc == MDS_DONE_WRITING);
 677                 epoch = req_capsule_client_get(&close_req->rq_pill,
 678                                                &RMF_MDT_EPOCH);
 679                 LASSERT(epoch);
 680
 681                 if (och != NULL)
 682                         LASSERT(!memcmp(&old, &epoch->handle, sizeof(old)));
 683                 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
 684                 epoch->handle = body->mbo_handle;
 685         }
 686         EXIT;
 687 }
 688
 689 void mdc_commit_open(struct ptlrpc_request *req)
 690 {
 691         struct md_open_data *mod = req->rq_cb_data;
 692         if (mod == NULL)
 693                 return;
 694
 695         /**
 696          * No need to touch md_open_data::mod_och, it holds a reference on
 697          * \var mod and will zero references to each other, \var mod will be
 698          * freed after that when md_open_data::mod_och will put the reference.
 699          */
 700
 701         /**
 702          * Do not let open request to disappear as it still may be needed
 703          * for close rpc to happen (it may happen on evict only, otherwise
 704          * ptlrpc_request::rq_replay does not let mdc_commit_open() to be
 705          * called), just mark this rpc as committed to distinguish these 2
 706          * cases, see mdc_close() for details. The open request reference will
 707          * be put along with freeing \var mod.
 708          */
 709         ptlrpc_request_addref(req);
 710         spin_lock(&req->rq_lock);
 711         req->rq_committed = 1;
 712         spin_unlock(&req->rq_lock);
 713         req->rq_cb_data = NULL;
 714         obd_mod_put(mod);
 715 }
 716
 717 int mdc_set_open_replay_data(struct obd_export *exp,
 718                              struct obd_client_handle *och,
 719                              struct lookup_intent *it)
 720 {
 721         struct md_open_data     *mod;
 722         struct mdt_rec_create   *rec;
 723         struct mdt_body         *body;
 724         struct ptlrpc_request   *open_req = it->d.lustre.it_data;
 725         struct obd_import       *imp = open_req->rq_import;
 726         ENTRY;
 727
 728         if (!open_req->rq_replay)
 729                 RETURN(0);
 730
 731         rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
 732         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
 733         LASSERT(rec != NULL);
 734         /* Incoming message in my byte order (it's been swabbed). */
 735         /* Outgoing messages always in my byte order. */
 736         LASSERT(body != NULL);
 737
 738         /* Only if the import is replayable, we set replay_open data */
 739         if (och && imp->imp_replayable) {
 740                 mod = obd_mod_alloc();
 741                 if (mod == NULL) {
 742                         DEBUG_REQ(D_ERROR, open_req,
 743                                   "Can't allocate md_open_data");
 744                         RETURN(0);
 745                 }
 746
 747                 /**
 748                  * Take a reference on \var mod, to be freed on mdc_close().
 749                  * It protects \var mod from being freed on eviction (commit
 750                  * callback is called despite rq_replay flag).
 751                  * Another reference for \var och.
 752                  */
 753                 obd_mod_get(mod);
 754                 obd_mod_get(mod);
 755
 756                 spin_lock(&open_req->rq_lock);
 757                 och->och_mod = mod;
 758                 mod->mod_och = och;
 759                 mod->mod_is_create = it_disposition(it, DISP_OPEN_CREATE) ||
 760                                      it_disposition(it, DISP_OPEN_STRIPE);
 761                 mod->mod_open_req = open_req;
 762                 open_req->rq_cb_data = mod;
 763                 open_req->rq_commit_cb = mdc_commit_open;
 764                 spin_unlock(&open_req->rq_lock);
 765         }
 766
 767         rec->cr_fid2 = body->mbo_fid1;
 768         rec->cr_ioepoch = body->mbo_ioepoch;
 769         rec->cr_old_handle.cookie = body->mbo_handle.cookie;
 770         open_req->rq_replay_cb = mdc_replay_open;
 771         if (!fid_is_sane(&body->mbo_fid1)) {
 772                 DEBUG_REQ(D_ERROR, open_req, "Saving replay request with "
 773                           "insane fid");
 774                 LBUG();
 775         }
 776
 777         DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
 778         RETURN(0);
 779 }
 780
 781 static void mdc_free_open(struct md_open_data *mod)
 782 {
 783         int committed = 0;
 784
 785         if (mod->mod_is_create == 0 &&
 786             imp_connect_disp_stripe(mod->mod_open_req->rq_import))
 787                 committed = 1;
 788
 789         LASSERT(mod->mod_open_req->rq_replay == 0);
 790
 791         DEBUG_REQ(D_RPCTRACE, mod->mod_open_req, "free open request\n");
 792
 793         ptlrpc_request_committed(mod->mod_open_req, committed);
 794         if (mod->mod_close_req)
 795                 ptlrpc_request_committed(mod->mod_close_req, committed);
 796 }
 797
 798 int mdc_clear_open_replay_data(struct obd_export *exp,
 799                                struct obd_client_handle *och)
 800 {
 801         struct md_open_data *mod = och->och_mod;
 802         ENTRY;
 803
 804         /**
 805          * It is possible to not have \var mod in a case of eviction between
 806          * lookup and ll_file_open().
 807          **/
 808         if (mod == NULL)
 809                 RETURN(0);
 810
 811         LASSERT(mod != LP_POISON);
 812         LASSERT(mod->mod_open_req != NULL);
 813         mdc_free_open(mod);
 814
 815         mod->mod_och = NULL;
 816         och->och_mod = NULL;
 817         obd_mod_put(mod);
 818
 819         RETURN(0);
 820 }
 821
 822 /* Prepares the request for the replay by the given reply */
 823 static void mdc_close_handle_reply(struct ptlrpc_request *req,
 824                                    struct md_op_data *op_data, int rc) {
 825         struct mdt_body  *repbody;
 826         struct mdt_ioepoch *epoch;
 827
 828         if (req && rc == -EAGAIN) {
 829                 repbody = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 830                 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
 831
 832                 epoch->flags |= MF_SOM_AU;
 833                 if (repbody->mbo_valid & OBD_MD_FLGETATTRLOCK)
 834                         op_data->op_flags |= MF_GETATTR_LOCK;
 835         }
 836 }
 837
 838 int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
 839               struct md_open_data *mod, struct ptlrpc_request **request)
 840 {
 841         struct obd_device     *obd = class_exp2obd(exp);
 842         struct ptlrpc_request *req;
 843         struct req_format     *req_fmt;
 844         int                    rc;
 845         int                    saved_rc = 0;
 846         ENTRY;
 847
 848         req_fmt = &RQF_MDS_CLOSE;
 849         if (op_data->op_bias & MDS_HSM_RELEASE) {
 850                 req_fmt = &RQF_MDS_RELEASE_CLOSE;
 851
 852                 /* allocate a FID for volatile file */
 853                 rc = mdc_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
 854                 if (rc < 0) {
 855                         CERROR("%s: "DFID" failed to allocate FID: %d\n",
 856                                obd->obd_name, PFID(&op_data->op_fid1), rc);
 857                         /* save the errcode and proceed to close */
 858                         saved_rc = rc;
 859                 }
 860         }
 861
 862         *request = NULL;
 863         req = ptlrpc_request_alloc(class_exp2cliimp(exp), req_fmt);
 864         if (req == NULL)
 865                 RETURN(-ENOMEM);
 866
 867         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 868
 869         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
 870         if (rc) {
 871                 ptlrpc_request_free(req);
 872                 RETURN(rc);
 873         }
 874
 875         /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
 876          * portal whose threads are not taking any DLM locks and are therefore
 877          * always progressing */
 878         req->rq_request_portal = MDS_READPAGE_PORTAL;
 879         ptlrpc_at_set_req_timeout(req);
 880
 881         /* Ensure that this close's handle is fixed up during replay. */
 882         if (likely(mod != NULL)) {
 883                 LASSERTF(mod->mod_open_req != NULL &&
 884                          mod->mod_open_req->rq_type != LI_POISON,
 885                          "POISONED open %p!\n", mod->mod_open_req);
 886
 887                 mod->mod_close_req = req;
 888
 889                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
 890                 /* We no longer want to preserve this open for replay even
 891                  * though the open was committed. b=3632, b=3633 */
 892                 spin_lock(&mod->mod_open_req->rq_lock);
 893                 mod->mod_open_req->rq_replay = 0;
 894                 spin_unlock(&mod->mod_open_req->rq_lock);
 895         } else {
 896                  CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
 897         }
 898
 899         mdc_close_pack(req, op_data);
 900
 901         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 902                              obd->u.cli.cl_default_mds_easize);
 903         req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
 904                              obd->u.cli.cl_default_mds_cookiesize);
 905
 906         ptlrpc_request_set_replen(req);
 907
 908         mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 909         rc = ptlrpc_queue_wait(req);
 910         mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 911
 912         if (req->rq_repmsg == NULL) {
 913                 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
 914                        req->rq_status);
 915                 if (rc == 0)
 916                         rc = req->rq_status ?: -EIO;
 917         } else if (rc == 0 || rc == -EAGAIN) {
 918                 struct mdt_body *body;
 919
 920                 rc = lustre_msg_get_status(req->rq_repmsg);
 921                 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
 922                         DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
 923                                   "= %d", rc);
 924                         if (rc > 0)
 925                                 rc = -rc;
 926                 }
 927                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 928                 if (body == NULL)
 929                         rc = -EPROTO;
 930         } else if (rc == -ESTALE) {
 931                 /**
 932                  * it can be allowed error after 3633 if open was committed and
 933                  * server failed before close was sent. Let's check if mod
 934                  * exists and return no error in that case
 935                  */
 936                 if (mod) {
 937                         DEBUG_REQ(D_HA, req, "Reset ESTALE = %d", rc);
 938                         LASSERT(mod->mod_open_req != NULL);
 939                         if (mod->mod_open_req->rq_committed)
 940                                 rc = 0;
 941                 }
 942         }
 943
 944         if (mod) {
 945                 if (rc != 0)
 946                         mod->mod_close_req = NULL;
 947                 /* Since now, mod is accessed through open_req only,
 948                  * thus close req does not keep a reference on mod anymore. */
 949                 obd_mod_put(mod);
 950         }
 951         *request = req;
 952         mdc_close_handle_reply(req, op_data, rc);
 953         RETURN(rc < 0 ? rc : saved_rc);
 954 }
 955
 956 int mdc_done_writing(struct obd_export *exp, struct md_op_data *op_data,
 957                      struct md_open_data *mod)
 958 {
 959         struct obd_device     *obd = class_exp2obd(exp);
 960         struct ptlrpc_request *req;
 961         int                    rc;
 962         ENTRY;
 963
 964         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 965                                    &RQF_MDS_DONE_WRITING);
 966         if (req == NULL)
 967                 RETURN(-ENOMEM);
 968
 969         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 970         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_DONE_WRITING);
 971         if (rc) {
 972                 ptlrpc_request_free(req);
 973                 RETURN(rc);
 974         }
 975
 976         if (mod != NULL) {
 977                 LASSERTF(mod->mod_open_req != NULL &&
 978                          mod->mod_open_req->rq_type != LI_POISON,
 979                          "POISONED setattr %p!\n", mod->mod_open_req);
 980
 981                 mod->mod_close_req = req;
 982                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched setattr");
 983                 /* We no longer want to preserve this setattr for replay even
 984                  * though the open was committed. b=3632, b=3633 */
 985                 spin_lock(&mod->mod_open_req->rq_lock);
 986                 mod->mod_open_req->rq_replay = 0;
 987                 spin_unlock(&mod->mod_open_req->rq_lock);
 988         }
 989
 990         mdc_close_pack(req, op_data);
 991         ptlrpc_request_set_replen(req);
 992
 993         mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 994         rc = ptlrpc_queue_wait(req);
 995         mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 996
 997         if (rc == -ESTALE) {
 998                 /**
 999                  * it can be allowed error after 3633 if open or setattr were
1000                  * committed and server failed before close was sent.
