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, 2014, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36
  37 #define DEBUG_SUBSYSTEM S_MDC
  38
  39 #include <linux/module.h>
  40 #include <linux/pagemap.h>
  41 #include <linux/miscdevice.h>
  42 #include <linux/init.h>
  43 #include <linux/utsname.h>
  44 #include <linux/kthread.h>
  45 #include <linux/user_namespace.h>
  46 #ifdef HAVE_UIDGID_HEADER
  47 # include <linux/uidgid.h>
  48 #endif
  49
  50 #include <lustre_acl.h>
  51 #include <lustre_ioctl.h>
  52 #include <obd_class.h>
  53 #include <lustre_lmv.h>
  54 #include <lustre_fid.h>
  55 #include <lprocfs_status.h>
  56 #include <lustre_param.h>
  57 #include <lustre_log.h>
  58 #include <lustre_kernelcomm.h>
  59 #include <cl_object.h>
  60
  61 #include "mdc_internal.h"
  62
  63 #define REQUEST_MINOR 244
  64
  65 struct mdc_renew_capa_args {
  66         struct obd_capa        *ra_oc;
  67         renew_capa_cb_t         ra_cb;
  68 };
  69
  70 static int mdc_cleanup(struct obd_device *obd);
  71
  72 static int mdc_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
  73                            const struct req_msg_field *field,
  74                            struct obd_capa **oc)
  75 {
  76         struct lustre_capa *capa;
  77         struct obd_capa *c;
  78         ENTRY;
  79
  80         /* swabbed already in mdc_enqueue */
  81         capa = req_capsule_server_get(&req->rq_pill, field);
  82         if (capa == NULL)
  83                 RETURN(-EPROTO);
  84
  85         c = alloc_capa(CAPA_SITE_CLIENT);
  86         if (IS_ERR(c)) {
  87                 CDEBUG(D_INFO, "alloc capa failed!\n");
  88                 RETURN(PTR_ERR(c));
  89         } else {
  90                 c->c_capa = *capa;
  91                 *oc = c;
  92                 RETURN(0);
  93         }
  94 }
  95
  96 static inline int mdc_queue_wait(struct ptlrpc_request *req)
  97 {
  98         struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
  99         int rc;
 100
 101         /* obd_get_request_slot() ensures that this client has no more
 102          * than cl_max_rpcs_in_flight RPCs simultaneously inf light
 103          * against an MDT. */
 104         rc = obd_get_request_slot(cli);
 105         if (rc != 0)
 106                 return rc;
 107
 108         rc = ptlrpc_queue_wait(req);
 109         obd_put_request_slot(cli);
 110
 111         return rc;
 112 }
 113
 114 /* Helper that implements most of mdc_getstatus and signal_completed_replay. */
 115 /* XXX this should become mdc_get_info("key"), sending MDS_GET_INFO RPC */
 116 static int send_getstatus(struct obd_import *imp, struct lu_fid *rootfid,
 117                           struct obd_capa **pc, int level, int msg_flags)
 118 {
 119         struct ptlrpc_request *req;
 120         struct mdt_body       *body;
 121         int                    rc;
 122         ENTRY;
 123
 124         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_GETSTATUS,
 125                                         LUSTRE_MDS_VERSION, MDS_GETSTATUS);
 126         if (req == NULL)
 127                 RETURN(-ENOMEM);
 128
 129         mdc_pack_body(req, NULL, NULL, 0, 0, -1, 0);
 130         lustre_msg_add_flags(req->rq_reqmsg, msg_flags);
 131         req->rq_send_state = level;
 132
 133         ptlrpc_request_set_replen(req);
 134
 135         rc = ptlrpc_queue_wait(req);
 136         if (rc)
 137                 GOTO(out, rc);
 138
 139         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 140         if (body == NULL)
 141                 GOTO(out, rc = -EPROTO);
 142
 143         if (body->mbo_valid & OBD_MD_FLMDSCAPA) {
 144                 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA1, pc);
 145                 if (rc)
 146                         GOTO(out, rc);
 147         }
 148
 149         *rootfid = body->mbo_fid1;
 150         CDEBUG(D_NET, "root fid="DFID", last_committed="LPU64"\n",
 151                PFID(rootfid), lustre_msg_get_last_committed(req->rq_repmsg));
 152         EXIT;
 153 out:
 154         ptlrpc_req_finished(req);
 155         return rc;
 156 }
 157
 158 /* This should be mdc_get_info("rootfid") */
 159 static int mdc_getstatus(struct obd_export *exp, struct lu_fid *rootfid,
 160                          struct obd_capa **pc)
 161 {
 162         return send_getstatus(class_exp2cliimp(exp), rootfid, pc,
 163                               LUSTRE_IMP_FULL, 0);
 164 }
 165
 166 /*
 167  * This function now is known to always saying that it will receive 4 buffers
 168  * from server. Even for cases when acl_size and md_size is zero, RPC header
 169  * will contain 4 fields and RPC itself will contain zero size fields. This is
 170  * because mdt_getattr*() _always_ returns 4 fields, but if acl is not needed
 171  * and thus zero, it shrinks it, making zero size. The same story about
 172  * md_size. And this is course of problem when client waits for smaller number
 173  * of fields. This issue will be fixed later when client gets aware of RPC
 174  * layouts.  --umka
 175  */
 176 static int mdc_getattr_common(struct obd_export *exp,
 177                               struct ptlrpc_request *req)
 178 {
 179         struct req_capsule *pill = &req->rq_pill;
 180         struct mdt_body    *body;
 181         void               *eadata;
 182         int                 rc;
 183         ENTRY;
 184
 185         /* Request message already built. */
 186         rc = ptlrpc_queue_wait(req);
 187         if (rc != 0)
 188                 RETURN(rc);
 189
 190         /* sanity check for the reply */
 191         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 192         if (body == NULL)
 193                 RETURN(-EPROTO);
 194
 195         CDEBUG(D_NET, "mode: %o\n", body->mbo_mode);
 196
 197         mdc_update_max_ea_from_body(exp, body);
 198         if (body->mbo_eadatasize != 0) {
 199                 eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
 200                                                       body->mbo_eadatasize);
 201                 if (eadata == NULL)
 202                         RETURN(-EPROTO);
 203         }
 204
 205         if (body->mbo_valid & OBD_MD_FLRMTPERM) {
 206                 struct mdt_remote_perm *perm;
 207
 208                 LASSERT(client_is_remote(exp));
 209                 perm = req_capsule_server_swab_get(pill, &RMF_ACL,
 210                                                 lustre_swab_mdt_remote_perm);
 211                 if (perm == NULL)
 212                         RETURN(-EPROTO);
 213         }
 214
 215         if (body->mbo_valid & OBD_MD_FLMDSCAPA) {
 216                 struct lustre_capa *capa;
 217                 capa = req_capsule_server_get(pill, &RMF_CAPA1);
 218                 if (capa == NULL)
 219                         RETURN(-EPROTO);
 220         }
 221
 222         RETURN(0);
 223 }
 224
 225 static int mdc_getattr(struct obd_export *exp, struct md_op_data *op_data,
 226                        struct ptlrpc_request **request)
 227 {
 228         struct ptlrpc_request *req;
 229         int                    rc;
 230         ENTRY;
 231
 232         /* Single MDS without an LMV case */
 233         if (op_data->op_flags & MF_GET_MDT_IDX) {
 234                 op_data->op_mds = 0;
 235                 RETURN(0);
 236         }
 237         *request = NULL;
 238         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
 239         if (req == NULL)
 240                 RETURN(-ENOMEM);
 241
 242         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 243
 244         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
 245         if (rc) {
 246                 ptlrpc_request_free(req);
 247                 RETURN(rc);
 248         }
 249
 250         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
 251                       op_data->op_valid, op_data->op_mode, -1, 0);
 252
 253         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 254                              op_data->op_mode);
 255         if (op_data->op_valid & OBD_MD_FLRMTPERM) {
 256                 LASSERT(client_is_remote(exp));
 257                 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
 258                                      sizeof(struct mdt_remote_perm));
 259         }
 260         ptlrpc_request_set_replen(req);
 261
 262         rc = mdc_getattr_common(exp, req);
 263         if (rc)
 264                 ptlrpc_req_finished(req);
 265         else
 266                 *request = req;
 267         RETURN(rc);
 268 }
 269
 270 static int mdc_getattr_name(struct obd_export *exp, struct md_op_data *op_data,
 271                             struct ptlrpc_request **request)
 272 {
 273         struct ptlrpc_request *req;
 274         int                    rc;
 275         ENTRY;
 276
 277         *request = NULL;
 278         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 279                                    &RQF_MDS_GETATTR_NAME);
 280         if (req == NULL)
 281                 RETURN(-ENOMEM);
 282
 283         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 284         req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 285                              op_data->op_namelen + 1);
 286
 287         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR_NAME);
 288         if (rc) {
 289                 ptlrpc_request_free(req);
 290                 RETURN(rc);
 291         }
 292
 293         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
 294                       op_data->op_valid, op_data->op_mode,
 295                       op_data->op_suppgids[0], 0);
 296
 297         if (op_data->op_name) {
 298                 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 299                 LASSERT(strnlen(op_data->op_name, op_data->op_namelen) ==
 300                                 op_data->op_namelen);
 301                 memcpy(name, op_data->op_name, op_data->op_namelen);
 302         }
 303
 304         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 305                              op_data->op_mode);
 306         ptlrpc_request_set_replen(req);
 307
 308         rc = mdc_getattr_common(exp, req);
 309         if (rc)
 310                 ptlrpc_req_finished(req);
 311         else
 312                 *request = req;
 313         RETURN(rc);
 314 }
 315
 316 static int mdc_xattr_common(struct obd_export *exp,const struct req_format *fmt,
 317                             const struct lu_fid *fid,
 318                             struct obd_capa *oc, int opcode, u64 valid,
 319                             const char *xattr_name, const char *input,
 320                             int input_size, int output_size, int flags,
 321                             __u32 suppgid, struct ptlrpc_request **request)
 322 {
 323         struct ptlrpc_request *req;
 324         int   xattr_namelen = 0;
 325         char *tmp;
 326         int   rc;
 327         ENTRY;
 328
 329         *request = NULL;
 330         req = ptlrpc_request_alloc(class_exp2cliimp(exp), fmt);
 331         if (req == NULL)
 332                 RETURN(-ENOMEM);
 333
 334         mdc_set_capa_size(req, &RMF_CAPA1, oc);
 335         if (xattr_name) {
 336                 xattr_namelen = strlen(xattr_name) + 1;
 337                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 338                                      xattr_namelen);
 339         }
 340         if (input_size) {
 341                 LASSERT(input);
 342                 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
 343                                      input_size);
 344         }
 345
 346         /* Flush local XATTR locks to get rid of a possible cancel RPC */
 347         if (opcode == MDS_REINT && fid_is_sane(fid) &&
 348             exp->exp_connect_data.ocd_ibits_known & MDS_INODELOCK_XATTR) {
 349                 struct list_head cancels = LIST_HEAD_INIT(cancels);
 350                 int count;
 351
 352                 /* Without that packing would fail */
 353                 if (input_size == 0)
 354                         req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
 355                                              RCL_CLIENT, 0);
 356
 357                 count = mdc_resource_get_unused(exp, fid,
 358                                                 &cancels, LCK_EX,
 359                                                 MDS_INODELOCK_XATTR);
 360
 361                 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
 362                 if (rc) {
 363                         ptlrpc_request_free(req);
 364                         RETURN(rc);
 365                 }
 366         } else {
 367                 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode);
 368                 if (rc) {
 369                         ptlrpc_request_free(req);
 370                         RETURN(rc);
 371                 }
 372         }
 373
 374         if (opcode == MDS_REINT) {
 375                 struct mdt_rec_setxattr *rec;
 376
 377                 CLASSERT(sizeof(struct mdt_rec_setxattr) ==
 378                          sizeof(struct mdt_rec_reint));
 379                 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
 380                 rec->sx_opcode = REINT_SETXATTR;
 381                 rec->sx_fsuid  = from_kuid(&init_user_ns, current_fsuid());
 382                 rec->sx_fsgid  = from_kgid(&init_user_ns, current_fsgid());
 383                 rec->sx_cap    = cfs_curproc_cap_pack();
 384                 rec->sx_suppgid1 = suppgid;
 385                 rec->sx_suppgid2 = -1;
 386                 rec->sx_fid    = *fid;
 387                 rec->sx_valid  = valid | OBD_MD_FLCTIME;
 388                 rec->sx_time   = cfs_time_current_sec();
 389                 rec->sx_size   = output_size;
 390                 rec->sx_flags  = flags;
 391
 392                 mdc_pack_capa(req, &RMF_CAPA1, oc);
 393         } else {
 394                 mdc_pack_body(req, fid, oc, valid, output_size, suppgid, flags);
 395         }
 396
 397         if (xattr_name) {
 398                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 399                 memcpy(tmp, xattr_name, xattr_namelen);
 400         }
 401         if (input_size) {
 402                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
 403                 memcpy(tmp, input, input_size);
 404         }
 405
 406         if (req_capsule_has_field(&req->rq_pill, &RMF_EADATA, RCL_SERVER))
 407                 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
 408                                      RCL_SERVER, output_size);
 409         ptlrpc_request_set_replen(req);
 410
 411         /* make rpc */
 412         if (opcode == MDS_REINT)
 413                 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 414
 415         rc = ptlrpc_queue_wait(req);
 416
 417         if (opcode == MDS_REINT)
 418                 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 419
 420         if (rc)
 421                 ptlrpc_req_finished(req);
 422         else
 423                 *request = req;
 424         RETURN(rc);
 425 }
 426
 427 static int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
 428                         struct obd_capa *oc, u64 valid,
 429                         const char *xattr_name,
