lustre/mdc/mdc_request.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2013, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36
  37 #define DEBUG_SUBSYSTEM S_MDC
  38
  39 #ifdef __KERNEL__
  40 # include <linux/module.h>
  41 # include <linux/pagemap.h>
  42 # include <linux/miscdevice.h>
  43 # include <linux/init.h>
  44 # include <linux/utsname.h>
  45 #else
  46 # include <liblustre.h>
  47 #endif
  48
  49 #include <lustre_acl.h>
  50 #include <obd_class.h>
  51 #include <lustre_lmv.h>
  52 #include <lustre_fid.h>
  53 #include <lprocfs_status.h>
  54 #include <lustre_param.h>
  55 #include <lustre_log.h>
  56 #include <cl_object.h>
  57 #include <lclient.h>
  58
  59 #include "mdc_internal.h"
  60
  61 #define REQUEST_MINOR 244
  62
  63 struct mdc_renew_capa_args {
  64         struct obd_capa        *ra_oc;
  65         renew_capa_cb_t         ra_cb;
  66 };
  67
  68 static int mdc_cleanup(struct obd_device *obd);
  69
  70 int mdc_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
  71                     const struct req_msg_field *field, struct obd_capa **oc)
  72 {
  73         struct lustre_capa *capa;
  74         struct obd_capa *c;
  75         ENTRY;
  76
  77         /* swabbed already in mdc_enqueue */
  78         capa = req_capsule_server_get(&req->rq_pill, field);
  79         if (capa == NULL)
  80                 RETURN(-EPROTO);
  81
  82         c = alloc_capa(CAPA_SITE_CLIENT);
  83         if (IS_ERR(c)) {
  84                 CDEBUG(D_INFO, "alloc capa failed!\n");
  85                 RETURN(PTR_ERR(c));
  86         } else {
  87                 c->c_capa = *capa;
  88                 *oc = c;
  89                 RETURN(0);
  90         }
  91 }
  92
  93 static inline int mdc_queue_wait(struct ptlrpc_request *req)
  94 {
  95         struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
  96         int rc;
  97
  98         /* mdc_enter_request() ensures that this client has no more
  99          * than cl_max_rpcs_in_flight RPCs simultaneously inf light
 100          * against an MDT. */
 101         rc = mdc_enter_request(cli);
 102         if (rc != 0)
 103                 return rc;
 104
 105         rc = ptlrpc_queue_wait(req);
 106         mdc_exit_request(cli);
 107
 108         return rc;
 109 }
 110
 111 /* Helper that implements most of mdc_getstatus and signal_completed_replay. */
 112 /* XXX this should become mdc_get_info("key"), sending MDS_GET_INFO RPC */
 113 static int send_getstatus(struct obd_import *imp, struct lu_fid *rootfid,
 114                           struct obd_capa **pc, int level, int msg_flags)
 115 {
 116         struct ptlrpc_request *req;
 117         struct mdt_body       *body;
 118         int                    rc;
 119         ENTRY;
 120
 121         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_GETSTATUS,
 122                                         LUSTRE_MDS_VERSION, MDS_GETSTATUS);
 123         if (req == NULL)
 124                 RETURN(-ENOMEM);
 125
 126         mdc_pack_body(req, NULL, NULL, 0, 0, -1, 0);
 127         lustre_msg_add_flags(req->rq_reqmsg, msg_flags);
 128         req->rq_send_state = level;
 129
 130         ptlrpc_request_set_replen(req);
 131
 132         rc = ptlrpc_queue_wait(req);
 133         if (rc)
 134                 GOTO(out, rc);
 135
 136         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 137         if (body == NULL)
 138                 GOTO(out, rc = -EPROTO);
 139
 140         if (body->valid & OBD_MD_FLMDSCAPA) {
 141                 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA1, pc);
 142                 if (rc)
 143                         GOTO(out, rc);
 144         }
 145
 146         *rootfid = body->fid1;
 147         CDEBUG(D_NET,
 148                "root fid="DFID", last_committed="LPU64"\n",
 149                PFID(rootfid),
 150                lustre_msg_get_last_committed(req->rq_repmsg));
 151         EXIT;
 152 out:
 153         ptlrpc_req_finished(req);
 154         return rc;
 155 }
 156
 157 /* This should be mdc_get_info("rootfid") */
 158 int mdc_getstatus(struct obd_export *exp, struct lu_fid *rootfid,
 159                   struct obd_capa **pc)
 160 {
 161         return send_getstatus(class_exp2cliimp(exp), rootfid, pc,
 162                               LUSTRE_IMP_FULL, 0);
 163 }
 164
 165 /*
 166  * This function now is known to always saying that it will receive 4 buffers
 167  * from server. Even for cases when acl_size and md_size is zero, RPC header
 168  * will contain 4 fields and RPC itself will contain zero size fields. This is
 169  * because mdt_getattr*() _always_ returns 4 fields, but if acl is not needed
 170  * and thus zero, it shrinks it, making zero size. The same story about
 171  * md_size. And this is course of problem when client waits for smaller number
 172  * of fields. This issue will be fixed later when client gets aware of RPC
 173  * layouts.  --umka
 174  */
 175 static int mdc_getattr_common(struct obd_export *exp,
 176                               struct ptlrpc_request *req)
 177 {
 178         struct req_capsule *pill = &req->rq_pill;
 179         struct mdt_body    *body;
 180         void               *eadata;
 181         int                 rc;
 182         ENTRY;
 183
 184         /* Request message already built. */
 185         rc = ptlrpc_queue_wait(req);
 186         if (rc != 0)
 187                 RETURN(rc);
 188
 189         /* sanity check for the reply */
 190         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 191         if (body == NULL)
 192                 RETURN(-EPROTO);
 193
 194         CDEBUG(D_NET, "mode: %o\n", body->mode);
 195
 196         mdc_update_max_ea_from_body(exp, body);
 197         if (body->eadatasize != 0) {
 198                 eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
 199                                                       body->eadatasize);
 200                 if (eadata == NULL)
 201                         RETURN(-EPROTO);
 202         }
 203
 204         if (body->valid & OBD_MD_FLRMTPERM) {
 205                 struct mdt_remote_perm *perm;
 206
 207                 LASSERT(client_is_remote(exp));
 208                 perm = req_capsule_server_swab_get(pill, &RMF_ACL,
 209                                                 lustre_swab_mdt_remote_perm);
 210                 if (perm == NULL)
 211                         RETURN(-EPROTO);
 212         }
 213
 214         if (body->valid & OBD_MD_FLMDSCAPA) {
 215                 struct lustre_capa *capa;
 216                 capa = req_capsule_server_get(pill, &RMF_CAPA1);
 217                 if (capa == NULL)
 218                         RETURN(-EPROTO);
 219         }
 220
 221         RETURN(0);
 222 }
 223
 224 int mdc_getattr(struct obd_export *exp, struct md_op_data *op_data,
 225                 struct ptlrpc_request **request)
 226 {
 227         struct ptlrpc_request *req;
 228         int                    rc;
 229         ENTRY;
 230
 231         /* Single MDS without an LMV case */
 232         if (op_data->op_flags & MF_GET_MDT_IDX) {
 233                 op_data->op_mds = 0;
 234                 RETURN(0);
 235         }
 236         *request = NULL;
 237         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
 238         if (req == NULL)
 239                 RETURN(-ENOMEM);
 240
 241         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 242
 243         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
 244         if (rc) {
 245                 ptlrpc_request_free(req);
 246                 RETURN(rc);
 247         }
 248
 249         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
 250                       op_data->op_valid, op_data->op_mode, -1, 0);
 251
 252         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 253                              op_data->op_mode);
 254         if (op_data->op_valid & OBD_MD_FLRMTPERM) {
 255                 LASSERT(client_is_remote(exp));
 256                 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
 257                                      sizeof(struct mdt_remote_perm));
 258         }
 259         ptlrpc_request_set_replen(req);
 260
 261         rc = mdc_getattr_common(exp, req);
 262         if (rc)
 263                 ptlrpc_req_finished(req);
 264         else
 265                 *request = req;
 266         RETURN(rc);
 267 }
 268
 269 int mdc_getattr_name(struct obd_export *exp, struct md_op_data *op_data,
 270                      struct ptlrpc_request **request)
 271 {
 272         struct ptlrpc_request *req;
 273         int                    rc;
 274         ENTRY;
 275
 276         *request = NULL;
 277         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 278                                    &RQF_MDS_GETATTR_NAME);
 279         if (req == NULL)
 280                 RETURN(-ENOMEM);
 281
 282         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 283         req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 284                              op_data->op_namelen + 1);
 285
 286         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR_NAME);
 287         if (rc) {
 288                 ptlrpc_request_free(req);
 289                 RETURN(rc);
 290         }
 291
 292         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
 293                       op_data->op_valid, op_data->op_mode,
 294                       op_data->op_suppgids[0], 0);
 295
 296         if (op_data->op_name) {
 297                 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 298                 LASSERT(strnlen(op_data->op_name, op_data->op_namelen) ==
 299                                 op_data->op_namelen);
 300                 memcpy(name, op_data->op_name, op_data->op_namelen);
 301         }
 302
 303         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 304                              op_data->op_mode);
 305         ptlrpc_request_set_replen(req);
 306
 307         rc = mdc_getattr_common(exp, req);
 308         if (rc)
 309                 ptlrpc_req_finished(req);
 310         else
 311                 *request = req;
 312         RETURN(rc);
 313 }
 314
 315 static int mdc_is_subdir(struct obd_export *exp,
 316                          const struct lu_fid *pfid,
 317                          const struct lu_fid *cfid,
 318                          struct ptlrpc_request **request)
 319 {
 320         struct ptlrpc_request  *req;
 321         int                     rc;
 322
 323         ENTRY;
 324
 325         *request = NULL;
 326         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
 327                                         &RQF_MDS_IS_SUBDIR, LUSTRE_MDS_VERSION,
 328                                         MDS_IS_SUBDIR);
 329         if (req == NULL)
 330                 RETURN(-ENOMEM);
 331
 332         mdc_is_subdir_pack(req, pfid, cfid, 0);
 333         ptlrpc_request_set_replen(req);
 334
 335         rc = ptlrpc_queue_wait(req);
 336         if (rc && rc != -EREMOTE)
 337                 ptlrpc_req_finished(req);
 338         else
 339                 *request = req;
 340         RETURN(rc);
 341 }
 342
 343 static int mdc_xattr_common(struct obd_export *exp,const struct req_format *fmt,