1001                  * Let's check if mod exists and return no error in that case
1002                  */
1003                 if (mod) {
1004                         LASSERT(mod->mod_open_req != NULL);
1005                         if (mod->mod_open_req->rq_committed)
1006                                 rc = 0;
1007                 }
1008         }
1009
1010         if (mod) {
1011                 if (rc != 0)
1012                         mod->mod_close_req = NULL;
1013                 LASSERT(mod->mod_open_req != NULL);
1014                 mdc_free_open(mod);
1015
1016                 /* Since now, mod is accessed through setattr req only,
1017                  * thus DW req does not keep a reference on mod anymore. */
1018                 obd_mod_put(mod);
1019         }
1020
1021         mdc_close_handle_reply(req, op_data, rc);
1022         ptlrpc_req_finished(req);
1023         RETURN(rc);
1024 }
1025
1026 #ifdef HAVE_SPLIT_SUPPORT
1027 int mdc_sendpage(struct obd_export *exp, const struct lu_fid *fid,
1028                  const struct page *page, int offset)
1029 {
1030         struct ptlrpc_request   *req;
1031         struct ptlrpc_bulk_desc *desc;
1032         int                      rc;
1033         ENTRY;
1034
1035         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_WRITEPAGE);
1036         if (req == NULL)
1037                 RETURN(-ENOMEM);
1038
1039         /* FIXME: capa doesn't support split yet */
1040         mdc_set_capa_size(req, &RMF_CAPA1, NULL);
1041
1042         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_WRITEPAGE);
1043         if (rc) {
1044                 ptlrpc_request_free(req);
1045                 RETURN(rc);
1046         }
1047
1048         req->rq_request_portal = MDS_READPAGE_PORTAL;
1049         ptlrpc_at_set_req_timeout(req);
1050
1051         desc = ptlrpc_prep_bulk_imp(req, 1, 1,BULK_GET_SOURCE, MDS_BULK_PORTAL);
1052         if (desc == NULL)
1053                 GOTO(out, rc = -ENOMEM);
1054
1055         /* NB req now owns desc and will free it when it gets freed. */
1056         ptlrpc_prep_bulk_page(desc, (struct page *)page, 0, offset);
1057         mdc_readdir_pack(req, 0, offset, fid, NULL);
1058
1059         ptlrpc_request_set_replen(req);
1060         rc = ptlrpc_queue_wait(req);
1061         if (rc)
1062                 GOTO(out, rc);
1063
1064         rc = sptlrpc_cli_unwrap_bulk_write(req, req->rq_bulk);
1065 out:
1066         ptlrpc_req_finished(req);
1067         return rc;
1068 }
1069 EXPORT_SYMBOL(mdc_sendpage);
1070 #endif
1071
1072 static int mdc_getpage(struct obd_export *exp, const struct lu_fid *fid,
1073                        __u64 offset, struct obd_capa *oc,
1074                        struct page **pages, int npages,
1075                        struct ptlrpc_request **request)
1076 {
1077         struct ptlrpc_request   *req;
1078         struct ptlrpc_bulk_desc *desc;
1079         int                      i;
1080         wait_queue_head_t        waitq;
1081         int                      resends = 0;
1082         struct l_wait_info       lwi;
1083         int                      rc;
1084         ENTRY;
1085
1086         *request = NULL;
1087         init_waitqueue_head(&waitq);
1088
1089 restart_bulk:
1090         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
1091         if (req == NULL)
1092                 RETURN(-ENOMEM);
1093
1094         mdc_set_capa_size(req, &RMF_CAPA1, oc);
1095
1096         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
1097         if (rc) {
1098                 ptlrpc_request_free(req);
1099                 RETURN(rc);
1100         }
1101
1102         req->rq_request_portal = MDS_READPAGE_PORTAL;
1103         ptlrpc_at_set_req_timeout(req);
1104
1105         desc = ptlrpc_prep_bulk_imp(req, npages, 1, BULK_PUT_SINK,
1106                                     MDS_BULK_PORTAL);
1107         if (desc == NULL) {
1108                 ptlrpc_request_free(req);
1109                 RETURN(-ENOMEM);
1110         }
1111
1112         /* NB req now owns desc and will free it when it gets freed */
1113         for (i = 0; i < npages; i++)
1114                 ptlrpc_prep_bulk_page_pin(desc, pages[i], 0, PAGE_CACHE_SIZE);
1115
1116         mdc_readdir_pack(req, offset, PAGE_CACHE_SIZE * npages, fid, oc);
1117
1118         ptlrpc_request_set_replen(req);
1119         rc = ptlrpc_queue_wait(req);
1120         if (rc) {
1121                 ptlrpc_req_finished(req);
1122                 if (rc != -ETIMEDOUT)
1123                         RETURN(rc);
1124
1125                 resends++;
1126                 if (!client_should_resend(resends, &exp->exp_obd->u.cli)) {
1127                         CERROR("%s: too many resend retries: rc = %d\n",
1128                                exp->exp_obd->obd_name, -EIO);
1129                         RETURN(-EIO);
1130                 }
1131                 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(resends), NULL, NULL,
1132                                        NULL);
1133                 l_wait_event(waitq, 0, &lwi);
1134
1135                 goto restart_bulk;
1136         }
1137
1138         rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
1139                                           req->rq_bulk->bd_nob_transferred);
1140         if (rc < 0) {
1141                 ptlrpc_req_finished(req);
1142                 RETURN(rc);
1143         }
1144
1145         if (req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK) {
1146                 CERROR("%s: unexpected bytes transferred: %d (%ld expected)\n",
1147                        exp->exp_obd->obd_name, req->rq_bulk->bd_nob_transferred,
1148                        PAGE_CACHE_SIZE * npages);
1149                 ptlrpc_req_finished(req);
1150                 RETURN(-EPROTO);
1151         }
1152
1153         *request = req;
1154         RETURN(0);
1155 }
1156
1157 #ifdef __KERNEL__
1158 static void mdc_release_page(struct page *page, int remove)
1159 {
1160         if (remove) {
1161                 lock_page(page);
1162                 if (likely(page->mapping != NULL))
1163                         truncate_complete_page(page->mapping, page);
1164                 unlock_page(page);
1165         }
1166         page_cache_release(page);
1167 }
1168
1169 static struct page *mdc_page_locate(struct address_space *mapping, __u64 *hash,
1170                                     __u64 *start, __u64 *end, int hash64)
1171 {
1172         /*
1173          * Complement of hash is used as an index so that
1174          * radix_tree_gang_lookup() can be used to find a page with starting
1175          * hash _smaller_ than one we are looking for.
1176          */
1177         unsigned long offset = hash_x_index(*hash, hash64);
1178         struct page *page;
1179         int found;
1180
1181         spin_lock_irq(&mapping->tree_lock);
1182         found = radix_tree_gang_lookup(&mapping->page_tree,
1183                                        (void **)&page, offset, 1);
1184         if (found > 0 && !radix_tree_exceptional_entry(page)) {
1185                 struct lu_dirpage *dp;
1186
1187                 page_cache_get(page);
1188                 spin_unlock_irq(&mapping->tree_lock);
1189                 /*
1190                  * In contrast to find_lock_page() we are sure that directory
1191                  * page cannot be truncated (while DLM lock is held) and,
1192                  * hence, can avoid restart.
1193                  *
1194                  * In fact, page cannot be locked here at all, because
1195                  * mdc_read_page_remote does synchronous io.
1196                  */
1197                 wait_on_page_locked(page);
1198                 if (PageUptodate(page)) {
1199                         dp = kmap(page);
1200                         if (BITS_PER_LONG == 32 && hash64) {
1201                                 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1202                                 *end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
1203                                 *hash  = *hash >> 32;
1204                         } else {
1205                                 *start = le64_to_cpu(dp->ldp_hash_start);
1206                                 *end   = le64_to_cpu(dp->ldp_hash_end);
1207                         }
1208                         if (unlikely(*start == 1 && *hash == 0))
1209                                 *hash = *start;
1210                         else
1211                                 LASSERTF(*start <= *hash, "start = "LPX64
1212                                          ",end = "LPX64",hash = "LPX64"\n",
1213                                          *start, *end, *hash);
1214                         CDEBUG(D_VFSTRACE, "offset %lx ["LPX64" "LPX64"],"
1215                               " hash "LPX64"\n", offset, *start, *end, *hash);
1216                         if (*hash > *end) {
1217                                 kunmap(page);
1218                                 mdc_release_page(page, 0);
1219                                 page = NULL;
1220                         } else if (*end != *start && *hash == *end) {
1221                                 /*
1222                                  * upon hash collision, remove this page,
1223                                  * otherwise put page reference, and
1224                                  * mdc_read_page_remote() will issue RPC to
1225                                  * fetch the page we want.
1226                                  */
1227                                 kunmap(page);
1228                                 mdc_release_page(page,
1229                                     le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
1230                                 page = NULL;
1231                         }
1232                 } else {
1233                         page_cache_release(page);
1234                         page = ERR_PTR(-EIO);
1235                 }
1236         } else {
1237                 spin_unlock_irq(&mapping->tree_lock);
1238                 page = NULL;
1239         }
1240         return page;
1241 }
1242
1243 /*
1244  * Adjust a set of pages, each page containing an array of lu_dirpages,
1245  * so that each page can be used as a single logical lu_dirpage.
1246  *
1247  * A lu_dirpage is laid out as follows, where s = ldp_hash_start,
1248  * e = ldp_hash_end, f = ldp_flags, p = padding, and each "ent" is a
1249  * struct lu_dirent.  It has size up to LU_PAGE_SIZE. The ldp_hash_end
1250  * value is used as a cookie to request the next lu_dirpage in a
1251  * directory listing that spans multiple pages (two in this example):
1252  *   ________
1253  *  |        |
1254  * .|--------v-------   -----.
1255  * |s|e|f|p|ent|ent| ... |ent|
1256  * '--|--------------   -----'   Each CFS_PAGE contains a single
1257  *    '------.                   lu_dirpage.
1258  * .---------v-------   -----.
1259  * |s|e|f|p|ent| 0 | ... | 0 |
1260  * '-----------------   -----'
1261  *
1262  * However, on hosts where the native VM page size (PAGE_CACHE_SIZE) is
1263  * larger than LU_PAGE_SIZE, a single host page may contain multiple
1264  * lu_dirpages. After reading the lu_dirpages from the MDS, the
1265  * ldp_hash_end of the first lu_dirpage refers to the one immediately
1266  * after it in the same CFS_PAGE (arrows simplified for brevity, but
1267  * in general e0==s1, e1==s2, etc.):
1268  *
1269  * .--------------------   -----.
1270  * |s0|e0|f0|p|ent|ent| ... |ent|
1271  * |---v----------------   -----|
1272  * |s1|e1|f1|p|ent|ent| ... |ent|
1273  * |---v----------------   -----|  Here, each CFS_PAGE contains
1274  *             ...                 multiple lu_dirpages.
1275  * |---v----------------   -----|
1276  * |s'|e'|f'|p|ent|ent| ... |ent|
1277  * '---|----------------   -----'
1278  *     v
1279  * .----------------------------.
1280  * |        next CFS_PAGE       |
1281  *
1282  * This structure is transformed into a single logical lu_dirpage as follows:
1283  *
1284  * - Replace e0 with e' so the request for the next lu_dirpage gets the page
1285  *   labeled 'next CFS_PAGE'.
1286  *
1287  * - Copy the LDF_COLLIDE flag from f' to f0 to correctly reflect whether
1288  *   a hash collision with the next page exists.
1289  *
1290  * - Adjust the lde_reclen of the ending entry of each lu_dirpage to span
1291  *   to the first entry of the next lu_dirpage.
