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->handle, sizeof(old)));
 686                 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
 687                 epoch->handle = body->mbo_handle;
 688         }
 689         EXIT;
 690 }
 691
 692 void mdc_commit_open(struct ptlrpc_request *req)
 693 {
 694         struct md_open_data *mod = req->rq_cb_data;
 695         if (mod == NULL)
 696                 return;
 697
 698         /**
 699          * No need to touch md_open_data::mod_och, it holds a reference on
 700          * \var mod and will zero references to each other, \var mod will be
 701          * freed after that when md_open_data::mod_och will put the reference.
 702          */
 703
 704         /**
 705          * Do not let open request to disappear as it still may be needed
 706          * for close rpc to happen (it may happen on evict only, otherwise
 707          * ptlrpc_request::rq_replay does not let mdc_commit_open() to be
 708          * called), just mark this rpc as committed to distinguish these 2
 709          * cases, see mdc_close() for details. The open request reference will
 710          * be put along with freeing \var mod.
 711          */
 712         ptlrpc_request_addref(req);
 713         spin_lock(&req->rq_lock);
 714         req->rq_committed = 1;
 715         spin_unlock(&req->rq_lock);
 716         req->rq_cb_data = NULL;
 717         obd_mod_put(mod);
 718 }
 719
 720 int mdc_set_open_replay_data(struct obd_export *exp,
 721                              struct obd_client_handle *och,
 722                              struct lookup_intent *it)
 723 {
 724         struct md_open_data     *mod;
 725         struct mdt_rec_create   *rec;
 726         struct mdt_body         *body;
 727         struct ptlrpc_request   *open_req = it->d.lustre.it_data;
 728         struct obd_import       *imp = open_req->rq_import;
 729         ENTRY;
 730
 731         if (!open_req->rq_replay)
 732                 RETURN(0);
 733
 734         rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
 735         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
 736         LASSERT(rec != NULL);
 737         /* Incoming message in my byte order (it's been swabbed). */
 738         /* Outgoing messages always in my byte order. */
 739         LASSERT(body != NULL);
 740
 741         /* Only if the import is replayable, we set replay_open data */
 742         if (och && imp->imp_replayable) {
 743                 mod = obd_mod_alloc();
 744                 if (mod == NULL) {
 745                         DEBUG_REQ(D_ERROR, open_req,
 746                                   "Can't allocate md_open_data");
 747                         RETURN(0);
 748                 }
 749
 750                 /**
 751                  * Take a reference on \var mod, to be freed on mdc_close().
 752                  * It protects \var mod from being freed on eviction (commit
 753                  * callback is called despite rq_replay flag).
 754                  * Another reference for \var och.
 755                  */
 756                 obd_mod_get(mod);
 757                 obd_mod_get(mod);
 758
 759                 spin_lock(&open_req->rq_lock);
 760                 och->och_mod = mod;
 761                 mod->mod_och = och;
 762                 mod->mod_is_create = it_disposition(it, DISP_OPEN_CREATE) ||
 763                                      it_disposition(it, DISP_OPEN_STRIPE);
 764                 mod->mod_open_req = open_req;
 765                 open_req->rq_cb_data = mod;
 766                 open_req->rq_commit_cb = mdc_commit_open;
 767                 spin_unlock(&open_req->rq_lock);
 768         }
 769
 770         rec->cr_fid2 = body->mbo_fid1;
 771         rec->cr_ioepoch = body->mbo_ioepoch;
 772         rec->cr_old_handle.cookie = body->mbo_handle.cookie;
 773         open_req->rq_replay_cb = mdc_replay_open;
 774         if (!fid_is_sane(&body->mbo_fid1)) {
 775                 DEBUG_REQ(D_ERROR, open_req, "Saving replay request with "
 776                           "insane fid");
 777                 LBUG();
 778         }
 779
 780         DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
 781         RETURN(0);
 782 }
 783
 784 static void mdc_free_open(struct md_open_data *mod)
 785 {
 786         int committed = 0;
 787
 788         if (mod->mod_is_create == 0 &&
 789             imp_connect_disp_stripe(mod->mod_open_req->rq_import))
 790                 committed = 1;
 791
 792         LASSERT(mod->mod_open_req->rq_replay == 0);
 793
 794         DEBUG_REQ(D_RPCTRACE, mod->mod_open_req, "free open request\n");
 795
 796         ptlrpc_request_committed(mod->mod_open_req, committed);
 797         if (mod->mod_close_req)
 798                 ptlrpc_request_committed(mod->mod_close_req, committed);
 799 }
 800
 801 int mdc_clear_open_replay_data(struct obd_export *exp,
 802                                struct obd_client_handle *och)
 803 {
 804         struct md_open_data *mod = och->och_mod;
 805         ENTRY;
 806
 807         /**
 808          * It is possible to not have \var mod in a case of eviction between
 809          * lookup and ll_file_open().
 810          **/
 811         if (mod == NULL)
 812                 RETURN(0);
 813
 814         LASSERT(mod != LP_POISON);
 815         LASSERT(mod->mod_open_req != NULL);
 816         mdc_free_open(mod);
 817
 818         mod->mod_och = NULL;
 819         och->och_mod = NULL;
 820         obd_mod_put(mod);
 821
 822         RETURN(0);
 823 }
 824
 825 static int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
 826                      struct md_open_data *mod, struct ptlrpc_request **request)
 827 {
 828         struct obd_device     *obd = class_exp2obd(exp);
 829         struct ptlrpc_request *req;
 830         struct req_format     *req_fmt;
 831         int                    rc;
 832         int                    saved_rc = 0;
 833         ENTRY;
 834
 835         req_fmt = &RQF_MDS_CLOSE;
 836         if (op_data->op_bias & MDS_HSM_RELEASE) {
 837                 req_fmt = &RQF_MDS_RELEASE_CLOSE;
 838
 839                 /* allocate a FID for volatile file */
 840                 rc = mdc_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
 841                 if (rc < 0) {
 842                         CERROR("%s: "DFID" failed to allocate FID: %d\n",
 843                                obd->obd_name, PFID(&op_data->op_fid1), rc);
 844                         /* save the errcode and proceed to close */
 845                         saved_rc = rc;
 846                 }
 847         }
 848
 849         *request = NULL;
 850         req = ptlrpc_request_alloc(class_exp2cliimp(exp), req_fmt);
 851         if (req == NULL)
 852                 RETURN(-ENOMEM);
 853
 854         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 855
 856         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
 857         if (rc) {
 858                 ptlrpc_request_free(req);
 859                 RETURN(rc);
 860         }
 861
 862         /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
 863          * portal whose threads are not taking any DLM locks and are therefore
 864          * always progressing */
 865         req->rq_request_portal = MDS_READPAGE_PORTAL;
 866         ptlrpc_at_set_req_timeout(req);
 867
 868         /* Ensure that this close's handle is fixed up during replay. */
 869         if (likely(mod != NULL)) {
 870                 LASSERTF(mod->mod_open_req != NULL &&
 871                          mod->mod_open_req->rq_type != LI_POISON,
 872                          "POISONED open %p!\n", mod->mod_open_req);
 873
 874                 mod->mod_close_req = req;
 875
 876                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
 877                 /* We no longer want to preserve this open for replay even
 878                  * though the open was committed. b=3632, b=3633 */
 879                 spin_lock(&mod->mod_open_req->rq_lock);
 880                 mod->mod_open_req->rq_replay = 0;
 881                 spin_unlock(&mod->mod_open_req->rq_lock);
 882         } else {
 883                  CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
 884         }
 885
 886         mdc_close_pack(req, op_data);
 887
 888         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 889                              obd->u.cli.cl_default_mds_easize);
 890         req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
 891                              obd->u.cli.cl_default_mds_cookiesize);
 892
 893         ptlrpc_request_set_replen(req);
 894
 895         mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 896         rc = ptlrpc_queue_wait(req);
 897         mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 898
 899         if (req->rq_repmsg == NULL) {
 900                 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
 901                        req->rq_status);
 902                 if (rc == 0)
 903                         rc = req->rq_status ?: -EIO;
 904         } else if (rc == 0 || rc == -EAGAIN) {
 905                 struct mdt_body *body;
 906
 907                 rc = lustre_msg_get_status(req->rq_repmsg);
 908                 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
 909                         DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
 910                                   "= %d", rc);
 911                         if (rc > 0)
 912                                 rc = -rc;
 913                 }
 914                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 915                 if (body == NULL)
 916                         rc = -EPROTO;
 917         } else if (rc == -ESTALE) {
 918                 /**
 919                  * it can be allowed error after 3633 if open was committed and
 920                  * server failed before close was sent. Let's check if mod
 921                  * exists and return no error in that case
 922                  */
 923                 if (mod) {
 924                         DEBUG_REQ(D_HA, req, "Reset ESTALE = %d", rc);
 925                         LASSERT(mod->mod_open_req != NULL);
 926                         if (mod->mod_open_req->rq_committed)
 927                                 rc = 0;
 928                 }
 929         }
 930
 931         if (mod) {
 932                 if (rc != 0)
 933                         mod->mod_close_req = NULL;
 934                 /* Since now, mod is accessed through open_req only,
 935                  * thus close req does not keep a reference on mod anymore. */
 936                 obd_mod_put(mod);
 937         }
 938         *request = req;
 939
 940         RETURN(rc < 0 ? rc : saved_rc);
 941 }
 942
 943 static int mdc_getpage(struct obd_export *exp, const struct lu_fid *fid,
 944                        __u64 offset, struct obd_capa *oc,
 945                        struct page **pages, int npages,
 946                        struct ptlrpc_request **request)
 947 {
 948         struct ptlrpc_request   *req;
 949         struct ptlrpc_bulk_desc *desc;
 950         int                      i;
 951         wait_queue_head_t        waitq;
 952         int                      resends = 0;
 953         struct l_wait_info       lwi;
 954         int                      rc;
 955         ENTRY;
 956
 957         *request = NULL;
 958         init_waitqueue_head(&waitq);
 959
 960 restart_bulk:
 961         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
 962         if (req == NULL)
 963                 RETURN(-ENOMEM);
 964
 965         mdc_set_capa_size(req, &RMF_CAPA1, oc);
 966
 967         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
 968         if (rc) {
 969                 ptlrpc_request_free(req);
 970                 RETURN(rc);
 971         }
 972
 973         req->rq_request_portal = MDS_READPAGE_PORTAL;
 974         ptlrpc_at_set_req_timeout(req);
 975
 976         desc = ptlrpc_prep_bulk_imp(req, npages, 1, BULK_PUT_SINK,
 977                                     MDS_BULK_PORTAL);
 978         if (desc == NULL) {
 979                 ptlrpc_request_free(req);
 980                 RETURN(-ENOMEM);
 981         }
 982
 983         /* NB req now owns desc and will free it when it gets freed */
 984         for (i = 0; i < npages; i++)
 985                 ptlrpc_prep_bulk_page_pin(desc, pages[i], 0, PAGE_CACHE_SIZE);
 986
 987         mdc_readdir_pack(req, offset, PAGE_CACHE_SIZE * npages, fid, oc);
 988
 989         ptlrpc_request_set_replen(req);
 990         rc = ptlrpc_queue_wait(req);
 991         if (rc) {
 992                 ptlrpc_req_finished(req);
 993                 if (rc != -ETIMEDOUT)
 994                         RETURN(rc);
 995
 996                 resends++;
 997                 if (!client_should_resend(resends, &exp->exp_obd->u.cli)) {
 998                         CERROR("%s: too many resend retries: rc = %d\n",
 999                                exp->exp_obd->obd_name, -EIO);
1000                         RETURN(-EIO);
1001                 }
1002                 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(resends), NULL, NULL,
1003                                        NULL);
1004                 l_wait_event(waitq, 0, &lwi);
1005
1006                 goto restart_bulk;
1007         }
1008
1009         rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
1010                                           req->rq_bulk->bd_nob_transferred);
1011         if (rc < 0) {
1012                 ptlrpc_req_finished(req);
1013                 RETURN(rc);
1014         }
1015
1016         if (req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK) {
1017                 CERROR("%s: unexpected bytes transferred: %d (%ld expected)\n",
1018                        exp->exp_obd->obd_name, req->rq_bulk->bd_nob_transferred,
1019                        PAGE_CACHE_SIZE * npages);
1020                 ptlrpc_req_finished(req);
1021                 RETURN(-EPROTO);
1022         }
1023
1024         *request = req;
1025         RETURN(0);
1026 }
1027
1028 static void mdc_release_page(struct page *page, int remove)
1029 {
1030         if (remove) {
1031                 lock_page(page);
1032                 if (likely(page->mapping != NULL))
1033                         truncate_complete_page(page->mapping, page);
1034                 unlock_page(page);
1035         }
1036         page_cache_release(page);
1037 }
1038
1039 static struct page *mdc_page_locate(struct address_space *mapping, __u64 *hash,
1040                                     __u64 *start, __u64 *end, int hash64)
1041 {
1042         /*
1043          * Complement of hash is used as an index so that
1044          * radix_tree_gang_lookup() can be used to find a page with starting
1045          * hash _smaller_ than one we are looking for.