 430                         const char *input, int input_size, int output_size,
 431                         int flags, __u32 suppgid,
 432                         struct ptlrpc_request **request)
 433 {
 434         return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
 435                                 fid, oc, MDS_REINT, valid, xattr_name,
 436                                 input, input_size, output_size, flags,
 437                                 suppgid, request);
 438 }
 439
 440 static int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
 441                         struct obd_capa *oc, u64 valid,
 442                         const char *xattr_name,
 443                         const char *input, int input_size, int output_size,
 444                         int flags, struct ptlrpc_request **request)
 445 {
 446         return mdc_xattr_common(exp, &RQF_MDS_GETXATTR,
 447                                 fid, oc, MDS_GETXATTR, valid, xattr_name,
 448                                 input, input_size, output_size, flags,
 449                                 -1, request);
 450 }
 451
 452 #ifdef CONFIG_FS_POSIX_ACL
 453 static int mdc_unpack_acl(struct ptlrpc_request *req, struct lustre_md *md)
 454 {
 455         struct req_capsule     *pill = &req->rq_pill;
 456         struct mdt_body        *body = md->body;
 457         struct posix_acl       *acl;
 458         void                   *buf;
 459         int                     rc;
 460         ENTRY;
 461
 462         if (!body->mbo_aclsize)
 463                 RETURN(0);
 464
 465         buf = req_capsule_server_sized_get(pill, &RMF_ACL, body->mbo_aclsize);
 466
 467         if (!buf)
 468                 RETURN(-EPROTO);
 469
 470         acl = posix_acl_from_xattr(&init_user_ns, buf, body->mbo_aclsize);
 471         if (acl == NULL)
 472                 RETURN(0);
 473         if (IS_ERR(acl)) {
 474                 rc = PTR_ERR(acl);
 475                 CERROR("convert xattr to acl: %d\n", rc);
 476                 RETURN(rc);
 477         }
 478
 479         rc = posix_acl_valid(acl);
 480         if (rc) {
 481                 CERROR("validate acl: %d\n", rc);
 482                 posix_acl_release(acl);
 483                 RETURN(rc);
 484         }
 485
 486         md->posix_acl = acl;
 487         RETURN(0);
 488 }
 489 #else
 490 #define mdc_unpack_acl(req, md) 0
 491 #endif
 492
 493 int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
 494                       struct obd_export *dt_exp, struct obd_export *md_exp,
 495                       struct lustre_md *md)
 496 {
 497         struct req_capsule *pill = &req->rq_pill;
 498         int rc;
 499         ENTRY;
 500
 501         LASSERT(md);
 502         memset(md, 0, sizeof(*md));
 503
 504         md->body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 505         LASSERT(md->body != NULL);
 506
 507         if (md->body->mbo_valid & OBD_MD_FLEASIZE) {
 508                 int lmmsize;
 509                 struct lov_mds_md *lmm;
 510
 511                 if (!S_ISREG(md->body->mbo_mode)) {
 512                         CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, should be a "
 513                                "regular file, but is not\n");
 514                         GOTO(out, rc = -EPROTO);
 515                 }
 516
 517                 if (md->body->mbo_eadatasize == 0) {
 518                         CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, "
 519                                "but eadatasize 0\n");
 520                         GOTO(out, rc = -EPROTO);
 521                 }
 522
 523                 lmmsize = md->body->mbo_eadatasize;
 524                 lmm = req_capsule_server_sized_get(pill, &RMF_MDT_MD, lmmsize);
 525                 if (!lmm)
 526                         GOTO(out, rc = -EPROTO);
 527
 528                 rc = obd_unpackmd(dt_exp, &md->lsm, lmm, lmmsize);
 529                 if (rc < 0)
 530                         GOTO(out, rc);
 531
 532                 if (rc < (typeof(rc))sizeof(*md->lsm)) {
 533                         CDEBUG(D_INFO, "lsm size too small: "
 534                                "rc < sizeof (*md->lsm) (%d < %d)\n",
 535                                rc, (int)sizeof(*md->lsm));
 536                         GOTO(out, rc = -EPROTO);
 537                 }
 538
 539         } else if (md->body->mbo_valid & OBD_MD_FLDIREA) {
 540                 int lmvsize;
 541                 struct lov_mds_md *lmv;
 542
 543                 if (!S_ISDIR(md->body->mbo_mode)) {
 544                         CDEBUG(D_INFO, "OBD_MD_FLDIREA set, should be a "
 545                                "directory, but is not\n");
 546                         GOTO(out, rc = -EPROTO);
 547                 }
 548
 549                 if (md->body->mbo_eadatasize == 0) {
 550                         CDEBUG(D_INFO, "OBD_MD_FLDIREA is set, "
 551                                "but eadatasize 0\n");
 552                         RETURN(-EPROTO);
 553                 }
 554
 555                 if (md->body->mbo_valid & OBD_MD_MEA) {
 556                         lmvsize = md->body->mbo_eadatasize;
 557                         lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
 558                                                            lmvsize);
 559                         if (!lmv)
 560                                 GOTO(out, rc = -EPROTO);
 561
 562                         rc = obd_unpackmd(md_exp, (void *)&md->lmv, lmv,
 563                                           lmvsize);
 564                         if (rc < 0)
 565                                 GOTO(out, rc);
 566
 567                         if (rc < (typeof(rc))sizeof(*md->lmv)) {
 568                                 CDEBUG(D_INFO, "size too small:  "
 569                                        "rc < sizeof(*md->lmv) (%d < %d)\n",
 570                                         rc, (int)sizeof(*md->lmv));
 571                                 GOTO(out, rc = -EPROTO);
 572                         }
 573                 }
 574         }
 575         rc = 0;
 576
 577         if (md->body->mbo_valid & OBD_MD_FLRMTPERM) {
 578                 /* remote permission */
 579                 LASSERT(client_is_remote(exp));
 580                 md->remote_perm = req_capsule_server_swab_get(pill, &RMF_ACL,
 581                                                 lustre_swab_mdt_remote_perm);
 582                 if (!md->remote_perm)
 583                         GOTO(out, rc = -EPROTO);
 584         } else if (md->body->mbo_valid & OBD_MD_FLACL) {
 585                 /* for ACL, it's possible that FLACL is set but aclsize is zero.
 586                  * only when aclsize != 0 there's an actual segment for ACL
 587                  * in reply buffer.
 588                  */
 589                 if (md->body->mbo_aclsize) {
 590                         rc = mdc_unpack_acl(req, md);
 591                         if (rc)
 592                                 GOTO(out, rc);
 593 #ifdef CONFIG_FS_POSIX_ACL
 594                 } else {
 595                         md->posix_acl = NULL;
 596 #endif
 597                 }
 598         }
 599         if (md->body->mbo_valid & OBD_MD_FLMDSCAPA) {
 600                 struct obd_capa *oc = NULL;
 601
 602                 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA1, &oc);
 603                 if (rc)
 604                         GOTO(out, rc);
 605                 md->mds_capa = oc;
 606         }
 607
 608         if (md->body->mbo_valid & OBD_MD_FLOSSCAPA) {
 609                 struct obd_capa *oc = NULL;
 610
 611                 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA2, &oc);
 612                 if (rc)
 613                         GOTO(out, rc);
 614                 md->oss_capa = oc;
 615         }
 616
 617         EXIT;
 618 out:
 619         if (rc) {
 620                 if (md->oss_capa) {
 621                         capa_put(md->oss_capa);
 622                         md->oss_capa = NULL;
 623                 }
 624                 if (md->mds_capa) {
 625                         capa_put(md->mds_capa);
 626                         md->mds_capa = NULL;
 627                 }
 628 #ifdef CONFIG_FS_POSIX_ACL
 629                 posix_acl_release(md->posix_acl);
 630 #endif
 631                 if (md->lsm)
 632                         obd_free_memmd(dt_exp, &md->lsm);
 633         }
 634         return rc;
 635 }
 636
 637 int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
 638 {
 639         ENTRY;
 640         RETURN(0);
 641 }
 642
 643 void mdc_replay_open(struct ptlrpc_request *req)
 644 {
 645         struct md_open_data *mod = req->rq_cb_data;
 646         struct ptlrpc_request *close_req;
 647         struct obd_client_handle *och;
 648         struct lustre_handle old;
 649         struct mdt_body *body;
 650         ENTRY;
 651
 652         if (mod == NULL) {
 653                 DEBUG_REQ(D_ERROR, req,
 654                           "Can't properly replay without open data.");
 655                 EXIT;
 656                 return;
 657         }
 658
 659         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 660         LASSERT(body != NULL);
 661
 662         och = mod->mod_och;
 663         if (och != NULL) {
 664                 struct lustre_handle *file_fh;
 665
 666                 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
 667
 668                 file_fh = &och->och_fh;
 669                 CDEBUG(D_HA, "updating handle from "LPX64" to "LPX64"\n",
 670                        file_fh->cookie, body->mbo_handle.cookie);
 671                 old = *file_fh;
 672                 *file_fh = body->mbo_handle;
 673         }
 674         close_req = mod->mod_close_req;
 675         if (close_req != NULL) {
 676                 __u32 opc = lustre_msg_get_opc(close_req->rq_reqmsg);
 677                 struct mdt_ioepoch *epoch;
 678
 679                 LASSERT(opc == MDS_CLOSE);
 680                 epoch = req_capsule_client_get(&close_req->rq_pill,
 681                                                &RMF_MDT_EPOCH);
 682                 LASSERT(epoch);
 683
 684                 if (och != NULL)
 685                         LASSERT(!memcmp(&old, &epoch->mio_handle, sizeof(old)));
 686
 687                 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
 688                 epoch->mio_handle = body->mbo_handle;
 689         }
 690         EXIT;
 691 }
 692
 693 void mdc_commit_open(struct ptlrpc_request *req)
 694 {
 695         struct md_open_data *mod = req->rq_cb_data;
 696         if (mod == NULL)
 697                 return;
 698
 699         /**
 700          * No need to touch md_open_data::mod_och, it holds a reference on
 701          * \var mod and will zero references to each other, \var mod will be
 702          * freed after that when md_open_data::mod_och will put the reference.
 703          */
 704
 705         /**
 706          * Do not let open request to disappear as it still may be needed
 707          * for close rpc to happen (it may happen on evict only, otherwise
 708          * ptlrpc_request::rq_replay does not let mdc_commit_open() to be
 709          * called), just mark this rpc as committed to distinguish these 2
 710          * cases, see mdc_close() for details. The open request reference will
 711          * be put along with freeing \var mod.
 712          */
 713         ptlrpc_request_addref(req);
 714         spin_lock(&req->rq_lock);
 715         req->rq_committed = 1;
 716         spin_unlock(&req->rq_lock);
 717         req->rq_cb_data = NULL;
 718         obd_mod_put(mod);
 719 }
 720
 721 int mdc_set_open_replay_data(struct obd_export *exp,
 722                              struct obd_client_handle *och,
 723                              struct lookup_intent *it)
 724 {
 725         struct md_open_data     *mod;
 726         struct mdt_rec_create   *rec;
 727         struct mdt_body         *body;
 728         struct ptlrpc_request   *open_req = it->d.lustre.it_data;
 729         struct obd_import       *imp = open_req->rq_import;
 730         ENTRY;
 731
 732         if (!open_req->rq_replay)
 733                 RETURN(0);
 734
 735         rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
 736         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
 737         LASSERT(rec != NULL);
 738         /* Incoming message in my byte order (it's been swabbed). */
 739         /* Outgoing messages always in my byte order. */
 740         LASSERT(body != NULL);
 741
 742         /* Only if the import is replayable, we set replay_open data */
 743         if (och && imp->imp_replayable) {
 744                 mod = obd_mod_alloc();
 745                 if (mod == NULL) {
 746                         DEBUG_REQ(D_ERROR, open_req,
 747                                   "Can't allocate md_open_data");
 748                         RETURN(0);
 749                 }
 750
 751                 /**
 752                  * Take a reference on \var mod, to be freed on mdc_close().
 753                  * It protects \var mod from being freed on eviction (commit
 754                  * callback is called despite rq_replay flag).
 755                  * Another reference for \var och.
 756                  */
 757                 obd_mod_get(mod);
 758                 obd_mod_get(mod);
 759
 760                 spin_lock(&open_req->rq_lock);
 761                 och->och_mod = mod;
 762                 mod->mod_och = och;
 763                 mod->mod_is_create = it_disposition(it, DISP_OPEN_CREATE) ||
 764                                      it_disposition(it, DISP_OPEN_STRIPE);
 765                 mod->mod_open_req = open_req;
 766                 open_req->rq_cb_data = mod;
 767                 open_req->rq_commit_cb = mdc_commit_open;
 768                 spin_unlock(&open_req->rq_lock);
 769         }
 770
 771         rec->cr_fid2 = body->mbo_fid1;
 772         rec->cr_ioepoch = body->mbo_ioepoch;
 773         rec->cr_old_handle.cookie = body->mbo_handle.cookie;
 774         open_req->rq_replay_cb = mdc_replay_open;
 775         if (!fid_is_sane(&body->mbo_fid1)) {
 776                 DEBUG_REQ(D_ERROR, open_req, "Saving replay request with "
 777                           "insane fid");
 778                 LBUG();
 779         }
 780
 781         DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
 782         RETURN(0);
 783 }
 784
 785 static void mdc_free_open(struct md_open_data *mod)
 786 {
 787         int committed = 0;
 788
 789         if (mod->mod_is_create == 0 &&
 790             imp_connect_disp_stripe(mod->mod_open_req->rq_import))
 791                 committed = 1;
 792
 793         LASSERT(mod->mod_open_req->rq_replay == 0);
 794
 795         DEBUG_REQ(D_RPCTRACE, mod->mod_open_req, "free open request\n");
 796
 797         ptlrpc_request_committed(mod->mod_open_req, committed);
 798         if (mod->mod_close_req)
 799                 ptlrpc_request_committed(mod->mod_close_req, committed);
 800 }
 801
 802 int mdc_clear_open_replay_data(struct obd_export *exp,
 803                                struct obd_client_handle *och)
 804 {
 805         struct md_open_data *mod = och->och_mod;
 806         ENTRY;
 807
 808         /**
 809          * It is possible to not have \var mod in a case of eviction between
 810          * lookup and ll_file_open().