1292  */
1293 #if PAGE_CACHE_SIZE > LU_PAGE_SIZE
1294 static void mdc_adjust_dirpages(struct page **pages, int cfs_pgs, int lu_pgs)
1295 {
1296         int i;
1297
1298         for (i = 0; i < cfs_pgs; i++) {
1299                 struct lu_dirpage       *dp = kmap(pages[i]);
1300                 struct lu_dirpage       *first = dp;
1301                 struct lu_dirent        *end_dirent = NULL;
1302                 struct lu_dirent        *ent;
1303                 __u64           hash_end = le64_to_cpu(dp->ldp_hash_end);
1304                 __u32           flags = le32_to_cpu(dp->ldp_flags);
1305
1306                 while (--lu_pgs > 0) {
1307                         ent = lu_dirent_start(dp);
1308                         for (end_dirent = ent; ent != NULL;
1309                              end_dirent = ent, ent = lu_dirent_next(ent));
1310
1311                         /* Advance dp to next lu_dirpage. */
1312                         dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE);
1313
1314                         /* Check if we've reached the end of the CFS_PAGE. */
1315                         if (!((unsigned long)dp & ~CFS_PAGE_MASK))
1316                                 break;
1317
1318                         /* Save the hash and flags of this lu_dirpage. */
1319                         hash_end = le64_to_cpu(dp->ldp_hash_end);
1320                         flags = le32_to_cpu(dp->ldp_flags);
1321
1322                         /* Check if lu_dirpage contains no entries. */
1323                         if (end_dirent == NULL)
1324                                 break;
1325
1326                         /* Enlarge the end entry lde_reclen from 0 to
1327                          * first entry of next lu_dirpage. */
1328                         LASSERT(le16_to_cpu(end_dirent->lde_reclen) == 0);
1329                         end_dirent->lde_reclen =
1330                                 cpu_to_le16((char *)(dp->ldp_entries) -
1331                                             (char *)end_dirent);
1332                 }
1333
1334                 first->ldp_hash_end = hash_end;
1335                 first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE);
1336                 first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE);
1337
1338                 kunmap(pages[i]);
1339         }
1340         LASSERTF(lu_pgs == 0, "left = %d", lu_pgs);
1341 }
1342 #else
1343 #define mdc_adjust_dirpages(pages, cfs_pgs, lu_pgs) do {} while (0)
1344 #endif  /* PAGE_CACHE_SIZE > LU_PAGE_SIZE */
1345
1346 /* parameters for readdir page */
1347 struct readpage_param {
1348         struct md_op_data       *rp_mod;
1349         __u64                   rp_off;
1350         int                     rp_hash64;
1351         struct obd_export       *rp_exp;
1352         struct md_callback      *rp_cb;
1353 };
1354
1355 /**
1356  * Read pages from server.
1357  *
1358  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
1359  * a header lu_dirpage which describes the start/end hash, and whether this
1360  * page is empty (contains no dir entry) or hash collide with next page.
1361  * After client receives reply, several pages will be integrated into dir page
1362  * in CFS_PAGE_SIZE (if CFS_PAGE_SIZE greater than LU_PAGE_SIZE), and the
1363  * lu_dirpage for this integrated page will be adjusted.
1364  **/
1365 static int mdc_read_page_remote(void *data, struct page *page0)
1366 {
1367         struct readpage_param   *rp = data;
1368         struct page             **page_pool;
1369         struct page             *page;
1370         struct lu_dirpage       *dp;
1371         int                     rd_pgs = 0; /* number of pages read actually */
1372         int                     npages;
1373         struct md_op_data       *op_data = rp->rp_mod;
1374         struct ptlrpc_request   *req;
1375         int                     max_pages = op_data->op_max_pages;
1376         struct inode            *inode;
1377         struct lu_fid           *fid;
1378         int                     i;
1379         int                     rc;
1380         ENTRY;
1381
1382         LASSERT(max_pages > 0 && max_pages <= PTLRPC_MAX_BRW_PAGES);
1383         inode = op_data->op_data;
1384         fid = &op_data->op_fid1;
1385         LASSERT(inode != NULL);
1386
1387         OBD_ALLOC(page_pool, sizeof(page_pool[0]) * max_pages);
1388         if (page_pool != NULL) {
1389                 page_pool[0] = page0;
1390         } else {
1391                 page_pool = &page0;
1392                 max_pages = 1;
1393         }
1394
1395         for (npages = 1; npages < max_pages; npages++) {
1396                 page = page_cache_alloc_cold(inode->i_mapping);
1397                 if (page == NULL)
1398                         break;
1399                 page_pool[npages] = page;
1400         }
1401
1402         rc = mdc_getpage(rp->rp_exp, fid, rp->rp_off, op_data->op_capa1,
1403                          page_pool, npages, &req);
1404         if (rc == 0) {
1405                 int lu_pgs;
1406
1407                 rd_pgs = (req->rq_bulk->bd_nob_transferred +
1408                             PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1409                 lu_pgs = req->rq_bulk->bd_nob_transferred >>
1410                                                         LU_PAGE_SHIFT;
1411                 LASSERT(!(req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK));
1412
1413                 CDEBUG(D_INODE, "read %d(%d)/%d pages\n", rd_pgs, lu_pgs,
1414                        op_data->op_npages);
1415
1416                 mdc_adjust_dirpages(page_pool, rd_pgs, lu_pgs);
1417
1418                 SetPageUptodate(page0);
1419         }
1420
1421         unlock_page(page0);
1422         ptlrpc_req_finished(req);
1423         CDEBUG(D_CACHE, "read %d/%d pages\n", rd_pgs, npages);
1424         for (i = 1; i < npages; i++) {
1425                 unsigned long   offset;
1426                 __u64           hash;
1427                 int ret;
1428
1429                 page = page_pool[i];
1430
1431                 if (rc < 0 || i >= rd_pgs) {
1432                         page_cache_release(page);
1433                         continue;
1434                 }
1435
1436                 SetPageUptodate(page);
1437
1438                 dp = kmap(page);
1439                 hash = le64_to_cpu(dp->ldp_hash_start);
1440                 kunmap(page);
1441
1442                 offset = hash_x_index(hash, rp->rp_hash64);
1443
1444                 prefetchw(&page->flags);
1445                 ret = add_to_page_cache_lru(page, inode->i_mapping, offset,
1446                                             GFP_KERNEL);
1447                 if (ret == 0)
1448                         unlock_page(page);
1449                 else
1450                         CDEBUG(D_VFSTRACE, "page %lu add to page cache failed:"
1451                                " rc = %d\n", offset, ret);
1452                 page_cache_release(page);
1453         }
1454
1455         if (page_pool != &page0)
1456                 OBD_FREE(page_pool, sizeof(page_pool[0]) * max_pages);
1457
1458         RETURN(rc);
1459 }
1460
1461 /**
1462  * Read dir page from cache first, if it can not find it, read it from
1463  * server and add into the cache.
1464  *
1465  * \param[in] exp       MDC export
1466  * \param[in] op_data   client MD stack parameters, transfering parameters
1467  *                      between different layers on client MD stack.
1468  * \param[in] cb_op     callback required for ldlm lock enqueue during
1469  *                      read page
1470  * \param[in] hash_offset the hash offset of the page to be read
1471  * \param[in] ppage     the page to be read
1472  *
1473  * retval               = 0 get the page successfully
1474  *                      errno(<0) get the page failed
1475  */
1476 static int mdc_read_page(struct obd_export *exp, struct md_op_data *op_data,
1477                          struct md_callback *cb_op, __u64 hash_offset,
1478                          struct page **ppage)
1479 {
1480         struct lookup_intent    it = { .it_op = IT_READDIR };
1481         struct page             *page;
1482         struct inode            *dir = op_data->op_data;
1483         struct address_space    *mapping;
1484         struct lu_dirpage       *dp;
1485         __u64                   start = 0;
1486         __u64                   end = 0;
1487         struct lustre_handle    lockh;
1488         struct ptlrpc_request   *enq_req = NULL;
1489         struct readpage_param   rp_param;
1490         int rc;
1491
1492         ENTRY;
1493
1494         *ppage = NULL;
1495
1496         LASSERT(dir != NULL);
1497         mapping = dir->i_mapping;
1498
1499         rc = mdc_intent_lock(exp, op_data, &it, &enq_req,
1500                              cb_op->md_blocking_ast, 0);
1501         if (enq_req != NULL)
1502                 ptlrpc_req_finished(enq_req);
1503
1504         if (rc < 0) {
1505                 CERROR("%s: "DFID" lock enqueue fails: rc = %d\n",
1506                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1), rc);
1507                 RETURN(rc);
1508         }
1509
1510         rc = 0;
1511         mdc_set_lock_data(exp, &it.d.lustre.it_lock_handle, dir, NULL);
1512
1513         rp_param.rp_off = hash_offset;
1514         rp_param.rp_hash64 = op_data->op_cli_flags & CLI_HASH64;
1515         page = mdc_page_locate(mapping, &rp_param.rp_off, &start, &end,
1516                                rp_param.rp_hash64);
1517         if (IS_ERR(page)) {
1518                 CERROR("%s: dir page locate: "DFID" at "LPU64": rc %ld\n",
1519                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1520                        rp_param.rp_off, PTR_ERR(page));
1521                 GOTO(out_unlock, rc = PTR_ERR(page));
1522         } else if (page != NULL) {
1523                 /*
1524                  * XXX nikita: not entirely correct handling of a corner case:
1525                  * suppose hash chain of entries with hash value HASH crosses
1526                  * border between pages P0 and P1. First both P0 and P1 are
1527                  * cached, seekdir() is called for some entry from the P0 part
1528                  * of the chain. Later P0 goes out of cache. telldir(HASH)
1529                  * happens and finds P1, as it starts with matching hash
1530                  * value. Remaining entries from P0 part of the chain are
1531                  * skipped. (Is that really a bug?)
1532                  *
1533                  * Possible solutions: 0. don't cache P1 is such case, handle
1534                  * it as an "overflow" page. 1. invalidate all pages at
1535                  * once. 2. use HASH|1 as an index for P1.
1536                  */
1537                 GOTO(hash_collision, page);
1538         }
1539
1540         rp_param.rp_exp = exp;
1541         rp_param.rp_mod = op_data;
1542         page = read_cache_page(mapping,
1543                                hash_x_index(rp_param.rp_off,
1544                                             rp_param.rp_hash64),
1545                                mdc_read_page_remote, &rp_param);
1546         if (IS_ERR(page)) {
1547                 CERROR("%s: read cache page: "DFID" at "LPU64": rc %ld\n",
1548                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1549                        rp_param.rp_off, PTR_ERR(page));
1550                 GOTO(out_unlock, rc = PTR_ERR(page));
1551         }
1552
1553         wait_on_page_locked(page);
1554         (void)kmap(page);
1555         if (!PageUptodate(page)) {
1556                 CERROR("%s: page not updated: "DFID" at "LPU64": rc %d\n",
1557                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1558                        rp_param.rp_off, -5);
1559                 goto fail;
1560         }
1561         if (!PageChecked(page))
1562                 SetPageChecked(page);
1563         if (PageError(page)) {
1564                 CERROR("%s: page error: "DFID" at "LPU64": rc %d\n",
1565                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1566                        rp_param.rp_off, -5);
1567                 goto fail;
1568         }
1569
1570 hash_collision:
1571         dp = page_address(page);
1572         if (BITS_PER_LONG == 32 && rp_param.rp_hash64) {
1573                 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1574                 end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
1575                 rp_param.rp_off = hash_offset >> 32;
1576         } else {
1577                 start = le64_to_cpu(dp->ldp_hash_start);
1578                 end   = le64_to_cpu(dp->ldp_hash_end);
1579                 rp_param.rp_off = hash_offset;
1580         }
1581         if (end == start) {
1582                 LASSERT(start == rp_param.rp_off);
1583                 CWARN("Page-wide hash collision: %#lx\n", (unsigned long)end);
1584 #if BITS_PER_LONG == 32
1585                 CWARN("Real page-wide hash collision at ["LPU64" "LPU64"] with "
1586                       "hash "LPU64"\n", le64_to_cpu(dp->ldp_hash_start),
1587                       le64_to_cpu(dp->ldp_hash_end), hash_offset);
1588 #endif
1589
1590                 /*
1591                  * Fetch whole overflow chain...
1592                  *
1593                  * XXX not yet.