1046          */
1047         unsigned long offset = hash_x_index(*hash, hash64);
1048         struct page *page;
1049         int found;
1050
1051         spin_lock_irq(&mapping->tree_lock);
1052         found = radix_tree_gang_lookup(&mapping->page_tree,
1053                                        (void **)&page, offset, 1);
1054         if (found > 0 && !radix_tree_exceptional_entry(page)) {
1055                 struct lu_dirpage *dp;
1056
1057                 page_cache_get(page);
1058                 spin_unlock_irq(&mapping->tree_lock);
1059                 /*
1060                  * In contrast to find_lock_page() we are sure that directory
1061                  * page cannot be truncated (while DLM lock is held) and,
1062                  * hence, can avoid restart.
1063                  *
1064                  * In fact, page cannot be locked here at all, because
1065                  * mdc_read_page_remote does synchronous io.
1066                  */
1067                 wait_on_page_locked(page);
1068                 if (PageUptodate(page)) {
1069                         dp = kmap(page);
1070                         if (BITS_PER_LONG == 32 && hash64) {
1071                                 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1072                                 *end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
1073                                 *hash  = *hash >> 32;
1074                         } else {
1075                                 *start = le64_to_cpu(dp->ldp_hash_start);
1076                                 *end   = le64_to_cpu(dp->ldp_hash_end);
1077                         }
1078                         if (unlikely(*start == 1 && *hash == 0))
1079                                 *hash = *start;
1080                         else
1081                                 LASSERTF(*start <= *hash, "start = "LPX64
1082                                          ",end = "LPX64",hash = "LPX64"\n",
1083                                          *start, *end, *hash);
1084                         CDEBUG(D_VFSTRACE, "offset %lx ["LPX64" "LPX64"],"
1085                               " hash "LPX64"\n", offset, *start, *end, *hash);
1086                         if (*hash > *end) {
1087                                 kunmap(page);
1088                                 mdc_release_page(page, 0);
1089                                 page = NULL;
1090                         } else if (*end != *start && *hash == *end) {
1091                                 /*
1092                                  * upon hash collision, remove this page,
1093                                  * otherwise put page reference, and
1094                                  * mdc_read_page_remote() will issue RPC to
1095                                  * fetch the page we want.
1096                                  */
1097                                 kunmap(page);
1098                                 mdc_release_page(page,
1099                                     le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
1100                                 page = NULL;
1101                         }
1102                 } else {
1103                         page_cache_release(page);
1104                         page = ERR_PTR(-EIO);
1105                 }
1106         } else {
1107                 spin_unlock_irq(&mapping->tree_lock);
1108                 page = NULL;
1109         }
1110         return page;
1111 }
1112
1113 /*
1114  * Adjust a set of pages, each page containing an array of lu_dirpages,
1115  * so that each page can be used as a single logical lu_dirpage.
1116  *
1117  * A lu_dirpage is laid out as follows, where s = ldp_hash_start,
1118  * e = ldp_hash_end, f = ldp_flags, p = padding, and each "ent" is a
1119  * struct lu_dirent.  It has size up to LU_PAGE_SIZE. The ldp_hash_end
1120  * value is used as a cookie to request the next lu_dirpage in a
1121  * directory listing that spans multiple pages (two in this example):
1122  *   ________
1123  *  |        |
1124  * .|--------v-------   -----.
1125  * |s|e|f|p|ent|ent| ... |ent|
1126  * '--|--------------   -----'   Each PAGE contains a single
1127  *    '------.                   lu_dirpage.
1128  * .---------v-------   -----.
1129  * |s|e|f|p|ent| 0 | ... | 0 |
1130  * '-----------------   -----'
1131  *
1132  * However, on hosts where the native VM page size (PAGE_CACHE_SIZE) is
1133  * larger than LU_PAGE_SIZE, a single host page may contain multiple
1134  * lu_dirpages. After reading the lu_dirpages from the MDS, the
1135  * ldp_hash_end of the first lu_dirpage refers to the one immediately
1136  * after it in the same PAGE (arrows simplified for brevity, but
1137  * in general e0==s1, e1==s2, etc.):
1138  *
1139  * .--------------------   -----.
1140  * |s0|e0|f0|p|ent|ent| ... |ent|
1141  * |---v----------------   -----|
1142  * |s1|e1|f1|p|ent|ent| ... |ent|
1143  * |---v----------------   -----|  Here, each PAGE contains
1144  *             ...                 multiple lu_dirpages.
1145  * |---v----------------   -----|
1146  * |s'|e'|f'|p|ent|ent| ... |ent|
1147  * '---|----------------   -----'
1148  *     v
1149  * .----------------------------.
1150  * |        next PAGE           |
1151  *
1152  * This structure is transformed into a single logical lu_dirpage as follows:
1153  *
1154  * - Replace e0 with e' so the request for the next lu_dirpage gets the page
1155  *   labeled 'next PAGE'.
1156  *
1157  * - Copy the LDF_COLLIDE flag from f' to f0 to correctly reflect whether
1158  *   a hash collision with the next page exists.
1159  *
1160  * - Adjust the lde_reclen of the ending entry of each lu_dirpage to span
1161  *   to the first entry of the next lu_dirpage.
1162  */
1163 #if PAGE_CACHE_SIZE > LU_PAGE_SIZE
1164 static void mdc_adjust_dirpages(struct page **pages, int cfs_pgs, int lu_pgs)
1165 {
1166         int i;
1167
1168         for (i = 0; i < cfs_pgs; i++) {
1169                 struct lu_dirpage       *dp = kmap(pages[i]);
1170                 struct lu_dirpage       *first = dp;
1171                 struct lu_dirent        *end_dirent = NULL;
1172                 struct lu_dirent        *ent;
1173                 __u64           hash_end = le64_to_cpu(dp->ldp_hash_end);
1174                 __u32           flags = le32_to_cpu(dp->ldp_flags);
1175
1176                 while (--lu_pgs > 0) {
1177                         ent = lu_dirent_start(dp);
1178                         for (end_dirent = ent; ent != NULL;
1179                              end_dirent = ent, ent = lu_dirent_next(ent));
1180
1181                         /* Advance dp to next lu_dirpage. */
1182                         dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE);
1183
1184                         /* Check if we've reached the end of the PAGE. */
1185                         if (!((unsigned long)dp & ~PAGE_MASK))
1186                                 break;
1187
1188                         /* Save the hash and flags of this lu_dirpage. */
1189                         hash_end = le64_to_cpu(dp->ldp_hash_end);
1190                         flags = le32_to_cpu(dp->ldp_flags);
1191
1192                         /* Check if lu_dirpage contains no entries. */
1193                         if (end_dirent == NULL)
1194                                 break;
1195
1196                         /* Enlarge the end entry lde_reclen from 0 to
1197                          * first entry of next lu_dirpage. */
1198                         LASSERT(le16_to_cpu(end_dirent->lde_reclen) == 0);
1199                         end_dirent->lde_reclen =
1200                                 cpu_to_le16((char *)(dp->ldp_entries) -
1201                                             (char *)end_dirent);
1202                 }
1203
1204                 first->ldp_hash_end = hash_end;
1205                 first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE);
1206                 first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE);
1207
1208                 kunmap(pages[i]);
1209         }
1210         LASSERTF(lu_pgs == 0, "left = %d\n", lu_pgs);
1211 }
1212 #else
1213 #define mdc_adjust_dirpages(pages, cfs_pgs, lu_pgs) do {} while (0)
1214 #endif  /* PAGE_CACHE_SIZE > LU_PAGE_SIZE */
1215
1216 /* parameters for readdir page */
1217 struct readpage_param {
1218         struct md_op_data       *rp_mod;
1219         __u64                   rp_off;
1220         int                     rp_hash64;
1221         struct obd_export       *rp_exp;
1222         struct md_callback      *rp_cb;
1223 };
1224
1225 #ifndef HAVE_DELETE_FROM_PAGE_CACHE
1226 static inline void delete_from_page_cache(struct page *page)
1227 {
1228         remove_from_page_cache(page);
1229         page_cache_release(page);
1230 }
1231 #endif
1232
1233 /**
1234  * Read pages from server.
1235  *
1236  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
1237  * a header lu_dirpage which describes the start/end hash, and whether this
1238  * page is empty (contains no dir entry) or hash collide with next page.
1239  * After client receives reply, several pages will be integrated into dir page
1240  * in PAGE_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the
1241  * lu_dirpage for this integrated page will be adjusted.