 811          **/
 812         if (mod == NULL)
 813                 RETURN(0);
 814
 815         LASSERT(mod != LP_POISON);
 816         LASSERT(mod->mod_open_req != NULL);
 817         mdc_free_open(mod);
 818
 819         mod->mod_och = NULL;
 820         och->och_mod = NULL;
 821         obd_mod_put(mod);
 822
 823         RETURN(0);
 824 }
 825
 826 static int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
 827                      struct md_open_data *mod, struct ptlrpc_request **request)
 828 {
 829         struct obd_device     *obd = class_exp2obd(exp);
 830         struct ptlrpc_request *req;
 831         struct req_format     *req_fmt;
 832         int                    rc;
 833         int                    saved_rc = 0;
 834         ENTRY;
 835
 836         req_fmt = &RQF_MDS_CLOSE;
 837         if (op_data->op_bias & MDS_HSM_RELEASE) {
 838                 req_fmt = &RQF_MDS_RELEASE_CLOSE;
 839
 840                 /* allocate a FID for volatile file */
 841                 rc = mdc_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
 842                 if (rc < 0) {
 843                         CERROR("%s: "DFID" failed to allocate FID: %d\n",
 844                                obd->obd_name, PFID(&op_data->op_fid1), rc);
 845                         /* save the errcode and proceed to close */
 846                         saved_rc = rc;
 847                 }
 848         }
 849
 850         *request = NULL;
 851         req = ptlrpc_request_alloc(class_exp2cliimp(exp), req_fmt);
 852         if (req == NULL)
 853                 RETURN(-ENOMEM);
 854
 855         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 856
 857         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
 858         if (rc) {
 859                 ptlrpc_request_free(req);
 860                 RETURN(rc);
 861         }
 862
 863         /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
 864          * portal whose threads are not taking any DLM locks and are therefore
 865          * always progressing */
 866         req->rq_request_portal = MDS_READPAGE_PORTAL;
 867         ptlrpc_at_set_req_timeout(req);
 868
 869         /* Ensure that this close's handle is fixed up during replay. */
 870         if (likely(mod != NULL)) {
 871                 LASSERTF(mod->mod_open_req != NULL &&
 872                          mod->mod_open_req->rq_type != LI_POISON,
 873                          "POISONED open %p!\n", mod->mod_open_req);
 874
 875                 mod->mod_close_req = req;
 876
 877                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
 878                 /* We no longer want to preserve this open for replay even
 879                  * though the open was committed. b=3632, b=3633 */
 880                 spin_lock(&mod->mod_open_req->rq_lock);
 881                 mod->mod_open_req->rq_replay = 0;
 882                 spin_unlock(&mod->mod_open_req->rq_lock);
 883         } else {
 884                  CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
 885         }
 886
 887         mdc_close_pack(req, op_data);
 888
 889         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 890                              obd->u.cli.cl_default_mds_easize);
 891         req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
 892                              obd->u.cli.cl_default_mds_cookiesize);
 893
 894         ptlrpc_request_set_replen(req);
 895
 896         mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 897         rc = ptlrpc_queue_wait(req);
 898         mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 899
 900         if (req->rq_repmsg == NULL) {
 901                 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
 902                        req->rq_status);
 903                 if (rc == 0)
 904                         rc = req->rq_status ?: -EIO;
 905         } else if (rc == 0 || rc == -EAGAIN) {
 906                 struct mdt_body *body;
 907
 908                 rc = lustre_msg_get_status(req->rq_repmsg);
 909                 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
 910                         DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
 911                                   "= %d", rc);
 912                         if (rc > 0)
 913                                 rc = -rc;
 914                 }
 915                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 916                 if (body == NULL)
 917                         rc = -EPROTO;
 918         } else if (rc == -ESTALE) {
 919                 /**
 920                  * it can be allowed error after 3633 if open was committed and
 921                  * server failed before close was sent. Let's check if mod
 922                  * exists and return no error in that case
 923                  */
 924                 if (mod) {
 925                         DEBUG_REQ(D_HA, req, "Reset ESTALE = %d", rc);
 926                         LASSERT(mod->mod_open_req != NULL);
 927                         if (mod->mod_open_req->rq_committed)
 928                                 rc = 0;
 929                 }
 930         }
 931
 932         if (mod) {
 933                 if (rc != 0)
 934                         mod->mod_close_req = NULL;
 935                 /* Since now, mod is accessed through open_req only,
 936                  * thus close req does not keep a reference on mod anymore. */
 937                 obd_mod_put(mod);
 938         }
 939         *request = req;
 940
 941         RETURN(rc < 0 ? rc : saved_rc);
 942 }
 943
 944 static int mdc_getpage(struct obd_export *exp, const struct lu_fid *fid,
 945                        __u64 offset, struct obd_capa *oc,
 946                        struct page **pages, int npages,
 947                        struct ptlrpc_request **request)
 948 {
 949         struct ptlrpc_request   *req;
 950         struct ptlrpc_bulk_desc *desc;
 951         int                      i;
 952         wait_queue_head_t        waitq;
 953         int                      resends = 0;
 954         struct l_wait_info       lwi;
 955         int                      rc;
 956         ENTRY;
 957
 958         *request = NULL;
 959         init_waitqueue_head(&waitq);
 960
 961 restart_bulk:
 962         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
 963         if (req == NULL)
 964                 RETURN(-ENOMEM);
 965
 966         mdc_set_capa_size(req, &RMF_CAPA1, oc);
 967
 968         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
 969         if (rc) {
 970                 ptlrpc_request_free(req);
 971                 RETURN(rc);
 972         }
 973
 974         req->rq_request_portal = MDS_READPAGE_PORTAL;
 975         ptlrpc_at_set_req_timeout(req);
 976
 977         desc = ptlrpc_prep_bulk_imp(req, npages, 1, BULK_PUT_SINK,
 978                                     MDS_BULK_PORTAL);
 979         if (desc == NULL) {
 980                 ptlrpc_request_free(req);
 981                 RETURN(-ENOMEM);
 982         }
 983
 984         /* NB req now owns desc and will free it when it gets freed */
 985         for (i = 0; i < npages; i++)
 986                 ptlrpc_prep_bulk_page_pin(desc, pages[i], 0, PAGE_CACHE_SIZE);
 987
 988         mdc_readdir_pack(req, offset, PAGE_CACHE_SIZE * npages, fid, oc);
 989
 990         ptlrpc_request_set_replen(req);
 991         rc = ptlrpc_queue_wait(req);
 992         if (rc) {
 993                 ptlrpc_req_finished(req);
 994                 if (rc != -ETIMEDOUT)
 995                         RETURN(rc);
 996
 997                 resends++;
 998                 if (!client_should_resend(resends, &exp->exp_obd->u.cli)) {
 999                         CERROR("%s: too many resend retries: rc = %d\n",
1000                                exp->exp_obd->obd_name, -EIO);
1001                         RETURN(-EIO);
1002                 }
1003                 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(resends), NULL, NULL,
1004                                        NULL);
1005                 l_wait_event(waitq, 0, &lwi);
1006
1007                 goto restart_bulk;
1008         }
1009
1010         rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
1011                                           req->rq_bulk->bd_nob_transferred);
1012         if (rc < 0) {
1013                 ptlrpc_req_finished(req);
1014                 RETURN(rc);
1015         }
1016
1017         if (req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK) {
1018                 CERROR("%s: unexpected bytes transferred: %d (%ld expected)\n",
1019                        exp->exp_obd->obd_name, req->rq_bulk->bd_nob_transferred,
1020                        PAGE_CACHE_SIZE * npages);
1021                 ptlrpc_req_finished(req);
1022                 RETURN(-EPROTO);
1023         }
1024
1025         *request = req;
1026         RETURN(0);
1027 }
1028
1029 static void mdc_release_page(struct page *page, int remove)
1030 {
1031         if (remove) {
1032                 lock_page(page);
1033                 if (likely(page->mapping != NULL))
1034                         truncate_complete_page(page->mapping, page);
1035                 unlock_page(page);
1036         }
1037         page_cache_release(page);
1038 }
1039
1040 static struct page *mdc_page_locate(struct address_space *mapping, __u64 *hash,
1041                                     __u64 *start, __u64 *end, int hash64)
1042 {
1043         /*
1044          * Complement of hash is used as an index so that
1045          * radix_tree_gang_lookup() can be used to find a page with starting
1046          * hash _smaller_ than one we are looking for.
1047          */
1048         unsigned long offset = hash_x_index(*hash, hash64);
1049         struct page *page;
1050         int found;
1051
1052         spin_lock_irq(&mapping->tree_lock);
1053         found = radix_tree_gang_lookup(&mapping->page_tree,
1054                                        (void **)&page, offset, 1);
1055         if (found > 0 && !radix_tree_exceptional_entry(page)) {
1056                 struct lu_dirpage *dp;
1057
1058                 page_cache_get(page);
1059                 spin_unlock_irq(&mapping->tree_lock);
1060                 /*
1061                  * In contrast to find_lock_page() we are sure that directory
1062                  * page cannot be truncated (while DLM lock is held) and,
1063                  * hence, can avoid restart.
1064                  *
1065                  * In fact, page cannot be locked here at all, because
1066                  * mdc_read_page_remote does synchronous io.
1067                  */
1068                 wait_on_page_locked(page);
1069                 if (PageUptodate(page)) {
1070                         dp = kmap(page);
1071                         if (BITS_PER_LONG == 32 && hash64) {
1072                                 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1073                                 *end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
1074                                 *hash  = *hash >> 32;
1075                         } else {
1076                                 *start = le64_to_cpu(dp->ldp_hash_start);
1077                                 *end   = le64_to_cpu(dp->ldp_hash_end);
1078                         }
1079                         if (unlikely(*start == 1 && *hash == 0))
1080                                 *hash = *start;
1081                         else
1082                                 LASSERTF(*start <= *hash, "start = "LPX64
1083                                          ",end = "LPX64",hash = "LPX64"\n",
1084                                          *start, *end, *hash);
1085                         CDEBUG(D_VFSTRACE, "offset %lx ["LPX64" "LPX64"],"
1086                               " hash "LPX64"\n", offset, *start, *end, *hash);
1087                         if (*hash > *end) {
1088                                 kunmap(page);
1089                                 mdc_release_page(page, 0);
1090                                 page = NULL;
1091                         } else if (*end != *start && *hash == *end) {
1092                                 /*
1093                                  * upon hash collision, remove this page,
1094                                  * otherwise put page reference, and
1095                                  * mdc_read_page_remote() will issue RPC to
1096                                  * fetch the page we want.
1097                                  */
1098                                 kunmap(page);
1099                                 mdc_release_page(page,
1100                                     le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
1101                                 page = NULL;
1102                         }
1103                 } else {
1104                         page_cache_release(page);
1105                         page = ERR_PTR(-EIO);
1106                 }
1107         } else {
1108                 spin_unlock_irq(&mapping->tree_lock);
1109                 page = NULL;
1110         }
1111         return page;
1112 }
1113
1114 /*
1115  * Adjust a set of pages, each page containing an array of lu_dirpages,
1116  * so that each page can be used as a single logical lu_dirpage.
1117  *
1118  * A lu_dirpage is laid out as follows, where s = ldp_hash_start,
1119  * e = ldp_hash_end, f = ldp_flags, p = padding, and each "ent" is a
1120  * struct lu_dirent.  It has size up to LU_PAGE_SIZE. The ldp_hash_end
1121  * value is used as a cookie to request the next lu_dirpage in a
1122  * directory listing that spans multiple pages (two in this example):
1123  *   ________
1124  *  |        |
1125  * .|--------v-------   -----.
1126  * |s|e|f|p|ent|ent| ... |ent|
1127  * '--|--------------   -----'   Each PAGE contains a single
1128  *    '------.                   lu_dirpage.
1129  * .---------v-------   -----.
1130  * |s|e|f|p|ent| 0 | ... | 0 |
1131  * '-----------------   -----'
1132  *
1133  * However, on hosts where the native VM page size (PAGE_CACHE_SIZE) is
1134  * larger than LU_PAGE_SIZE, a single host page may contain multiple
1135  * lu_dirpages. After reading the lu_dirpages from the MDS, the
1136  * ldp_hash_end of the first lu_dirpage refers to the one immediately
1137  * after it in the same PAGE (arrows simplified for brevity, but
1138  * in general e0==s1, e1==s2, etc.):
1139  *
1140  * .--------------------   -----.
1141  * |s0|e0|f0|p|ent|ent| ... |ent|
1142  * |---v----------------   -----|
1143  * |s1|e1|f1|p|ent|ent| ... |ent|
1144  * |---v----------------   -----|  Here, each PAGE contains
1145  *             ...                 multiple lu_dirpages.
1146  * |---v----------------   -----|
1147  * |s'|e'|f'|p|ent|ent| ... |ent|
1148  * '---|----------------   -----'
1149  *     v
1150  * .----------------------------.
1151  * |        next PAGE           |
1152  *
1153  * This structure is transformed into a single logical lu_dirpage as follows:
1154  *
1155  * - Replace e0 with e' so the request for the next lu_dirpage gets the page
1156  *   labeled 'next PAGE'.
1157  *
1158  * - Copy the LDF_COLLIDE flag from f' to f0 to correctly reflect whether
1159  *   a hash collision with the next page exists.
1160  *
1161  * - Adjust the lde_reclen of the ending entry of each lu_dirpage to span
1162  *   to the first entry of the next lu_dirpage.