 344                             const struct lu_fid *fid,
 345                             struct obd_capa *oc, int opcode, obd_valid valid,
 346                             const char *xattr_name, const char *input,
 347                             int input_size, int output_size, int flags,
 348                             __u32 suppgid, struct ptlrpc_request **request)
 349 {
 350         struct ptlrpc_request *req;
 351         int   xattr_namelen = 0;
 352         char *tmp;
 353         int   rc;
 354         ENTRY;
 355
 356         *request = NULL;
 357         req = ptlrpc_request_alloc(class_exp2cliimp(exp), fmt);
 358         if (req == NULL)
 359                 RETURN(-ENOMEM);
 360
 361         mdc_set_capa_size(req, &RMF_CAPA1, oc);
 362         if (xattr_name) {
 363                 xattr_namelen = strlen(xattr_name) + 1;
 364                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 365                                      xattr_namelen);
 366         }
 367         if (input_size) {
 368                 LASSERT(input);
 369                 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
 370                                      input_size);
 371         }
 372
 373         /* Flush local XATTR locks to get rid of a possible cancel RPC */
 374         if (opcode == MDS_REINT && fid_is_sane(fid) &&
 375             exp->exp_connect_data.ocd_ibits_known & MDS_INODELOCK_XATTR) {
 376                 CFS_LIST_HEAD(cancels);
 377                 int count;
 378
 379                 /* Without that packing would fail */
 380                 if (input_size == 0)
 381                         req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
 382                                              RCL_CLIENT, 0);
 383
 384                 count = mdc_resource_get_unused(exp, fid,
 385                                                 &cancels, LCK_EX,
 386                                                 MDS_INODELOCK_XATTR);
 387
 388                 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
 389                 if (rc) {
 390                         ptlrpc_request_free(req);
 391                         RETURN(rc);
 392                 }
 393         } else {
 394                 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode);
 395                 if (rc) {
 396                         ptlrpc_request_free(req);
 397                         RETURN(rc);
 398                 }
 399         }
 400
 401         if (opcode == MDS_REINT) {
 402                 struct mdt_rec_setxattr *rec;
 403
 404                 CLASSERT(sizeof(struct mdt_rec_setxattr) ==
 405                          sizeof(struct mdt_rec_reint));
 406                 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
 407                 rec->sx_opcode = REINT_SETXATTR;
 408                 rec->sx_fsuid  = current_fsuid();
 409                 rec->sx_fsgid  = current_fsgid();
 410                 rec->sx_cap    = cfs_curproc_cap_pack();
 411                 rec->sx_suppgid1 = suppgid;
 412                 rec->sx_suppgid2 = -1;
 413                 rec->sx_fid    = *fid;
 414                 rec->sx_valid  = valid | OBD_MD_FLCTIME;
 415                 rec->sx_time   = cfs_time_current_sec();
 416                 rec->sx_size   = output_size;
 417                 rec->sx_flags  = flags;
 418
 419                 mdc_pack_capa(req, &RMF_CAPA1, oc);
 420         } else {
 421                 mdc_pack_body(req, fid, oc, valid, output_size, suppgid, flags);
 422         }
 423
 424         if (xattr_name) {
 425                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 426                 memcpy(tmp, xattr_name, xattr_namelen);
 427         }
 428         if (input_size) {
 429                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
 430                 memcpy(tmp, input, input_size);
 431         }
 432
 433         if (req_capsule_has_field(&req->rq_pill, &RMF_EADATA, RCL_SERVER))
 434                 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
 435                                      RCL_SERVER, output_size);
 436         ptlrpc_request_set_replen(req);
 437
 438         /* make rpc */
 439         if (opcode == MDS_REINT)
 440                 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 441
 442         rc = ptlrpc_queue_wait(req);
 443
 444         if (opcode == MDS_REINT)
 445                 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 446
 447         if (rc)
 448                 ptlrpc_req_finished(req);
 449         else
 450                 *request = req;
 451         RETURN(rc);
 452 }
 453
 454 int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
 455                  struct obd_capa *oc, obd_valid valid, const char *xattr_name,
 456                  const char *input, int input_size, int output_size,
 457                  int flags, __u32 suppgid, struct ptlrpc_request **request)
 458 {
 459         return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
 460                                 fid, oc, MDS_REINT, valid, xattr_name,
 461                                 input, input_size, output_size, flags,
 462                                 suppgid, request);
 463 }
 464
 465 int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
 466                  struct obd_capa *oc, obd_valid valid, const char *xattr_name,
 467                  const char *input, int input_size, int output_size,
 468                  int flags, struct ptlrpc_request **request)
 469 {
 470         return mdc_xattr_common(exp, &RQF_MDS_GETXATTR,
 471                                 fid, oc, MDS_GETXATTR, valid, xattr_name,
 472                                 input, input_size, output_size, flags,
 473                                 -1, request);
 474 }
 475
 476 #ifdef CONFIG_FS_POSIX_ACL
 477 static int mdc_unpack_acl(struct ptlrpc_request *req, struct lustre_md *md)
 478 {
 479         struct req_capsule     *pill = &req->rq_pill;
 480         struct mdt_body        *body = md->body;
 481         struct posix_acl       *acl;
 482         void                   *buf;
 483         int                     rc;
 484         ENTRY;
 485
 486         if (!body->aclsize)
 487                 RETURN(0);
 488
 489         buf = req_capsule_server_sized_get(pill, &RMF_ACL, body->aclsize);
 490
 491         if (!buf)
 492                 RETURN(-EPROTO);
 493
 494         acl = posix_acl_from_xattr(&init_user_ns, buf, body->aclsize);
 495         if (IS_ERR(acl)) {
 496                 rc = PTR_ERR(acl);
 497                 CERROR("convert xattr to acl: %d\n", rc);
 498                 RETURN(rc);
 499         }
 500
 501         rc = posix_acl_valid(acl);
 502         if (rc) {
 503                 CERROR("validate acl: %d\n", rc);
 504                 posix_acl_release(acl);
 505                 RETURN(rc);
 506         }
 507
 508         md->posix_acl = acl;
 509         RETURN(0);
 510 }
 511 #else
 512 #define mdc_unpack_acl(req, md) 0
 513 #endif
 514
 515 int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
 516                       struct obd_export *dt_exp, struct obd_export *md_exp,
 517                       struct lustre_md *md)
 518 {
 519         struct req_capsule *pill = &req->rq_pill;
 520         int rc;
 521         ENTRY;
 522
 523         LASSERT(md);
 524         memset(md, 0, sizeof(*md));
 525
 526         md->body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 527         LASSERT(md->body != NULL);
 528
 529         if (md->body->valid & OBD_MD_FLEASIZE) {
 530                 int lmmsize;
 531                 struct lov_mds_md *lmm;
 532
 533                 if (!S_ISREG(md->body->mode)) {
 534                         CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, should be a "
 535                                "regular file, but is not\n");
 536                         GOTO(out, rc = -EPROTO);
 537                 }
 538
 539                 if (md->body->eadatasize == 0) {
 540                         CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, "
 541                                "but eadatasize 0\n");
 542                         GOTO(out, rc = -EPROTO);
 543                 }
 544                 lmmsize = md->body->eadatasize;
 545                 lmm = req_capsule_server_sized_get(pill, &RMF_MDT_MD, lmmsize);
 546                 if (!lmm)
 547                         GOTO(out, rc = -EPROTO);
 548
 549                 rc = obd_unpackmd(dt_exp, &md->lsm, lmm, lmmsize);
 550                 if (rc < 0)
 551                         GOTO(out, rc);
 552
 553                 if (rc < sizeof(*md->lsm)) {
 554                         CDEBUG(D_INFO, "lsm size too small: "
 555                                "rc < sizeof (*md->lsm) (%d < %d)\n",
 556                                rc, (int)sizeof(*md->lsm));
 557                         GOTO(out, rc = -EPROTO);
 558                 }
 559
 560         } else if (md->body->valid & OBD_MD_FLDIREA) {
 561                 int lmvsize;
 562                 struct lov_mds_md *lmv;
 563
 564                 if(!S_ISDIR(md->body->mode)) {
 565                         CDEBUG(D_INFO, "OBD_MD_FLDIREA set, should be a "
 566                                "directory, but is not\n");
 567                         GOTO(out, rc = -EPROTO);
 568                 }
 569
 570                 if (md->body->eadatasize == 0) {
 571                         CDEBUG(D_INFO, "OBD_MD_FLDIREA is set, "
 572                                "but eadatasize 0\n");
 573                         RETURN(-EPROTO);
 574                 }
 575                 if (md->body->valid & OBD_MD_MEA) {
 576                         lmvsize = md->body->eadatasize;
 577                         lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
 578                                                            lmvsize);
 579                         if (!lmv)
 580                                 GOTO(out, rc = -EPROTO);
 581
 582                         rc = obd_unpackmd(md_exp, (void *)&md->lmv, lmv,
 583                                           lmvsize);
 584                         if (rc < 0)
 585                                 GOTO(out, rc);
 586
 587                         if (rc < sizeof(*md->lmv)) {
 588                                 CDEBUG(D_INFO, "size too small:  "
 589                                        "rc < sizeof(*md->lmv) (%d < %d)\n",
 590                                         rc, (int)sizeof(*md->lmv));
 591                                 GOTO(out, rc = -EPROTO);
 592                         }
 593                 }
 594         }
 595         rc = 0;
 596
 597         if (md->body->valid & OBD_MD_FLRMTPERM) {
 598                 /* remote permission */
 599                 LASSERT(client_is_remote(exp));
 600                 md->remote_perm = req_capsule_server_swab_get(pill, &RMF_ACL,
 601                                                 lustre_swab_mdt_remote_perm);
 602                 if (!md->remote_perm)
 603                         GOTO(out, rc = -EPROTO);
 604         }
 605         else if (md->body->valid & OBD_MD_FLACL) {
 606                 /* for ACL, it's possible that FLACL is set but aclsize is zero.