1242  **/
1243 static int mdc_read_page_remote(void *data, struct page *page0)
1244 {
1245         struct readpage_param   *rp = data;
1246         struct page             **page_pool;
1247         struct page             *page;
1248         struct lu_dirpage       *dp;
1249         int                     rd_pgs = 0; /* number of pages read actually */
1250         int                     npages;
1251         struct md_op_data       *op_data = rp->rp_mod;
1252         struct ptlrpc_request   *req;
1253         int                     max_pages = op_data->op_max_pages;
1254         struct inode            *inode;
1255         struct lu_fid           *fid;
1256         int                     i;
1257         int                     rc;
1258         ENTRY;
1259
1260         LASSERT(max_pages > 0 && max_pages <= PTLRPC_MAX_BRW_PAGES);
1261         inode = op_data->op_data;
1262         fid = &op_data->op_fid1;
1263         LASSERT(inode != NULL);
1264
1265         OBD_ALLOC(page_pool, sizeof(page_pool[0]) * max_pages);
1266         if (page_pool != NULL) {
1267                 page_pool[0] = page0;
1268         } else {
1269                 page_pool = &page0;
1270                 max_pages = 1;
1271         }
1272
1273         for (npages = 1; npages < max_pages; npages++) {
1274                 page = page_cache_alloc_cold(inode->i_mapping);
1275                 if (page == NULL)
1276                         break;
1277                 page_pool[npages] = page;
1278         }
1279
1280         rc = mdc_getpage(rp->rp_exp, fid, rp->rp_off, op_data->op_capa1,
1281                          page_pool, npages, &req);
1282         if (rc < 0) {
1283                 /* page0 is special, which was added into page cache early */
1284                 delete_from_page_cache(page0);
1285         } else {
1286                 int lu_pgs;
1287
1288                 rd_pgs = (req->rq_bulk->bd_nob_transferred +
1289                             PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1290                 lu_pgs = req->rq_bulk->bd_nob_transferred >>
1291                                                         LU_PAGE_SHIFT;
1292                 LASSERT(!(req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK));
1293
1294                 CDEBUG(D_INODE, "read %d(%d) pages\n", rd_pgs, lu_pgs);
1295
1296                 mdc_adjust_dirpages(page_pool, rd_pgs, lu_pgs);
1297
1298                 SetPageUptodate(page0);
1299         }
1300         unlock_page(page0);
1301
1302         ptlrpc_req_finished(req);
1303         CDEBUG(D_CACHE, "read %d/%d pages\n", rd_pgs, npages);
1304         for (i = 1; i < npages; i++) {
1305                 unsigned long   offset;
1306                 __u64           hash;
1307                 int ret;
1308
1309                 page = page_pool[i];
1310
1311                 if (rc < 0 || i >= rd_pgs) {
1312                         page_cache_release(page);
1313                         continue;
1314                 }
1315
1316                 SetPageUptodate(page);
1317
1318                 dp = kmap(page);
1319                 hash = le64_to_cpu(dp->ldp_hash_start);
1320                 kunmap(page);
1321
1322                 offset = hash_x_index(hash, rp->rp_hash64);
1323
1324                 prefetchw(&page->flags);
1325                 ret = add_to_page_cache_lru(page, inode->i_mapping, offset,
1326                                             GFP_KERNEL);
1327                 if (ret == 0)
1328                         unlock_page(page);
1329                 else
1330                         CDEBUG(D_VFSTRACE, "page %lu add to page cache failed:"
1331                                " rc = %d\n", offset, ret);
1332                 page_cache_release(page);
1333         }
1334
1335         if (page_pool != &page0)
1336                 OBD_FREE(page_pool, sizeof(page_pool[0]) * max_pages);
1337
1338         RETURN(rc);
1339 }
1340
1341 /**
1342  * Read dir page from cache first, if it can not find it, read it from
1343  * server and add into the cache.
1344  *
1345  * \param[in] exp       MDC export
1346  * \param[in] op_data   client MD stack parameters, transfering parameters
1347  *                      between different layers on client MD stack.
1348  * \param[in] cb_op     callback required for ldlm lock enqueue during
1349  *                      read page
1350  * \param[in] hash_offset the hash offset of the page to be read
1351  * \param[in] ppage     the page to be read
1352  *
1353  * retval               = 0 get the page successfully
1354  *                      errno(<0) get the page failed
1355  */
1356 static int mdc_read_page(struct obd_export *exp, struct md_op_data *op_data,
1357                          struct md_callback *cb_op, __u64 hash_offset,
1358                          struct page **ppage)
1359 {
1360         struct lookup_intent    it = { .it_op = IT_READDIR };
1361         struct page             *page;
1362         struct inode            *dir = op_data->op_data;
1363         struct address_space    *mapping;
1364         struct lu_dirpage       *dp;
1365         __u64                   start = 0;
1366         __u64                   end = 0;
1367         struct lustre_handle    lockh;
1368         struct ptlrpc_request   *enq_req = NULL;
1369         struct readpage_param   rp_param;
1370         int rc;
1371
1372         ENTRY;
1373
1374         *ppage = NULL;
1375
1376         LASSERT(dir != NULL);
1377         mapping = dir->i_mapping;
1378
1379         rc = mdc_intent_lock(exp, op_data, &it, &enq_req,
1380                              cb_op->md_blocking_ast, 0);
1381         if (enq_req != NULL)
1382                 ptlrpc_req_finished(enq_req);
1383
1384         if (rc < 0) {
1385                 CERROR("%s: "DFID" lock enqueue fails: rc = %d\n",
1386                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1), rc);
1387                 RETURN(rc);
1388         }
1389
1390         rc = 0;
1391         mdc_set_lock_data(exp, &it.d.lustre.it_lock_handle, dir, NULL);
1392
1393         rp_param.rp_off = hash_offset;
1394         rp_param.rp_hash64 = op_data->op_cli_flags & CLI_HASH64;
1395         page = mdc_page_locate(mapping, &rp_param.rp_off, &start, &end,
1396                                rp_param.rp_hash64);
1397         if (IS_ERR(page)) {
1398                 CERROR("%s: dir page locate: "DFID" at "LPU64": rc %ld\n",
1399                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1400                        rp_param.rp_off, PTR_ERR(page));
1401                 GOTO(out_unlock, rc = PTR_ERR(page));
1402         } else if (page != NULL) {
1403                 /*
1404                  * XXX nikita: not entirely correct handling of a corner case:
1405                  * suppose hash chain of entries with hash value HASH crosses
1406                  * border between pages P0 and P1. First both P0 and P1 are
1407                  * cached, seekdir() is called for some entry from the P0 part
1408                  * of the chain. Later P0 goes out of cache. telldir(HASH)
1409                  * happens and finds P1, as it starts with matching hash
1410                  * value. Remaining entries from P0 part of the chain are
1411                  * skipped. (Is that really a bug?)
1412                  *
1413                  * Possible solutions: 0. don't cache P1 is such case, handle
1414                  * it as an "overflow" page. 1. invalidate all pages at
1415                  * once. 2. use HASH|1 as an index for P1.
1416                  */
1417                 GOTO(hash_collision, page);
1418         }
1419
1420         rp_param.rp_exp = exp;
1421         rp_param.rp_mod = op_data;
1422         page = read_cache_page(mapping,
1423                                hash_x_index(rp_param.rp_off,
1424                                             rp_param.rp_hash64),
1425                                mdc_read_page_remote, &rp_param);
1426         if (IS_ERR(page)) {
1427                 CDEBUG(D_INFO, "%s: read cache page: "DFID" at "LPU64": %ld\n",
1428                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1429                        rp_param.rp_off, PTR_ERR(page));
1430                 GOTO(out_unlock, rc = PTR_ERR(page));
1431         }
1432
1433         wait_on_page_locked(page);
1434         (void)kmap(page);
1435         if (!PageUptodate(page)) {
1436                 CERROR("%s: page not updated: "DFID" at "LPU64": rc %d\n",
1437                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1438                        rp_param.rp_off, -5);
1439                 goto fail;
1440         }
1441         if (!PageChecked(page))
1442                 SetPageChecked(page);
1443         if (PageError(page)) {
1444                 CERROR("%s: page error: "DFID" at "LPU64": rc %d\n",
1445                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1446                        rp_param.rp_off, -5);
1447                 goto fail;
1448         }
1449
1450 hash_collision:
1451         dp = page_address(page);
1452         if (BITS_PER_LONG == 32 && rp_param.rp_hash64) {
1453                 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1454                 end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
1455                 rp_param.rp_off = hash_offset >> 32;
1456         } else {
1457                 start = le64_to_cpu(dp->ldp_hash_start);
1458                 end   = le64_to_cpu(dp->ldp_hash_end);
1459                 rp_param.rp_off = hash_offset;
1460         }
1461         if (end == start) {
1462                 LASSERT(start == rp_param.rp_off);
1463                 CWARN("Page-wide hash collision: %#lx\n", (unsigned long)end);
1464 #if BITS_PER_LONG == 32
1465                 CWARN("Real page-wide hash collision at ["LPU64" "LPU64"] with "
1466                       "hash "LPU64"\n", le64_to_cpu(dp->ldp_hash_start),
1467                       le64_to_cpu(dp->ldp_hash_end), hash_offset);
1468 #endif
1469
1470                 /*
1471                  * Fetch whole overflow chain...
1472                  *
1473                  * XXX not yet.