1163  */
1164 #if PAGE_CACHE_SIZE > LU_PAGE_SIZE
1165 static void mdc_adjust_dirpages(struct page **pages, int cfs_pgs, int lu_pgs)
1166 {
1167         int i;
1168
1169         for (i = 0; i < cfs_pgs; i++) {
1170                 struct lu_dirpage       *dp = kmap(pages[i]);
1171                 struct lu_dirpage       *first = dp;
1172                 struct lu_dirent        *end_dirent = NULL;
1173                 struct lu_dirent        *ent;
1174                 __u64           hash_end = le64_to_cpu(dp->ldp_hash_end);
1175                 __u32           flags = le32_to_cpu(dp->ldp_flags);
1176
1177                 while (--lu_pgs > 0) {
1178                         ent = lu_dirent_start(dp);
1179                         for (end_dirent = ent; ent != NULL;
1180                              end_dirent = ent, ent = lu_dirent_next(ent));
1181
1182                         /* Advance dp to next lu_dirpage. */
1183                         dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE);
1184
1185                         /* Check if we've reached the end of the PAGE. */
1186                         if (!((unsigned long)dp & ~PAGE_MASK))
1187                                 break;
1188
1189                         /* Save the hash and flags of this lu_dirpage. */
1190                         hash_end = le64_to_cpu(dp->ldp_hash_end);
1191                         flags = le32_to_cpu(dp->ldp_flags);
1192
1193                         /* Check if lu_dirpage contains no entries. */
1194                         if (end_dirent == NULL)
1195                                 break;
1196
1197                         /* Enlarge the end entry lde_reclen from 0 to
1198                          * first entry of next lu_dirpage. */
1199                         LASSERT(le16_to_cpu(end_dirent->lde_reclen) == 0);
1200                         end_dirent->lde_reclen =
1201                                 cpu_to_le16((char *)(dp->ldp_entries) -
1202                                             (char *)end_dirent);
1203                 }
1204
1205                 first->ldp_hash_end = hash_end;
1206                 first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE);
1207                 first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE);
1208
1209                 kunmap(pages[i]);
1210         }
1211         LASSERTF(lu_pgs == 0, "left = %d\n", lu_pgs);
1212 }
1213 #else
1214 #define mdc_adjust_dirpages(pages, cfs_pgs, lu_pgs) do {} while (0)
1215 #endif  /* PAGE_CACHE_SIZE > LU_PAGE_SIZE */
1216
1217 /* parameters for readdir page */
1218 struct readpage_param {
1219         struct md_op_data       *rp_mod;
1220         __u64                   rp_off;
1221         int                     rp_hash64;
1222         struct obd_export       *rp_exp;
1223         struct md_callback      *rp_cb;
1224 };
1225
1226 #ifndef HAVE_DELETE_FROM_PAGE_CACHE
1227 static inline void delete_from_page_cache(struct page *page)
1228 {
1229         remove_from_page_cache(page);
1230         page_cache_release(page);
1231 }
1232 #endif
1233
1234 /**
1235  * Read pages from server.
1236  *
1237  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
1238  * a header lu_dirpage which describes the start/end hash, and whether this
1239  * page is empty (contains no dir entry) or hash collide with next page.
1240  * After client receives reply, several pages will be integrated into dir page
1241  * in PAGE_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the
1242  * lu_dirpage for this integrated page will be adjusted.
1243  **/
1244 static int mdc_read_page_remote(void *data, struct page *page0)
1245 {
1246         struct readpage_param   *rp = data;
1247         struct page             **page_pool;
1248         struct page             *page;
1249         struct lu_dirpage       *dp;
1250         int                     rd_pgs = 0; /* number of pages read actually */
1251         int                     npages;
1252         struct md_op_data       *op_data = rp->rp_mod;
1253         struct ptlrpc_request   *req;
1254         int                     max_pages = op_data->op_max_pages;
1255         struct inode            *inode;
1256         struct lu_fid           *fid;
1257         int                     i;
1258         int                     rc;
1259         ENTRY;
1260
1261         LASSERT(max_pages > 0 && max_pages <= PTLRPC_MAX_BRW_PAGES);
1262         inode = op_data->op_data;
1263         fid = &op_data->op_fid1;
1264         LASSERT(inode != NULL);
1265
1266         OBD_ALLOC(page_pool, sizeof(page_pool[0]) * max_pages);
1267         if (page_pool != NULL) {
1268                 page_pool[0] = page0;
1269         } else {
1270                 page_pool = &page0;
1271                 max_pages = 1;
1272         }
1273
1274         for (npages = 1; npages < max_pages; npages++) {
1275                 page = page_cache_alloc_cold(inode->i_mapping);
1276                 if (page == NULL)
1277                         break;
1278                 page_pool[npages] = page;
1279         }
1280
1281         rc = mdc_getpage(rp->rp_exp, fid, rp->rp_off, op_data->op_capa1,
1282                          page_pool, npages, &req);
1283         if (rc < 0) {
1284                 /* page0 is special, which was added into page cache early */
1285                 delete_from_page_cache(page0);
1286         } else {
1287                 int lu_pgs;
1288
1289                 rd_pgs = (req->rq_bulk->bd_nob_transferred +
1290                             PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1291                 lu_pgs = req->rq_bulk->bd_nob_transferred >>
1292                                                         LU_PAGE_SHIFT;
1293                 LASSERT(!(req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK));
1294
1295                 CDEBUG(D_INODE, "read %d(%d) pages\n", rd_pgs, lu_pgs);
1296
1297                 mdc_adjust_dirpages(page_pool, rd_pgs, lu_pgs);
1298
1299                 SetPageUptodate(page0);
1300         }
1301         unlock_page(page0);
1302
1303         ptlrpc_req_finished(req);
1304         CDEBUG(D_CACHE, "read %d/%d pages\n", rd_pgs, npages);
1305         for (i = 1; i < npages; i++) {
1306                 unsigned long   offset;
1307                 __u64           hash;
1308                 int ret;
1309
1310                 page = page_pool[i];
1311
1312                 if (rc < 0 || i >= rd_pgs) {
1313                         page_cache_release(page);
1314                         continue;
1315                 }
1316
1317                 SetPageUptodate(page);
1318
1319                 dp = kmap(page);
1320                 hash = le64_to_cpu(dp->ldp_hash_start);
1321                 kunmap(page);
1322
1323                 offset = hash_x_index(hash, rp->rp_hash64);
1324
1325                 prefetchw(&page->flags);
1326                 ret = add_to_page_cache_lru(page, inode->i_mapping, offset,
1327                                             GFP_KERNEL);
1328                 if (ret == 0)
1329                         unlock_page(page);
1330                 else
1331                         CDEBUG(D_VFSTRACE, "page %lu add to page cache failed:"
1332                                " rc = %d\n", offset, ret);
1333                 page_cache_release(page);
1334         }
1335
1336         if (page_pool != &page0)
1337                 OBD_FREE(page_pool, sizeof(page_pool[0]) * max_pages);
1338
1339         RETURN(rc);
1340 }
1341
1342 /**
1343  * Read dir page from cache first, if it can not find it, read it from
1344  * server and add into the cache.
1345  *
1346  * \param[in] exp       MDC export
1347  * \param[in] op_data   client MD stack parameters, transfering parameters
1348  *                      between different layers on client MD stack.
1349  * \param[in] cb_op     callback required for ldlm lock enqueue during
1350  *                      read page
1351  * \param[in] hash_offset the hash offset of the page to be read
1352  * \param[in] ppage     the page to be read
1353  *
1354  * retval               = 0 get the page successfully
1355  *                      errno(<0) get the page failed
1356  */
1357 static int mdc_read_page(struct obd_export *exp, struct md_op_data *op_data,
1358                          struct md_callback *cb_op, __u64 hash_offset,
1359                          struct page **ppage)
1360 {
1361         struct lookup_intent    it = { .it_op = IT_READDIR };
1362         struct page             *page;
1363         struct inode            *dir = op_data->op_data;
1364         struct address_space    *mapping;
1365         struct lu_dirpage       *dp;
1366         __u64                   start = 0;
1367         __u64                   end = 0;
1368         struct lustre_handle    lockh;
1369         struct ptlrpc_request   *enq_req = NULL;
1370         struct readpage_param   rp_param;
1371         int rc;
1372
1373         ENTRY;
1374
1375         *ppage = NULL;
1376
1377         LASSERT(dir != NULL);
1378         mapping = dir->i_mapping;
1379
1380         rc = mdc_intent_lock(exp, op_data, &it, &enq_req,
1381                              cb_op->md_blocking_ast, 0);
1382         if (enq_req != NULL)
1383                 ptlrpc_req_finished(enq_req);
1384
1385         if (rc < 0) {
1386                 CERROR("%s: "DFID" lock enqueue fails: rc = %d\n",
1387                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1), rc);
1388                 RETURN(rc);
1389         }
1390
1391         rc = 0;
1392         mdc_set_lock_data(exp, &it.d.lustre.it_lock_handle, dir, NULL);
1393
1394         rp_param.rp_off = hash_offset;
1395         rp_param.rp_hash64 = op_data->op_cli_flags & CLI_HASH64;
1396         page = mdc_page_locate(mapping, &rp_param.rp_off, &start, &end,
1397                                rp_param.rp_hash64);
1398         if (IS_ERR(page)) {
1399                 CERROR("%s: dir page locate: "DFID" at "LPU64": rc %ld\n",
1400                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1401                        rp_param.rp_off, PTR_ERR(page));
1402                 GOTO(out_unlock, rc = PTR_ERR(page));
1403         } else if (page != NULL) {
1404                 /*
1405                  * XXX nikita: not entirely correct handling of a corner case:
1406                  * suppose hash chain of entries with hash value HASH crosses
1407                  * border between pages P0 and P1. First both P0 and P1 are
1408                  * cached, seekdir() is called for some entry from the P0 part
1409                  * of the chain. Later P0 goes out of cache. telldir(HASH)
1410                  * happens and finds P1, as it starts with matching hash
1411                  * value. Remaining entries from P0 part of the chain are
1412                  * skipped. (Is that really a bug?)
1413                  *
1414                  * Possible solutions: 0. don't cache P1 is such case, handle
1415                  * it as an "overflow" page. 1. invalidate all pages at
1416                  * once. 2. use HASH|1 as an index for P1.
1417                  */
1418                 GOTO(hash_collision, page);
1419         }
1420
1421         rp_param.rp_exp = exp;
1422         rp_param.rp_mod = op_data;
1423         page = read_cache_page(mapping,
1424                                hash_x_index(rp_param.rp_off,
1425                                             rp_param.rp_hash64),
1426                                mdc_read_page_remote, &rp_param);
1427         if (IS_ERR(page)) {
1428                 CDEBUG(D_INFO, "%s: read cache page: "DFID" at "LPU64": %ld\n",
1429                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1430                        rp_param.rp_off, PTR_ERR(page));
1431                 GOTO(out_unlock, rc = PTR_ERR(page));
1432         }
1433
1434         wait_on_page_locked(page);
1435         (void)kmap(page);
1436         if (!PageUptodate(page)) {
1437                 CERROR("%s: page not updated: "DFID" at "LPU64": rc %d\n",
1438                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1439                        rp_param.rp_off, -5);
1440                 goto fail;
1441         }
1442         if (!PageChecked(page))
1443                 SetPageChecked(page);
1444         if (PageError(page)) {
1445                 CERROR("%s: page error: "DFID" at "LPU64": rc %d\n",
1446                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1447                        rp_param.rp_off, -5);
1448                 goto fail;
1449         }
1450
1451 hash_collision:
1452         dp = page_address(page);
1453         if (BITS_PER_LONG == 32 && rp_param.rp_hash64) {
1454                 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1455                 end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
1456                 rp_param.rp_off = hash_offset >> 32;
1457         } else {
1458                 start = le64_to_cpu(dp->ldp_hash_start);
1459                 end   = le64_to_cpu(dp->ldp_hash_end);
1460                 rp_param.rp_off = hash_offset;
1461         }
1462         if (end == start) {
1463                 LASSERT(start == rp_param.rp_off);
1464                 CWARN("Page-wide hash collision: %#lx\n", (unsigned long)end);
1465 #if BITS_PER_LONG == 32
1466                 CWARN("Real page-wide hash collision at ["LPU64" "LPU64"] with "
1467                       "hash "LPU64"\n", le64_to_cpu(dp->ldp_hash_start),
1468                       le64_to_cpu(dp->ldp_hash_end), hash_offset);
1469 #endif
1470
1471                 /*
1472                  * Fetch whole overflow chain...
1473                  *
1474                  * XXX not yet.