 607                  * only when aclsize != 0 there's an actual segment for ACL
 608                  * in reply buffer.
 609                  */
 610                 if (md->body->aclsize) {
 611                         rc = mdc_unpack_acl(req, md);
 612                         if (rc)
 613                                 GOTO(out, rc);
 614 #ifdef CONFIG_FS_POSIX_ACL
 615                 } else {
 616                         md->posix_acl = NULL;
 617 #endif
 618                 }
 619         }
 620         if (md->body->valid & OBD_MD_FLMDSCAPA) {
 621                 struct obd_capa *oc = NULL;
 622
 623                 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA1, &oc);
 624                 if (rc)
 625                         GOTO(out, rc);
 626                 md->mds_capa = oc;
 627         }
 628
 629         if (md->body->valid & OBD_MD_FLOSSCAPA) {
 630                 struct obd_capa *oc = NULL;
 631
 632                 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA2, &oc);
 633                 if (rc)
 634                         GOTO(out, rc);
 635                 md->oss_capa = oc;
 636         }
 637
 638         EXIT;
 639 out:
 640         if (rc) {
 641                 if (md->oss_capa) {
 642                         capa_put(md->oss_capa);
 643                         md->oss_capa = NULL;
 644                 }
 645                 if (md->mds_capa) {
 646                         capa_put(md->mds_capa);
 647                         md->mds_capa = NULL;
 648                 }
 649 #ifdef CONFIG_FS_POSIX_ACL
 650                 posix_acl_release(md->posix_acl);
 651 #endif
 652                 if (md->lsm)
 653                         obd_free_memmd(dt_exp, &md->lsm);
 654         }
 655         return rc;
 656 }
 657
 658 int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
 659 {
 660         ENTRY;
 661         RETURN(0);
 662 }
 663
 664 /**
 665  * Handles both OPEN and SETATTR RPCs for OPEN-CLOSE and SETATTR-DONE_WRITING
 666  * RPC chains.
 667  */
 668 void mdc_replay_open(struct ptlrpc_request *req)
 669 {
 670         struct md_open_data *mod = req->rq_cb_data;
 671         struct ptlrpc_request *close_req;
 672         struct obd_client_handle *och;
 673         struct lustre_handle old;
 674         struct mdt_body *body;
 675         ENTRY;
 676
 677         if (mod == NULL) {
 678                 DEBUG_REQ(D_ERROR, req,
 679                           "Can't properly replay without open data.");
 680                 EXIT;
 681                 return;
 682         }
 683
 684         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 685         LASSERT(body != NULL);
 686
 687         och = mod->mod_och;
 688         if (och != NULL) {
 689                 struct lustre_handle *file_fh;
 690
 691                 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
 692
 693                 file_fh = &och->och_fh;
 694                 CDEBUG(D_HA, "updating handle from "LPX64" to "LPX64"\n",
 695                        file_fh->cookie, body->handle.cookie);
 696                 old = *file_fh;
 697                 *file_fh = body->handle;
 698         }
 699         close_req = mod->mod_close_req;
 700         if (close_req != NULL) {
 701                 __u32 opc = lustre_msg_get_opc(close_req->rq_reqmsg);
 702                 struct mdt_ioepoch *epoch;
 703
 704                 LASSERT(opc == MDS_CLOSE || opc == MDS_DONE_WRITING);
 705                 epoch = req_capsule_client_get(&close_req->rq_pill,
 706                                                &RMF_MDT_EPOCH);
 707                 LASSERT(epoch);
 708
 709                 if (och != NULL)
 710                         LASSERT(!memcmp(&old, &epoch->handle, sizeof(old)));
 711                 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
 712                 epoch->handle = body->handle;
 713         }
 714         EXIT;
 715 }
 716
 717 void mdc_commit_open(struct ptlrpc_request *req)
 718 {
 719         struct md_open_data *mod = req->rq_cb_data;
 720         if (mod == NULL)
 721                 return;
 722
 723         /**
 724          * No need to touch md_open_data::mod_och, it holds a reference on
 725          * \var mod and will zero references to each other, \var mod will be
 726          * freed after that when md_open_data::mod_och will put the reference.
 727          */
 728
 729         /**
 730          * Do not let open request to disappear as it still may be needed
 731          * for close rpc to happen (it may happen on evict only, otherwise
 732          * ptlrpc_request::rq_replay does not let mdc_commit_open() to be
 733          * called), just mark this rpc as committed to distinguish these 2
 734          * cases, see mdc_close() for details. The open request reference will
 735          * be put along with freeing \var mod.
 736          */
 737         ptlrpc_request_addref(req);
 738         spin_lock(&req->rq_lock);
 739         req->rq_committed = 1;
 740         spin_unlock(&req->rq_lock);
 741         req->rq_cb_data = NULL;
 742         obd_mod_put(mod);
 743 }
 744
 745 int mdc_set_open_replay_data(struct obd_export *exp,
 746                              struct obd_client_handle *och,
 747                              struct lookup_intent *it)
 748 {
 749         struct md_open_data     *mod;
 750         struct mdt_rec_create   *rec;
 751         struct mdt_body         *body;
 752         struct ptlrpc_request   *open_req = it->d.lustre.it_data;
 753         struct obd_import       *imp = open_req->rq_import;
 754         ENTRY;
 755
 756         if (!open_req->rq_replay)
 757                 RETURN(0);
 758
 759         rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
 760         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
 761         LASSERT(rec != NULL);
 762         /* Incoming message in my byte order (it's been swabbed). */
 763         /* Outgoing messages always in my byte order. */
 764         LASSERT(body != NULL);
 765
 766         /* Only if the import is replayable, we set replay_open data */
 767         if (och && imp->imp_replayable) {
 768                 mod = obd_mod_alloc();
 769                 if (mod == NULL) {
 770                         DEBUG_REQ(D_ERROR, open_req,
 771                                   "Can't allocate md_open_data");
 772                         RETURN(0);
 773                 }
 774
 775                 /**
 776                  * Take a reference on \var mod, to be freed on mdc_close().
 777                  * It protects \var mod from being freed on eviction (commit
 778                  * callback is called despite rq_replay flag).
 779                  * Another reference for \var och.
 780                  */
 781                 obd_mod_get(mod);
 782                 obd_mod_get(mod);
 783
 784                 spin_lock(&open_req->rq_lock);
 785                 och->och_mod = mod;
 786                 mod->mod_och = och;
 787                 mod->mod_is_create = it_disposition(it, DISP_OPEN_CREATE) ||
 788                                      it_disposition(it, DISP_OPEN_STRIPE);
 789                 mod->mod_open_req = open_req;
 790                 open_req->rq_cb_data = mod;
 791                 open_req->rq_commit_cb = mdc_commit_open;
 792                 spin_unlock(&open_req->rq_lock);
 793         }
 794
 795         rec->cr_fid2 = body->fid1;
 796         rec->cr_ioepoch = body->ioepoch;
 797         rec->cr_old_handle.cookie = body->handle.cookie;
 798         open_req->rq_replay_cb = mdc_replay_open;
 799         if (!fid_is_sane(&body->fid1)) {
 800                 DEBUG_REQ(D_ERROR, open_req, "Saving replay request with "
 801                           "insane fid");
 802                 LBUG();
 803         }
 804
 805         DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
 806         RETURN(0);
 807 }
 808
 809 static void mdc_free_open(struct md_open_data *mod)
 810 {
 811         int committed = 0;
 812
 813         if (mod->mod_is_create == 0 &&
 814             imp_connect_disp_stripe(mod->mod_open_req->rq_import))
 815                 committed = 1;
 816
 817         LASSERT(mod->mod_open_req->rq_replay == 0);
 818
 819         DEBUG_REQ(D_RPCTRACE, mod->mod_open_req, "free open request\n");
 820
 821         ptlrpc_request_committed(mod->mod_open_req, committed);
 822         if (mod->mod_close_req)
 823                 ptlrpc_request_committed(mod->mod_close_req, committed);
 824 }
 825
 826 int mdc_clear_open_replay_data(struct obd_export *exp,
 827                                struct obd_client_handle *och)
 828 {
 829         struct md_open_data *mod = och->och_mod;
 830         ENTRY;
 831
 832         /**
 833          * It is possible to not have \var mod in a case of eviction between
 834          * lookup and ll_file_open().