1594                  */
1595                 goto fail;
1596         }
1597         *ppage = page;
1598 out_unlock:
1599         lockh.cookie = it.d.lustre.it_lock_handle;
1600         ldlm_lock_decref(&lockh, it.d.lustre.it_lock_mode);
1601         it.d.lustre.it_lock_handle = 0;
1602         return rc;
1603 fail:
1604         kunmap(page);
1605         mdc_release_page(page, 1);
1606         rc = -EIO;
1607         goto out_unlock;
1608 }
1609
1610 #else /* __KERNEL__ */
1611
1612 static struct page
1613 *mdc_read_page_remote(struct obd_export *exp, const struct lmv_oinfo *lmo,
1614                       const __u64 hash, struct obd_capa *oc)
1615 {
1616         struct ptlrpc_request *req = NULL;
1617         struct page *page;
1618         int rc;
1619
1620         OBD_PAGE_ALLOC(page, 0);
1621         if (page == NULL)
1622                 return ERR_PTR(-ENOMEM);
1623
1624         rc = mdc_getpage(exp, &lmo->lmo_fid, hash, oc, &page, 1, &req);
1625         if (req != NULL)
1626                 ptlrpc_req_finished(req);
1627
1628         if (unlikely(rc)) {
1629                 OBD_PAGE_FREE(page);
1630                 return ERR_PTR(rc);
1631         }
1632         return page;
1633 }
1634
1635
1636 static int mdc_read_page(struct obd_export *exp, struct md_op_data *op_data,
1637                         struct md_callback *cb_op, __u64 hash_offset,
1638                         struct page **ppage)
1639 {
1640         struct page *page;
1641         struct lmv_oinfo *lmo;
1642         int rc = 0;
1643
1644         /* No local cache for liblustre, always read entry remotely */
1645         lmo = &op_data->op_mea1->lsm_md_oinfo[op_data->op_stripe_offset];
1646         page = mdc_read_page_remote(exp, lmo, hash_offset,
1647                                     op_data->op_capa1);
1648         if (IS_ERR(page))
1649                 return PTR_ERR(page);
1650
1651         *ppage = page;
1652
1653         return rc;
1654 }
1655
1656 #endif
1657
1658 static int mdc_statfs(const struct lu_env *env,
1659                       struct obd_export *exp, struct obd_statfs *osfs,
1660                       __u64 max_age, __u32 flags)
1661 {
1662         struct obd_device     *obd = class_exp2obd(exp);
1663         struct ptlrpc_request *req;
1664         struct obd_statfs     *msfs;
1665         struct obd_import     *imp = NULL;
1666         int                    rc;
1667         ENTRY;
1668
1669         /*
1670          * Since the request might also come from lprocfs, so we need
1671          * sync this with client_disconnect_export Bug15684
1672          */
1673         down_read(&obd->u.cli.cl_sem);
1674         if (obd->u.cli.cl_import)
1675                 imp = class_import_get(obd->u.cli.cl_import);
1676         up_read(&obd->u.cli.cl_sem);
1677         if (!imp)
1678                 RETURN(-ENODEV);
1679
1680         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_STATFS,
1681                                         LUSTRE_MDS_VERSION, MDS_STATFS);
1682         if (req == NULL)
1683                 GOTO(output, rc = -ENOMEM);
1684
1685         ptlrpc_request_set_replen(req);
1686
1687         if (flags & OBD_STATFS_NODELAY) {
1688                 /* procfs requests not want stay in wait for avoid deadlock */
1689                 req->rq_no_resend = 1;
1690                 req->rq_no_delay = 1;
1691         }
1692
1693         rc = ptlrpc_queue_wait(req);
1694         if (rc) {
1695                 /* check connection error first */
1696                 if (imp->imp_connect_error)
1697                         rc = imp->imp_connect_error;
1698                 GOTO(out, rc);
1699         }
1700
1701         msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1702         if (msfs == NULL)
1703                 GOTO(out, rc = -EPROTO);
1704
1705         *osfs = *msfs;
1706         EXIT;
1707 out:
1708         ptlrpc_req_finished(req);
1709 output:
1710         class_import_put(imp);
1711         return rc;
1712 }
1713
1714 static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
1715 {
1716         __u32 keylen, vallen;
1717         void *key;
1718         int rc;
1719
1720         if (gf->gf_pathlen > PATH_MAX)
1721                 RETURN(-ENAMETOOLONG);
1722         if (gf->gf_pathlen < 2)
1723                 RETURN(-EOVERFLOW);
1724
1725         /* Key is KEY_FID2PATH + getinfo_fid2path description */
1726         keylen = cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf);
1727         OBD_ALLOC(key, keylen);
1728         if (key == NULL)
1729                 RETURN(-ENOMEM);
1730         memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
1731         memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf));
1732
1733         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
1734                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
1735
1736         if (!fid_is_sane(&gf->gf_fid))
1737                 GOTO(out, rc = -EINVAL);
1738
1739         /* Val is struct getinfo_fid2path result plus path */
1740         vallen = sizeof(*gf) + gf->gf_pathlen;
1741
1742         rc = obd_get_info(NULL, exp, keylen, key, &vallen, gf, NULL);
1743         if (rc != 0 && rc != -EREMOTE)
1744                 GOTO(out, rc);
1745
1746         if (vallen <= sizeof(*gf))
1747                 GOTO(out, rc = -EPROTO);
1748         else if (vallen > sizeof(*gf) + gf->gf_pathlen)
1749                 GOTO(out, rc = -EOVERFLOW);
1750
1751         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n%s\n",
1752                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno, gf->gf_path);
1753
1754 out:
1755         OBD_FREE(key, keylen);
1756         return rc;
1757 }
1758
1759 static int mdc_ioc_hsm_progress(struct obd_export *exp,
1760                                 struct hsm_progress_kernel *hpk)
1761 {
1762         struct obd_import               *imp = class_exp2cliimp(exp);
1763         struct hsm_progress_kernel      *req_hpk;
1764         struct ptlrpc_request           *req;
1765         int                              rc;
1766         ENTRY;
1767
1768         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_PROGRESS,
1769                                         LUSTRE_MDS_VERSION, MDS_HSM_PROGRESS);
1770         if (req == NULL)
1771                 GOTO(out, rc = -ENOMEM);
1772
1773         mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
1774
1775         /* Copy hsm_progress struct */
1776         req_hpk = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_PROGRESS);
1777         if (req_hpk == NULL)
1778                 GOTO(out, rc = -EPROTO);
1779
1780         *req_hpk = *hpk;
1781         req_hpk->hpk_errval = lustre_errno_hton(hpk->hpk_errval);
1782
1783         ptlrpc_request_set_replen(req);
1784
1785         rc = mdc_queue_wait(req);
1786         GOTO(out, rc);
1787 out:
1788         ptlrpc_req_finished(req);
1789         return rc;
1790 }
1791
1792 static int mdc_ioc_hsm_ct_register(struct obd_import *imp, __u32 archives)
1793 {
1794         __u32                   *archive_mask;
1795         struct ptlrpc_request   *req;
1796         int                      rc;
1797         ENTRY;
1798
1799         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_REGISTER,
1800                                         LUSTRE_MDS_VERSION,
1801                                         MDS_HSM_CT_REGISTER);
1802         if (req == NULL)
1803                 GOTO(out, rc = -ENOMEM);
1804
1805         mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
1806
1807         /* Copy hsm_progress struct */
1808         archive_mask = req_capsule_client_get(&req->rq_pill,
1809                                               &RMF_MDS_HSM_ARCHIVE);
1810         if (archive_mask == NULL)
1811                 GOTO(out, rc = -EPROTO);
1812
1813         *archive_mask = archives;
1814
1815         ptlrpc_request_set_replen(req);
1816
1817         rc = mdc_queue_wait(req);
1818         GOTO(out, rc);
1819 out:
1820         ptlrpc_req_finished(req);
1821         return rc;
1822 }
1823
1824 static int mdc_ioc_hsm_current_action(struct obd_export *exp,
1825                                       struct md_op_data *op_data)
1826 {
1827         struct hsm_current_action       *hca = op_data->op_data;
1828         struct hsm_current_action       *req_hca;
1829         struct ptlrpc_request           *req;
1830         int                              rc;
1831         ENTRY;
1832
1833         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1834                                    &RQF_MDS_HSM_ACTION);
1835         if (req == NULL)
1836                 RETURN(-ENOMEM);
1837
1838         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1839
1840         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_ACTION);
1841         if (rc) {
1842                 ptlrpc_request_free(req);
1843                 RETURN(rc);
1844         }
1845
1846         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
1847                       OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
1848
1849         ptlrpc_request_set_replen(req);
1850
1851         rc = mdc_queue_wait(req);
1852         if (rc)
1853                 GOTO(out, rc);
1854
1855         req_hca = req_capsule_server_get(&req->rq_pill,
1856                                          &RMF_MDS_HSM_CURRENT_ACTION);
1857         if (req_hca == NULL)
1858                 GOTO(out, rc = -EPROTO);
1859
1860         *hca = *req_hca;
1861
1862         EXIT;
1863 out:
1864         ptlrpc_req_finished(req);
1865         return rc;
1866 }
1867
1868 static int mdc_ioc_hsm_ct_unregister(struct obd_import *imp)
1869 {
1870         struct ptlrpc_request   *req;
1871         int                      rc;
1872         ENTRY;
1873
1874         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_UNREGISTER,
1875                                         LUSTRE_MDS_VERSION,
1876                                         MDS_HSM_CT_UNREGISTER);
1877         if (req == NULL)
1878                 GOTO(out, rc = -ENOMEM);
1879
1880         mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
1881
1882         ptlrpc_request_set_replen(req);
1883
1884         rc = mdc_queue_wait(req);
1885         GOTO(out, rc);
1886 out:
1887         ptlrpc_req_finished(req);
1888         return rc;
1889 }
1890
1891 static int mdc_ioc_hsm_state_get(struct obd_export *exp,
1892                                  struct md_op_data *op_data)
1893 {
1894         struct hsm_user_state   *hus = op_data->op_data;
1895         struct hsm_user_state   *req_hus;
1896         struct ptlrpc_request   *req;
1897         int                      rc;
1898         ENTRY;
1899
1900         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1901                                    &RQF_MDS_HSM_STATE_GET);
1902         if (req == NULL)
1903                 RETURN(-ENOMEM);
1904
1905         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1906
1907         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_GET);
1908         if (rc != 0) {
1909                 ptlrpc_request_free(req);
1910                 RETURN(rc);
1911         }
1912
1913         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
1914                       OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
1915
1916         ptlrpc_request_set_replen(req);
1917
1918         rc = mdc_queue_wait(req);
1919         if (rc)
1920                 GOTO(out, rc);
1921
1922         req_hus = req_capsule_server_get(&req->rq_pill, &RMF_HSM_USER_STATE);
1923         if (req_hus == NULL)
1924                 GOTO(out, rc = -EPROTO);
1925
1926         *hus = *req_hus;
1927
1928         EXIT;
1929 out:
1930         ptlrpc_req_finished(req);
1931         return rc;
1932 }
1933
1934 static int mdc_ioc_hsm_state_set(struct obd_export *exp,
1935                                  struct md_op_data *op_data)
1936 {
1937         struct hsm_state_set    *hss = op_data->op_data;
1938         struct hsm_state_set    *req_hss;
1939         struct ptlrpc_request   *req;
1940         int                      rc;
1941         ENTRY;
1942
1943         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1944                                    &RQF_MDS_HSM_STATE_SET);
1945         if (req == NULL)
1946                 RETURN(-ENOMEM);
1947
1948         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1949
1950         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_SET);
1951         if (rc) {
1952                 ptlrpc_request_free(req);
1953                 RETURN(rc);
1954         }
1955
1956         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