1475                  */
1476                 goto fail;
1477         }
1478         *ppage = page;
1479 out_unlock:
1480         lockh.cookie = it.d.lustre.it_lock_handle;
1481         ldlm_lock_decref(&lockh, it.d.lustre.it_lock_mode);
1482         it.d.lustre.it_lock_handle = 0;
1483         return rc;
1484 fail:
1485         kunmap(page);
1486         mdc_release_page(page, 1);
1487         rc = -EIO;
1488         goto out_unlock;
1489 }
1490
1491
1492 static int mdc_statfs(const struct lu_env *env,
1493                       struct obd_export *exp, struct obd_statfs *osfs,
1494                       __u64 max_age, __u32 flags)
1495 {
1496         struct obd_device     *obd = class_exp2obd(exp);
1497         struct ptlrpc_request *req;
1498         struct obd_statfs     *msfs;
1499         struct obd_import     *imp = NULL;
1500         int                    rc;
1501         ENTRY;
1502
1503         /*
1504          * Since the request might also come from lprocfs, so we need
1505          * sync this with client_disconnect_export Bug15684
1506          */
1507         down_read(&obd->u.cli.cl_sem);
1508         if (obd->u.cli.cl_import)
1509                 imp = class_import_get(obd->u.cli.cl_import);
1510         up_read(&obd->u.cli.cl_sem);
1511         if (!imp)
1512                 RETURN(-ENODEV);
1513
1514         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_STATFS,
1515                                         LUSTRE_MDS_VERSION, MDS_STATFS);
1516         if (req == NULL)
1517                 GOTO(output, rc = -ENOMEM);
1518
1519         ptlrpc_request_set_replen(req);
1520
1521         if (flags & OBD_STATFS_NODELAY) {
1522                 /* procfs requests not want stay in wait for avoid deadlock */
1523                 req->rq_no_resend = 1;
1524                 req->rq_no_delay = 1;
1525         }
1526
1527         rc = ptlrpc_queue_wait(req);
1528         if (rc) {
1529                 /* check connection error first */
1530                 if (imp->imp_connect_error)
1531                         rc = imp->imp_connect_error;
1532                 GOTO(out, rc);
1533         }
1534
1535         msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1536         if (msfs == NULL)
1537                 GOTO(out, rc = -EPROTO);
1538
1539         *osfs = *msfs;
1540         EXIT;
1541 out:
1542         ptlrpc_req_finished(req);
1543 output:
1544         class_import_put(imp);
1545         return rc;
1546 }
1547
1548 static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
1549 {
1550         __u32 keylen, vallen;
1551         void *key;
1552         int rc;
1553
1554         if (gf->gf_pathlen > PATH_MAX)
1555                 RETURN(-ENAMETOOLONG);
1556         if (gf->gf_pathlen < 2)
1557                 RETURN(-EOVERFLOW);
1558
1559         /* Key is KEY_FID2PATH + getinfo_fid2path description */
1560         keylen = cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf);
1561         OBD_ALLOC(key, keylen);
1562         if (key == NULL)
1563                 RETURN(-ENOMEM);
1564         memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
1565         memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf));
1566
1567         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
1568                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
1569
1570         if (!fid_is_sane(&gf->gf_fid))
1571                 GOTO(out, rc = -EINVAL);
1572
1573         /* Val is struct getinfo_fid2path result plus path */
1574         vallen = sizeof(*gf) + gf->gf_pathlen;
1575
1576         rc = obd_get_info(NULL, exp, keylen, key, &vallen, gf);
1577         if (rc != 0 && rc != -EREMOTE)
1578                 GOTO(out, rc);
1579
1580         if (vallen <= sizeof(*gf))
1581                 GOTO(out, rc = -EPROTO);
1582         else if (vallen > sizeof(*gf) + gf->gf_pathlen)
1583                 GOTO(out, rc = -EOVERFLOW);
1584
1585         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n%s\n",
1586                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno, gf->gf_path);
1587
1588 out:
1589         OBD_FREE(key, keylen);
1590         return rc;
1591 }
1592
1593 static int mdc_ioc_hsm_progress(struct obd_export *exp,
1594                                 struct hsm_progress_kernel *hpk)
1595 {
1596         struct obd_import               *imp = class_exp2cliimp(exp);
1597         struct hsm_progress_kernel      *req_hpk;
1598         struct ptlrpc_request           *req;
1599         int                              rc;
1600         ENTRY;
1601
1602         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_PROGRESS,
1603                                         LUSTRE_MDS_VERSION, MDS_HSM_PROGRESS);
1604         if (req == NULL)
1605                 GOTO(out, rc = -ENOMEM);
1606
1607         mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
1608
1609         /* Copy hsm_progress struct */
1610         req_hpk = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_PROGRESS);
1611         if (req_hpk == NULL)
1612                 GOTO(out, rc = -EPROTO);
1613
1614         *req_hpk = *hpk;
1615         req_hpk->hpk_errval = lustre_errno_hton(hpk->hpk_errval);
1616
1617         ptlrpc_request_set_replen(req);
1618
1619         rc = mdc_queue_wait(req);
1620         GOTO(out, rc);
1621 out:
1622         ptlrpc_req_finished(req);
1623         return rc;
1624 }
1625
1626 static int mdc_ioc_hsm_ct_register(struct obd_import *imp, __u32 archives)
1627 {
1628         __u32                   *archive_mask;
1629         struct ptlrpc_request   *req;
1630         int                      rc;
1631         ENTRY;
1632
1633         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_REGISTER,
1634                                         LUSTRE_MDS_VERSION,
1635                                         MDS_HSM_CT_REGISTER);
1636         if (req == NULL)
1637                 GOTO(out, rc = -ENOMEM);
1638
1639         mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
1640
1641         /* Copy hsm_progress struct */
1642         archive_mask = req_capsule_client_get(&req->rq_pill,
1643                                               &RMF_MDS_HSM_ARCHIVE);
1644         if (archive_mask == NULL)
1645                 GOTO(out, rc = -EPROTO);
1646
1647         *archive_mask = archives;
1648
1649         ptlrpc_request_set_replen(req);
1650
1651         rc = mdc_queue_wait(req);
1652         GOTO(out, rc);
1653 out:
1654         ptlrpc_req_finished(req);
1655         return rc;
1656 }
1657
1658 static int mdc_ioc_hsm_current_action(struct obd_export *exp,
1659                                       struct md_op_data *op_data)
1660 {
1661         struct hsm_current_action       *hca = op_data->op_data;
1662         struct hsm_current_action       *req_hca;
1663         struct ptlrpc_request           *req;
1664         int                              rc;
1665         ENTRY;
1666
1667         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1668                                    &RQF_MDS_HSM_ACTION);
1669         if (req == NULL)
1670                 RETURN(-ENOMEM);
1671
1672         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1673
1674         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_ACTION);
1675         if (rc) {
1676                 ptlrpc_request_free(req);
1677                 RETURN(rc);
1678         }
1679
1680         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
1681                       OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
1682
1683         ptlrpc_request_set_replen(req);
1684
1685         rc = mdc_queue_wait(req);
1686         if (rc)
1687                 GOTO(out, rc);
1688
1689         req_hca = req_capsule_server_get(&req->rq_pill,
1690                                          &RMF_MDS_HSM_CURRENT_ACTION);
1691         if (req_hca == NULL)
1692                 GOTO(out, rc = -EPROTO);
1693
1694         *hca = *req_hca;
1695
1696         EXIT;
1697 out:
1698         ptlrpc_req_finished(req);
1699         return rc;
1700 }
1701
1702 static int mdc_ioc_hsm_ct_unregister(struct obd_import *imp)
1703 {
1704         struct ptlrpc_request   *req;
1705         int                      rc;
1706         ENTRY;
1707
1708         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_UNREGISTER,
1709                                         LUSTRE_MDS_VERSION,
1710                                         MDS_HSM_CT_UNREGISTER);
1711         if (req == NULL)
1712                 GOTO(out, rc = -ENOMEM);
1713
1714         mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
1715
1716         ptlrpc_request_set_replen(req);
1717
1718         rc = mdc_queue_wait(req);
1719         GOTO(out, rc);
1720 out:
1721         ptlrpc_req_finished(req);
1722         return rc;
1723 }
1724
1725 static int mdc_ioc_hsm_state_get(struct obd_export *exp,
1726                                  struct md_op_data *op_data)
1727 {
1728         struct hsm_user_state   *hus = op_data->op_data;
1729         struct hsm_user_state   *req_hus;
1730         struct ptlrpc_request   *req;
1731         int                      rc;
1732         ENTRY;
1733
1734         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1735                                    &RQF_MDS_HSM_STATE_GET);
1736         if (req == NULL)
1737                 RETURN(-ENOMEM);
1738
1739         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1740
1741         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_GET);
1742         if (rc != 0) {
1743                 ptlrpc_request_free(req);
1744                 RETURN(rc);
1745         }
1746
1747         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
1748                       OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
1749
1750         ptlrpc_request_set_replen(req);
1751
1752         rc = mdc_queue_wait(req);
1753         if (rc)
1754                 GOTO(out, rc);
1755
1756         req_hus = req_capsule_server_get(&req->rq_pill, &RMF_HSM_USER_STATE);
1757         if (req_hus == NULL)
1758                 GOTO(out, rc = -EPROTO);
1759
1760         *hus = *req_hus;
1761
1762         EXIT;
1763 out:
1764         ptlrpc_req_finished(req);
1765         return rc;
1766 }
1767
1768 static int mdc_ioc_hsm_state_set(struct obd_export *exp,
1769                                  struct md_op_data *op_data)
1770 {
1771         struct hsm_state_set    *hss = op_data->op_data;
1772         struct hsm_state_set    *req_hss;
1773         struct ptlrpc_request   *req;
1774         int                      rc;
1775         ENTRY;
1776
1777         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1778                                    &RQF_MDS_HSM_STATE_SET);
1779         if (req == NULL)
1780                 RETURN(-ENOMEM);
1781
1782         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1783
1784         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_SET);
1785         if (rc) {
1786                 ptlrpc_request_free(req);
1787                 RETURN(rc);
1788         }
1789
1790         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
1791                       OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
1792
1793         /* Copy states */
1794         req_hss = req_capsule_client_get(&req->rq_pill, &RMF_HSM_STATE_SET);
1795         if (req_hss == NULL)
1796                 GOTO(out, rc = -EPROTO);
1797         *req_hss = *hss;
1798
1799         ptlrpc_request_set_replen(req);
1800
1801         rc = mdc_queue_wait(req);
1802         GOTO(out, rc);
1803
1804         EXIT;
1805 out:
1806         ptlrpc_req_finished(req);
1807         return rc;
1808 }
1809
1810 static int mdc_ioc_hsm_request(struct obd_export *exp,
1811                                struct hsm_user_request *hur)
1812 {
1813         struct obd_import       *imp = class_exp2cliimp(exp);
1814         struct ptlrpc_request   *req;
1815         struct hsm_request      *req_hr;
1816         struct hsm_user_item    *req_hui;
1817         char                    *req_opaque;
1818         int                      rc;
1819         ENTRY;
1820
1821         req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_REQUEST);
1822         if (req == NULL)
1823                 GOTO(out, rc = -ENOMEM);
1824
1825         req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM, RCL_CLIENT,
1826                              hur->hur_request.hr_itemcount
1827                              * sizeof(struct hsm_user_item));
1828         req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA, RCL_CLIENT,
1829                              hur->hur_request.hr_data_len);
1830
1831         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_REQUEST);
1832         if (rc) {
1833                 ptlrpc_request_free(req);
1834                 RETURN(rc);
1835         }
1836
1837         mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
1838
1839         /* Copy hsm_request struct */
1840         req_hr = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_REQUEST);
1841         if (req_hr == NULL)
1842                 GOTO(out, rc = -EPROTO);
1843         *req_hr = hur->hur_request;
1844
1845         /* Copy hsm_user_item structs */
1846         req_hui = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM);
1847         if (req_hui == NULL)
1848                 GOTO(out, rc = -EPROTO);
1849         memcpy(req_hui, hur->hur_user_item,
1850                hur->hur_request.hr_itemcount * sizeof(struct hsm_user_item));
1851
1852         /* Copy opaque field */
1853         req_opaque = req_capsule_client_get(&req->rq_pill, &RMF_GENERIC_DATA);
1854         if (req_opaque == NULL)
1855                 GOTO(out, rc = -EPROTO);
1856         memcpy(req_opaque, hur_data(hur), hur->hur_request.hr_data_len);
1857
1858         ptlrpc_request_set_replen(req);
1859
1860         rc = mdc_queue_wait(req);
1861         GOTO(out, rc);
1862
1863 out:
1864         ptlrpc_req_finished(req);
1865         return rc;
1866 }
1867
1868 static struct kuc_hdr *changelog_kuc_hdr(char *buf, size_t len, __u32 flags)
1869 {
1870         struct kuc_hdr *lh = (struct kuc_hdr *)buf;
1871
1872         LASSERT(len <= KUC_CHANGELOG_MSG_MAXSIZE);
1873
1874         lh->kuc_magic = KUC_MAGIC;
1875         lh->kuc_transport = KUC_TRANSPORT_CHANGELOG;
1876         lh->kuc_flags = flags;
1877         lh->kuc_msgtype = CL_RECORD;
1878         lh->kuc_msglen = len;
1879         return lh;
1880 }
1881
1882 struct changelog_show {
1883         __u64                            cs_startrec;