 835          **/
 836         if (mod == NULL)
 837                 RETURN(0);
 838
 839         LASSERT(mod != LP_POISON);
 840         LASSERT(mod->mod_open_req != NULL);
 841         mdc_free_open(mod);
 842
 843         mod->mod_och = NULL;
 844         och->och_mod = NULL;
 845         obd_mod_put(mod);
 846
 847         RETURN(0);
 848 }
 849
 850 /* Prepares the request for the replay by the given reply */
 851 static void mdc_close_handle_reply(struct ptlrpc_request *req,
 852                                    struct md_op_data *op_data, int rc) {
 853         struct mdt_body  *repbody;
 854         struct mdt_ioepoch *epoch;
 855
 856         if (req && rc == -EAGAIN) {
 857                 repbody = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 858                 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
 859
 860                 epoch->flags |= MF_SOM_AU;
 861                 if (repbody->valid & OBD_MD_FLGETATTRLOCK)
 862                         op_data->op_flags |= MF_GETATTR_LOCK;
 863         }
 864 }
 865
 866 int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
 867               struct md_open_data *mod, struct ptlrpc_request **request)
 868 {
 869         struct obd_device     *obd = class_exp2obd(exp);
 870         struct ptlrpc_request *req;
 871         struct req_format     *req_fmt;
 872         int                    rc;
 873         int                    saved_rc = 0;
 874         ENTRY;
 875
 876         req_fmt = &RQF_MDS_CLOSE;
 877         if (op_data->op_bias & MDS_HSM_RELEASE) {
 878                 req_fmt = &RQF_MDS_RELEASE_CLOSE;
 879
 880                 /* allocate a FID for volatile file */
 881                 rc = mdc_fid_alloc(exp, &op_data->op_fid2, op_data);
 882                 if (rc < 0) {
 883                         CERROR("%s: "DFID" failed to allocate FID: %d\n",
 884                                obd->obd_name, PFID(&op_data->op_fid1), rc);
 885                         /* save the errcode and proceed to close */
 886                         saved_rc = rc;
 887                 }
 888         }
 889
 890         *request = NULL;
 891         req = ptlrpc_request_alloc(class_exp2cliimp(exp), req_fmt);
 892         if (req == NULL)
 893                 RETURN(-ENOMEM);
 894
 895         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 896
 897         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
 898         if (rc) {
 899                 ptlrpc_request_free(req);
 900                 RETURN(rc);
 901         }
 902
 903         /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
 904          * portal whose threads are not taking any DLM locks and are therefore
 905          * always progressing */
 906         req->rq_request_portal = MDS_READPAGE_PORTAL;
 907         ptlrpc_at_set_req_timeout(req);
 908
 909         /* Ensure that this close's handle is fixed up during replay. */
 910         if (likely(mod != NULL)) {
 911                 LASSERTF(mod->mod_open_req != NULL &&
 912                          mod->mod_open_req->rq_type != LI_POISON,
 913                          "POISONED open %p!\n", mod->mod_open_req);
 914
 915                 mod->mod_close_req = req;
 916
 917                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
 918                 /* We no longer want to preserve this open for replay even
 919                  * though the open was committed. b=3632, b=3633 */
 920                 spin_lock(&mod->mod_open_req->rq_lock);
 921                 mod->mod_open_req->rq_replay = 0;
 922                 spin_unlock(&mod->mod_open_req->rq_lock);
 923         } else {
 924                  CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
 925         }
 926
 927         mdc_close_pack(req, op_data);
 928
 929         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 930                              obd->u.cli.cl_max_mds_easize);
 931         req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
 932                              obd->u.cli.cl_max_mds_cookiesize);
 933
 934         ptlrpc_request_set_replen(req);
 935
 936         mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 937         rc = ptlrpc_queue_wait(req);
 938         mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 939
 940         if (req->rq_repmsg == NULL) {
 941                 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
 942                        req->rq_status);
 943                 if (rc == 0)
 944                         rc = req->rq_status ?: -EIO;
 945         } else if (rc == 0 || rc == -EAGAIN) {
 946                 struct mdt_body *body;
 947
 948                 rc = lustre_msg_get_status(req->rq_repmsg);
 949                 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
 950                         DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
 951                                   "= %d", rc);
 952                         if (rc > 0)
 953                                 rc = -rc;
 954                 }
 955                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 956                 if (body == NULL)
 957                         rc = -EPROTO;
 958         } else if (rc == -ESTALE) {
 959                 /**
 960                  * it can be allowed error after 3633 if open was committed and
 961                  * server failed before close was sent. Let's check if mod
 962                  * exists and return no error in that case
 963                  */
 964                 if (mod) {
 965                         DEBUG_REQ(D_HA, req, "Reset ESTALE = %d", rc);
 966                         LASSERT(mod->mod_open_req != NULL);
 967                         if (mod->mod_open_req->rq_committed)
 968                                 rc = 0;
 969                 }
 970         }
 971
 972         if (mod) {
 973                 if (rc != 0)
 974                         mod->mod_close_req = NULL;
 975                 /* Since now, mod is accessed through open_req only,
 976                  * thus close req does not keep a reference on mod anymore. */
 977                 obd_mod_put(mod);
 978         }
 979         *request = req;
 980         mdc_close_handle_reply(req, op_data, rc);
 981         RETURN(rc < 0 ? rc : saved_rc);
 982 }
 983
 984 int mdc_done_writing(struct obd_export *exp, struct md_op_data *op_data,
 985                      struct md_open_data *mod)
 986 {
 987         struct obd_device     *obd = class_exp2obd(exp);
 988         struct ptlrpc_request *req;
 989         int                    rc;
 990         ENTRY;
 991
 992         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 993                                    &RQF_MDS_DONE_WRITING);
 994         if (req == NULL)
 995                 RETURN(-ENOMEM);
 996
 997         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 998         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_DONE_WRITING);
 999         if (rc) {
1000                 ptlrpc_request_free(req);
1001                 RETURN(rc);
1002         }
1003
1004         if (mod != NULL) {
1005                 LASSERTF(mod->mod_open_req != NULL &&
1006                          mod->mod_open_req->rq_type != LI_POISON,
1007                          "POISONED setattr %p!\n", mod->mod_open_req);
1008
1009                 mod->mod_close_req = req;
1010                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched setattr");
1011                 /* We no longer want to preserve this setattr for replay even
1012                  * though the open was committed. b=3632, b=3633 */
1013                 spin_lock(&mod->mod_open_req->rq_lock);
1014                 mod->mod_open_req->rq_replay = 0;
1015                 spin_unlock(&mod->mod_open_req->rq_lock);
1016         }
1017
1018         mdc_close_pack(req, op_data);
1019         ptlrpc_request_set_replen(req);
1020
1021         mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
1022         rc = ptlrpc_queue_wait(req);
1023         mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
1024
1025         if (rc == -ESTALE) {
1026                 /**
1027                  * it can be allowed error after 3633 if open or setattr were
1028                  * committed and server failed before close was sent.