1474                  */
1475                 goto fail;
1476         }
1477         *ppage = page;
1478 out_unlock:
1479         lockh.cookie = it.d.lustre.it_lock_handle;
1480         ldlm_lock_decref(&lockh, it.d.lustre.it_lock_mode);
1481         it.d.lustre.it_lock_handle = 0;
1482         return rc;
1483 fail:
1484         kunmap(page);
1485         mdc_release_page(page, 1);
1486         rc = -EIO;
1487         goto out_unlock;
1488 }
1489
1490
1491 static int mdc_statfs(const struct lu_env *env,
1492                       struct obd_export *exp, struct obd_statfs *osfs,
1493                       __u64 max_age, __u32 flags)
1494 {
1495         struct obd_device     *obd = class_exp2obd(exp);
1496         struct ptlrpc_request *req;
1497         struct obd_statfs     *msfs;
1498         struct obd_import     *imp = NULL;
1499         int                    rc;
1500         ENTRY;
1501
1502         /*
1503          * Since the request might also come from lprocfs, so we need
1504          * sync this with client_disconnect_export Bug15684
1505          */
1506         down_read(&obd->u.cli.cl_sem);
1507         if (obd->u.cli.cl_import)
1508                 imp = class_import_get(obd->u.cli.cl_import);
1509         up_read(&obd->u.cli.cl_sem);
1510         if (!imp)
1511                 RETURN(-ENODEV);
1512
1513         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_STATFS,
1514                                         LUSTRE_MDS_VERSION, MDS_STATFS);
1515         if (req == NULL)
1516                 GOTO(output, rc = -ENOMEM);
1517
1518         ptlrpc_request_set_replen(req);
1519
1520         if (flags & OBD_STATFS_NODELAY) {
1521                 /* procfs requests not want stay in wait for avoid deadlock */
1522                 req->rq_no_resend = 1;
1523                 req->rq_no_delay = 1;
1524         }
1525
1526         rc = ptlrpc_queue_wait(req);
1527         if (rc) {
1528                 /* check connection error first */
1529                 if (imp->imp_connect_error)
1530                         rc = imp->imp_connect_error;
1531                 GOTO(out, rc);
1532         }
1533
1534         msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1535         if (msfs == NULL)
1536                 GOTO(out, rc = -EPROTO);
1537
1538         *osfs = *msfs;
1539         EXIT;
1540 out:
1541         ptlrpc_req_finished(req);
1542 output:
1543         class_import_put(imp);
1544         return rc;
1545 }
1546
1547 static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
1548 {
1549         __u32 keylen, vallen;
1550         void *key;
1551         int rc;
1552
1553         if (gf->gf_pathlen > PATH_MAX)
1554                 RETURN(-ENAMETOOLONG);
1555         if (gf->gf_pathlen < 2)
1556                 RETURN(-EOVERFLOW);
1557
1558         /* Key is KEY_FID2PATH + getinfo_fid2path description */
1559         keylen = cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf);
1560         OBD_ALLOC(key, keylen);
1561         if (key == NULL)
1562                 RETURN(-ENOMEM);
1563         memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
1564         memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf));
1565
1566         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
1567                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
1568
1569         if (!fid_is_sane(&gf->gf_fid))
1570                 GOTO(out, rc = -EINVAL);
1571
1572         /* Val is struct getinfo_fid2path result plus path */
1573         vallen = sizeof(*gf) + gf->gf_pathlen;
1574
1575         rc = obd_get_info(NULL, exp, keylen, key, &vallen, gf, NULL);
1576         if (rc != 0 && rc != -EREMOTE)
1577                 GOTO(out, rc);
1578
1579         if (vallen <= sizeof(*gf))
1580                 GOTO(out, rc = -EPROTO);
1581         else if (vallen > sizeof(*gf) + gf->gf_pathlen)
1582                 GOTO(out, rc = -EOVERFLOW);
1583
1584         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n%s\n",
1585                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno, gf->gf_path);
1586
1587 out:
1588         OBD_FREE(key, keylen);
1589         return rc;
1590 }
1591
1592 static int mdc_ioc_hsm_progress(struct obd_export *exp,
1593                                 struct hsm_progress_kernel *hpk)
1594 {
1595         struct obd_import               *imp = class_exp2cliimp(exp);
1596         struct hsm_progress_kernel      *req_hpk;
1597         struct ptlrpc_request           *req;
1598         int                              rc;
1599         ENTRY;
1600
1601         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_PROGRESS,
1602                                         LUSTRE_MDS_VERSION, MDS_HSM_PROGRESS);
1603         if (req == NULL)
1604                 GOTO(out, rc = -ENOMEM);
1605
1606         mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
1607
1608         /* Copy hsm_progress struct */
1609         req_hpk = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_PROGRESS);
1610         if (req_hpk == NULL)
1611                 GOTO(out, rc = -EPROTO);
1612
1613         *req_hpk = *hpk;
1614         req_hpk->hpk_errval = lustre_errno_hton(hpk->hpk_errval);
1615
1616         ptlrpc_request_set_replen(req);
1617
1618         rc = mdc_queue_wait(req);
1619         GOTO(out, rc);
1620 out:
1621         ptlrpc_req_finished(req);
1622         return rc;
1623 }
1624
1625 static int mdc_ioc_hsm_ct_register(struct obd_import *imp, __u32 archives)
1626 {
1627         __u32                   *archive_mask;
1628         struct ptlrpc_request   *req;
1629         int                      rc;
1630         ENTRY;
1631
1632         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_REGISTER,
1633                                         LUSTRE_MDS_VERSION,
1634                                         MDS_HSM_CT_REGISTER);
1635         if (req == NULL)
1636                 GOTO(out, rc = -ENOMEM);
1637
1638         mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
1639
1640         /* Copy hsm_progress struct */
1641         archive_mask = req_capsule_client_get(&req->rq_pill,
1642                                               &RMF_MDS_HSM_ARCHIVE);
1643         if (archive_mask == NULL)
1644                 GOTO(out, rc = -EPROTO);
1645
1646         *archive_mask = archives;
1647
1648         ptlrpc_request_set_replen(req);
1649
1650         rc = mdc_queue_wait(req);
1651         GOTO(out, rc);
1652 out:
1653         ptlrpc_req_finished(req);
1654         return rc;
1655 }
1656
1657 static int mdc_ioc_hsm_current_action(struct obd_export *exp,
1658                                       struct md_op_data *op_data)
1659 {
1660         struct hsm_current_action       *hca = op_data->op_data;
1661         struct hsm_current_action       *req_hca;
1662         struct ptlrpc_request           *req;
1663         int                              rc;
1664         ENTRY;
1665
1666         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1667                                    &RQF_MDS_HSM_ACTION);
1668         if (req == NULL)
1669                 RETURN(-ENOMEM);
1670
1671         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1672
1673         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_ACTION);
1674         if (rc) {
1675                 ptlrpc_request_free(req);
1676                 RETURN(rc);
1677         }
1678
1679         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
1680                       OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
1681
1682         ptlrpc_request_set_replen(req);
1683
1684         rc = mdc_queue_wait(req);
1685         if (rc)
1686                 GOTO(out, rc);
1687
1688         req_hca = req_capsule_server_get(&req->rq_pill,
1689                                          &RMF_MDS_HSM_CURRENT_ACTION);
1690         if (req_hca == NULL)
1691                 GOTO(out, rc = -EPROTO);
1692
1693         *hca = *req_hca;
1694
1695         EXIT;
1696 out:
1697         ptlrpc_req_finished(req);
1698         return rc;
1699 }
1700
1701 static int mdc_ioc_hsm_ct_unregister(struct obd_import *imp)
1702 {
1703         struct ptlrpc_request   *req;
1704         int                      rc;
1705         ENTRY;
1706
1707         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_UNREGISTER,
1708                                         LUSTRE_MDS_VERSION,
1709                                         MDS_HSM_CT_UNREGISTER);
1710         if (req == NULL)
1711                 GOTO(out, rc = -ENOMEM);
1712
1713         mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
1714
1715         ptlrpc_request_set_replen(req);
1716
1717         rc = mdc_queue_wait(req);
1718         GOTO(out, rc);
1719 out:
1720         ptlrpc_req_finished(req);
1721         return rc;
1722 }
1723
1724 static int mdc_ioc_hsm_state_get(struct obd_export *exp,
1725                                  struct md_op_data *op_data)
1726 {
1727         struct hsm_user_state   *hus = op_data->op_data;
1728         struct hsm_user_state   *req_hus;
1729         struct ptlrpc_request   *req;
1730         int                      rc;
1731         ENTRY;
1732
1733         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1734                                    &RQF_MDS_HSM_STATE_GET);
1735         if (req == NULL)
1736                 RETURN(-ENOMEM);
1737
1738         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1739
1740         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_GET);
1741         if (rc != 0) {
1742                 ptlrpc_request_free(req);
1743                 RETURN(rc);
1744         }
1745
1746         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
1747                       OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
1748
1749         ptlrpc_request_set_replen(req);
1750
1751         rc = mdc_queue_wait(req);
1752         if (rc)
1753                 GOTO(out, rc);
1754
1755         req_hus = req_capsule_server_get(&req->rq_pill, &RMF_HSM_USER_STATE);
1756         if (req_hus == NULL)
1757                 GOTO(out, rc = -EPROTO);
1758
1759         *hus = *req_hus;
1760
1761         EXIT;
1762 out:
1763         ptlrpc_req_finished(req);
1764         return rc;
1765 }
1766
1767 static int mdc_ioc_hsm_state_set(struct obd_export *exp,
1768                                  struct md_op_data *op_data)
1769 {
1770         struct hsm_state_set    *hss = op_data->op_data;
1771         struct hsm_state_set    *req_hss;
1772         struct ptlrpc_request   *req;
1773         int                      rc;
1774         ENTRY;
1775
1776         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1777                                    &RQF_MDS_HSM_STATE_SET);
1778         if (req == NULL)
1779                 RETURN(-ENOMEM);
1780
1781         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1782
1783         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_SET);
1784         if (rc) {
1785                 ptlrpc_request_free(req);
1786                 RETURN(rc);
1787         }
1788
1789         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
1790                       OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
1791
1792         /* Copy states */
1793         req_hss = req_capsule_client_get(&req->rq_pill, &RMF_HSM_STATE_SET);
1794         if (req_hss == NULL)
1795                 GOTO(out, rc = -EPROTO);
1796         *req_hss = *hss;
1797
1798         ptlrpc_request_set_replen(req);
1799
1800         rc = mdc_queue_wait(req);
1801         GOTO(out, rc);
1802
1803         EXIT;
1804 out:
1805         ptlrpc_req_finished(req);
1806         return rc;
1807 }
1808
1809 static int mdc_ioc_hsm_request(struct obd_export *exp,
1810                                struct hsm_user_request *hur)
1811 {
1812         struct obd_import       *imp = class_exp2cliimp(exp);
1813         struct ptlrpc_request   *req;
1814         struct hsm_request      *req_hr;
1815         struct hsm_user_item    *req_hui;
1816         char                    *req_opaque;
1817         int                      rc;