1957                       OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
1958
1959         /* Copy states */
1960         req_hss = req_capsule_client_get(&req->rq_pill, &RMF_HSM_STATE_SET);
1961         if (req_hss == NULL)
1962                 GOTO(out, rc = -EPROTO);
1963         *req_hss = *hss;
1964
1965         ptlrpc_request_set_replen(req);
1966
1967         rc = mdc_queue_wait(req);
1968         GOTO(out, rc);
1969
1970         EXIT;
1971 out:
1972         ptlrpc_req_finished(req);
1973         return rc;
1974 }
1975
1976 static int mdc_ioc_hsm_request(struct obd_export *exp,
1977                                struct hsm_user_request *hur)
1978 {
1979         struct obd_import       *imp = class_exp2cliimp(exp);
1980         struct ptlrpc_request   *req;
1981         struct hsm_request      *req_hr;
1982         struct hsm_user_item    *req_hui;
1983         char                    *req_opaque;
1984         int                      rc;
1985         ENTRY;
1986
1987         req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_REQUEST);
1988         if (req == NULL)
1989                 GOTO(out, rc = -ENOMEM);
1990
1991         req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM, RCL_CLIENT,
1992                              hur->hur_request.hr_itemcount
1993                              * sizeof(struct hsm_user_item));
1994         req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA, RCL_CLIENT,
1995                              hur->hur_request.hr_data_len);
1996
1997         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_REQUEST);
1998         if (rc) {
1999                 ptlrpc_request_free(req);
2000                 RETURN(rc);
2001         }
2002
2003         mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
2004
2005         /* Copy hsm_request struct */
2006         req_hr = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_REQUEST);
2007         if (req_hr == NULL)
2008                 GOTO(out, rc = -EPROTO);
2009         *req_hr = hur->hur_request;
2010
2011         /* Copy hsm_user_item structs */
2012         req_hui = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM);
2013         if (req_hui == NULL)
2014                 GOTO(out, rc = -EPROTO);
2015         memcpy(req_hui, hur->hur_user_item,
2016                hur->hur_request.hr_itemcount * sizeof(struct hsm_user_item));
2017
2018         /* Copy opaque field */
2019         req_opaque = req_capsule_client_get(&req->rq_pill, &RMF_GENERIC_DATA);
2020         if (req_opaque == NULL)
2021                 GOTO(out, rc = -EPROTO);
2022         memcpy(req_opaque, hur_data(hur), hur->hur_request.hr_data_len);
2023
2024         ptlrpc_request_set_replen(req);
2025
2026         rc = mdc_queue_wait(req);
2027         GOTO(out, rc);
2028
2029 out:
2030         ptlrpc_req_finished(req);
2031         return rc;
2032 }
2033
2034 static struct kuc_hdr *changelog_kuc_hdr(char *buf, int len, int flags)
2035 {
2036         struct kuc_hdr *lh = (struct kuc_hdr *)buf;
2037
2038         LASSERT(len <= KUC_CHANGELOG_MSG_MAXSIZE);
2039
2040         lh->kuc_magic = KUC_MAGIC;
2041         lh->kuc_transport = KUC_TRANSPORT_CHANGELOG;
2042         lh->kuc_flags = flags;
2043         lh->kuc_msgtype = CL_RECORD;
2044         lh->kuc_msglen = len;
2045         return lh;
2046 }
2047
2048 #define D_CHANGELOG 0
2049
2050 struct changelog_show {
2051         __u64           cs_startrec;
2052         __u32           cs_flags;
2053         struct file     *cs_fp;
2054         char            *cs_buf;
2055         struct obd_device *cs_obd;
2056 };
2057
2058 static int changelog_kkuc_cb(const struct lu_env *env, struct llog_handle *llh,
2059                              struct llog_rec_hdr *hdr, void *data)
2060 {
2061         struct changelog_show *cs = data;
2062         struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
2063         struct kuc_hdr *lh;
2064         int len, rc;
2065         ENTRY;
2066
2067         if (rec->cr_hdr.lrh_type != CHANGELOG_REC) {
2068                 rc = -EINVAL;
2069                 CERROR("%s: not a changelog rec %x/%d: rc = %d\n",
2070                        cs->cs_obd->obd_name, rec->cr_hdr.lrh_type,
2071                        rec->cr.cr_type, rc);
2072                 RETURN(rc);
2073         }
2074
2075         if (rec->cr.cr_index < cs->cs_startrec) {
2076                 /* Skip entries earlier than what we are interested in */
2077                 CDEBUG(D_CHANGELOG, "rec="LPU64" start="LPU64"\n",
2078                        rec->cr.cr_index, cs->cs_startrec);
2079                 RETURN(0);
2080         }
2081
2082         CDEBUG(D_CHANGELOG, LPU64" %02d%-5s "LPU64" 0x%x t="DFID" p="DFID
2083                 " %.*s\n", rec->cr.cr_index, rec->cr.cr_type,
2084                 changelog_type2str(rec->cr.cr_type), rec->cr.cr_time,
2085                 rec->cr.cr_flags & CLF_FLAGMASK,
2086                 PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid),
2087                 rec->cr.cr_namelen, changelog_rec_name(&rec->cr));
2088
2089         len = sizeof(*lh) + changelog_rec_size(&rec->cr) + rec->cr.cr_namelen;
2090
2091         /* Set up the message */
2092         lh = changelog_kuc_hdr(cs->cs_buf, len, cs->cs_flags);
2093         memcpy(lh + 1, &rec->cr, len - sizeof(*lh));
2094
2095         rc = libcfs_kkuc_msg_put(cs->cs_fp, lh);
2096         CDEBUG(D_CHANGELOG, "kucmsg fp %p len %d rc %d\n", cs->cs_fp, len,rc);
2097
2098         RETURN(rc);
2099 }
2100
2101 static int mdc_changelog_send_thread(void *csdata)
2102 {
2103         struct changelog_show *cs = csdata;
2104         struct llog_ctxt *ctxt = NULL;
2105         struct llog_handle *llh = NULL;
2106         struct kuc_hdr *kuch;
2107         int rc;
2108
2109         CDEBUG(D_CHANGELOG, "changelog to fp=%p start "LPU64"\n",
2110                cs->cs_fp, cs->cs_startrec);
2111
2112         OBD_ALLOC(cs->cs_buf, KUC_CHANGELOG_MSG_MAXSIZE);
2113         if (cs->cs_buf == NULL)
2114                 GOTO(out, rc = -ENOMEM);
2115
2116         /* Set up the remote catalog handle */
2117         ctxt = llog_get_context(cs->cs_obd, LLOG_CHANGELOG_REPL_CTXT);
2118         if (ctxt == NULL)
2119                 GOTO(out, rc = -ENOENT);
2120         rc = llog_open(NULL, ctxt, &llh, NULL, CHANGELOG_CATALOG,
2121                        LLOG_OPEN_EXISTS);
2122         if (rc) {
2123                 CERROR("%s: fail to open changelog catalog: rc = %d\n",
2124                        cs->cs_obd->obd_name, rc);
2125                 GOTO(out, rc);
2126         }
2127         rc = llog_init_handle(NULL, llh, LLOG_F_IS_CAT, NULL);
2128         if (rc) {
2129                 CERROR("llog_init_handle failed %d\n", rc);
2130                 GOTO(out, rc);
2131         }
2132
2133         rc = llog_cat_process(NULL, llh, changelog_kkuc_cb, cs, 0, 0);
2134
2135         /* Send EOF no matter what our result */
2136         if ((kuch = changelog_kuc_hdr(cs->cs_buf, sizeof(*kuch),
2137                                       cs->cs_flags))) {
2138                 kuch->kuc_msgtype = CL_EOF;
2139                 libcfs_kkuc_msg_put(cs->cs_fp, kuch);
2140         }
2141
2142 out:
2143         fput(cs->cs_fp);
2144         if (llh)
2145                 llog_cat_close(NULL, llh);
2146         if (ctxt)
2147                 llog_ctxt_put(ctxt);
2148         if (cs->cs_buf)
2149                 OBD_FREE(cs->cs_buf, KUC_CHANGELOG_MSG_MAXSIZE);
2150         OBD_FREE_PTR(cs);
2151         return rc;
2152 }
2153
2154 static int mdc_ioc_changelog_send(struct obd_device *obd,
2155                                   struct ioc_changelog *icc)
2156 {
2157         struct changelog_show *cs;
2158         struct task_struct *task;
2159         int rc;
2160
2161         /* Freed in mdc_changelog_send_thread */
2162         OBD_ALLOC_PTR(cs);
2163         if (!cs)
2164                 return -ENOMEM;
2165
2166         cs->cs_obd = obd;
2167         cs->cs_startrec = icc->icc_recno;
2168         /* matching fput in mdc_changelog_send_thread */
2169         cs->cs_fp = fget(icc->icc_id);
2170         cs->cs_flags = icc->icc_flags;
2171
2172         /*
2173          * New thread because we should return to user app before
2174          * writing into our pipe
2175          */
2176         task = kthread_run(mdc_changelog_send_thread, cs,
2177                            "mdc_clg_send_thread");
2178         if (IS_ERR(task)) {
2179                 rc = PTR_ERR(task);
2180                 CERROR("%s: cannot start changelog thread: rc = %d\n",
2181                        obd->obd_name, rc);
2182                 OBD_FREE_PTR(cs);
2183         } else {
2184                 rc = 0;
2185                 CDEBUG(D_CHANGELOG, "%s: started changelog thread\n",
2186                        obd->obd_name);
2187         }
2188
2189         return rc;
2190 }
2191
2192 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2193                                 struct lustre_kernelcomm *lk);
2194
2195 static int mdc_quotacheck(struct obd_device *unused, struct obd_export *exp,
2196                           struct obd_quotactl *oqctl)
2197 {
2198         struct client_obd       *cli = &exp->exp_obd->u.cli;
2199         struct ptlrpc_request   *req;
2200         struct obd_quotactl     *body;
2201         int                      rc;
2202         ENTRY;
2203
2204         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
2205                                         &RQF_MDS_QUOTACHECK, LUSTRE_MDS_VERSION,
2206                                         MDS_QUOTACHECK);
2207         if (req == NULL)
2208                 RETURN(-ENOMEM);
2209
2210         body = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
2211         *body = *oqctl;
2212
2213         ptlrpc_request_set_replen(req);
2214
2215         /* the next poll will find -ENODATA, that means quotacheck is
2216          * going on */
2217         cli->cl_qchk_stat = -ENODATA;
2218         rc = ptlrpc_queue_wait(req);
2219         if (rc)
2220                 cli->cl_qchk_stat = rc;
2221         ptlrpc_req_finished(req);
2222         RETURN(rc);
2223 }
2224
2225 static int mdc_quota_poll_check(struct obd_export *exp,
2226                                 struct if_quotacheck *qchk)
2227 {
2228         struct client_obd *cli = &exp->exp_obd->u.cli;
2229         int rc;
2230         ENTRY;
2231
2232         qchk->obd_uuid = cli->cl_target_uuid;
2233         memcpy(qchk->obd_type, LUSTRE_MDS_NAME, strlen(LUSTRE_MDS_NAME));
2234
2235         rc = cli->cl_qchk_stat;
2236         /* the client is not the previous one */
2237         if (rc == CL_NOT_QUOTACHECKED)
2238                 rc = -EINTR;
2239         RETURN(rc);
2240 }
2241
2242 static int mdc_quotactl(struct obd_device *unused, struct obd_export *exp,
2243                         struct obd_quotactl *oqctl)
2244 {
2245         struct ptlrpc_request   *req;
2246         struct obd_quotactl     *oqc;
2247         int                      rc;
2248         ENTRY;
2249
2250         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
2251                                         &RQF_MDS_QUOTACTL, LUSTRE_MDS_VERSION,
2252                                         MDS_QUOTACTL);
2253         if (req == NULL)
2254                 RETURN(-ENOMEM);
2255
2256         oqc = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
2257         *oqc = *oqctl;
2258
2259         ptlrpc_request_set_replen(req);
2260         ptlrpc_at_set_req_timeout(req);
2261         req->rq_no_resend = 1;
2262
2263         rc = ptlrpc_queue_wait(req);
2264         if (rc)
2265                 CERROR("ptlrpc_queue_wait failed, rc: %d\n", rc);
2266
2267         if (req->rq_repmsg &&
2268             (oqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL))) {
2269                 *oqctl = *oqc;
2270         } else if (!rc) {
2271                 CERROR ("Can't unpack obd_quotactl\n");
2272                 rc = -EPROTO;
2273         }
2274         ptlrpc_req_finished(req);
2275
2276         RETURN(rc);
2277 }
2278
2279 static int mdc_ioc_swap_layouts(struct obd_export *exp,
2280                                 struct md_op_data *op_data)
2281 {
2282         struct list_head cancels = LIST_HEAD_INIT(cancels);
2283         struct ptlrpc_request   *req;
2284         int                      rc, count;
2285         struct mdc_swap_layouts *msl, *payload;
2286         ENTRY;
2287
2288         msl = op_data->op_data;
2289
2290         /* When the MDT will get the MDS_SWAP_LAYOUTS RPC the
2291          * first thing it will do is to cancel the 2 layout
2292          * locks hold by this client.