1029                  * Let's check if mod exists and return no error in that case
1030                  */
1031                 if (mod) {
1032                         LASSERT(mod->mod_open_req != NULL);
1033                         if (mod->mod_open_req->rq_committed)
1034                                 rc = 0;
1035                 }
1036         }
1037
1038         if (mod) {
1039                 if (rc != 0)
1040                         mod->mod_close_req = NULL;
1041                 LASSERT(mod->mod_open_req != NULL);
1042                 mdc_free_open(mod);
1043
1044                 /* Since now, mod is accessed through setattr req only,
1045                  * thus DW req does not keep a reference on mod anymore. */
1046                 obd_mod_put(mod);
1047         }
1048
1049         mdc_close_handle_reply(req, op_data, rc);
1050         ptlrpc_req_finished(req);
1051         RETURN(rc);
1052 }
1053
1054 #ifdef HAVE_SPLIT_SUPPORT
1055 int mdc_sendpage(struct obd_export *exp, const struct lu_fid *fid,
1056                  const struct page *page, int offset)
1057 {
1058         struct ptlrpc_request   *req;
1059         struct ptlrpc_bulk_desc *desc;
1060         int                      rc;
1061         ENTRY;
1062
1063         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_WRITEPAGE);
1064         if (req == NULL)
1065                 RETURN(-ENOMEM);
1066
1067         /* FIXME: capa doesn't support split yet */
1068         mdc_set_capa_size(req, &RMF_CAPA1, NULL);
1069
1070         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_WRITEPAGE);
1071         if (rc) {
1072                 ptlrpc_request_free(req);
1073                 RETURN(rc);
1074         }
1075
1076         req->rq_request_portal = MDS_READPAGE_PORTAL;
1077         ptlrpc_at_set_req_timeout(req);
1078
1079         desc = ptlrpc_prep_bulk_imp(req, 1, 1,BULK_GET_SOURCE, MDS_BULK_PORTAL);
1080         if (desc == NULL)
1081                 GOTO(out, rc = -ENOMEM);
1082
1083         /* NB req now owns desc and will free it when it gets freed. */
1084         ptlrpc_prep_bulk_page(desc, (struct page *)page, 0, offset);
1085         mdc_readdir_pack(req, 0, offset, fid, NULL);
1086
1087         ptlrpc_request_set_replen(req);
1088         rc = ptlrpc_queue_wait(req);
1089         if (rc)
1090                 GOTO(out, rc);
1091
1092         rc = sptlrpc_cli_unwrap_bulk_write(req, req->rq_bulk);
1093 out:
1094         ptlrpc_req_finished(req);
1095         return rc;
1096 }
1097 EXPORT_SYMBOL(mdc_sendpage);
1098 #endif
1099
1100 static int mdc_getpage(struct obd_export *exp, const struct lu_fid *fid,
1101                        __u64 offset, struct obd_capa *oc,
1102                        struct page **pages, int npages,
1103                        struct ptlrpc_request **request)
1104 {
1105         struct ptlrpc_request   *req;
1106         struct ptlrpc_bulk_desc *desc;
1107         int                      i;
1108         wait_queue_head_t        waitq;
1109         int                      resends = 0;
1110         struct l_wait_info       lwi;
1111         int                      rc;
1112         ENTRY;
1113
1114         *request = NULL;
1115         init_waitqueue_head(&waitq);
1116
1117 restart_bulk:
1118         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
1119         if (req == NULL)
1120                 RETURN(-ENOMEM);
1121
1122         mdc_set_capa_size(req, &RMF_CAPA1, oc);
1123
1124         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
1125         if (rc) {
1126                 ptlrpc_request_free(req);
1127                 RETURN(rc);
1128         }
1129
1130         req->rq_request_portal = MDS_READPAGE_PORTAL;
1131         ptlrpc_at_set_req_timeout(req);
1132
1133         desc = ptlrpc_prep_bulk_imp(req, npages, 1, BULK_PUT_SINK,
1134                                     MDS_BULK_PORTAL);
1135         if (desc == NULL) {
1136                 ptlrpc_request_free(req);
1137                 RETURN(-ENOMEM);
1138         }
1139
1140         /* NB req now owns desc and will free it when it gets freed */
1141         for (i = 0; i < npages; i++)
1142                 ptlrpc_prep_bulk_page_pin(desc, pages[i], 0, PAGE_CACHE_SIZE);
1143
1144         mdc_readdir_pack(req, offset, PAGE_CACHE_SIZE * npages, fid, oc);
1145
1146         ptlrpc_request_set_replen(req);
1147         rc = ptlrpc_queue_wait(req);
1148         if (rc) {
1149                 ptlrpc_req_finished(req);
1150                 if (rc != -ETIMEDOUT)
1151                         RETURN(rc);
1152
1153                 resends++;
1154                 if (!client_should_resend(resends, &exp->exp_obd->u.cli)) {
1155                         CERROR("%s: too many resend retries: rc = %d\n",
1156                                exp->exp_obd->obd_name, -EIO);
1157                         RETURN(-EIO);
1158                 }
1159                 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(resends), NULL, NULL,
1160                                        NULL);
1161                 l_wait_event(waitq, 0, &lwi);
1162
1163                 goto restart_bulk;
1164         }
1165
1166         rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
1167                                           req->rq_bulk->bd_nob_transferred);
1168         if (rc < 0) {
1169                 ptlrpc_req_finished(req);
1170                 RETURN(rc);
1171         }
1172
1173         if (req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK) {
1174                 CERROR("%s: unexpected bytes transferred: %d (%ld expected)\n",
1175                        exp->exp_obd->obd_name, req->rq_bulk->bd_nob_transferred,
1176                        PAGE_CACHE_SIZE * npages);
1177                 ptlrpc_req_finished(req);
1178                 RETURN(-EPROTO);
1179         }
1180
1181         *request = req;
1182         RETURN(0);
1183 }
1184
1185 #ifdef __KERNEL__
1186 static void mdc_release_page(struct page *page, int remove)
1187 {
1188         if (remove) {
1189                 lock_page(page);
1190                 if (likely(page->mapping != NULL))
1191                         truncate_complete_page(page->mapping, page);
1192                 unlock_page(page);
1193         }
1194         page_cache_release(page);
1195 }
1196
1197 static struct page *mdc_page_locate(struct address_space *mapping, __u64 *hash,
1198                                     __u64 *start, __u64 *end, int hash64)
1199 {
1200         /*
1201          * Complement of hash is used as an index so that
1202          * radix_tree_gang_lookup() can be used to find a page with starting
1203          * hash _smaller_ than one we are looking for.
1204          */
1205         unsigned long offset = hash_x_index(*hash, hash64);
1206         struct page *page;
1207         int found;
1208
1209         spin_lock_irq(&mapping->tree_lock);
1210         found = radix_tree_gang_lookup(&mapping->page_tree,
1211                                        (void **)&page, offset, 1);
1212         if (found > 0) {
1213                 struct lu_dirpage *dp;
1214
1215                 page_cache_get(page);
1216                 spin_unlock_irq(&mapping->tree_lock);
1217                 /*
1218                  * In contrast to find_lock_page() we are sure that directory
1219                  * page cannot be truncated (while DLM lock is held) and,
1220                  * hence, can avoid restart.
1221                  *
1222                  * In fact, page cannot be locked here at all, because
1223                  * mdc_read_page_remote does synchronous io.
1224                  */
1225                 wait_on_page_locked(page);
1226                 if (PageUptodate(page)) {
1227                         dp = kmap(page);
1228                         if (BITS_PER_LONG == 32 && hash64) {
1229                                 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1230                                 *end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
1231                                 *hash  = *hash >> 32;
1232                         } else {
1233                                 *start = le64_to_cpu(dp->ldp_hash_start);
1234                                 *end   = le64_to_cpu(dp->ldp_hash_end);
1235                         }
1236                         LASSERTF(*start <= *hash, "start = "LPX64",end = "
1237                                  LPX64",hash = "LPX64"\n", *start, *end, *hash);
1238                         CDEBUG(D_VFSTRACE, "page%lu [%llu %llu], hash"LPU64"\n",
1239                                offset, *start, *end, *hash);
1240                         if (*hash > *end) {
1241                                 kunmap(page);
1242                                 mdc_release_page(page, 0);
1243                                 page = NULL;
1244                         } else if (*end != *start && *hash == *end) {
1245                                 /*
1246                                  * upon hash collision, remove this page,
1247                                  * otherwise put page reference, and
1248                                  * ll_get_dir_page() will issue RPC to fetch
1249                                  * the page we want.
1250                                  */
1251                                 kunmap(page);
1252                                 mdc_release_page(page,
1253                                     le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
1254                                 page = NULL;
1255                         }
1256                 } else {
1257                         page_cache_release(page);
1258                         page = ERR_PTR(-EIO);
1259                 }
1260         } else {
1261                 spin_unlock_irq(&mapping->tree_lock);
1262                 page = NULL;
1263         }
1264         return page;
1265 }
1266
1267 /*
1268  * Adjust a set of pages, each page containing an array of lu_dirpages,
1269  * so that each page can be used as a single logical lu_dirpage.
1270  *
1271  * A lu_dirpage is laid out as follows, where s = ldp_hash_start,
1272  * e = ldp_hash_end, f = ldp_flags, p = padding, and each "ent" is a
1273  * struct lu_dirent.  It has size up to LU_PAGE_SIZE. The ldp_hash_end
1274  * value is used as a cookie to request the next lu_dirpage in a
1275  * directory listing that spans multiple pages (two in this example):
1276  *   ________
1277  *  |        |
1278  * .|--------v-------   -----.
1279  * |s|e|f|p|ent|ent| ... |ent|
1280  * '--|--------------   -----'   Each CFS_PAGE contains a single
1281  *    '------.                   lu_dirpage.
1282  * .---------v-------   -----.
1283  * |s|e|f|p|ent| 0 | ... | 0 |
1284  * '-----------------   -----'
1285  *
1286  * However, on hosts where the native VM page size (PAGE_CACHE_SIZE) is
1287  * larger than LU_PAGE_SIZE, a single host page may contain multiple
1288  * lu_dirpages. After reading the lu_dirpages from the MDS, the
1289  * ldp_hash_end of the first lu_dirpage refers to the one immediately
1290  * after it in the same CFS_PAGE (arrows simplified for brevity, but
1291  * in general e0==s1, e1==s2, etc.):
1292  *
1293  * .--------------------   -----.
1294  * |s0|e0|f0|p|ent|ent| ... |ent|
1295  * |---v----------------   -----|
1296  * |s1|e1|f1|p|ent|ent| ... |ent|
1297  * |---v----------------   -----|  Here, each CFS_PAGE contains
1298  *             ...                 multiple lu_dirpages.
1299  * |---v----------------   -----|
1300  * |s'|e'|f'|p|ent|ent| ... |ent|
1301  * '---|----------------   -----'
1302  *     v
1303  * .----------------------------.