1818         ENTRY;
1819
1820         req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_REQUEST);
1821         if (req == NULL)
1822                 GOTO(out, rc = -ENOMEM);
1823
1824         req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM, RCL_CLIENT,
1825                              hur->hur_request.hr_itemcount
1826                              * sizeof(struct hsm_user_item));
1827         req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA, RCL_CLIENT,
1828                              hur->hur_request.hr_data_len);
1829
1830         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_REQUEST);
1831         if (rc) {
1832                 ptlrpc_request_free(req);
1833                 RETURN(rc);
1834         }
1835
1836         mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
1837
1838         /* Copy hsm_request struct */
1839         req_hr = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_REQUEST);
1840         if (req_hr == NULL)
1841                 GOTO(out, rc = -EPROTO);
1842         *req_hr = hur->hur_request;
1843
1844         /* Copy hsm_user_item structs */
1845         req_hui = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM);
1846         if (req_hui == NULL)
1847                 GOTO(out, rc = -EPROTO);
1848         memcpy(req_hui, hur->hur_user_item,
1849                hur->hur_request.hr_itemcount * sizeof(struct hsm_user_item));
1850
1851         /* Copy opaque field */
1852         req_opaque = req_capsule_client_get(&req->rq_pill, &RMF_GENERIC_DATA);
1853         if (req_opaque == NULL)
1854                 GOTO(out, rc = -EPROTO);
1855         memcpy(req_opaque, hur_data(hur), hur->hur_request.hr_data_len);
1856
1857         ptlrpc_request_set_replen(req);
1858
1859         rc = mdc_queue_wait(req);
1860         GOTO(out, rc);
1861
1862 out:
1863         ptlrpc_req_finished(req);
1864         return rc;
1865 }
1866
1867 static struct kuc_hdr *changelog_kuc_hdr(char *buf, size_t len, __u32 flags)
1868 {
1869         struct kuc_hdr *lh = (struct kuc_hdr *)buf;
1870
1871         LASSERT(len <= KUC_CHANGELOG_MSG_MAXSIZE);
1872
1873         lh->kuc_magic = KUC_MAGIC;
1874         lh->kuc_transport = KUC_TRANSPORT_CHANGELOG;
1875         lh->kuc_flags = flags;
1876         lh->kuc_msgtype = CL_RECORD;
1877         lh->kuc_msglen = len;
1878         return lh;
1879 }
1880
1881 struct changelog_show {
1882         __u64                            cs_startrec;
1883         enum changelog_send_flag         cs_flags;
1884         struct file                     *cs_fp;
1885         char                            *cs_buf;
1886         struct obd_device               *cs_obd;
1887 };
1888
1889 static inline char *cs_obd_name(struct changelog_show *cs)
1890 {
1891         return cs->cs_obd->obd_name;
1892 }
1893
1894 static int changelog_kkuc_cb(const struct lu_env *env, struct llog_handle *llh,
1895                              struct llog_rec_hdr *hdr, void *data)
1896 {
1897         struct changelog_show           *cs = data;
1898         struct llog_changelog_rec       *rec = (struct llog_changelog_rec *)hdr;
1899         struct kuc_hdr                  *lh;
1900         size_t                           len;
1901         int                              rc;
1902         ENTRY;
1903
1904         if (rec->cr_hdr.lrh_type != CHANGELOG_REC) {
1905                 rc = -EINVAL;
1906                 CERROR("%s: not a changelog rec %x/%d: rc = %d\n",
1907                        cs_obd_name(cs), rec->cr_hdr.lrh_type,
1908                        rec->cr.cr_type, rc);
1909                 RETURN(rc);
1910         }
1911
1912         if (rec->cr.cr_index < cs->cs_startrec) {
1913                 /* Skip entries earlier than what we are interested in */
1914                 CDEBUG(D_HSM, "rec="LPU64" start="LPU64"\n",
1915                        rec->cr.cr_index, cs->cs_startrec);
1916                 RETURN(0);
1917         }
1918
1919         CDEBUG(D_HSM, LPU64" %02d%-5s "LPU64" 0x%x t="DFID" p="DFID" %.*s\n",
1920                rec->cr.cr_index, rec->cr.cr_type,
1921                changelog_type2str(rec->cr.cr_type), rec->cr.cr_time,
1922                rec->cr.cr_flags & CLF_FLAGMASK,
1923                PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid),
1924                rec->cr.cr_namelen, changelog_rec_name(&rec->cr));
1925
1926         len = sizeof(*lh) + changelog_rec_size(&rec->cr) + rec->cr.cr_namelen;
1927
1928         /* Set up the message */
1929         lh = changelog_kuc_hdr(cs->cs_buf, len, cs->cs_flags);
1930         memcpy(lh + 1, &rec->cr, len - sizeof(*lh));
1931
1932         rc = libcfs_kkuc_msg_put(cs->cs_fp, lh);
1933         CDEBUG(D_HSM, "kucmsg fp %p len %zu rc %d\n", cs->cs_fp, len, rc);
1934
1935         RETURN(rc);
1936 }
1937
1938 static int mdc_changelog_send_thread(void *csdata)
1939 {
1940         struct changelog_show   *cs = csdata;
1941         struct llog_ctxt        *ctxt = NULL;
1942         struct llog_handle      *llh = NULL;
1943         struct kuc_hdr          *kuch;
1944         enum llog_flag           flags = LLOG_F_IS_CAT;
1945         int                      rc;
1946
1947         CDEBUG(D_HSM, "changelog to fp=%p start "LPU64"\n",
1948                cs->cs_fp, cs->cs_startrec);
1949
1950         OBD_ALLOC(cs->cs_buf, KUC_CHANGELOG_MSG_MAXSIZE);
1951         if (cs->cs_buf == NULL)
1952                 GOTO(out, rc = -ENOMEM);
1953
1954         /* Set up the remote catalog handle */
1955         ctxt = llog_get_context(cs->cs_obd, LLOG_CHANGELOG_REPL_CTXT);
1956         if (ctxt == NULL)
1957                 GOTO(out, rc = -ENOENT);
1958         rc = llog_open(NULL, ctxt, &llh, NULL, CHANGELOG_CATALOG,
1959                        LLOG_OPEN_EXISTS);
1960         if (rc) {
1961                 CERROR("%s: fail to open changelog catalog: rc = %d\n",
1962                        cs_obd_name(cs), rc);
1963                 GOTO(out, rc);
1964         }
1965
1966         if (cs->cs_flags & CHANGELOG_FLAG_JOBID)
1967                 flags |= LLOG_F_EXT_JOBID;
1968
1969         rc = llog_init_handle(NULL, llh, flags, NULL);
1970         if (rc) {
1971                 CERROR("llog_init_handle failed %d\n", rc);
1972                 GOTO(out, rc);
1973         }
1974
1975         rc = llog_cat_process(NULL, llh, changelog_kkuc_cb, cs, 0, 0);
1976
1977         /* Send EOF no matter what our result */
1978         if ((kuch = changelog_kuc_hdr(cs->cs_buf, sizeof(*kuch),
1979                                       cs->cs_flags))) {
1980                 kuch->kuc_msgtype = CL_EOF;
1981                 libcfs_kkuc_msg_put(cs->cs_fp, kuch);
1982         }
1983
1984 out:
1985         fput(cs->cs_fp);
1986         if (llh)
1987                 llog_cat_close(NULL, llh);
1988         if (ctxt)
1989                 llog_ctxt_put(ctxt);
1990         if (cs->cs_buf)
1991                 OBD_FREE(cs->cs_buf, KUC_CHANGELOG_MSG_MAXSIZE);
1992         OBD_FREE_PTR(cs);
1993         return rc;
1994 }
1995
1996 static int mdc_ioc_changelog_send(struct obd_device *obd,
1997                                   struct ioc_changelog *icc)
1998 {
1999         struct changelog_show *cs;
2000         struct task_struct *task;
2001         int rc;
2002
2003         /* Freed in mdc_changelog_send_thread */
2004         OBD_ALLOC_PTR(cs);
2005         if (!cs)
2006                 return -ENOMEM;
2007
2008         cs->cs_obd = obd;
2009         cs->cs_startrec = icc->icc_recno;
2010         /* matching fput in mdc_changelog_send_thread */
2011         cs->cs_fp = fget(icc->icc_id);
2012         cs->cs_flags = icc->icc_flags;
2013
2014         /*
2015          * New thread because we should return to user app before
2016          * writing into our pipe
2017          */
2018         task = kthread_run(mdc_changelog_send_thread, cs,
2019                            "mdc_clg_send_thread");
2020         if (IS_ERR(task)) {
2021                 rc = PTR_ERR(task);
2022                 CERROR("%s: cannot start changelog thread: rc = %d\n",
2023                        cs_obd_name(cs), rc);
2024                 OBD_FREE_PTR(cs);
2025         } else {
2026                 rc = 0;
2027                 CDEBUG(D_HSM, "%s: started changelog thread\n",
2028                        cs_obd_name(cs));
2029         }
2030
2031         return rc;
2032 }
2033
2034 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2035                                 struct lustre_kernelcomm *lk);
2036
2037 static int mdc_quotacheck(struct obd_device *unused, struct obd_export *exp,
2038                           struct obd_quotactl *oqctl)
2039 {
2040         struct client_obd       *cli = &exp->exp_obd->u.cli;
2041         struct ptlrpc_request   *req;
2042         struct obd_quotactl     *body;
2043         int                      rc;
2044         ENTRY;
2045
2046         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
2047                                         &RQF_MDS_QUOTACHECK, LUSTRE_MDS_VERSION,
2048                                         MDS_QUOTACHECK);
2049         if (req == NULL)
2050                 RETURN(-ENOMEM);
2051
2052         body = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
2053         *body = *oqctl;
2054
2055         ptlrpc_request_set_replen(req);
2056
2057         /* the next poll will find -ENODATA, that means quotacheck is
2058          * going on */
2059         cli->cl_qchk_stat = -ENODATA;
2060         rc = ptlrpc_queue_wait(req);
2061         if (rc)
2062                 cli->cl_qchk_stat = rc;
2063         ptlrpc_req_finished(req);
2064         RETURN(rc);
2065 }
2066
2067 static int mdc_quota_poll_check(struct obd_export *exp,
2068                                 struct if_quotacheck *qchk)
2069 {
2070         struct client_obd *cli = &exp->exp_obd->u.cli;
2071         int rc;
2072         ENTRY;
2073
2074         qchk->obd_uuid = cli->cl_target_uuid;
2075         memcpy(qchk->obd_type, LUSTRE_MDS_NAME, strlen(LUSTRE_MDS_NAME));
2076
2077         rc = cli->cl_qchk_stat;
2078         /* the client is not the previous one */
2079         if (rc == CL_NOT_QUOTACHECKED)
2080                 rc = -EINTR;
2081         RETURN(rc);
2082 }
2083
2084 static int mdc_quotactl(struct obd_device *unused, struct obd_export *exp,
2085                         struct obd_quotactl *oqctl)
2086 {
2087         struct ptlrpc_request   *req;
2088         struct obd_quotactl     *oqc;
2089         int                      rc;
2090         ENTRY;
2091
2092         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
2093                                         &RQF_MDS_QUOTACTL, LUSTRE_MDS_VERSION,
2094                                         MDS_QUOTACTL);
2095         if (req == NULL)
2096                 RETURN(-ENOMEM);
2097
2098         oqc = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
2099         *oqc = *oqctl;
2100
2101         ptlrpc_request_set_replen(req);
2102         ptlrpc_at_set_req_timeout(req);
2103         req->rq_no_resend = 1;
2104
2105         rc = ptlrpc_queue_wait(req);
2106         if (rc)
2107                 CERROR("ptlrpc_queue_wait failed, rc: %d\n", rc);
2108
2109         if (req->rq_repmsg &&
2110             (oqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL))) {
2111                 *oqctl = *oqc;
2112         } else if (!rc) {
2113                 CERROR ("Can't unpack obd_quotactl\n");
2114                 rc = -EPROTO;
2115         }
2116         ptlrpc_req_finished(req);
2117
2118         RETURN(rc);
2119 }
2120
2121 static int mdc_ioc_swap_layouts(struct obd_export *exp,
2122                                 struct md_op_data *op_data)
2123 {
2124         struct list_head cancels = LIST_HEAD_INIT(cancels);
2125         struct ptlrpc_request   *req;
2126         int                      rc, count;
2127         struct mdc_swap_layouts *msl, *payload;
2128         ENTRY;
2129
2130         msl = op_data->op_data;
2131
2132         /* When the MDT will get the MDS_SWAP_LAYOUTS RPC the
2133          * first thing it will do is to cancel the 2 layout
2134          * locks hold by this client.