2293          * So the client must cancel its layout locks on the 2 fids
2294          * with the request RPC to avoid extra RPC round trips
2295          */
2296         count = mdc_resource_get_unused(exp, &op_data->op_fid1, &cancels,
2297                                         LCK_EX, MDS_INODELOCK_LAYOUT |
2298                                         MDS_INODELOCK_XATTR);
2299         count += mdc_resource_get_unused(exp, &op_data->op_fid2, &cancels,
2300                                          LCK_EX, MDS_INODELOCK_LAYOUT |
2301                                          MDS_INODELOCK_XATTR);
2302
2303         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
2304                                    &RQF_MDS_SWAP_LAYOUTS);
2305         if (req == NULL) {
2306                 ldlm_lock_list_put(&cancels, l_bl_ast, count);
2307                 RETURN(-ENOMEM);
2308         }
2309
2310         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
2311         mdc_set_capa_size(req, &RMF_CAPA2, op_data->op_capa2);
2312
2313         rc = mdc_prep_elc_req(exp, req, MDS_SWAP_LAYOUTS, &cancels, count);
2314         if (rc) {
2315                 ptlrpc_request_free(req);
2316                 RETURN(rc);
2317         }
2318
2319         mdc_swap_layouts_pack(req, op_data);
2320
2321         payload = req_capsule_client_get(&req->rq_pill, &RMF_SWAP_LAYOUTS);
2322         LASSERT(payload);
2323
2324         *payload = *msl;
2325
2326         ptlrpc_request_set_replen(req);
2327
2328         rc = ptlrpc_queue_wait(req);
2329         if (rc)
2330                 GOTO(out, rc);
2331         EXIT;
2332
2333 out:
2334         ptlrpc_req_finished(req);
2335         return rc;
2336 }
2337
2338 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2339                          void *karg, void *uarg)
2340 {
2341         struct obd_device *obd = exp->exp_obd;
2342         struct obd_ioctl_data *data = karg;
2343         struct obd_import *imp = obd->u.cli.cl_import;
2344         int rc;
2345         ENTRY;
2346
2347         if (!try_module_get(THIS_MODULE)) {
2348                 CERROR("Can't get module. Is it alive?");
2349                 return -EINVAL;
2350         }
2351         switch (cmd) {
2352         case OBD_IOC_CHANGELOG_SEND:
2353                 rc = mdc_ioc_changelog_send(obd, karg);
2354                 GOTO(out, rc);
2355         case OBD_IOC_CHANGELOG_CLEAR: {
2356                 struct ioc_changelog *icc = karg;
2357                 struct changelog_setinfo cs =
2358                         {.cs_recno = icc->icc_recno, .cs_id = icc->icc_id};
2359                 rc = obd_set_info_async(NULL, exp, strlen(KEY_CHANGELOG_CLEAR),
2360                                         KEY_CHANGELOG_CLEAR, sizeof(cs), &cs,
2361                                         NULL);
2362                 GOTO(out, rc);
2363         }
2364         case OBD_IOC_FID2PATH:
2365                 rc = mdc_ioc_fid2path(exp, karg);
2366                 GOTO(out, rc);
2367         case LL_IOC_HSM_CT_START:
2368                 rc = mdc_ioc_hsm_ct_start(exp, karg);
2369                 /* ignore if it was already registered on this MDS. */
2370                 if (rc == -EEXIST)
2371                         rc = 0;
2372                 GOTO(out, rc);
2373         case LL_IOC_HSM_PROGRESS:
2374                 rc = mdc_ioc_hsm_progress(exp, karg);
2375                 GOTO(out, rc);
2376         case LL_IOC_HSM_STATE_GET:
2377                 rc = mdc_ioc_hsm_state_get(exp, karg);
2378                 GOTO(out, rc);
2379         case LL_IOC_HSM_STATE_SET:
2380                 rc = mdc_ioc_hsm_state_set(exp, karg);
2381                 GOTO(out, rc);
2382         case LL_IOC_HSM_ACTION:
2383                 rc = mdc_ioc_hsm_current_action(exp, karg);
2384                 GOTO(out, rc);
2385         case LL_IOC_HSM_REQUEST:
2386                 rc = mdc_ioc_hsm_request(exp, karg);
2387                 GOTO(out, rc);
2388         case OBD_IOC_CLIENT_RECOVER:
2389                 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1, 0);
2390                 if (rc < 0)
2391                         GOTO(out, rc);
2392                 GOTO(out, rc = 0);
2393         case IOC_OSC_SET_ACTIVE:
2394                 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
2395                 GOTO(out, rc);
2396         case OBD_IOC_POLL_QUOTACHECK:
2397                 rc = mdc_quota_poll_check(exp, (struct if_quotacheck *)karg);
2398                 GOTO(out, rc);
2399         case OBD_IOC_PING_TARGET:
2400                 rc = ptlrpc_obd_ping(obd);
2401                 GOTO(out, rc);
2402         /*
2403          * Normally IOC_OBD_STATFS, OBD_IOC_QUOTACTL iocontrol are handled by
2404          * LMV instead of MDC. But when the cluster is upgraded from 1.8,
2405          * there'd be no LMV layer thus we might be called here. Eventually
2406          * this code should be removed.
2407          * bz20731, LU-592.
2408          */
2409         case IOC_OBD_STATFS: {
2410                 struct obd_statfs stat_buf = {0};
2411
2412                 if (*((__u32 *) data->ioc_inlbuf2) != 0)
2413                         GOTO(out, rc = -ENODEV);
2414
2415                 /* copy UUID */
2416                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd),
2417                                  min((int)data->ioc_plen2,
2418                                      (int)sizeof(struct obd_uuid))))
2419                         GOTO(out, rc = -EFAULT);
2420
2421                 rc = mdc_statfs(NULL, obd->obd_self_export, &stat_buf,
2422                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
2423                                 0);
2424                 if (rc != 0)
2425                         GOTO(out, rc);
2426
2427                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
2428                                      min((int) data->ioc_plen1,
2429                                          (int) sizeof(stat_buf))))
2430                         GOTO(out, rc = -EFAULT);
2431
2432                 GOTO(out, rc = 0);
2433         }
2434         case OBD_IOC_QUOTACTL: {
2435                 struct if_quotactl *qctl = karg;
2436                 struct obd_quotactl *oqctl;
2437
2438                 OBD_ALLOC_PTR(oqctl);
2439                 if (oqctl == NULL)
2440                         GOTO(out, rc = -ENOMEM);
2441
2442                 QCTL_COPY(oqctl, qctl);
2443                 rc = obd_quotactl(exp, oqctl);
2444                 if (rc == 0) {
2445                         QCTL_COPY(qctl, oqctl);
2446                         qctl->qc_valid = QC_MDTIDX;
2447                         qctl->obd_uuid = obd->u.cli.cl_target_uuid;
2448                 }
2449
2450                 OBD_FREE_PTR(oqctl);
2451                 GOTO(out, rc);
2452         }
2453         case LL_IOC_GET_CONNECT_FLAGS:
2454                 if (copy_to_user(uarg, exp_connect_flags_ptr(exp),
2455                                  sizeof(*exp_connect_flags_ptr(exp))))
2456                         GOTO(out, rc = -EFAULT);
2457
2458                 GOTO(out, rc = 0);
2459         case LL_IOC_LOV_SWAP_LAYOUTS:
2460                 rc = mdc_ioc_swap_layouts(exp, karg);
2461                 GOTO(out, rc);
2462         default:
2463                 CERROR("unrecognised ioctl: cmd = %#x\n", cmd);
2464                 GOTO(out, rc = -ENOTTY);
2465         }
2466 out:
2467         module_put(THIS_MODULE);
2468
2469         return rc;
2470 }
2471
2472 int mdc_get_info_rpc(struct obd_export *exp,
2473                      obd_count keylen, void *key,
2474                      int vallen, void *val)
2475 {
2476         struct obd_import      *imp = class_exp2cliimp(exp);
2477         struct ptlrpc_request  *req;
2478         char                   *tmp;
2479         int                     rc = -EINVAL;
2480         ENTRY;
2481
2482         req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
2483         if (req == NULL)
2484                 RETURN(-ENOMEM);
2485
2486         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
2487                              RCL_CLIENT, keylen);
2488         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
2489                              RCL_CLIENT, sizeof(__u32));
2490
2491         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
2492         if (rc) {
2493                 ptlrpc_request_free(req);
2494                 RETURN(rc);
2495         }
2496
2497         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
2498         memcpy(tmp, key, keylen);
2499         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
2500         memcpy(tmp, &vallen, sizeof(__u32));
2501
2502         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
2503                              RCL_SERVER, vallen);
2504         ptlrpc_request_set_replen(req);
2505
2506         rc = ptlrpc_queue_wait(req);
2507         /* -EREMOTE means the get_info result is partial, and it needs to
2508          * continue on another MDT, see fid2path part in lmv_iocontrol */
2509         if (rc == 0 || rc == -EREMOTE) {
2510                 tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
2511                 memcpy(val, tmp, vallen);
2512                 if (ptlrpc_rep_need_swab(req)) {
2513                         if (KEY_IS(KEY_FID2PATH))
2514                                 lustre_swab_fid2path(val);
2515                 }
2516         }
2517         ptlrpc_req_finished(req);
2518
2519         RETURN(rc);
2520 }
2521
2522 static void lustre_swab_hai(struct hsm_action_item *h)
2523 {
2524         __swab32s(&h->hai_len);
2525         __swab32s(&h->hai_action);
2526         lustre_swab_lu_fid(&h->hai_fid);
2527         lustre_swab_lu_fid(&h->hai_dfid);
2528         __swab64s(&h->hai_cookie);
2529         __swab64s(&h->hai_extent.offset);
2530         __swab64s(&h->hai_extent.length);
2531         __swab64s(&h->hai_gid);
2532 }
2533
2534 static void lustre_swab_hal(struct hsm_action_list *h)
2535 {
2536         struct hsm_action_item  *hai;
2537         int                      i;
2538
2539         __swab32s(&h->hal_version);
2540         __swab32s(&h->hal_count);
2541         __swab32s(&h->hal_archive_id);
2542         __swab64s(&h->hal_flags);
2543         hai = hai_first(h);
2544         for (i = 0; i < h->hal_count; i++, hai = hai_next(hai))
2545                 lustre_swab_hai(hai);
2546 }
2547
2548 static void lustre_swab_kuch(struct kuc_hdr *l)
2549 {
2550         __swab16s(&l->kuc_magic);
2551         /* __u8 l->kuc_transport */
2552         __swab16s(&l->kuc_msgtype);
2553         __swab16s(&l->kuc_msglen);
2554 }
2555
2556 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2557                                 struct lustre_kernelcomm *lk)
2558 {
2559         struct obd_import  *imp = class_exp2cliimp(exp);
2560         __u32               archive = lk->lk_data;
2561         int                 rc = 0;
2562
2563         if (lk->lk_group != KUC_GRP_HSM) {
2564                 CERROR("Bad copytool group %d\n", lk->lk_group);
2565                 return -EINVAL;
2566         }
2567
2568         CDEBUG(D_HSM, "CT start r%d w%d u%d g%d f%#x\n", lk->lk_rfd, lk->lk_wfd,
2569                lk->lk_uid, lk->lk_group, lk->lk_flags);
2570
2571         if (lk->lk_flags & LK_FLG_STOP) {
2572                 /* Unregister with the coordinator */
2573                 rc = mdc_ioc_hsm_ct_unregister(imp);
2574         } else {
2575                 rc = mdc_ioc_hsm_ct_register(imp, archive);
2576         }
2577
2578         return rc;
2579 }
2580
2581 /**
2582  * Send a message to any listening copytools
2583  * @param val KUC message (kuc_hdr + hsm_action_list)
2584  * @param len total length of message
2585  */
2586 static int mdc_hsm_copytool_send(int len, void *val)
2587 {
2588         struct kuc_hdr          *lh = (struct kuc_hdr *)val;
2589         struct hsm_action_list  *hal = (struct hsm_action_list *)(lh + 1);
2590         int                      rc;
2591         ENTRY;
2592
2593         if (len < sizeof(*lh) + sizeof(*hal)) {
2594                 CERROR("Short HSM message %d < %d\n", len,
2595                        (int) (sizeof(*lh) + sizeof(*hal)));
2596                 RETURN(-EPROTO);
2597         }
2598         if (lh->kuc_magic == __swab16(KUC_MAGIC)) {
2599                 lustre_swab_kuch(lh);
2600                 lustre_swab_hal(hal);
2601         } else if (lh->kuc_magic != KUC_MAGIC) {
2602                 CERROR("Bad magic %x!=%x\n", lh->kuc_magic, KUC_MAGIC);
2603                 RETURN(-EPROTO);
2604         }
2605
2606         CDEBUG(D_HSM, " Received message mg=%x t=%d m=%d l=%d actions=%d "
2607                "on %s\n",
2608                lh->kuc_magic, lh->kuc_transport, lh->kuc_msgtype,
2609                lh->kuc_msglen, hal->hal_count, hal->hal_fsname);
2610
2611         /* Broadcast to HSM listeners */
2612         rc = libcfs_kkuc_group_put(KUC_GRP_HSM, lh);
2613
2614         RETURN(rc);
2615 }
2616
2617 /**
2618  * callback function passed to kuc for re-registering each HSM copytool
2619  * running on MDC, after MDT shutdown/recovery.