1304  * |        next CFS_PAGE       |
1305  *
1306  * This structure is transformed into a single logical lu_dirpage as follows:
1307  *
1308  * - Replace e0 with e' so the request for the next lu_dirpage gets the page
1309  *   labeled 'next CFS_PAGE'.
1310  *
1311  * - Copy the LDF_COLLIDE flag from f' to f0 to correctly reflect whether
1312  *   a hash collision with the next page exists.
1313  *
1314  * - Adjust the lde_reclen of the ending entry of each lu_dirpage to span
1315  *   to the first entry of the next lu_dirpage.
1316  */
1317 #if PAGE_CACHE_SIZE > LU_PAGE_SIZE
1318 static void mdc_adjust_dirpages(struct page **pages, int cfs_pgs, int lu_pgs)
1319 {
1320         int i;
1321
1322         for (i = 0; i < cfs_pgs; i++) {
1323                 struct lu_dirpage       *dp = kmap(pages[i]);
1324                 struct lu_dirpage       *first = dp;
1325                 struct lu_dirent        *end_dirent = NULL;
1326                 struct lu_dirent        *ent;
1327                 __u64                   hash_end = dp->ldp_hash_end;
1328                 __u32                   flags = dp->ldp_flags;
1329
1330                 while (--lu_pgs > 0) {
1331                         ent = lu_dirent_start(dp);
1332                         for (end_dirent = ent; ent != NULL;
1333                              end_dirent = ent, ent = lu_dirent_next(ent));
1334
1335                         /* Advance dp to next lu_dirpage. */
1336                         dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE);
1337
1338                         /* Check if we've reached the end of the CFS_PAGE. */
1339                         if (!((unsigned long)dp & ~CFS_PAGE_MASK))
1340                                 break;
1341
1342                         /* Save the hash and flags of this lu_dirpage. */
1343                         hash_end = dp->ldp_hash_end;
1344                         flags = dp->ldp_flags;
1345
1346                         /* Check if lu_dirpage contains no entries. */
1347                         if (end_dirent == NULL)
1348                                 break;
1349
1350                         /* Enlarge the end entry lde_reclen from 0 to
1351                          * first entry of next lu_dirpage. */
1352                         LASSERT(le16_to_cpu(end_dirent->lde_reclen) == 0);
1353                         end_dirent->lde_reclen =
1354                                 cpu_to_le16((char *)(dp->ldp_entries) -
1355                                             (char *)end_dirent);
1356                 }
1357
1358                 first->ldp_hash_end = hash_end;
1359                 first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE);
1360                 first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE);
1361
1362                 kunmap(pages[i]);
1363         }
1364         LASSERTF(lu_pgs == 0, "left = %d", lu_pgs);
1365 }
1366 #else
1367 #define mdc_adjust_dirpages(pages, cfs_pgs, lu_pgs) do {} while (0)
1368 #endif  /* PAGE_CACHE_SIZE > LU_PAGE_SIZE */
1369
1370 /* parameters for readdir page */
1371 struct readpage_param {
1372         struct md_op_data       *rp_mod;
1373         __u64                   rp_off;
1374         int                     rp_hash64;
1375         struct obd_export       *rp_exp;
1376         struct md_callback      *rp_cb;
1377 };
1378
1379 /**
1380  * Read pages from server.
1381  *
1382  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
1383  * a header lu_dirpage which describes the start/end hash, and whether this
1384  * page is empty (contains no dir entry) or hash collide with next page.
1385  * After client receives reply, several pages will be integrated into dir page
1386  * in CFS_PAGE_SIZE (if CFS_PAGE_SIZE greater than LU_PAGE_SIZE), and the
1387  * lu_dirpage for this integrated page will be adjusted.
1388  **/
1389 static int mdc_read_page_remote(void *data, struct page *page0)
1390 {
1391         struct readpage_param   *rp = data;
1392         struct page             **page_pool;
1393         struct page             *page;
1394         struct lu_dirpage       *dp;
1395         int                     rd_pgs = 0; /* number of pages read actually */
1396         int                     npages;
1397         struct md_op_data       *op_data = rp->rp_mod;
1398         struct ptlrpc_request   *req;
1399         int                     max_pages = op_data->op_max_pages;
1400         struct inode            *inode;
1401         struct lu_fid           *fid;
1402         int                     i;
1403         int                     rc;
1404         ENTRY;
1405
1406         LASSERT(max_pages > 0 && max_pages <= PTLRPC_MAX_BRW_PAGES);
1407         if (op_data->op_mea1 != NULL) {
1408                 __u32 index = op_data->op_stripe_offset;
1409
1410                 inode = op_data->op_mea1->lsm_md_oinfo[index].lmo_root;
1411                 fid = &op_data->op_mea1->lsm_md_oinfo[index].lmo_fid;
1412         } else {
1413                 inode = op_data->op_data;
1414                 fid = &op_data->op_fid1;
1415         }
1416         LASSERT(inode != NULL);
1417
1418         OBD_ALLOC(page_pool, sizeof(page_pool[0]) * max_pages);
1419         if (page_pool != NULL) {
1420                 page_pool[0] = page0;
1421         } else {
1422                 page_pool = &page0;
1423                 max_pages = 1;
1424         }
1425
1426         for (npages = 1; npages < max_pages; npages++) {
1427                 page = page_cache_alloc_cold(inode->i_mapping);
1428                 if (page == NULL)
1429                         break;
1430                 page_pool[npages] = page;
1431         }
1432
1433         rc = mdc_getpage(rp->rp_exp, fid, rp->rp_off, op_data->op_capa1,
1434                          page_pool, npages, &req);
1435         if (rc == 0) {
1436                 int lu_pgs;
1437
1438                 rd_pgs = (req->rq_bulk->bd_nob_transferred +
1439                             PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1440                 lu_pgs = req->rq_bulk->bd_nob_transferred >>
1441                                                         LU_PAGE_SHIFT;
1442                 LASSERT(!(req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK));
1443
1444                 CDEBUG(D_INODE, "read %d(%d)/%d pages\n", rd_pgs, lu_pgs,
1445                        op_data->op_npages);
1446
1447                 mdc_adjust_dirpages(page_pool, rd_pgs, lu_pgs);
1448
1449                 SetPageUptodate(page0);
1450         }
1451
1452         unlock_page(page0);
1453         ptlrpc_req_finished(req);
1454         CDEBUG(D_CACHE, "read %d/%d pages\n", rd_pgs, npages);
1455         for (i = 1; i < npages; i++) {
1456                 unsigned long   offset;
1457                 __u64           hash;
1458                 int ret;
1459
1460                 page = page_pool[i];
1461
1462                 if (rc < 0 || i >= rd_pgs) {
1463                         page_cache_release(page);
1464                         continue;
1465                 }
1466
1467                 SetPageUptodate(page);
1468
1469                 dp = kmap(page);
1470                 hash = le64_to_cpu(dp->ldp_hash_start);
1471                 kunmap(page);
1472
1473                 offset = hash_x_index(hash, rp->rp_hash64);
1474
1475                 prefetchw(&page->flags);
1476                 ret = add_to_page_cache_lru(page, inode->i_mapping, offset,
1477                                             GFP_KERNEL);
1478                 if (ret == 0)
1479                         unlock_page(page);
1480                 else
1481                         CDEBUG(D_VFSTRACE, "page %lu add to page cache failed:"
1482                                " rc = %d\n", offset, ret);
1483                 page_cache_release(page);
1484         }
1485
1486         if (page_pool != &page0)
1487                 OBD_FREE(page_pool, sizeof(page_pool[0]) * max_pages);
1488
1489         RETURN(rc);
1490 }
1491
1492 /**
1493  * Read dir page from cache first, if it can not find it, read it from
1494  * server and add into the cache.
1495  */
1496 static int mdc_read_page(struct obd_export *exp, struct md_op_data *op_data,
1497                          struct md_callback *cb_op, struct page **ppage)
1498 {
1499         struct lookup_intent    it = { .it_op = IT_READDIR };
1500         struct page             *page;
1501         struct inode            *dir = NULL;
1502         struct address_space    *mapping;
1503         struct lu_dirpage       *dp;
1504         __u64                   start = 0;
1505         __u64                   end = 0;
1506         struct lustre_handle    lockh;
1507         struct ptlrpc_request   *enq_req = NULL;
1508         struct readpage_param   rp_param;
1509         int rc;
1510
1511         ENTRY;
1512
1513         *ppage = NULL;
1514
1515         if (op_data->op_mea1 != NULL) {
1516                 __u32 index = op_data->op_stripe_offset;
1517
1518                 dir = op_data->op_mea1->lsm_md_oinfo[index].lmo_root;
1519         } else {
1520                 dir = op_data->op_data;
1521         }
1522         LASSERT(dir != NULL);
1523
1524         mapping = dir->i_mapping;
1525
1526         rc = mdc_intent_lock(exp, op_data, NULL, 0, &it, 0, &enq_req,
1527                              cb_op->md_blocking_ast, 0);
1528         if (enq_req != NULL)
1529                 ptlrpc_req_finished(enq_req);
1530
1531         if (rc < 0) {
1532                 CERROR("%s: "DFID" lock enqueue fails: rc = %d\n",
1533                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1), rc);
1534                 RETURN(rc);
1535         }
1536
1537         rc = 0;
1538         mdc_set_lock_data(exp, &it.d.lustre.it_lock_handle, dir, NULL);
1539
1540         rp_param.rp_off = op_data->op_hash_offset;
1541         rp_param.rp_hash64 = op_data->op_cli_flags & CLI_HASH64;
1542         page = mdc_page_locate(mapping, &rp_param.rp_off, &start, &end,
1543                                rp_param.rp_hash64);
1544         if (IS_ERR(page)) {
1545                 CERROR("%s: dir page locate: "DFID" at "LPU64": rc %ld\n",
1546                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1547                        rp_param.rp_off, PTR_ERR(page));
1548                 GOTO(out_unlock, rc = PTR_ERR(page));
1549         } else if (page != NULL) {
1550                 /*
1551                  * XXX nikita: not entirely correct handling of a corner case:
1552                  * suppose hash chain of entries with hash value HASH crosses
1553                  * border between pages P0 and P1. First both P0 and P1 are
1554                  * cached, seekdir() is called for some entry from the P0 part
1555                  * of the chain. Later P0 goes out of cache. telldir(HASH)
1556                  * happens and finds P1, as it starts with matching hash
1557                  * value. Remaining entries from P0 part of the chain are
1558                  * skipped. (Is that really a bug?)