2135          * So the client must cancel its layout locks on the 2 fids
2136          * with the request RPC to avoid extra RPC round trips
2137          */
2138         count = mdc_resource_get_unused(exp, &op_data->op_fid1, &cancels,
2139                                         LCK_EX, MDS_INODELOCK_LAYOUT |
2140                                         MDS_INODELOCK_XATTR);
2141         count += mdc_resource_get_unused(exp, &op_data->op_fid2, &cancels,
2142                                          LCK_EX, MDS_INODELOCK_LAYOUT |
2143                                          MDS_INODELOCK_XATTR);
2144
2145         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
2146                                    &RQF_MDS_SWAP_LAYOUTS);
2147         if (req == NULL) {
2148                 ldlm_lock_list_put(&cancels, l_bl_ast, count);
2149                 RETURN(-ENOMEM);
2150         }
2151
2152         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
2153         mdc_set_capa_size(req, &RMF_CAPA2, op_data->op_capa2);
2154
2155         rc = mdc_prep_elc_req(exp, req, MDS_SWAP_LAYOUTS, &cancels, count);
2156         if (rc) {
2157                 ptlrpc_request_free(req);
2158                 RETURN(rc);
2159         }
2160
2161         mdc_swap_layouts_pack(req, op_data);
2162
2163         payload = req_capsule_client_get(&req->rq_pill, &RMF_SWAP_LAYOUTS);
2164         LASSERT(payload);
2165
2166         *payload = *msl;
2167
2168         ptlrpc_request_set_replen(req);
2169
2170         rc = ptlrpc_queue_wait(req);
2171         if (rc)
2172                 GOTO(out, rc);
2173         EXIT;
2174
2175 out:
2176         ptlrpc_req_finished(req);
2177         return rc;
2178 }
2179
2180 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2181                          void *karg, void __user *uarg)
2182 {
2183         struct obd_device *obd = exp->exp_obd;
2184         struct obd_ioctl_data *data = karg;
2185         struct obd_import *imp = obd->u.cli.cl_import;
2186         int rc;
2187         ENTRY;
2188
2189         if (!try_module_get(THIS_MODULE)) {
2190                 CERROR("%s: cannot get module '%s'\n", obd->obd_name,
2191                        module_name(THIS_MODULE));
2192                 return -EINVAL;
2193         }
2194         switch (cmd) {
2195         case OBD_IOC_CHANGELOG_SEND:
2196                 rc = mdc_ioc_changelog_send(obd, karg);
2197                 GOTO(out, rc);
2198         case OBD_IOC_CHANGELOG_CLEAR: {
2199                 struct ioc_changelog *icc = karg;
2200                 struct changelog_setinfo cs =
2201                         {.cs_recno = icc->icc_recno, .cs_id = icc->icc_id};
2202                 rc = obd_set_info_async(NULL, exp, strlen(KEY_CHANGELOG_CLEAR),
2203                                         KEY_CHANGELOG_CLEAR, sizeof(cs), &cs,
2204                                         NULL);
2205                 GOTO(out, rc);
2206         }
2207         case OBD_IOC_FID2PATH:
2208                 rc = mdc_ioc_fid2path(exp, karg);
2209                 GOTO(out, rc);
2210         case LL_IOC_HSM_CT_START:
2211                 rc = mdc_ioc_hsm_ct_start(exp, karg);
2212                 /* ignore if it was already registered on this MDS. */
2213                 if (rc == -EEXIST)
2214                         rc = 0;
2215                 GOTO(out, rc);
2216         case LL_IOC_HSM_PROGRESS:
2217                 rc = mdc_ioc_hsm_progress(exp, karg);
2218                 GOTO(out, rc);
2219         case LL_IOC_HSM_STATE_GET:
2220                 rc = mdc_ioc_hsm_state_get(exp, karg);
2221                 GOTO(out, rc);
2222         case LL_IOC_HSM_STATE_SET:
2223                 rc = mdc_ioc_hsm_state_set(exp, karg);
2224                 GOTO(out, rc);
2225         case LL_IOC_HSM_ACTION:
2226                 rc = mdc_ioc_hsm_current_action(exp, karg);
2227                 GOTO(out, rc);
2228         case LL_IOC_HSM_REQUEST:
2229                 rc = mdc_ioc_hsm_request(exp, karg);
2230                 GOTO(out, rc);
2231         case OBD_IOC_CLIENT_RECOVER:
2232                 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1, 0);
2233                 if (rc < 0)
2234                         GOTO(out, rc);
2235                 GOTO(out, rc = 0);
2236         case IOC_OSC_SET_ACTIVE:
2237                 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
2238                 GOTO(out, rc);
2239         case OBD_IOC_POLL_QUOTACHECK:
2240                 rc = mdc_quota_poll_check(exp, (struct if_quotacheck *)karg);
2241                 GOTO(out, rc);
2242         case OBD_IOC_PING_TARGET:
2243                 rc = ptlrpc_obd_ping(obd);
2244                 GOTO(out, rc);
2245         /*
2246          * Normally IOC_OBD_STATFS, OBD_IOC_QUOTACTL iocontrol are handled by
2247          * LMV instead of MDC. But when the cluster is upgraded from 1.8,
2248          * there'd be no LMV layer thus we might be called here. Eventually
2249          * this code should be removed.
2250          * bz20731, LU-592.
2251          */
2252         case IOC_OBD_STATFS: {
2253                 struct obd_statfs stat_buf = {0};
2254
2255                 if (*((__u32 *) data->ioc_inlbuf2) != 0)
2256                         GOTO(out, rc = -ENODEV);
2257
2258                 /* copy UUID */
2259                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd),
2260                                  min((int)data->ioc_plen2,
2261                                      (int)sizeof(struct obd_uuid))))
2262                         GOTO(out, rc = -EFAULT);
2263
2264                 rc = mdc_statfs(NULL, obd->obd_self_export, &stat_buf,
2265                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
2266                                 0);
2267                 if (rc != 0)
2268                         GOTO(out, rc);
2269
2270                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
2271                                      min((int) data->ioc_plen1,
2272                                          (int) sizeof(stat_buf))))
2273                         GOTO(out, rc = -EFAULT);
2274
2275                 GOTO(out, rc = 0);
2276         }
2277         case OBD_IOC_QUOTACTL: {
2278                 struct if_quotactl *qctl = karg;
2279                 struct obd_quotactl *oqctl;
2280
2281                 OBD_ALLOC_PTR(oqctl);
2282                 if (oqctl == NULL)
2283                         GOTO(out, rc = -ENOMEM);
2284
2285                 QCTL_COPY(oqctl, qctl);
2286                 rc = obd_quotactl(exp, oqctl);
2287                 if (rc == 0) {
2288                         QCTL_COPY(qctl, oqctl);
2289                         qctl->qc_valid = QC_MDTIDX;
2290                         qctl->obd_uuid = obd->u.cli.cl_target_uuid;
2291                 }
2292
2293                 OBD_FREE_PTR(oqctl);
2294                 GOTO(out, rc);
2295         }
2296         case LL_IOC_GET_CONNECT_FLAGS:
2297                 if (copy_to_user(uarg, exp_connect_flags_ptr(exp),
2298                                  sizeof(*exp_connect_flags_ptr(exp))))
2299                         GOTO(out, rc = -EFAULT);
2300
2301                 GOTO(out, rc = 0);
2302         case LL_IOC_LOV_SWAP_LAYOUTS:
2303                 rc = mdc_ioc_swap_layouts(exp, karg);
2304                 GOTO(out, rc);
2305         default:
2306                 CERROR("unrecognised ioctl: cmd = %#x\n", cmd);
2307                 GOTO(out, rc = -ENOTTY);
2308         }
2309 out:
2310         module_put(THIS_MODULE);
2311
2312         return rc;
2313 }
2314
2315 static int mdc_get_info_rpc(struct obd_export *exp,
2316                             u32 keylen, void *key,
2317                             u32 vallen, void *val)
2318 {
2319         struct obd_import      *imp = class_exp2cliimp(exp);
2320         struct ptlrpc_request  *req;
2321         char                   *tmp;
2322         int                     rc = -EINVAL;
2323         ENTRY;
2324
2325         req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
2326         if (req == NULL)
2327                 RETURN(-ENOMEM);
2328
2329         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
2330                              RCL_CLIENT, keylen);
2331         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
2332                              RCL_CLIENT, sizeof(vallen));
2333
2334         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
2335         if (rc) {
2336                 ptlrpc_request_free(req);
2337                 RETURN(rc);
2338         }
2339
2340         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
2341         memcpy(tmp, key, keylen);
2342         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
2343         memcpy(tmp, &vallen, sizeof(vallen));
2344
2345         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
2346                              RCL_SERVER, vallen);
2347         ptlrpc_request_set_replen(req);
2348
2349         rc = ptlrpc_queue_wait(req);
2350         /* -EREMOTE means the get_info result is partial, and it needs to
2351          * continue on another MDT, see fid2path part in lmv_iocontrol */
2352         if (rc == 0 || rc == -EREMOTE) {
2353                 tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
2354                 memcpy(val, tmp, vallen);
2355                 if (ptlrpc_rep_need_swab(req)) {
2356                         if (KEY_IS(KEY_FID2PATH))
2357                                 lustre_swab_fid2path(val);
2358                 }
2359         }
2360         ptlrpc_req_finished(req);
2361
2362         RETURN(rc);
2363 }
2364
2365 static void lustre_swab_hai(struct hsm_action_item *h)
2366 {
2367         __swab32s(&h->hai_len);
2368         __swab32s(&h->hai_action);
2369         lustre_swab_lu_fid(&h->hai_fid);
2370         lustre_swab_lu_fid(&h->hai_dfid);
2371         __swab64s(&h->hai_cookie);
2372         __swab64s(&h->hai_extent.offset);
2373         __swab64s(&h->hai_extent.length);
2374         __swab64s(&h->hai_gid);
2375 }
2376
2377 static void lustre_swab_hal(struct hsm_action_list *h)
2378 {
2379         struct hsm_action_item  *hai;
2380         __u32                    i;
2381
2382         __swab32s(&h->hal_version);
2383         __swab32s(&h->hal_count);
2384         __swab32s(&h->hal_archive_id);
2385         __swab64s(&h->hal_flags);
2386         hai = hai_first(h);
2387         for (i = 0; i < h->hal_count; i++, hai = hai_next(hai))
2388                 lustre_swab_hai(hai);
2389 }
2390
2391 static void lustre_swab_kuch(struct kuc_hdr *l)
2392 {
2393         __swab16s(&l->kuc_magic);
2394         /* __u8 l->kuc_transport */
2395         __swab16s(&l->kuc_msgtype);
2396         __swab16s(&l->kuc_msglen);
2397 }
2398
2399 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2400                                 struct lustre_kernelcomm *lk)
2401 {
2402         struct obd_import  *imp = class_exp2cliimp(exp);
2403         __u32               archive = lk->lk_data;
2404         int                 rc = 0;
2405
2406         if (lk->lk_group != KUC_GRP_HSM) {
2407                 CERROR("Bad copytool group %d\n", lk->lk_group);
2408                 return -EINVAL;
2409         }
2410
2411         CDEBUG(D_HSM, "CT start r%d w%d u%d g%d f%#x\n", lk->lk_rfd, lk->lk_wfd,
2412                lk->lk_uid, lk->lk_group, lk->lk_flags);
2413
2414         if (lk->lk_flags & LK_FLG_STOP) {
2415                 /* Unregister with the coordinator */
2416                 rc = mdc_ioc_hsm_ct_unregister(imp);
2417         } else {
2418                 rc = mdc_ioc_hsm_ct_register(imp, archive);
2419         }
2420
2421         return rc;
2422 }
2423
2424 /**
2425  * Send a message to any listening copytools
2426  * @param val KUC message (kuc_hdr + hsm_action_list)
2427  * @param len total length of message
2428  */
2429 static int mdc_hsm_copytool_send(size_t len, void *val)
2430 {
2431         struct kuc_hdr          *lh = (struct kuc_hdr *)val;
2432         struct hsm_action_list  *hal = (struct hsm_action_list *)(lh + 1);
2433         int                      rc;
2434         ENTRY;
2435
2436         if (len < sizeof(*lh) + sizeof(*hal)) {
2437                 CERROR("Short HSM message %zu < %zu\n", len,
2438                        sizeof(*lh) + sizeof(*hal));
2439                 RETURN(-EPROTO);
2440         }
2441         if (lh->kuc_magic == __swab16(KUC_MAGIC)) {
2442                 lustre_swab_kuch(lh);
2443                 lustre_swab_hal(hal);
2444         } else if (lh->kuc_magic != KUC_MAGIC) {
2445                 CERROR("Bad magic %x!=%x\n", lh->kuc_magic, KUC_MAGIC);
2446                 RETURN(-EPROTO);
2447         }
2448
2449         CDEBUG(D_HSM, " Received message mg=%x t=%d m=%d l=%d actions=%d "
2450                "on %s\n",
2451                lh->kuc_magic, lh->kuc_transport, lh->kuc_msgtype,
2452                lh->kuc_msglen, hal->hal_count, hal->hal_fsname);
2453
2454         /* Broadcast to HSM listeners */
2455         rc = libcfs_kkuc_group_put(KUC_GRP_HSM, lh);
2456
2457         RETURN(rc);
2458 }
2459
2460 /**
2461  * callback function passed to kuc for re-registering each HSM copytool
2462  * running on MDC, after MDT shutdown/recovery.