2620  * @param data copytool registration data
2621  * @param cb_arg callback argument (obd_import)
2622  */
2623 static int mdc_hsm_ct_reregister(void *data, void *cb_arg)
2624 {
2625         struct kkuc_ct_data     *kcd = data;
2626         struct obd_import       *imp = (struct obd_import *)cb_arg;
2627         int                      rc;
2628
2629         if (kcd == NULL || kcd->kcd_magic != KKUC_CT_DATA_MAGIC)
2630                 return -EPROTO;
2631
2632         if (!obd_uuid_equals(&kcd->kcd_uuid, &imp->imp_obd->obd_uuid))
2633                 return 0;
2634
2635         CDEBUG(D_HA, "%s: recover copytool registration to MDT (archive=%#x)\n",
2636                imp->imp_obd->obd_name, kcd->kcd_archive);
2637         rc = mdc_ioc_hsm_ct_register(imp, kcd->kcd_archive);
2638
2639         /* ignore error if the copytool is already registered */
2640         return (rc == -EEXIST) ? 0 : rc;
2641 }
2642
2643 /**
2644  * Re-establish all kuc contexts with MDT
2645  * after MDT shutdown/recovery.
2646  */
2647 static int mdc_kuc_reregister(struct obd_import *imp)
2648 {
2649         /* re-register HSM agents */
2650         return libcfs_kkuc_group_foreach(KUC_GRP_HSM, mdc_hsm_ct_reregister,
2651                                          (void *)imp);
2652 }
2653
2654 int mdc_set_info_async(const struct lu_env *env,
2655                        struct obd_export *exp,
2656                        obd_count keylen, void *key,
2657                        obd_count vallen, void *val,
2658                        struct ptlrpc_request_set *set)
2659 {
2660         struct obd_import       *imp = class_exp2cliimp(exp);
2661         int                      rc;
2662         ENTRY;
2663
2664         if (KEY_IS(KEY_READ_ONLY)) {
2665                 if (vallen != sizeof(int))
2666                         RETURN(-EINVAL);
2667
2668                 spin_lock(&imp->imp_lock);
2669                 if (*((int *)val)) {
2670                         imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
2671                         imp->imp_connect_data.ocd_connect_flags |=
2672                                                         OBD_CONNECT_RDONLY;
2673                 } else {
2674                         imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
2675                         imp->imp_connect_data.ocd_connect_flags &=
2676                                                         ~OBD_CONNECT_RDONLY;
2677                 }
2678                 spin_unlock(&imp->imp_lock);
2679
2680                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2681                                        keylen, key, vallen, val, set);
2682                 RETURN(rc);
2683         }
2684         if (KEY_IS(KEY_SPTLRPC_CONF)) {
2685                 sptlrpc_conf_client_adapt(exp->exp_obd);
2686                 RETURN(0);
2687         }
2688         if (KEY_IS(KEY_FLUSH_CTX)) {
2689                 sptlrpc_import_flush_my_ctx(imp);
2690                 RETURN(0);
2691         }
2692         if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2693                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2694                                        keylen, key, vallen, val, set);
2695                 RETURN(rc);
2696         }
2697         if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) {
2698                 rc = mdc_hsm_copytool_send(vallen, val);
2699                 RETURN(rc);
2700         }
2701
2702         CERROR("Unknown key %s\n", (char *)key);
2703         RETURN(-EINVAL);
2704 }
2705
2706 int mdc_get_info(const struct lu_env *env, struct obd_export *exp,
2707                  __u32 keylen, void *key, __u32 *vallen, void *val,
2708                  struct lov_stripe_md *lsm)
2709 {
2710         int rc = -EINVAL;
2711
2712         if (KEY_IS(KEY_MAX_EASIZE)) {
2713                 int mdsize, *max_easize;
2714
2715                 if (*vallen != sizeof(int))
2716                         RETURN(-EINVAL);
2717                 mdsize = *(int *)val;
2718                 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
2719                         exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
2720                 max_easize = val;
2721                 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
2722                 RETURN(0);
2723         } else if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2724                 int *default_easize;
2725
2726                 if (*vallen != sizeof(int))
2727                         RETURN(-EINVAL);
2728                 default_easize = val;
2729                 *default_easize = exp->exp_obd->u.cli.cl_default_mds_easize;
2730                 RETURN(0);
2731         } else if (KEY_IS(KEY_MAX_COOKIESIZE)) {
2732                 int mdsize, *max_cookiesize;
2733
2734                 if (*vallen != sizeof(int))
2735                         RETURN(-EINVAL);
2736                 mdsize = *(int *)val;
2737                 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_cookiesize)
2738                         exp->exp_obd->u.cli.cl_max_mds_cookiesize = mdsize;
2739                 max_cookiesize = val;
2740                 *max_cookiesize = exp->exp_obd->u.cli.cl_max_mds_cookiesize;
2741                 RETURN(0);
2742         } else if (KEY_IS(KEY_DEFAULT_COOKIESIZE)) {
2743                 int *default_cookiesize;
2744
2745                 if (*vallen != sizeof(int))
2746                         RETURN(-EINVAL);
2747                 default_cookiesize = val;
2748                 *default_cookiesize =
2749                         exp->exp_obd->u.cli.cl_default_mds_cookiesize;
2750                 RETURN(0);
2751         } else if (KEY_IS(KEY_CONN_DATA)) {
2752                 struct obd_import *imp = class_exp2cliimp(exp);
2753                 struct obd_connect_data *data = val;
2754
2755                 if (*vallen != sizeof(*data))
2756                         RETURN(-EINVAL);
2757
2758                 *data = imp->imp_connect_data;
2759                 RETURN(0);
2760         } else if (KEY_IS(KEY_TGT_COUNT)) {
2761                 *((int *)val) = 1;
2762                 RETURN(0);
2763         }
2764
2765         rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
2766
2767         RETURN(rc);
2768 }
2769
2770 int mdc_fsync(struct obd_export *exp, const struct lu_fid *fid,
2771               struct obd_capa *oc, struct ptlrpc_request **request)
2772 {
2773         struct ptlrpc_request *req;
2774         int                    rc;
2775         ENTRY;
2776
2777         *request = NULL;
2778         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
2779         if (req == NULL)
2780                 RETURN(-ENOMEM);
2781
2782         mdc_set_capa_size(req, &RMF_CAPA1, oc);
2783
2784         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
2785         if (rc) {
2786                 ptlrpc_request_free(req);
2787                 RETURN(rc);
2788         }
2789
2790         mdc_pack_body(req, fid, oc, 0, 0, -1, 0);
2791
2792         ptlrpc_request_set_replen(req);
2793
2794         rc = ptlrpc_queue_wait(req);
2795         if (rc)
2796                 ptlrpc_req_finished(req);
2797         else
2798                 *request = req;
2799         RETURN(rc);
2800 }
2801
2802 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
2803                             enum obd_import_event event)
2804 {
2805         int rc = 0;
2806
2807         LASSERT(imp->imp_obd == obd);
2808
2809         switch (event) {
2810         case IMP_EVENT_DISCON: {
2811 #if 0
2812                 /* XXX Pass event up to OBDs stack. used only for FLD now */
2813                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DISCON, NULL);
2814 #endif
2815                 break;
2816         }
2817         case IMP_EVENT_INACTIVE: {
2818                 struct client_obd *cli = &obd->u.cli;
2819                 /*
2820                  * Flush current sequence to make client obtain new one
2821                  * from server in case of disconnect/reconnect.
2822                  */
2823                 if (cli->cl_seq != NULL)
2824                         seq_client_flush(cli->cl_seq);
2825
2826                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
2827                 break;
2828         }
2829         case IMP_EVENT_INVALIDATE: {
2830                 struct ldlm_namespace *ns = obd->obd_namespace;
2831
2832                 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2833
2834                 break;
2835         }
2836         case IMP_EVENT_ACTIVE:
2837                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
2838                 /* redo the kuc registration after reconnecting */
2839                 if (rc == 0)
2840                         rc = mdc_kuc_reregister(imp);
2841                 break;
2842         case IMP_EVENT_OCD:
2843                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
2844                 break;
2845         case IMP_EVENT_DEACTIVATE:
2846         case IMP_EVENT_ACTIVATE:
2847                 break;
2848         default:
2849                 CERROR("Unknown import event %x\n", event);
2850                 LBUG();
2851         }
2852         RETURN(rc);
2853 }
2854
2855 int mdc_fid_alloc(const struct lu_env *env, struct obd_export *exp,
2856                   struct lu_fid *fid, struct md_op_data *op_data)
2857 {
2858         struct client_obd *cli = &exp->exp_obd->u.cli;
2859         struct lu_client_seq *seq = cli->cl_seq;
2860         ENTRY;
2861         RETURN(seq_client_alloc_fid(env, seq, fid));
2862 }
2863
2864 struct obd_uuid *mdc_get_uuid(struct obd_export *exp) {
2865         struct client_obd *cli = &exp->exp_obd->u.cli;
2866         return &cli->cl_target_uuid;
2867 }
2868
2869 /**
2870  * Determine whether the lock can be canceled before replaying it during
2871  * recovery, non zero value will be return if the lock can be canceled,
2872  * or zero returned for not
2873  */
2874 static int mdc_cancel_weight(struct ldlm_lock *lock)
2875 {
2876         if (lock->l_resource->lr_type != LDLM_IBITS)
2877                 RETURN(0);
2878
2879         /* FIXME: if we ever get into a situation where there are too many
2880          * opened files with open locks on a single node, then we really
2881          * should replay these open locks to reget it */
2882         if (lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_OPEN)
2883                 RETURN(0);
2884
2885         RETURN(1);
2886 }
2887
2888 static int mdc_resource_inode_free(struct ldlm_resource *res)
2889 {
2890         if (res->lr_lvb_inode)
2891                 res->lr_lvb_inode = NULL;
2892
2893         return 0;
2894 }
2895
2896 struct ldlm_valblock_ops inode_lvbo = {
2897         .lvbo_free = mdc_resource_inode_free
2898 };
2899
2900 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
2901 {
2902         struct client_obd               *cli = &obd->u.cli;
2903         int                             rc;
2904         ENTRY;
2905
2906         OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
2907         if (!cli->cl_rpc_lock)
2908                 RETURN(-ENOMEM);
2909         mdc_init_rpc_lock(cli->cl_rpc_lock);
2910
2911         rc = ptlrpcd_addref();
2912         if (rc < 0)
2913                 GOTO(err_rpc_lock, rc);
2914
2915         OBD_ALLOC(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
2916         if (!cli->cl_close_lock)
2917                 GOTO(err_ptlrpcd_decref, rc = -ENOMEM);
2918         mdc_init_rpc_lock(cli->cl_close_lock);
2919
2920         rc = client_obd_setup(obd, cfg);
2921         if (rc)
2922                 GOTO(err_close_lock, rc);
2923 #ifdef LPROCFS
2924         obd->obd_vars = lprocfs_mdc_obd_vars;
2925         lprocfs_seq_obd_setup(obd);
2926         lprocfs_alloc_md_stats(obd, 0);
2927 #endif
2928         sptlrpc_lprocfs_cliobd_attach(obd);
2929         ptlrpc_lprocfs_register_obd(obd);
2930
2931         ns_register_cancel(obd->obd_namespace, mdc_cancel_weight);
2932
2933         obd->obd_namespace->ns_lvbo = &inode_lvbo;
2934
2935         rc = obd_llog_init(obd, &obd->obd_olg, obd, NULL);
2936         if (rc) {
2937                 mdc_cleanup(obd);
2938                 CERROR("failed to setup llogging subsystems\n");
2939         }
2940
2941         RETURN(rc);
2942
2943 err_close_lock:
2944         OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
2945 err_ptlrpcd_decref:
2946         ptlrpcd_decref();
2947 err_rpc_lock:
2948         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
2949         RETURN(rc);
2950 }
2951
2952 /* Initialize the default and maximum LOV EA and cookie sizes.  This allows
2953  * us to make MDS RPCs with large enough reply buffers to hold a default
2954  * sized EA and cookie without having to calculate this (via a call into the
2955  * LOV + OSCs) each time we make an RPC.  The maximum size is also tracked
2956  * but not used to avoid wastefully vmalloc()'ing large reply buffers when
2957  * a large number of stripes is possible.  If a larger reply buffer is
2958  * required it will be reallocated in the ptlrpc layer due to overflow.