1559                  *
1560                  * Possible solutions: 0. don't cache P1 is such case, handle
1561                  * it as an "overflow" page. 1. invalidate all pages at
1562                  * once. 2. use HASH|1 as an index for P1.
1563                  */
1564                 GOTO(hash_collision, page);
1565         }
1566
1567         rp_param.rp_exp = exp;
1568         rp_param.rp_mod = op_data;
1569         page = read_cache_page(mapping,
1570                                hash_x_index(rp_param.rp_off,
1571                                             rp_param.rp_hash64),
1572                                mdc_read_page_remote, &rp_param);
1573         if (IS_ERR(page)) {
1574                 CERROR("%s: read cache page: "DFID" at "LPU64": rc %ld\n",
1575                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1576                        rp_param.rp_off, PTR_ERR(page));
1577                 GOTO(out_unlock, rc = PTR_ERR(page));
1578         }
1579
1580         wait_on_page_locked(page);
1581         (void)kmap(page);
1582         if (!PageUptodate(page)) {
1583                 CERROR("%s: page not updated: "DFID" at "LPU64": rc %d\n",
1584                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1585                        rp_param.rp_off, -5);
1586                 goto fail;
1587         }
1588         if (!PageChecked(page))
1589                 SetPageChecked(page);
1590         if (PageError(page)) {
1591                 CERROR("%s: page error: "DFID" at "LPU64": rc %d\n",
1592                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1593                        rp_param.rp_off, -5);
1594                 goto fail;
1595         }
1596
1597 hash_collision:
1598         dp = page_address(page);
1599         if (BITS_PER_LONG == 32 && rp_param.rp_hash64) {
1600                 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1601                 end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
1602                 rp_param.rp_off = op_data->op_hash_offset >> 32;
1603         } else {
1604                 start = le64_to_cpu(dp->ldp_hash_start);
1605                 end   = le64_to_cpu(dp->ldp_hash_end);
1606                 rp_param.rp_off = op_data->op_hash_offset;
1607         }
1608         if (end == start) {
1609                 LASSERT(start == rp_param.rp_off);
1610                 CWARN("Page-wide hash collision: %#lx\n", (unsigned long)end);
1611 #if BITS_PER_LONG == 32
1612                 CWARN("Real page-wide hash collision at ["LPU64" "LPU64"] with "
1613                       "hash "LPU64"\n", le64_to_cpu(dp->ldp_hash_start),
1614                       le64_to_cpu(dp->ldp_hash_end), op_data->op_hash_offset);
1615 #endif
1616
1617                 /*
1618                  * Fetch whole overflow chain...
1619                  *
1620                  * XXX not yet.
1621                  */
1622                 goto fail;
1623         }
1624         *ppage = page;
1625 out_unlock:
1626         lockh.cookie = it.d.lustre.it_lock_handle;
1627         ldlm_lock_decref(&lockh, it.d.lustre.it_lock_mode);
1628         it.d.lustre.it_lock_handle = 0;
1629         return rc;
1630 fail:
1631         kunmap(page);
1632         mdc_release_page(page, 1);
1633         rc = -EIO;
1634         goto out_unlock;
1635 }
1636
1637 /**
1638  * Read one directory entry from the cache.
1639  */
1640 int mdc_read_entry(struct obd_export *exp, struct md_op_data *op_data,
1641                    struct md_callback *cb_op, struct lu_dirent **entp,
1642                    struct page **ppage)
1643 {
1644         struct page             *page = NULL;
1645         struct lu_dirpage       *dp;
1646         struct lu_dirent        *ent;
1647         int                     rc = 0;
1648         ENTRY;
1649
1650         CDEBUG(D_INFO, DFID "offset = "LPU64"\n", PFID(&op_data->op_fid1),
1651                op_data->op_hash_offset);
1652
1653         *ppage = NULL;
1654         *entp = NULL;
1655
1656         if (op_data->op_hash_offset == MDS_DIR_END_OFF)
1657                 RETURN(0);
1658
1659         rc = mdc_read_page(exp, op_data, cb_op, &page);
1660         if (rc != 0)
1661                 RETURN(rc);
1662
1663         dp = page_address(page);
1664         for (ent = lu_dirent_start(dp); ent != NULL;
1665              ent = lu_dirent_next(ent)) {
1666                 /* Skip dummy entry */
1667                 if (le16_to_cpu(ent->lde_namelen) == 0)
1668                         continue;
1669
1670                 if (le64_to_cpu(ent->lde_hash) > op_data->op_hash_offset)
1671                         break;
1672         }
1673
1674         /* If it can not find entry in current page, try next page. */
1675         if (ent == NULL) {
1676                 __u64 orig_offset = op_data->op_hash_offset;
1677
1678                 if (dp->ldp_hash_end == MDS_DIR_END_OFF) {
1679                         mdc_release_page(page, 0);
1680                         RETURN(0);
1681                 }
1682
1683                 op_data->op_hash_offset = dp->ldp_hash_end;
1684                 mdc_release_page(page,
1685                                  le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
1686                 rc = mdc_read_page(exp, op_data, cb_op, &page);
1687                 if (rc != 0)
1688                         RETURN(rc);
1689
1690                 if (page != NULL) {
1691                         dp = page_address(page);
1692                         ent = lu_dirent_start(dp);
1693                 }
1694
1695                 op_data->op_hash_offset = orig_offset;
1696         }
1697
1698         *ppage = page;
1699         *entp = ent;
1700
1701         RETURN(rc);
1702 }
1703
1704 #else /* __KERNEL__ */
1705
1706 static struct page
1707 *mdc_read_page_remote(struct obd_export *exp, const struct lmv_oinfo *lmo,
1708                       const __u64 hash, struct obd_capa *oc)
1709 {
1710         struct ptlrpc_request *req = NULL;
1711         struct page *page;
1712         int rc;
1713
1714         OBD_PAGE_ALLOC(page, 0);
1715         if (page == NULL)
1716                 return ERR_PTR(-ENOMEM);
1717
1718         rc = mdc_getpage(exp, &lmo->lmo_fid, hash, oc, &page, 1, &req);
1719         if (req != NULL)
1720                 ptlrpc_req_finished(req);
1721
1722         if (unlikely(rc)) {
1723                 OBD_PAGE_FREE(page);
1724                 return ERR_PTR(rc);
1725         }
1726         return page;
1727 }
1728
1729
1730 static int mdc_read_page(struct obd_export *exp, struct md_op_data *op_data,
1731                         struct md_callback *cb_op,
1732                         struct page **ppage)
1733 {
1734         struct page *page;
1735         struct lmv_oinfo *lmo;
1736         int rc = 0;
1737
1738         /* No local cache for liblustre, always read entry remotely */
1739         lmo = &op_data->op_mea1->lsm_md_oinfo[op_data->op_stripe_offset];
1740         page = mdc_read_page_remote(exp, lmo, op_data->op_hash_offset,
1741                                     op_data->op_capa1);
1742         if (IS_ERR(page))
1743                 return PTR_ERR(page);
1744
1745         *ppage = page;
1746
1747         return rc;
1748 }
1749
1750 int mdc_read_entry(struct obd_export *exp, struct md_op_data *op_data,
1751                    struct md_callback *cb_op, struct lu_dirent **entp,
1752                    struct page **ppage)
1753 {
1754         struct page             *page = NULL;
1755         struct lu_dirpage       *dp;
1756         struct lu_dirent        *ent;
1757         int                     rc;
1758         ENTRY;
1759
1760         rc = mdc_read_page(exp, op_data, cb_op, &page);
1761         if (rc != 0)
1762                 RETURN(rc);
1763
1764         dp = page_address(page);
1765         if (dp->ldp_hash_end < op_data->op_hash_offset)
1766                 GOTO(out, *entp = NULL);
1767
1768         for (ent = lu_dirent_start(dp); ent != NULL;
1769              ent = lu_dirent_next(ent))
1770                 if (ent->lde_hash >= op_data->op_hash_offset)
1771                         break;
1772         *entp = ent;
1773 out:
1774
1775         OBD_PAGE_FREE(page);
1776         RETURN(rc);
1777 }
1778
1779 #endif
1780
1781 static int mdc_statfs(const struct lu_env *env,
1782                       struct obd_export *exp, struct obd_statfs *osfs,
1783                       __u64 max_age, __u32 flags)
1784 {
1785         struct obd_device     *obd = class_exp2obd(exp);
1786         struct ptlrpc_request *req;
1787         struct obd_statfs     *msfs;
1788         struct obd_import     *imp = NULL;
1789         int                    rc;
1790         ENTRY;
1791
1792         /*
1793          * Since the request might also come from lprocfs, so we need
1794          * sync this with client_disconnect_export Bug15684
1795          */
1796         down_read(&obd->u.cli.cl_sem);
1797         if (obd->u.cli.cl_import)
1798                 imp = class_import_get(obd->u.cli.cl_import);
1799         up_read(&obd->u.cli.cl_sem);
1800         if (!imp)
1801                 RETURN(-ENODEV);
1802
1803         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_STATFS,