2463  * @param data copytool registration data
2464  * @param cb_arg callback argument (obd_import)
2465  */
2466 static int mdc_hsm_ct_reregister(void *data, void *cb_arg)
2467 {
2468         struct kkuc_ct_data     *kcd = data;
2469         struct obd_import       *imp = (struct obd_import *)cb_arg;
2470         int                      rc;
2471
2472         if (kcd == NULL || kcd->kcd_magic != KKUC_CT_DATA_MAGIC)
2473                 return -EPROTO;
2474
2475         if (!obd_uuid_equals(&kcd->kcd_uuid, &imp->imp_obd->obd_uuid))
2476                 return 0;
2477
2478         CDEBUG(D_HA, "%s: recover copytool registration to MDT (archive=%#x)\n",
2479                imp->imp_obd->obd_name, kcd->kcd_archive);
2480         rc = mdc_ioc_hsm_ct_register(imp, kcd->kcd_archive);
2481
2482         /* ignore error if the copytool is already registered */
2483         return (rc == -EEXIST) ? 0 : rc;
2484 }
2485
2486 /**
2487  * Re-establish all kuc contexts with MDT
2488  * after MDT shutdown/recovery.
2489  */
2490 static int mdc_kuc_reregister(struct obd_import *imp)
2491 {
2492         /* re-register HSM agents */
2493         return libcfs_kkuc_group_foreach(KUC_GRP_HSM, mdc_hsm_ct_reregister,
2494                                          (void *)imp);
2495 }
2496
2497 static int mdc_set_info_async(const struct lu_env *env,
2498                               struct obd_export *exp,
2499                               u32 keylen, void *key,
2500                               u32 vallen, void *val,
2501                               struct ptlrpc_request_set *set)
2502 {
2503         struct obd_import       *imp = class_exp2cliimp(exp);
2504         int                      rc;
2505         ENTRY;
2506
2507         if (KEY_IS(KEY_READ_ONLY)) {
2508                 if (vallen != sizeof(int))
2509                         RETURN(-EINVAL);
2510
2511                 spin_lock(&imp->imp_lock);
2512                 if (*((int *)val)) {
2513                         imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
2514                         imp->imp_connect_data.ocd_connect_flags |=
2515                                                         OBD_CONNECT_RDONLY;
2516                 } else {
2517                         imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
2518                         imp->imp_connect_data.ocd_connect_flags &=
2519                                                         ~OBD_CONNECT_RDONLY;
2520                 }
2521                 spin_unlock(&imp->imp_lock);
2522
2523                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2524                                        keylen, key, vallen, val, set);
2525                 RETURN(rc);
2526         }
2527         if (KEY_IS(KEY_SPTLRPC_CONF)) {
2528                 sptlrpc_conf_client_adapt(exp->exp_obd);
2529                 RETURN(0);
2530         }
2531         if (KEY_IS(KEY_FLUSH_CTX)) {
2532                 sptlrpc_import_flush_my_ctx(imp);
2533                 RETURN(0);
2534         }
2535         if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2536                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2537                                        keylen, key, vallen, val, set);
2538                 RETURN(rc);
2539         }
2540         if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) {
2541                 rc = mdc_hsm_copytool_send(vallen, val);
2542                 RETURN(rc);
2543         }
2544
2545         if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2546                 __u32 *default_easize = val;
2547
2548                 exp->exp_obd->u.cli.cl_default_mds_easize = *default_easize;
2549                 RETURN(0);
2550         }
2551
2552         CERROR("Unknown key %s\n", (char *)key);
2553         RETURN(-EINVAL);
2554 }
2555
2556 static int mdc_get_info(const struct lu_env *env, struct obd_export *exp,
2557                         __u32 keylen, void *key,
2558                         __u32 *vallen, void *val,
2559                         struct lov_stripe_md *lsm)
2560 {
2561         int rc = -EINVAL;
2562
2563         if (KEY_IS(KEY_MAX_EASIZE)) {
2564                 __u32 mdsize, *max_easize;
2565
2566                 if (*vallen != sizeof(int))
2567                         RETURN(-EINVAL);
2568                 mdsize = *(__u32 *)val;
2569                 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
2570                         exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
2571                 max_easize = val;
2572                 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
2573                 RETURN(0);
2574         } else if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2575                 __u32 *default_easize;
2576
2577                 if (*vallen != sizeof(int))
2578                         RETURN(-EINVAL);
2579                 default_easize = val;
2580                 *default_easize = exp->exp_obd->u.cli.cl_default_mds_easize;
2581                 RETURN(0);
2582         } else if (KEY_IS(KEY_MAX_COOKIESIZE)) {
2583                 __u32 *max_cookiesize;
2584
2585                 if (*vallen != sizeof(int))
2586                         RETURN(-EINVAL);
2587                 max_cookiesize = val;
2588                 *max_cookiesize = exp->exp_obd->u.cli.cl_max_mds_cookiesize;
2589                 RETURN(0);
2590         } else if (KEY_IS(KEY_DEFAULT_COOKIESIZE)) {
2591                 __u32 *default_cookiesize;
2592
2593                 if (*vallen != sizeof(int))
2594                         RETURN(-EINVAL);
2595                 default_cookiesize = val;
2596                 *default_cookiesize =
2597                         exp->exp_obd->u.cli.cl_default_mds_cookiesize;
2598                 RETURN(0);
2599         } else if (KEY_IS(KEY_CONN_DATA)) {
2600                 struct obd_import *imp = class_exp2cliimp(exp);
2601                 struct obd_connect_data *data = val;
2602
2603                 if (*vallen != sizeof(*data))
2604                         RETURN(-EINVAL);
2605
2606                 *data = imp->imp_connect_data;
2607                 RETURN(0);
2608         } else if (KEY_IS(KEY_TGT_COUNT)) {
2609                 *((__u32 *)val) = 1;
2610                 RETURN(0);
2611         }
2612
2613         rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
2614
2615         RETURN(rc);
2616 }
2617
2618 static int mdc_fsync(struct obd_export *exp, const struct lu_fid *fid,
2619                      struct obd_capa *oc, struct ptlrpc_request **request)
2620 {
2621         struct ptlrpc_request *req;
2622         int                    rc;
2623         ENTRY;
2624
2625         *request = NULL;
2626         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
2627         if (req == NULL)
2628                 RETURN(-ENOMEM);
2629
2630         mdc_set_capa_size(req, &RMF_CAPA1, oc);
2631
2632         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
2633         if (rc) {
2634                 ptlrpc_request_free(req);
2635                 RETURN(rc);
2636         }
2637
2638         mdc_pack_body(req, fid, oc, 0, 0, -1, 0);
2639
2640         ptlrpc_request_set_replen(req);
2641
2642         rc = ptlrpc_queue_wait(req);
2643         if (rc)
2644                 ptlrpc_req_finished(req);
2645         else
2646                 *request = req;
2647         RETURN(rc);
2648 }
2649
2650 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
2651                             enum obd_import_event event)
2652 {
2653         int rc = 0;
2654
2655         LASSERT(imp->imp_obd == obd);
2656
2657         switch (event) {
2658         case IMP_EVENT_DISCON: {
2659 #if 0
2660                 /* XXX Pass event up to OBDs stack. used only for FLD now */
2661                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DISCON, NULL);
2662 #endif
2663                 break;
2664         }
2665         case IMP_EVENT_INACTIVE: {
2666                 struct client_obd *cli = &obd->u.cli;
2667                 /*
2668                  * Flush current sequence to make client obtain new one
2669                  * from server in case of disconnect/reconnect.
2670                  */
2671                 if (cli->cl_seq != NULL)
2672                         seq_client_flush(cli->cl_seq);
2673
2674                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
2675                 break;
2676         }
2677         case IMP_EVENT_INVALIDATE: {
2678                 struct ldlm_namespace *ns = obd->obd_namespace;
2679
2680                 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2681
2682                 break;
2683         }
2684         case IMP_EVENT_ACTIVE:
2685                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
2686                 /* redo the kuc registration after reconnecting */
2687                 if (rc == 0)
2688                         rc = mdc_kuc_reregister(imp);
2689                 break;
2690         case IMP_EVENT_OCD:
2691                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
2692                 break;
2693         case IMP_EVENT_DEACTIVATE:
2694         case IMP_EVENT_ACTIVATE:
2695                 break;
2696         default:
2697                 CERROR("Unknown import event %x\n", event);
2698                 LBUG();
2699         }
2700         RETURN(rc);
2701 }
2702
2703 int mdc_fid_alloc(const struct lu_env *env, struct obd_export *exp,
2704                   struct lu_fid *fid, struct md_op_data *op_data)
2705 {
2706         struct client_obd *cli = &exp->exp_obd->u.cli;
2707         struct lu_client_seq *seq = cli->cl_seq;
2708         ENTRY;
2709         RETURN(seq_client_alloc_fid(env, seq, fid));
2710 }
2711
2712 static struct obd_uuid *mdc_get_uuid(struct obd_export *exp)
2713 {
2714         struct client_obd *cli = &exp->exp_obd->u.cli;
2715         return &cli->cl_target_uuid;
2716 }
2717
2718 /**
2719  * Determine whether the lock can be canceled before replaying it during
2720  * recovery, non zero value will be return if the lock can be canceled,
2721  * or zero returned for not
2722  */
2723 static int mdc_cancel_weight(struct ldlm_lock *lock)
2724 {
2725         if (lock->l_resource->lr_type != LDLM_IBITS)
2726                 RETURN(0);
2727
2728         /* FIXME: if we ever get into a situation where there are too many
2729          * opened files with open locks on a single node, then we really
2730          * should replay these open locks to reget it */
2731         if (lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_OPEN)
2732                 RETURN(0);
2733
2734         RETURN(1);
2735 }
2736
2737 static int mdc_resource_inode_free(struct ldlm_resource *res)
2738 {
2739         if (res->lr_lvb_inode)
2740                 res->lr_lvb_inode = NULL;
2741
2742         return 0;
2743 }
2744
2745 static struct ldlm_valblock_ops inode_lvbo = {
2746         .lvbo_free = mdc_resource_inode_free
2747 };
2748
2749 static int mdc_llog_init(struct obd_device *obd)
2750 {
2751         struct obd_llog_group   *olg = &obd->obd_olg;
2752         struct llog_ctxt        *ctxt;
2753         int                      rc;
2754
2755         ENTRY;
2756
2757         rc = llog_setup(NULL, obd, olg, LLOG_CHANGELOG_REPL_CTXT, obd,
2758                         &llog_client_ops);
2759         if (rc < 0)
2760                 RETURN(rc);
2761
2762         ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
2763         llog_initiator_connect(ctxt);
2764         llog_ctxt_put(ctxt);
2765
2766         RETURN(0);
2767 }
2768
2769 static void mdc_llog_finish(struct obd_device *obd)
2770 {
2771         struct llog_ctxt *ctxt;
2772
2773         ENTRY;
2774
2775         ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
2776         if (ctxt != NULL)
2777                 llog_cleanup(NULL, ctxt);
2778
2779         EXIT;
2780 }
2781
2782 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
2783 {
2784         struct client_obd               *cli = &obd->u.cli;
2785         int                             rc;
2786         ENTRY;
2787
2788         OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
2789         if (!cli->cl_rpc_lock)
2790                 RETURN(-ENOMEM);
2791         mdc_init_rpc_lock(cli->cl_rpc_lock);
2792
2793         rc = ptlrpcd_addref();
2794         if (rc < 0)
2795                 GOTO(err_rpc_lock, rc);
2796
2797         OBD_ALLOC(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
2798         if (!cli->cl_close_lock)
2799                 GOTO(err_ptlrpcd_decref, rc = -ENOMEM);
2800         mdc_init_rpc_lock(cli->cl_close_lock);
2801
2802         rc = client_obd_setup(obd, cfg);
2803         if (rc)
2804                 GOTO(err_close_lock, rc);
2805 #ifdef CONFIG_PROC_FS
2806         obd->obd_vars = lprocfs_mdc_obd_vars;
2807         lprocfs_obd_setup(obd);
2808         lprocfs_alloc_md_stats(obd, 0);
2809 #endif
2810         sptlrpc_lprocfs_cliobd_attach(obd);
2811         ptlrpc_lprocfs_register_obd(obd);
2812
2813         ns_register_cancel(obd->obd_namespace, mdc_cancel_weight);
2814
2815         obd->obd_namespace->ns_lvbo = &inode_lvbo;
2816
2817         rc = mdc_llog_init(obd);
2818         if (rc) {
2819                 mdc_cleanup(obd);
2820                 CERROR("failed to setup llogging subsystems\n");
2821         }
2822
2823         RETURN(rc);
2824
2825 err_close_lock:
2826         OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
2827 err_ptlrpcd_decref:
2828         ptlrpcd_decref();
2829 err_rpc_lock:
2830         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
2831         RETURN(rc);
2832 }
2833
2834 /* Initialize the default and maximum LOV EA and cookie sizes.  This allows
2835  * us to make MDS RPCs with large enough reply buffers to hold a default
2836  * sized EA and cookie without having to calculate this (via a call into the
2837  * LOV + OSCs) each time we make an RPC.  The maximum size is also tracked
2838  * but not used to avoid wastefully vmalloc()'ing large reply buffers when
2839  * a large number of stripes is possible.  If a larger reply buffer is
2840  * required it will be reallocated in the ptlrpc layer due to overflow.