2959  */
2960 static int mdc_init_ea_size(struct obd_export *exp, int easize,
2961                             int def_easize, int cookiesize, int def_cookiesize)
2962 {
2963         struct obd_device *obd = exp->exp_obd;
2964         struct client_obd *cli = &obd->u.cli;
2965         ENTRY;
2966
2967         if (cli->cl_max_mds_easize < easize)
2968                 cli->cl_max_mds_easize = easize;
2969
2970         if (cli->cl_default_mds_easize < def_easize)
2971                 cli->cl_default_mds_easize = def_easize;
2972
2973         if (cli->cl_max_mds_cookiesize < cookiesize)
2974                 cli->cl_max_mds_cookiesize = cookiesize;
2975
2976         if (cli->cl_default_mds_cookiesize < def_cookiesize)
2977                 cli->cl_default_mds_cookiesize = def_cookiesize;
2978
2979         RETURN(0);
2980 }
2981
2982 static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2983 {
2984         int rc = 0;
2985         ENTRY;
2986
2987         switch (stage) {
2988         case OBD_CLEANUP_EARLY:
2989                 break;
2990         case OBD_CLEANUP_EXPORTS:
2991                 /* Failsafe, ok if racy */
2992                 if (obd->obd_type->typ_refcnt <= 1)
2993                         libcfs_kkuc_group_rem(0, KUC_GRP_HSM, NULL);
2994
2995                 obd_cleanup_client_import(obd);
2996                 ptlrpc_lprocfs_unregister_obd(obd);
2997                 lprocfs_obd_cleanup(obd);
2998                 lprocfs_free_md_stats(obd);
2999
3000                 rc = obd_llog_finish(obd, 0);
3001                 if (rc != 0)
3002                         CERROR("failed to cleanup llogging subsystems\n");
3003                 break;
3004         }
3005         RETURN(rc);
3006 }
3007
3008 static int mdc_cleanup(struct obd_device *obd)
3009 {
3010         struct client_obd *cli = &obd->u.cli;
3011
3012         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
3013         OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
3014
3015         ptlrpcd_decref();
3016
3017         return client_obd_cleanup(obd);
3018 }
3019
3020
3021 static int mdc_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
3022                          struct obd_device *tgt, int *index)
3023 {
3024         struct llog_ctxt        *ctxt;
3025         int                      rc;
3026
3027         ENTRY;
3028
3029         LASSERT(olg == &obd->obd_olg);
3030
3031         rc = llog_setup(NULL, obd, olg, LLOG_CHANGELOG_REPL_CTXT, tgt,
3032                         &llog_client_ops);
3033         if (rc)
3034                 RETURN(rc);
3035
3036         ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
3037         llog_initiator_connect(ctxt);
3038         llog_ctxt_put(ctxt);
3039
3040         RETURN(0);
3041 }
3042
3043 static int mdc_llog_finish(struct obd_device *obd, int count)
3044 {
3045         struct llog_ctxt *ctxt;
3046
3047         ENTRY;
3048
3049         ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
3050         if (ctxt)
3051                 llog_cleanup(NULL, ctxt);
3052
3053         RETURN(0);
3054 }
3055
3056 static int mdc_process_config(struct obd_device *obd, obd_count len, void *buf)
3057 {
3058         struct lustre_cfg *lcfg = buf;
3059         int rc = class_process_proc_seq_param(PARAM_MDC, obd->obd_vars,
3060                                               lcfg, obd);
3061         return (rc > 0 ? 0: rc);
3062 }
3063
3064
3065 /* get remote permission for current user on fid */
3066 int mdc_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
3067                         struct obd_capa *oc, __u32 suppgid,
3068                         struct ptlrpc_request **request)
3069 {
3070         struct ptlrpc_request  *req;
3071         int                    rc;
3072         ENTRY;
3073
3074         LASSERT(client_is_remote(exp));
3075
3076         *request = NULL;
3077         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
3078         if (req == NULL)
3079                 RETURN(-ENOMEM);
3080
3081         mdc_set_capa_size(req, &RMF_CAPA1, oc);
3082
3083         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
3084         if (rc) {
3085                 ptlrpc_request_free(req);
3086                 RETURN(rc);
3087         }
3088
3089         mdc_pack_body(req, fid, oc, OBD_MD_FLRMTPERM, 0, suppgid, 0);
3090
3091         req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
3092                              sizeof(struct mdt_remote_perm));
3093
3094         ptlrpc_request_set_replen(req);
3095
3096         rc = ptlrpc_queue_wait(req);
3097         if (rc)
3098                 ptlrpc_req_finished(req);
3099         else
3100                 *request = req;
3101         RETURN(rc);
3102 }
3103
3104 static int mdc_interpret_renew_capa(const struct lu_env *env,
3105                                     struct ptlrpc_request *req, void *args,
3106                                     int status)
3107 {
3108         struct mdc_renew_capa_args *ra = args;
3109         struct mdt_body *body = NULL;
3110         struct lustre_capa *capa;
3111         ENTRY;
3112
3113         if (status)
3114                 GOTO(out, capa = ERR_PTR(status));
3115
3116         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
3117         if (body == NULL)
3118                 GOTO(out, capa = ERR_PTR(-EFAULT));
3119
3120         if ((body->mbo_valid & OBD_MD_FLOSSCAPA) == 0)
3121                 GOTO(out, capa = ERR_PTR(-ENOENT));
3122
3123         capa = req_capsule_server_get(&req->rq_pill, &RMF_CAPA2);
3124         if (!capa)
3125                 GOTO(out, capa = ERR_PTR(-EFAULT));
3126         EXIT;
3127 out:
3128         ra->ra_cb(ra->ra_oc, capa);
3129         return 0;
3130 }
3131
3132 static int mdc_renew_capa(struct obd_export *exp, struct obd_capa *oc,
3133                           renew_capa_cb_t cb)
3134 {
3135         struct ptlrpc_request *req;
3136         struct mdc_renew_capa_args *ra;
3137         ENTRY;
3138
3139         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_GETATTR,
3140                                         LUSTRE_MDS_VERSION, MDS_GETATTR);
3141         if (req == NULL)
3142                 RETURN(-ENOMEM);
3143
3144         /* NB, OBD_MD_FLOSSCAPA is set here, but it doesn't necessarily mean the
3145          * capa to renew is oss capa.
3146          */
3147         mdc_pack_body(req, &oc->c_capa.lc_fid, oc, OBD_MD_FLOSSCAPA, 0, -1, 0);
3148         ptlrpc_request_set_replen(req);
3149
3150         CLASSERT(sizeof(*ra) <= sizeof(req->rq_async_args));
3151         ra = ptlrpc_req_async_args(req);
3152         ra->ra_oc = oc;
3153         ra->ra_cb = cb;
3154         req->rq_interpret_reply = mdc_interpret_renew_capa;
3155         ptlrpcd_add_req(req, PDL_POLICY_LOCAL, -1);
3156         RETURN(0);
3157 }
3158
3159 struct obd_ops mdc_obd_ops = {
3160         .o_owner            = THIS_MODULE,
3161         .o_setup            = mdc_setup,
3162         .o_precleanup       = mdc_precleanup,
3163         .o_cleanup          = mdc_cleanup,
3164         .o_add_conn         = client_import_add_conn,
3165         .o_del_conn         = client_import_del_conn,
3166         .o_connect          = client_connect_import,
3167         .o_disconnect       = client_disconnect_export,
3168         .o_iocontrol        = mdc_iocontrol,
3169         .o_set_info_async   = mdc_set_info_async,
3170         .o_statfs           = mdc_statfs,
3171         .o_fid_init         = client_fid_init,
3172         .o_fid_fini         = client_fid_fini,
3173         .o_fid_alloc        = mdc_fid_alloc,
3174         .o_import_event     = mdc_import_event,
3175         .o_llog_init        = mdc_llog_init,
3176         .o_llog_finish      = mdc_llog_finish,
3177         .o_get_info         = mdc_get_info,
3178         .o_process_config   = mdc_process_config,
3179         .o_get_uuid         = mdc_get_uuid,
3180         .o_quotactl         = mdc_quotactl,
3181         .o_quotacheck       = mdc_quotacheck
3182 };
3183
3184 struct md_ops mdc_md_ops = {
3185         .m_getstatus        = mdc_getstatus,
3186         .m_null_inode       = mdc_null_inode,
3187         .m_find_cbdata      = mdc_find_cbdata,
3188         .m_close            = mdc_close,
3189         .m_create           = mdc_create,
3190         .m_done_writing     = mdc_done_writing,
3191         .m_enqueue          = mdc_enqueue,
3192         .m_getattr          = mdc_getattr,
3193         .m_getattr_name     = mdc_getattr_name,
3194         .m_intent_lock      = mdc_intent_lock,
3195         .m_link             = mdc_link,
3196         .m_rename           = mdc_rename,
3197         .m_setattr          = mdc_setattr,
3198         .m_setxattr         = mdc_setxattr,
3199         .m_getxattr         = mdc_getxattr,
3200         .m_fsync                = mdc_fsync,
3201         .m_read_page            = mdc_read_page,
3202         .m_unlink           = mdc_unlink,
3203         .m_cancel_unused    = mdc_cancel_unused,
3204         .m_init_ea_size     = mdc_init_ea_size,
3205         .m_set_lock_data    = mdc_set_lock_data,
3206         .m_lock_match       = mdc_lock_match,
3207         .m_get_lustre_md    = mdc_get_lustre_md,
3208         .m_free_lustre_md   = mdc_free_lustre_md,
3209         .m_set_open_replay_data = mdc_set_open_replay_data,
3210         .m_clear_open_replay_data = mdc_clear_open_replay_data,
3211         .m_renew_capa       = mdc_renew_capa,
3212         .m_unpack_capa      = mdc_unpack_capa,
3213         .m_get_remote_perm  = mdc_get_remote_perm,
3214         .m_intent_getattr_async = mdc_intent_getattr_async,
3215         .m_revalidate_lock      = mdc_revalidate_lock
3216 };
3217
3218 int __init mdc_init(void)
3219 {
3220         return class_register_type(&mdc_obd_ops, &mdc_md_ops, true, NULL,
3221 #ifndef HAVE_ONLY_PROCFS_SEQ
3222                                    NULL,
3223 #endif
3224                                    LUSTRE_MDC_NAME, NULL);
3225 }
3226
3227 #ifdef __KERNEL__
3228 static void /*__exit*/ mdc_exit(void)
3229 {
3230         class_unregister_type(LUSTRE_MDC_NAME);
3231 }
3232
3233 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3234 MODULE_DESCRIPTION("Lustre Metadata Client");
3235 MODULE_LICENSE("GPL");
3236
3237 module_init(mdc_init);
3238 module_exit(mdc_exit);
3239 #endif