1884         enum changelog_send_flag         cs_flags;
1885         struct file                     *cs_fp;
1886         char                            *cs_buf;
1887         struct obd_device               *cs_obd;
1888 };
1889
1890 static inline char *cs_obd_name(struct changelog_show *cs)
1891 {
1892         return cs->cs_obd->obd_name;
1893 }
1894
1895 static int changelog_kkuc_cb(const struct lu_env *env, struct llog_handle *llh,
1896                              struct llog_rec_hdr *hdr, void *data)
1897 {
1898         struct changelog_show           *cs = data;
1899         struct llog_changelog_rec       *rec = (struct llog_changelog_rec *)hdr;
1900         struct kuc_hdr                  *lh;
1901         size_t                           len;
1902         int                              rc;
1903         ENTRY;
1904
1905         if (rec->cr_hdr.lrh_type != CHANGELOG_REC) {
1906                 rc = -EINVAL;
1907                 CERROR("%s: not a changelog rec %x/%d: rc = %d\n",
1908                        cs_obd_name(cs), rec->cr_hdr.lrh_type,
1909                        rec->cr.cr_type, rc);
1910                 RETURN(rc);
1911         }
1912
1913         if (rec->cr.cr_index < cs->cs_startrec) {
1914                 /* Skip entries earlier than what we are interested in */
1915                 CDEBUG(D_HSM, "rec="LPU64" start="LPU64"\n",
1916                        rec->cr.cr_index, cs->cs_startrec);
1917                 RETURN(0);
1918         }
1919
1920         CDEBUG(D_HSM, LPU64" %02d%-5s "LPU64" 0x%x t="DFID" p="DFID" %.*s\n",
1921                rec->cr.cr_index, rec->cr.cr_type,
1922                changelog_type2str(rec->cr.cr_type), rec->cr.cr_time,
1923                rec->cr.cr_flags & CLF_FLAGMASK,
1924                PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid),
1925                rec->cr.cr_namelen, changelog_rec_name(&rec->cr));
1926
1927         len = sizeof(*lh) + changelog_rec_size(&rec->cr) + rec->cr.cr_namelen;
1928
1929         /* Set up the message */
1930         lh = changelog_kuc_hdr(cs->cs_buf, len, cs->cs_flags);
1931         memcpy(lh + 1, &rec->cr, len - sizeof(*lh));
1932
1933         rc = libcfs_kkuc_msg_put(cs->cs_fp, lh);
1934         CDEBUG(D_HSM, "kucmsg fp %p len %zu rc %d\n", cs->cs_fp, len, rc);
1935
1936         RETURN(rc);
1937 }
1938
1939 static int mdc_changelog_send_thread(void *csdata)
1940 {
1941         struct changelog_show   *cs = csdata;
1942         struct llog_ctxt        *ctxt = NULL;
1943         struct llog_handle      *llh = NULL;
1944         struct kuc_hdr          *kuch;
1945         enum llog_flag           flags = LLOG_F_IS_CAT;
1946         int                      rc;
1947
1948         CDEBUG(D_HSM, "changelog to fp=%p start "LPU64"\n",
1949                cs->cs_fp, cs->cs_startrec);
1950
1951         OBD_ALLOC(cs->cs_buf, KUC_CHANGELOG_MSG_MAXSIZE);
1952         if (cs->cs_buf == NULL)
1953                 GOTO(out, rc = -ENOMEM);
1954
1955         /* Set up the remote catalog handle */
1956         ctxt = llog_get_context(cs->cs_obd, LLOG_CHANGELOG_REPL_CTXT);
1957         if (ctxt == NULL)
1958                 GOTO(out, rc = -ENOENT);
1959         rc = llog_open(NULL, ctxt, &llh, NULL, CHANGELOG_CATALOG,
1960                        LLOG_OPEN_EXISTS);
1961         if (rc) {
1962                 CERROR("%s: fail to open changelog catalog: rc = %d\n",
1963                        cs_obd_name(cs), rc);
1964                 GOTO(out, rc);
1965         }
1966
1967         if (cs->cs_flags & CHANGELOG_FLAG_JOBID)
1968                 flags |= LLOG_F_EXT_JOBID;
1969
1970         rc = llog_init_handle(NULL, llh, flags, NULL);
1971         if (rc) {
1972                 CERROR("llog_init_handle failed %d\n", rc);
1973                 GOTO(out, rc);
1974         }
1975
1976         rc = llog_cat_process(NULL, llh, changelog_kkuc_cb, cs, 0, 0);
1977
1978         /* Send EOF no matter what our result */
1979         kuch = changelog_kuc_hdr(cs->cs_buf, sizeof(*kuch), cs->cs_flags);
1980         kuch->kuc_msgtype = CL_EOF;
1981         libcfs_kkuc_msg_put(cs->cs_fp, kuch);
1982
1983 out:
1984         fput(cs->cs_fp);
1985         if (llh)
1986                 llog_cat_close(NULL, llh);
1987         if (ctxt)
1988                 llog_ctxt_put(ctxt);
1989         if (cs->cs_buf)
1990                 OBD_FREE(cs->cs_buf, KUC_CHANGELOG_MSG_MAXSIZE);
1991         OBD_FREE_PTR(cs);
1992         return rc;
1993 }
1994
1995 static int mdc_ioc_changelog_send(struct obd_device *obd,
1996                                   struct ioc_changelog *icc)
1997 {
1998         struct changelog_show *cs;
1999         struct task_struct *task;
2000         int rc;
2001
2002         /* Freed in mdc_changelog_send_thread */
2003         OBD_ALLOC_PTR(cs);
2004         if (!cs)
2005                 return -ENOMEM;
2006
2007         cs->cs_obd = obd;
2008         cs->cs_startrec = icc->icc_recno;
2009         /* matching fput in mdc_changelog_send_thread */
2010         cs->cs_fp = fget(icc->icc_id);
2011         cs->cs_flags = icc->icc_flags;
2012
2013         /*
2014          * New thread because we should return to user app before
2015          * writing into our pipe
2016          */
2017         task = kthread_run(mdc_changelog_send_thread, cs,
2018                            "mdc_clg_send_thread");
2019         if (IS_ERR(task)) {
2020                 rc = PTR_ERR(task);
2021                 CERROR("%s: cannot start changelog thread: rc = %d\n",
2022                        cs_obd_name(cs), rc);
2023                 OBD_FREE_PTR(cs);
2024         } else {
2025                 rc = 0;
2026                 CDEBUG(D_HSM, "%s: started changelog thread\n",
2027                        cs_obd_name(cs));
2028         }
2029
2030         return rc;
2031 }
2032
2033 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2034                                 struct lustre_kernelcomm *lk);
2035
2036 static int mdc_quotacheck(struct obd_device *unused, struct obd_export *exp,
2037                           struct obd_quotactl *oqctl)
2038 {
2039         struct client_obd       *cli = &exp->exp_obd->u.cli;
2040         struct ptlrpc_request   *req;
2041         struct obd_quotactl     *body;
2042         int                      rc;
2043         ENTRY;
2044
2045         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
2046                                         &RQF_MDS_QUOTACHECK, LUSTRE_MDS_VERSION,
2047                                         MDS_QUOTACHECK);
2048         if (req == NULL)
2049                 RETURN(-ENOMEM);
2050
2051         body = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
2052         *body = *oqctl;
2053
2054         ptlrpc_request_set_replen(req);
2055
2056         /* the next poll will find -ENODATA, that means quotacheck is
2057          * going on */
2058         cli->cl_qchk_stat = -ENODATA;
2059         rc = ptlrpc_queue_wait(req);
2060         if (rc)
2061                 cli->cl_qchk_stat = rc;
2062         ptlrpc_req_finished(req);
2063         RETURN(rc);
2064 }
2065
2066 static int mdc_quota_poll_check(struct obd_export *exp,
2067                                 struct if_quotacheck *qchk)
2068 {
2069         struct client_obd *cli = &exp->exp_obd->u.cli;
2070         int rc;
2071         ENTRY;
2072
2073         qchk->obd_uuid = cli->cl_target_uuid;
2074         memcpy(qchk->obd_type, LUSTRE_MDS_NAME, strlen(LUSTRE_MDS_NAME));
2075
2076         rc = cli->cl_qchk_stat;
2077         /* the client is not the previous one */
2078         if (rc == CL_NOT_QUOTACHECKED)
2079                 rc = -EINTR;
2080         RETURN(rc);
2081 }
2082
2083 static int mdc_quotactl(struct obd_device *unused, struct obd_export *exp,
2084                         struct obd_quotactl *oqctl)
2085 {
2086         struct ptlrpc_request   *req;
2087         struct obd_quotactl     *oqc;
2088         int                      rc;
2089         ENTRY;
2090
2091         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
2092                                         &RQF_MDS_QUOTACTL, LUSTRE_MDS_VERSION,
2093                                         MDS_QUOTACTL);
2094         if (req == NULL)
2095                 RETURN(-ENOMEM);
2096
2097         oqc = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
2098         *oqc = *oqctl;
2099
2100         ptlrpc_request_set_replen(req);
2101         ptlrpc_at_set_req_timeout(req);
2102         req->rq_no_resend = 1;
2103
2104         rc = ptlrpc_queue_wait(req);
2105         if (rc)
2106                 CERROR("ptlrpc_queue_wait failed, rc: %d\n", rc);
2107
2108         if (req->rq_repmsg &&
2109             (oqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL))) {
2110                 *oqctl = *oqc;
2111         } else if (!rc) {
2112                 CERROR ("Can't unpack obd_quotactl\n");
2113                 rc = -EPROTO;
2114         }
2115         ptlrpc_req_finished(req);
2116
2117         RETURN(rc);
2118 }
2119
2120 static int mdc_ioc_swap_layouts(struct obd_export *exp,
2121                                 struct md_op_data *op_data)
2122 {
2123         struct list_head cancels = LIST_HEAD_INIT(cancels);
2124         struct ptlrpc_request   *req;
2125         int                      rc, count;
2126         struct mdc_swap_layouts *msl, *payload;
2127         ENTRY;
2128
2129         msl = op_data->op_data;
2130
2131         /* When the MDT will get the MDS_SWAP_LAYOUTS RPC the
2132          * first thing it will do is to cancel the 2 layout
2133          * locks hold by this client.
2134          * So the client must cancel its layout locks on the 2 fids
2135          * with the request RPC to avoid extra RPC round trips
2136          */
2137         count = mdc_resource_get_unused(exp, &op_data->op_fid1, &cancels,
2138                                         LCK_EX, MDS_INODELOCK_LAYOUT |
2139                                         MDS_INODELOCK_XATTR);
2140         count += mdc_resource_get_unused(exp, &op_data->op_fid2, &cancels,
2141                                          LCK_EX, MDS_INODELOCK_LAYOUT |
2142                                          MDS_INODELOCK_XATTR);
2143
2144         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
2145                                    &RQF_MDS_SWAP_LAYOUTS);
2146         if (req == NULL) {
2147                 ldlm_lock_list_put(&cancels, l_bl_ast, count);
2148                 RETURN(-ENOMEM);
2149         }
2150
2151         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
2152         mdc_set_capa_size(req, &RMF_CAPA2, op_data->op_capa2);
2153
2154         rc = mdc_prep_elc_req(exp, req, MDS_SWAP_LAYOUTS, &cancels, count);
2155         if (rc) {
2156                 ptlrpc_request_free(req);
2157                 RETURN(rc);
2158         }
2159
2160         mdc_swap_layouts_pack(req, op_data);
2161
2162         payload = req_capsule_client_get(&req->rq_pill, &RMF_SWAP_LAYOUTS);
2163         LASSERT(payload);
2164
2165         *payload = *msl;
2166
2167         ptlrpc_request_set_replen(req);
2168
2169         rc = ptlrpc_queue_wait(req);
2170         if (rc)
2171                 GOTO(out, rc);
2172         EXIT;
2173
2174 out:
2175         ptlrpc_req_finished(req);
2176         return rc;
2177 }
2178
2179 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2180                          void *karg, void __user *uarg)
2181 {
2182         struct obd_device *obd = exp->exp_obd;
2183         struct obd_ioctl_data *data = karg;
2184         struct obd_import *imp = obd->u.cli.cl_import;
2185         int rc;
2186         ENTRY;
2187
2188         if (!try_module_get(THIS_MODULE)) {
2189                 CERROR("%s: cannot get module '%s'\n", obd->obd_name,
2190                        module_name(THIS_MODULE));
2191                 return -EINVAL;
2192         }
2193         switch (cmd) {
2194         case OBD_IOC_CHANGELOG_SEND:
2195                 rc = mdc_ioc_changelog_send(obd, karg);
2196                 GOTO(out, rc);
2197         case OBD_IOC_CHANGELOG_CLEAR: {
2198                 struct ioc_changelog *icc = karg;
2199                 struct changelog_setinfo cs =
2200                         {.cs_recno = icc->icc_recno, .cs_id = icc->icc_id};
2201                 rc = obd_set_info_async(NULL, exp, strlen(KEY_CHANGELOG_CLEAR),
2202                                         KEY_CHANGELOG_CLEAR, sizeof(cs), &cs,
2203                                         NULL);
2204                 GOTO(out, rc);
2205         }
2206         case OBD_IOC_FID2PATH:
2207                 rc = mdc_ioc_fid2path(exp, karg);
2208                 GOTO(out, rc);
2209         case LL_IOC_HSM_CT_START:
2210                 rc = mdc_ioc_hsm_ct_start(exp, karg);
2211                 /* ignore if it was already registered on this MDS. */
2212                 if (rc == -EEXIST)
2213                         rc = 0;
2214                 GOTO(out, rc);
2215         case LL_IOC_HSM_PROGRESS:
2216                 rc = mdc_ioc_hsm_progress(exp, karg);
2217                 GOTO(out, rc);
2218         case LL_IOC_HSM_STATE_GET:
2219                 rc = mdc_ioc_hsm_state_get(exp, karg);
2220                 GOTO(out, rc);
2221         case LL_IOC_HSM_STATE_SET:
2222                 rc = mdc_ioc_hsm_state_set(exp, karg);
2223                 GOTO(out, rc);
2224         case LL_IOC_HSM_ACTION:
2225                 rc = mdc_ioc_hsm_current_action(exp, karg);
2226                 GOTO(out, rc);
2227         case LL_IOC_HSM_REQUEST:
2228                 rc = mdc_ioc_hsm_request(exp, karg);
2229                 GOTO(out, rc);
2230         case OBD_IOC_CLIENT_RECOVER:
2231                 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1, 0);
2232                 if (rc < 0)
2233                         GOTO(out, rc);
2234                 GOTO(out, rc = 0);
2235         case IOC_OSC_SET_ACTIVE:
2236                 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
2237                 GOTO(out, rc);
2238         case OBD_IOC_POLL_QUOTACHECK:
2239                 rc = mdc_quota_poll_check(exp, (struct if_quotacheck *)karg);
2240                 GOTO(out, rc);
2241         case OBD_IOC_PING_TARGET:
2242                 rc = ptlrpc_obd_ping(obd);
2243                 GOTO(out, rc);
2244         /*
2245          * Normally IOC_OBD_STATFS, OBD_IOC_QUOTACTL iocontrol are handled by
2246          * LMV instead of MDC. But when the cluster is upgraded from 1.8,
2247          * there'd be no LMV layer thus we might be called here. Eventually
2248          * this code should be removed.
2249          * bz20731, LU-592.