2841  */
2842 static int mdc_init_ea_size(struct obd_export *exp, __u32 easize,
2843                             __u32 def_easize, __u32 cookiesize,
2844                             __u32 def_cookiesize)
2845 {
2846         struct obd_device *obd = exp->exp_obd;
2847         struct client_obd *cli = &obd->u.cli;
2848         ENTRY;
2849
2850         if (cli->cl_max_mds_easize < easize)
2851                 cli->cl_max_mds_easize = easize;
2852
2853         if (cli->cl_default_mds_easize < def_easize)
2854                 cli->cl_default_mds_easize = def_easize;
2855
2856         if (cli->cl_max_mds_cookiesize < cookiesize)
2857                 cli->cl_max_mds_cookiesize = cookiesize;
2858
2859         if (cli->cl_default_mds_cookiesize < def_cookiesize)
2860                 cli->cl_default_mds_cookiesize = def_cookiesize;
2861
2862         RETURN(0);
2863 }
2864
2865 static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2866 {
2867         int rc = 0;
2868         ENTRY;
2869
2870         switch (stage) {
2871         case OBD_CLEANUP_EARLY:
2872                 break;
2873         case OBD_CLEANUP_EXPORTS:
2874                 /* Failsafe, ok if racy */
2875                 if (obd->obd_type->typ_refcnt <= 1)
2876                         libcfs_kkuc_group_rem(0, KUC_GRP_HSM, NULL);
2877
2878                 obd_cleanup_client_import(obd);
2879                 ptlrpc_lprocfs_unregister_obd(obd);
2880                 lprocfs_obd_cleanup(obd);
2881                 lprocfs_free_md_stats(obd);
2882                 mdc_llog_finish(obd);
2883                 break;
2884         }
2885         RETURN(rc);
2886 }
2887
2888 static int mdc_cleanup(struct obd_device *obd)
2889 {
2890         struct client_obd *cli = &obd->u.cli;
2891
2892         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
2893         OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
2894
2895         ptlrpcd_decref();
2896
2897         return client_obd_cleanup(obd);
2898 }
2899
2900 static int mdc_process_config(struct obd_device *obd, size_t len, void *buf)
2901 {
2902         struct lustre_cfg *lcfg = buf;
2903         int rc = class_process_proc_param(PARAM_MDC, obd->obd_vars, lcfg, obd);
2904         return (rc > 0 ? 0: rc);
2905 }
2906
2907
2908 /* get remote permission for current user on fid */
2909 static int mdc_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
2910                                struct obd_capa *oc, __u32 suppgid,
2911                                struct ptlrpc_request **request)
2912 {
2913         struct ptlrpc_request  *req;
2914         int                    rc;
2915         ENTRY;
2916
2917         LASSERT(client_is_remote(exp));
2918
2919         *request = NULL;
2920         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
2921         if (req == NULL)
2922                 RETURN(-ENOMEM);
2923
2924         mdc_set_capa_size(req, &RMF_CAPA1, oc);
2925
2926         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
2927         if (rc) {
2928                 ptlrpc_request_free(req);
2929                 RETURN(rc);
2930         }
2931
2932         mdc_pack_body(req, fid, oc, OBD_MD_FLRMTPERM, 0, suppgid, 0);
2933
2934         req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
2935                              sizeof(struct mdt_remote_perm));
2936
2937         ptlrpc_request_set_replen(req);
2938
2939         rc = ptlrpc_queue_wait(req);
2940         if (rc)
2941                 ptlrpc_req_finished(req);
2942         else
2943                 *request = req;
2944         RETURN(rc);
2945 }
2946
2947 static int mdc_interpret_renew_capa(const struct lu_env *env,
2948                                     struct ptlrpc_request *req, void *args,
2949                                     int status)
2950 {
2951         struct mdc_renew_capa_args *ra = args;
2952         struct mdt_body *body = NULL;
2953         struct lustre_capa *capa;
2954         ENTRY;
2955
2956         if (status)
2957                 GOTO(out, capa = ERR_PTR(status));
2958
2959         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2960         if (body == NULL)
2961                 GOTO(out, capa = ERR_PTR(-EFAULT));
2962
2963         if ((body->mbo_valid & OBD_MD_FLOSSCAPA) == 0)
2964                 GOTO(out, capa = ERR_PTR(-ENOENT));
2965
2966         capa = req_capsule_server_get(&req->rq_pill, &RMF_CAPA2);
2967         if (!capa)
2968                 GOTO(out, capa = ERR_PTR(-EFAULT));
2969         EXIT;
2970 out:
2971         ra->ra_cb(ra->ra_oc, capa);
2972         return 0;
2973 }
2974
2975 static int mdc_renew_capa(struct obd_export *exp, struct obd_capa *oc,
2976                           renew_capa_cb_t cb)
2977 {
2978         struct ptlrpc_request *req;
2979         struct mdc_renew_capa_args *ra;
2980         ENTRY;
2981
2982         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_GETATTR,
2983                                         LUSTRE_MDS_VERSION, MDS_GETATTR);
2984         if (req == NULL)
2985                 RETURN(-ENOMEM);
2986
2987         /* NB, OBD_MD_FLOSSCAPA is set here, but it doesn't necessarily mean the
2988          * capa to renew is oss capa.
2989          */
2990         mdc_pack_body(req, &oc->c_capa.lc_fid, oc, OBD_MD_FLOSSCAPA, 0, -1, 0);
2991         ptlrpc_request_set_replen(req);
2992
2993         CLASSERT(sizeof(*ra) <= sizeof(req->rq_async_args));
2994         ra = ptlrpc_req_async_args(req);
2995         ra->ra_oc = oc;
2996         ra->ra_cb = cb;
2997         req->rq_interpret_reply = mdc_interpret_renew_capa;
2998         ptlrpcd_add_req(req, PDL_POLICY_LOCAL, -1);
2999         RETURN(0);
3000 }
3001
3002 static struct obd_ops mdc_obd_ops = {
3003         .o_owner            = THIS_MODULE,
3004         .o_setup            = mdc_setup,
3005         .o_precleanup       = mdc_precleanup,
3006         .o_cleanup          = mdc_cleanup,
3007         .o_add_conn         = client_import_add_conn,
3008         .o_del_conn         = client_import_del_conn,
3009         .o_connect          = client_connect_import,
3010         .o_disconnect       = client_disconnect_export,
3011         .o_iocontrol        = mdc_iocontrol,
3012         .o_set_info_async   = mdc_set_info_async,
3013         .o_statfs           = mdc_statfs,
3014         .o_fid_init         = client_fid_init,
3015         .o_fid_fini         = client_fid_fini,
3016         .o_fid_alloc        = mdc_fid_alloc,
3017         .o_import_event     = mdc_import_event,
3018         .o_get_info         = mdc_get_info,
3019         .o_process_config   = mdc_process_config,
3020         .o_get_uuid         = mdc_get_uuid,
3021         .o_quotactl         = mdc_quotactl,
3022         .o_quotacheck       = mdc_quotacheck
3023 };
3024
3025 static struct md_ops mdc_md_ops = {
3026         .m_getstatus        = mdc_getstatus,
3027         .m_null_inode       = mdc_null_inode,
3028         .m_find_cbdata      = mdc_find_cbdata,
3029         .m_close            = mdc_close,
3030         .m_create           = mdc_create,
3031         .m_enqueue          = mdc_enqueue,
3032         .m_getattr          = mdc_getattr,
3033         .m_getattr_name     = mdc_getattr_name,
3034         .m_intent_lock      = mdc_intent_lock,
3035         .m_link             = mdc_link,
3036         .m_rename           = mdc_rename,
3037         .m_setattr          = mdc_setattr,
3038         .m_setxattr         = mdc_setxattr,
3039         .m_getxattr         = mdc_getxattr,
3040         .m_fsync                = mdc_fsync,
3041         .m_read_page            = mdc_read_page,
3042         .m_unlink           = mdc_unlink,
3043         .m_cancel_unused    = mdc_cancel_unused,
3044         .m_init_ea_size     = mdc_init_ea_size,
3045         .m_set_lock_data    = mdc_set_lock_data,
3046         .m_lock_match       = mdc_lock_match,
3047         .m_get_lustre_md    = mdc_get_lustre_md,
3048         .m_free_lustre_md   = mdc_free_lustre_md,
3049         .m_set_open_replay_data = mdc_set_open_replay_data,
3050         .m_clear_open_replay_data = mdc_clear_open_replay_data,
3051         .m_renew_capa       = mdc_renew_capa,
3052         .m_unpack_capa      = mdc_unpack_capa,
3053         .m_get_remote_perm  = mdc_get_remote_perm,
3054         .m_intent_getattr_async = mdc_intent_getattr_async,
3055         .m_revalidate_lock      = mdc_revalidate_lock
3056 };
3057
3058 static int __init mdc_init(void)
3059 {
3060         return class_register_type(&mdc_obd_ops, &mdc_md_ops, true, NULL,
3061                                    LUSTRE_MDC_NAME, NULL);
3062 }
3063
3064 static void /*__exit*/ mdc_exit(void)
3065 {
3066         class_unregister_type(LUSTRE_MDC_NAME);
3067 }
3068
3069 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3070 MODULE_DESCRIPTION("Lustre Metadata Client");
3071 MODULE_LICENSE("GPL");
3072
3073 module_init(mdc_init);
3074 module_exit(mdc_exit);