2250          */
2251         case IOC_OBD_STATFS: {
2252                 struct obd_statfs stat_buf = {0};
2253
2254                 if (*((__u32 *) data->ioc_inlbuf2) != 0)
2255                         GOTO(out, rc = -ENODEV);
2256
2257                 /* copy UUID */
2258                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd),
2259                                  min((int)data->ioc_plen2,
2260                                      (int)sizeof(struct obd_uuid))))
2261                         GOTO(out, rc = -EFAULT);
2262
2263                 rc = mdc_statfs(NULL, obd->obd_self_export, &stat_buf,
2264                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
2265                                 0);
2266                 if (rc != 0)
2267                         GOTO(out, rc);
2268
2269                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
2270                                      min((int) data->ioc_plen1,
2271                                          (int) sizeof(stat_buf))))
2272                         GOTO(out, rc = -EFAULT);
2273
2274                 GOTO(out, rc = 0);
2275         }
2276         case OBD_IOC_QUOTACTL: {
2277                 struct if_quotactl *qctl = karg;
2278                 struct obd_quotactl *oqctl;
2279
2280                 OBD_ALLOC_PTR(oqctl);
2281                 if (oqctl == NULL)
2282                         GOTO(out, rc = -ENOMEM);
2283
2284                 QCTL_COPY(oqctl, qctl);
2285                 rc = obd_quotactl(exp, oqctl);
2286                 if (rc == 0) {
2287                         QCTL_COPY(qctl, oqctl);
2288                         qctl->qc_valid = QC_MDTIDX;
2289                         qctl->obd_uuid = obd->u.cli.cl_target_uuid;
2290                 }
2291
2292                 OBD_FREE_PTR(oqctl);
2293                 GOTO(out, rc);
2294         }
2295         case LL_IOC_GET_CONNECT_FLAGS:
2296                 if (copy_to_user(uarg, exp_connect_flags_ptr(exp),
2297                                  sizeof(*exp_connect_flags_ptr(exp))))
2298                         GOTO(out, rc = -EFAULT);
2299
2300                 GOTO(out, rc = 0);
2301         case LL_IOC_LOV_SWAP_LAYOUTS:
2302                 rc = mdc_ioc_swap_layouts(exp, karg);
2303                 GOTO(out, rc);
2304         default:
2305                 CERROR("unrecognised ioctl: cmd = %#x\n", cmd);
2306                 GOTO(out, rc = -ENOTTY);
2307         }
2308 out:
2309         module_put(THIS_MODULE);
2310
2311         return rc;
2312 }
2313
2314 static int mdc_get_info_rpc(struct obd_export *exp,
2315                             u32 keylen, void *key,
2316                             u32 vallen, void *val)
2317 {
2318         struct obd_import      *imp = class_exp2cliimp(exp);
2319         struct ptlrpc_request  *req;
2320         char                   *tmp;
2321         int                     rc = -EINVAL;
2322         ENTRY;
2323
2324         req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
2325         if (req == NULL)
2326                 RETURN(-ENOMEM);
2327
2328         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
2329                              RCL_CLIENT, keylen);
2330         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
2331                              RCL_CLIENT, sizeof(vallen));
2332
2333         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
2334         if (rc) {
2335                 ptlrpc_request_free(req);
2336                 RETURN(rc);
2337         }
2338
2339         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
2340         memcpy(tmp, key, keylen);
2341         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
2342         memcpy(tmp, &vallen, sizeof(vallen));
2343
2344         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
2345                              RCL_SERVER, vallen);
2346         ptlrpc_request_set_replen(req);
2347
2348         rc = ptlrpc_queue_wait(req);
2349         /* -EREMOTE means the get_info result is partial, and it needs to
2350          * continue on another MDT, see fid2path part in lmv_iocontrol */
2351         if (rc == 0 || rc == -EREMOTE) {
2352                 tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
2353                 memcpy(val, tmp, vallen);
2354                 if (ptlrpc_rep_need_swab(req)) {
2355                         if (KEY_IS(KEY_FID2PATH))
2356                                 lustre_swab_fid2path(val);
2357                 }
2358         }
2359         ptlrpc_req_finished(req);
2360
2361         RETURN(rc);
2362 }
2363
2364 static void lustre_swab_hai(struct hsm_action_item *h)
2365 {
2366         __swab32s(&h->hai_len);
2367         __swab32s(&h->hai_action);
2368         lustre_swab_lu_fid(&h->hai_fid);
2369         lustre_swab_lu_fid(&h->hai_dfid);
2370         __swab64s(&h->hai_cookie);
2371         __swab64s(&h->hai_extent.offset);
2372         __swab64s(&h->hai_extent.length);
2373         __swab64s(&h->hai_gid);
2374 }
2375
2376 static void lustre_swab_hal(struct hsm_action_list *h)
2377 {
2378         struct hsm_action_item  *hai;
2379         __u32                    i;
2380
2381         __swab32s(&h->hal_version);
2382         __swab32s(&h->hal_count);
2383         __swab32s(&h->hal_archive_id);
2384         __swab64s(&h->hal_flags);
2385         hai = hai_first(h);
2386         for (i = 0; i < h->hal_count; i++, hai = hai_next(hai))
2387                 lustre_swab_hai(hai);
2388 }
2389
2390 static void lustre_swab_kuch(struct kuc_hdr *l)
2391 {
2392         __swab16s(&l->kuc_magic);
2393         /* __u8 l->kuc_transport */
2394         __swab16s(&l->kuc_msgtype);
2395         __swab16s(&l->kuc_msglen);
2396 }
2397
2398 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2399                                 struct lustre_kernelcomm *lk)
2400 {
2401         struct obd_import  *imp = class_exp2cliimp(exp);
2402         __u32               archive = lk->lk_data;
2403         int                 rc = 0;
2404
2405         if (lk->lk_group != KUC_GRP_HSM) {
2406                 CERROR("Bad copytool group %d\n", lk->lk_group);
2407                 return -EINVAL;
2408         }
2409
2410         CDEBUG(D_HSM, "CT start r%d w%d u%d g%d f%#x\n", lk->lk_rfd, lk->lk_wfd,
2411                lk->lk_uid, lk->lk_group, lk->lk_flags);
2412
2413         if (lk->lk_flags & LK_FLG_STOP) {
2414                 /* Unregister with the coordinator */
2415                 rc = mdc_ioc_hsm_ct_unregister(imp);
2416         } else {
2417                 rc = mdc_ioc_hsm_ct_register(imp, archive);
2418         }
2419
2420         return rc;
2421 }
2422
2423 /**
2424  * Send a message to any listening copytools
2425  * @param val KUC message (kuc_hdr + hsm_action_list)
2426  * @param len total length of message
2427  */
2428 static int mdc_hsm_copytool_send(size_t len, void *val)
2429 {
2430         struct kuc_hdr          *lh = (struct kuc_hdr *)val;
2431         struct hsm_action_list  *hal = (struct hsm_action_list *)(lh + 1);
2432         int                      rc;
2433         ENTRY;
2434
2435         if (len < sizeof(*lh) + sizeof(*hal)) {
2436                 CERROR("Short HSM message %zu < %zu\n", len,
2437                        sizeof(*lh) + sizeof(*hal));
2438                 RETURN(-EPROTO);
2439         }
2440         if (lh->kuc_magic == __swab16(KUC_MAGIC)) {
2441                 lustre_swab_kuch(lh);
2442                 lustre_swab_hal(hal);
2443         } else if (lh->kuc_magic != KUC_MAGIC) {
2444                 CERROR("Bad magic %x!=%x\n", lh->kuc_magic, KUC_MAGIC);
2445                 RETURN(-EPROTO);
2446         }
2447
2448         CDEBUG(D_HSM, " Received message mg=%x t=%d m=%d l=%d actions=%d "
2449                "on %s\n",
2450                lh->kuc_magic, lh->kuc_transport, lh->kuc_msgtype,
2451                lh->kuc_msglen, hal->hal_count, hal->hal_fsname);
2452
2453         /* Broadcast to HSM listeners */
2454         rc = libcfs_kkuc_group_put(KUC_GRP_HSM, lh);
2455
2456         RETURN(rc);
2457 }
2458
2459 /**
2460  * callback function passed to kuc for re-registering each HSM copytool
2461  * running on MDC, after MDT shutdown/recovery.
2462  * @param data copytool registration data
2463  * @param cb_arg callback argument (obd_import)
2464  */
2465 static int mdc_hsm_ct_reregister(void *data, void *cb_arg)
2466 {
2467         struct kkuc_ct_data     *kcd = data;
2468         struct obd_import       *imp = (struct obd_import *)cb_arg;
2469         int                      rc;
2470
2471         if (kcd == NULL || kcd->kcd_magic != KKUC_CT_DATA_MAGIC)
2472                 return -EPROTO;
2473
2474         if (!obd_uuid_equals(&kcd->kcd_uuid, &imp->imp_obd->obd_uuid))
2475                 return 0;
2476
2477         CDEBUG(D_HA, "%s: recover copytool registration to MDT (archive=%#x)\n",
2478                imp->imp_obd->obd_name, kcd->kcd_archive);
2479         rc = mdc_ioc_hsm_ct_register(imp, kcd->kcd_archive);
2480
2481         /* ignore error if the copytool is already registered */
2482         return (rc == -EEXIST) ? 0 : rc;
2483 }
2484
2485 /**
2486  * Re-establish all kuc contexts with MDT
2487  * after MDT shutdown/recovery.
2488  */
2489 static int mdc_kuc_reregister(struct obd_import *imp)
2490 {
2491         /* re-register HSM agents */
2492         return libcfs_kkuc_group_foreach(KUC_GRP_HSM, mdc_hsm_ct_reregister,
2493                                          (void *)imp);
2494 }
2495
2496 static int mdc_set_info_async(const struct lu_env *env,
2497                               struct obd_export *exp,
2498                               u32 keylen, void *key,
2499                               u32 vallen, void *val,
2500                               struct ptlrpc_request_set *set)
2501 {
2502         struct obd_import       *imp = class_exp2cliimp(exp);
2503         int                      rc;
2504         ENTRY;
2505
2506         if (KEY_IS(KEY_READ_ONLY)) {
2507                 if (vallen != sizeof(int))
2508                         RETURN(-EINVAL);
2509
2510                 spin_lock(&imp->imp_lock);
2511                 if (*((int *)val)) {
2512                         imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
2513                         imp->imp_connect_data.ocd_connect_flags |=
2514                                                         OBD_CONNECT_RDONLY;
2515                 } else {
2516                         imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
2517                         imp->imp_connect_data.ocd_connect_flags &=
2518                                                         ~OBD_CONNECT_RDONLY;
2519                 }
2520                 spin_unlock(&imp->imp_lock);
2521
2522                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2523                                        keylen, key, vallen, val, set);
2524                 RETURN(rc);
2525         }
2526         if (KEY_IS(KEY_SPTLRPC_CONF)) {
2527                 sptlrpc_conf_client_adapt(exp->exp_obd);
2528                 RETURN(0);
2529         }
2530         if (KEY_IS(KEY_FLUSH_CTX)) {
2531                 sptlrpc_import_flush_my_ctx(imp);
2532                 RETURN(0);
2533         }
2534         if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2535                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2536                                        keylen, key, vallen, val, set);
2537                 RETURN(rc);
2538         }
2539         if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) {
2540                 rc = mdc_hsm_copytool_send(vallen, val);
2541                 RETURN(rc);
2542         }
2543
2544         if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2545                 __u32 *default_easize = val;
2546
2547                 exp->exp_obd->u.cli.cl_default_mds_easize = *default_easize;
2548                 RETURN(0);
2549         }
2550
2551         CERROR("Unknown key %s\n", (char *)key);
2552         RETURN(-EINVAL);
2553 }
2554
2555 static int mdc_get_info(const struct lu_env *env, struct obd_export *exp,
2556                         __u32 keylen, void *key, __u32 *vallen, void *val)
2557 {
2558         int rc = -EINVAL;
2559
2560         if (KEY_IS(KEY_MAX_EASIZE)) {
2561                 __u32 mdsize, *max_easize;
2562
2563                 if (*vallen != sizeof(int))
2564                         RETURN(-EINVAL);
2565                 mdsize = *(__u32 *)val;
2566                 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
2567                         exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
2568                 max_easize = val;
2569                 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
2570                 RETURN(0);
2571         } else if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2572                 __u32 *default_easize;
2573
2574                 if (*vallen != sizeof(int))
2575                         RETURN(-EINVAL);
2576                 default_easize = val;
2577                 *default_easize = exp->exp_obd->u.cli.cl_default_mds_easize;
2578                 RETURN(0);
2579         } else if (KEY_IS(KEY_MAX_COOKIESIZE)) {
2580                 __u32 *max_cookiesize;
2581
2582                 if (*vallen != sizeof(int))
2583                         RETURN(-EINVAL);
2584                 max_cookiesize = val;
2585                 *max_cookiesize = exp->exp_obd->u.cli.cl_max_mds_cookiesize;
2586                 RETURN(0);
2587         } else if (KEY_IS(KEY_DEFAULT_COOKIESIZE)) {
2588                 __u32 *default_cookiesize;
2589
2590                 if (*vallen != sizeof(int))
2591                         RETURN(-EINVAL);
2592                 default_cookiesize = val;
2593                 *default_cookiesize =
2594                         exp->exp_obd->u.cli.cl_default_mds_cookiesize;
2595                 RETURN(0);
2596         } else if (KEY_IS(KEY_CONN_DATA)) {
2597                 struct obd_import *imp = class_exp2cliimp(exp);
2598                 struct obd_connect_data *data = val;
2599
2600                 if (*vallen != sizeof(*data))
2601                         RETURN(-EINVAL);
2602
2603                 *data = imp->imp_connect_data;
2604                 RETURN(0);
2605         } else if (KEY_IS(KEY_TGT_COUNT)) {
2606                 *((__u32 *)val) = 1;
2607                 RETURN(0);
2608         }
2609
2610         rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
2611
2612         RETURN(rc);
2613 }
2614
2615 static int mdc_fsync(struct obd_export *exp, const struct lu_fid *fid,
2616                      struct obd_capa *oc, struct ptlrpc_request **request)
2617 {
2618         struct ptlrpc_request *req;
2619         int                    rc;
2620         ENTRY;
2621
2622         *request = NULL;
2623         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
2624         if (req == NULL)
2625                 RETURN(-ENOMEM);
2626
2627         mdc_set_capa_size(req, &RMF_CAPA1, oc);
2628
2629         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
2630         if (rc) {
2631                 ptlrpc_request_free(req);
2632                 RETURN(rc);
2633         }
2634
2635         mdc_pack_body(req, fid, oc, 0, 0, -1, 0);
2636
2637         ptlrpc_request_set_replen(req);
2638
2639         rc = ptlrpc_queue_wait(req);
2640         if (rc)
2641                 ptlrpc_req_finished(req);
2642         else
2643                 *request = req;
2644         RETURN(rc);
2645 }
2646
2647 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
2648                             enum obd_import_event event)
2649 {
2650         int rc = 0;
2651
2652         LASSERT(imp->imp_obd == obd);
2653
2654         switch (event) {
2655         case IMP_EVENT_DISCON: {
2656 #if 0
2657                 /* XXX Pass event up to OBDs stack. used only for FLD now */
2658                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DISCON, NULL);
2659 #endif
2660                 break;
2661         }
2662         case IMP_EVENT_INACTIVE: {
2663                 struct client_obd *cli = &obd->u.cli;
2664                 /*
2665                  * Flush current sequence to make client obtain new one
2666                  * from server in case of disconnect/reconnect.
2667                  */
2668                 if (cli->cl_seq != NULL)
2669                         seq_client_flush(cli->cl_seq);
2670
2671                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
2672                 break;
2673         }
2674         case IMP_EVENT_INVALIDATE: {
2675                 struct ldlm_namespace *ns = obd->obd_namespace;
2676
2677                 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2678
2679                 break;
2680         }
2681         case IMP_EVENT_ACTIVE:
2682                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
2683                 /* redo the kuc registration after reconnecting */
2684                 if (rc == 0)
2685                         rc = mdc_kuc_reregister(imp);
2686                 break;
2687         case IMP_EVENT_OCD:
2688                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
2689                 break;
2690         case IMP_EVENT_DEACTIVATE:
2691         case IMP_EVENT_ACTIVATE:
2692                 break;
2693         default:
2694                 CERROR("Unknown import event %x\n", event);
2695                 LBUG();
2696         }
2697         RETURN(rc);
2698 }
2699
2700 int mdc_fid_alloc(const struct lu_env *env, struct obd_export *exp,
2701                   struct lu_fid *fid, struct md_op_data *op_data)
2702 {
2703         struct client_obd *cli = &exp->exp_obd->u.cli;
2704         struct lu_client_seq *seq = cli->cl_seq;
2705         ENTRY;
2706         RETURN(seq_client_alloc_fid(env, seq, fid));
2707 }
2708
2709 static struct obd_uuid *mdc_get_uuid(struct obd_export *exp)
2710 {
2711         struct client_obd *cli = &exp->exp_obd->u.cli;
2712         return &cli->cl_target_uuid;
2713 }
2714
2715 /**
2716  * Determine whether the lock can be canceled before replaying it during
2717  * recovery, non zero value will be return if the lock can be canceled,
2718  * or zero returned for not
2719  */
2720 static int mdc_cancel_weight(struct ldlm_lock *lock)
2721 {
2722         if (lock->l_resource->lr_type != LDLM_IBITS)
2723                 RETURN(0);
2724
2725         /* FIXME: if we ever get into a situation where there are too many
2726          * opened files with open locks on a single node, then we really
2727          * should replay these open locks to reget it */
2728         if (lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_OPEN)
2729                 RETURN(0);
2730
2731         RETURN(1);
2732 }
2733
2734 static int mdc_resource_inode_free(struct ldlm_resource *res)
2735 {
2736         if (res->lr_lvb_inode)
2737                 res->lr_lvb_inode = NULL;
2738
2739         return 0;
2740 }
2741
2742 static struct ldlm_valblock_ops inode_lvbo = {
2743         .lvbo_free = mdc_resource_inode_free
2744 };
2745
2746 static int mdc_llog_init(struct obd_device *obd)
2747 {
2748         struct obd_llog_group   *olg = &obd->obd_olg;
2749         struct llog_ctxt        *ctxt;
2750         int                      rc;
2751
2752         ENTRY;
2753
2754         rc = llog_setup(NULL, obd, olg, LLOG_CHANGELOG_REPL_CTXT, obd,
2755                         &llog_client_ops);
2756         if (rc < 0)
2757                 RETURN(rc);
2758
2759         ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
2760         llog_initiator_connect(ctxt);
2761         llog_ctxt_put(ctxt);
2762
2763         RETURN(0);
2764 }
2765
2766 static void mdc_llog_finish(struct obd_device *obd)
2767 {
2768         struct llog_ctxt *ctxt;
2769
2770         ENTRY;
2771
2772         ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
2773         if (ctxt != NULL)
2774                 llog_cleanup(NULL, ctxt);
2775
2776         EXIT;
2777 }
2778
2779 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
2780 {
2781         struct client_obd               *cli = &obd->u.cli;
2782         int                             rc;
2783         ENTRY;
2784
2785         OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
2786         if (!cli->cl_rpc_lock)
2787                 RETURN(-ENOMEM);
2788         mdc_init_rpc_lock(cli->cl_rpc_lock);
2789
2790         rc = ptlrpcd_addref();
2791         if (rc < 0)
2792                 GOTO(err_rpc_lock, rc);
2793
2794         OBD_ALLOC(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
2795         if (!cli->cl_close_lock)
2796                 GOTO(err_ptlrpcd_decref, rc = -ENOMEM);
2797         mdc_init_rpc_lock(cli->cl_close_lock);
2798
2799         rc = client_obd_setup(obd, cfg);
2800         if (rc)
2801                 GOTO(err_close_lock, rc);
2802 #ifdef CONFIG_PROC_FS
2803         obd->obd_vars = lprocfs_mdc_obd_vars;
2804         lprocfs_obd_setup(obd);
2805         lprocfs_alloc_md_stats(obd, 0);
2806 #endif
2807         sptlrpc_lprocfs_cliobd_attach(obd);
2808         ptlrpc_lprocfs_register_obd(obd);
2809
2810         ns_register_cancel(obd->obd_namespace, mdc_cancel_weight);
2811
2812         obd->obd_namespace->ns_lvbo = &inode_lvbo;
2813
2814         rc = mdc_llog_init(obd);
2815         if (rc) {
2816                 mdc_cleanup(obd);
2817                 CERROR("failed to setup llogging subsystems\n");
2818         }
2819
2820         RETURN(rc);
2821
2822 err_close_lock:
2823         OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
2824 err_ptlrpcd_decref:
2825         ptlrpcd_decref();
2826 err_rpc_lock:
2827         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
2828         RETURN(rc);
2829 }
2830
2831 /* Initialize the default and maximum LOV EA and cookie sizes.  This allows
2832  * us to make MDS RPCs with large enough reply buffers to hold a default
2833  * sized EA and cookie without having to calculate this (via a call into the
2834  * LOV + OSCs) each time we make an RPC.  The maximum size is also tracked
2835  * but not used to avoid wastefully vmalloc()'ing large reply buffers when
2836  * a large number of stripes is possible.  If a larger reply buffer is
2837  * required it will be reallocated in the ptlrpc layer due to overflow.
2838  */
2839 static int mdc_init_ea_size(struct obd_export *exp, __u32 easize,
2840                             __u32 def_easize, __u32 cookiesize,
2841                             __u32 def_cookiesize)
2842 {
2843         struct obd_device *obd = exp->exp_obd;
2844         struct client_obd *cli = &obd->u.cli;
2845         ENTRY;
2846
2847         if (cli->cl_max_mds_easize < easize)
2848                 cli->cl_max_mds_easize = easize;
2849
2850         if (cli->cl_default_mds_easize < def_easize)
2851                 cli->cl_default_mds_easize = def_easize;
2852
2853         if (cli->cl_max_mds_cookiesize < cookiesize)
2854                 cli->cl_max_mds_cookiesize = cookiesize;
2855
2856         if (cli->cl_default_mds_cookiesize < def_cookiesize)
2857                 cli->cl_default_mds_cookiesize = def_cookiesize;
2858
2859         RETURN(0);
2860 }
2861
2862 static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2863 {
2864         int rc = 0;
2865         ENTRY;
2866
2867         switch (stage) {
2868         case OBD_CLEANUP_EARLY:
2869                 break;
2870         case OBD_CLEANUP_EXPORTS:
2871                 /* Failsafe, ok if racy */
2872                 if (obd->obd_type->typ_refcnt <= 1)
2873                         libcfs_kkuc_group_rem(0, KUC_GRP_HSM, NULL);
2874
2875                 obd_cleanup_client_import(obd);
2876                 ptlrpc_lprocfs_unregister_obd(obd);
2877                 lprocfs_obd_cleanup(obd);
2878                 lprocfs_free_md_stats(obd);
2879                 mdc_llog_finish(obd);
2880                 break;
2881         }
2882         RETURN(rc);
2883 }
2884
2885 static int mdc_cleanup(struct obd_device *obd)
2886 {
2887         struct client_obd *cli = &obd->u.cli;
2888
2889         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
2890         OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
2891
2892         ptlrpcd_decref();
2893
2894         return client_obd_cleanup(obd);
2895 }
2896
2897 static int mdc_process_config(struct obd_device *obd, size_t len, void *buf)
2898 {
2899         struct lustre_cfg *lcfg = buf;
2900         int rc = class_process_proc_param(PARAM_MDC, obd->obd_vars, lcfg, obd);
2901         return (rc > 0 ? 0: rc);
2902 }
2903
2904
2905 /* get remote permission for current user on fid */
2906 static int mdc_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
2907                                struct obd_capa *oc, __u32 suppgid,
2908                                struct ptlrpc_request **request)
2909 {
2910         struct ptlrpc_request  *req;
2911         int                    rc;
2912         ENTRY;
2913
2914         LASSERT(client_is_remote(exp));
2915
2916         *request = NULL;
2917         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
2918         if (req == NULL)
2919                 RETURN(-ENOMEM);
2920
2921         mdc_set_capa_size(req, &RMF_CAPA1, oc);
2922
2923         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
2924         if (rc) {
2925                 ptlrpc_request_free(req);
2926                 RETURN(rc);
2927         }
2928
2929         mdc_pack_body(req, fid, oc, OBD_MD_FLRMTPERM, 0, suppgid, 0);
2930
2931         req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
2932                              sizeof(struct mdt_remote_perm));
2933
2934         ptlrpc_request_set_replen(req);
2935
2936         rc = ptlrpc_queue_wait(req);
2937         if (rc)
2938                 ptlrpc_req_finished(req);
2939         else
2940                 *request = req;
2941         RETURN(rc);
2942 }
2943
2944 static int mdc_interpret_renew_capa(const struct lu_env *env,
2945                                     struct ptlrpc_request *req, void *args,
2946                                     int status)
2947 {
2948         struct mdc_renew_capa_args *ra = args;
2949         struct mdt_body *body = NULL;
2950         struct lustre_capa *capa;
2951         ENTRY;
2952
2953         if (status)
2954                 GOTO(out, capa = ERR_PTR(status));
2955
2956         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2957         if (body == NULL)
2958                 GOTO(out, capa = ERR_PTR(-EFAULT));
2959
2960         if ((body->mbo_valid & OBD_MD_FLOSSCAPA) == 0)
2961                 GOTO(out, capa = ERR_PTR(-ENOENT));
2962
2963         capa = req_capsule_server_get(&req->rq_pill, &RMF_CAPA2);
2964         if (!capa)
2965                 GOTO(out, capa = ERR_PTR(-EFAULT));
2966         EXIT;
2967 out:
2968         ra->ra_cb(ra->ra_oc, capa);
2969         return 0;
2970 }
2971
2972 static int mdc_renew_capa(struct obd_export *exp, struct obd_capa *oc,
2973                           renew_capa_cb_t cb)
2974 {
2975         struct ptlrpc_request *req;
2976         struct mdc_renew_capa_args *ra;
2977         ENTRY;
2978
2979         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_GETATTR,
2980                                         LUSTRE_MDS_VERSION, MDS_GETATTR);
2981         if (req == NULL)
2982                 RETURN(-ENOMEM);
2983
2984         /* NB, OBD_MD_FLOSSCAPA is set here, but it doesn't necessarily mean the
2985          * capa to renew is oss capa.
2986          */
2987         mdc_pack_body(req, &oc->c_capa.lc_fid, oc, OBD_MD_FLOSSCAPA, 0, -1, 0);
2988         ptlrpc_request_set_replen(req);
2989
2990         CLASSERT(sizeof(*ra) <= sizeof(req->rq_async_args));
2991         ra = ptlrpc_req_async_args(req);
2992         ra->ra_oc = oc;
2993         ra->ra_cb = cb;
2994         req->rq_interpret_reply = mdc_interpret_renew_capa;
2995         ptlrpcd_add_req(req, PDL_POLICY_LOCAL, -1);
2996         RETURN(0);
2997 }
2998
2999 static struct obd_ops mdc_obd_ops = {
3000         .o_owner            = THIS_MODULE,
3001         .o_setup            = mdc_setup,
3002         .o_precleanup       = mdc_precleanup,
3003         .o_cleanup          = mdc_cleanup,
3004         .o_add_conn         = client_import_add_conn,
3005         .o_del_conn         = client_import_del_conn,
3006         .o_connect          = client_connect_import,
3007         .o_disconnect       = client_disconnect_export,
3008         .o_iocontrol        = mdc_iocontrol,
3009         .o_set_info_async   = mdc_set_info_async,
3010         .o_statfs           = mdc_statfs,
3011         .o_fid_init         = client_fid_init,
3012         .o_fid_fini         = client_fid_fini,
3013         .o_fid_alloc        = mdc_fid_alloc,
3014         .o_import_event     = mdc_import_event,
3015         .o_get_info         = mdc_get_info,
3016         .o_process_config   = mdc_process_config,
3017         .o_get_uuid         = mdc_get_uuid,
3018         .o_quotactl         = mdc_quotactl,
3019         .o_quotacheck       = mdc_quotacheck
3020 };
3021
3022 static struct md_ops mdc_md_ops = {
3023         .m_getstatus        = mdc_getstatus,
3024         .m_null_inode       = mdc_null_inode,
3025         .m_find_cbdata      = mdc_find_cbdata,
3026         .m_close            = mdc_close,
3027         .m_create           = mdc_create,
3028         .m_enqueue          = mdc_enqueue,
3029         .m_getattr          = mdc_getattr,
3030         .m_getattr_name     = mdc_getattr_name,
3031         .m_intent_lock      = mdc_intent_lock,
3032         .m_link             = mdc_link,
3033         .m_rename           = mdc_rename,
3034         .m_setattr          = mdc_setattr,
3035         .m_setxattr         = mdc_setxattr,
3036         .m_getxattr         = mdc_getxattr,
3037         .m_fsync                = mdc_fsync,
3038         .m_read_page            = mdc_read_page,
3039         .m_unlink           = mdc_unlink,
3040         .m_cancel_unused    = mdc_cancel_unused,
3041         .m_init_ea_size     = mdc_init_ea_size,
3042         .m_set_lock_data    = mdc_set_lock_data,
3043         .m_lock_match       = mdc_lock_match,
3044         .m_get_lustre_md    = mdc_get_lustre_md,
3045         .m_free_lustre_md   = mdc_free_lustre_md,
3046         .m_set_open_replay_data = mdc_set_open_replay_data,
3047         .m_clear_open_replay_data = mdc_clear_open_replay_data,
3048         .m_renew_capa       = mdc_renew_capa,
3049         .m_unpack_capa      = mdc_unpack_capa,
3050         .m_get_remote_perm  = mdc_get_remote_perm,
3051         .m_intent_getattr_async = mdc_intent_getattr_async,
3052         .m_revalidate_lock      = mdc_revalidate_lock
3053 };
3054
3055 static int __init mdc_init(void)
3056 {
3057         return class_register_type(&mdc_obd_ops, &mdc_md_ops, true, NULL,
3058                                    LUSTRE_MDC_NAME, NULL);
3059 }
3060
3061 static void /*__exit*/ mdc_exit(void)
3062 {
3063         class_unregister_type(LUSTRE_MDC_NAME);
3064 }
3065
3066 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3067 MODULE_DESCRIPTION("Lustre Metadata Client");
3068 MODULE_LICENSE("GPL");
3069
3070 module_init(mdc_init);
3071 module_exit(mdc_exit);