1804                                         LUSTRE_MDS_VERSION, MDS_STATFS);
1805         if (req == NULL)
1806                 GOTO(output, rc = -ENOMEM);
1807
1808         ptlrpc_request_set_replen(req);
1809
1810         if (flags & OBD_STATFS_NODELAY) {
1811                 /* procfs requests not want stay in wait for avoid deadlock */
1812                 req->rq_no_resend = 1;
1813                 req->rq_no_delay = 1;
1814         }
1815
1816         rc = ptlrpc_queue_wait(req);
1817         if (rc) {
1818                 /* check connection error first */
1819                 if (imp->imp_connect_error)
1820                         rc = imp->imp_connect_error;
1821                 GOTO(out, rc);
1822         }
1823
1824         msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1825         if (msfs == NULL)
1826                 GOTO(out, rc = -EPROTO);
1827
1828         *osfs = *msfs;
1829         EXIT;
1830 out:
1831         ptlrpc_req_finished(req);
1832 output:
1833         class_import_put(imp);
1834         return rc;
1835 }
1836
1837 static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
1838 {
1839         __u32 keylen, vallen;
1840         void *key;
1841         int rc;
1842
1843         if (gf->gf_pathlen > PATH_MAX)
1844                 RETURN(-ENAMETOOLONG);
1845         if (gf->gf_pathlen < 2)
1846                 RETURN(-EOVERFLOW);
1847
1848         /* Key is KEY_FID2PATH + getinfo_fid2path description */
1849         keylen = cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf);
1850         OBD_ALLOC(key, keylen);
1851         if (key == NULL)
1852                 RETURN(-ENOMEM);
1853         memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
1854         memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf));
1855
1856         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
1857                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
1858
1859         if (!fid_is_sane(&gf->gf_fid))
1860                 GOTO(out, rc = -EINVAL);
1861
1862         /* Val is struct getinfo_fid2path result plus path */
1863         vallen = sizeof(*gf) + gf->gf_pathlen;
1864
1865         rc = obd_get_info(NULL, exp, keylen, key, &vallen, gf, NULL);
1866         if (rc != 0 && rc != -EREMOTE)
1867                 GOTO(out, rc);
1868
1869         if (vallen <= sizeof(*gf))
1870                 GOTO(out, rc = -EPROTO);
1871         else if (vallen > sizeof(*gf) + gf->gf_pathlen)
1872                 GOTO(out, rc = -EOVERFLOW);
1873
1874         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n%s\n",
1875                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno, gf->gf_path);
1876
1877 out:
1878         OBD_FREE(key, keylen);
1879         return rc;
1880 }
1881
1882 static int mdc_ioc_hsm_progress(struct obd_export *exp,
1883                                 struct hsm_progress_kernel *hpk)
1884 {
1885         struct obd_import               *imp = class_exp2cliimp(exp);
1886         struct hsm_progress_kernel      *req_hpk;
1887         struct ptlrpc_request           *req;
1888         int                              rc;
1889         ENTRY;
1890
1891         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_PROGRESS,
1892                                         LUSTRE_MDS_VERSION, MDS_HSM_PROGRESS);
1893         if (req == NULL)
1894                 GOTO(out, rc = -ENOMEM);
1895
1896         mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
1897
1898         /* Copy hsm_progress struct */
1899         req_hpk = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_PROGRESS);
1900         if (req_hpk == NULL)
1901                 GOTO(out, rc = -EPROTO);
1902
1903         *req_hpk = *hpk;
1904         req_hpk->hpk_errval = lustre_errno_hton(hpk->hpk_errval);
1905
1906         ptlrpc_request_set_replen(req);
1907
1908         rc = mdc_queue_wait(req);
1909         GOTO(out, rc);
1910 out:
1911         ptlrpc_req_finished(req);
1912         return rc;
1913 }
1914
1915 static int mdc_ioc_hsm_ct_register(struct obd_import *imp, __u32 archives)
1916 {
1917         __u32                   *archive_mask;
1918         struct ptlrpc_request   *req;
1919         int                      rc;
1920         ENTRY;
1921
1922         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_REGISTER,
1923                                         LUSTRE_MDS_VERSION,
1924                                         MDS_HSM_CT_REGISTER);
1925         if (req == NULL)
1926                 GOTO(out, rc = -ENOMEM);
1927
1928         mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
1929
1930         /* Copy hsm_progress struct */
1931         archive_mask = req_capsule_client_get(&req->rq_pill,
1932                                               &RMF_MDS_HSM_ARCHIVE);
1933         if (archive_mask == NULL)
1934                 GOTO(out, rc = -EPROTO);
1935
1936         *archive_mask = archives;
1937
1938         ptlrpc_request_set_replen(req);
1939
1940         rc = mdc_queue_wait(req);
1941         GOTO(out, rc);
1942 out:
1943         ptlrpc_req_finished(req);
1944         return rc;
1945 }
1946
1947 static int mdc_ioc_hsm_current_action(struct obd_export *exp,
1948                                       struct md_op_data *op_data)
1949 {
1950         struct hsm_current_action       *hca = op_data->op_data;
1951         struct hsm_current_action       *req_hca;
1952         struct ptlrpc_request           *req;
1953         int                              rc;
1954         ENTRY;
1955
1956         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1957                                    &RQF_MDS_HSM_ACTION);
1958         if (req == NULL)
1959                 RETURN(-ENOMEM);
1960
1961         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1962
1963         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_ACTION);
1964         if (rc) {
1965                 ptlrpc_request_free(req);
1966                 RETURN(rc);
1967         }
1968
1969         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
1970                       OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
1971
1972         ptlrpc_request_set_replen(req);
1973
1974         rc = mdc_queue_wait(req);
1975         if (rc)
1976                 GOTO(out, rc);
1977
1978         req_hca = req_capsule_server_get(&req->rq_pill,
1979                                          &RMF_MDS_HSM_CURRENT_ACTION);
1980         if (req_hca == NULL)
1981                 GOTO(out, rc = -EPROTO);
1982
1983         *hca = *req_hca;
1984
1985         EXIT;
1986 out:
1987         ptlrpc_req_finished(req);
1988         return rc;
1989 }
1990
1991 static int mdc_ioc_hsm_ct_unregister(struct obd_import *imp)
1992 {
1993         struct ptlrpc_request   *req;
1994         int                      rc;
1995         ENTRY;
1996
1997         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_UNREGISTER,
1998                                         LUSTRE_MDS_VERSION,
1999                                         MDS_HSM_CT_UNREGISTER);
2000         if (req == NULL)
2001                 GOTO(out, rc = -ENOMEM);
2002
2003         mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
2004
2005         ptlrpc_request_set_replen(req);
2006
2007         rc = mdc_queue_wait(req);
2008         GOTO(out, rc);
2009 out:
2010         ptlrpc_req_finished(req);
2011         return rc;
2012 }
2013
2014 static int mdc_ioc_hsm_state_get(struct obd_export *exp,
2015                                  struct md_op_data *op_data)
2016 {
2017         struct hsm_user_state   *hus = op_data->op_data;
2018         struct hsm_user_state   *req_hus;
2019         struct ptlrpc_request   *req;
2020         int                      rc;
2021         ENTRY;
2022
2023         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
2024                                    &RQF_MDS_HSM_STATE_GET);
2025         if (req == NULL)
2026                 RETURN(-ENOMEM);
2027
2028         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
2029
2030         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_GET);
2031         if (rc != 0) {
2032                 ptlrpc_request_free(req);
2033                 RETURN(rc);
2034         }
2035
2036         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
2037                       OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
2038
2039         ptlrpc_request_set_replen(req);
2040
2041         rc = mdc_queue_wait(req);
2042         if (rc)
2043                 GOTO(out, rc);
2044
2045         req_hus = req_capsule_server_get(&req->rq_pill, &RMF_HSM_USER_STATE);
2046         if (req_hus == NULL)
2047                 GOTO(out, rc = -EPROTO);
2048
2049         *hus = *req_hus;
2050
2051         EXIT;
2052 out:
2053         ptlrpc_req_finished(req);
2054         return rc;
2055 }
2056
2057 static int mdc_ioc_hsm_state_set(struct obd_export *exp,
2058                                  struct md_op_data *op_data)
2059 {
2060         struct hsm_state_set    *hss = op_data->op_data;
2061         struct hsm_state_set    *req_hss;
2062         struct ptlrpc_request   *req;
2063         int                      rc;
2064         ENTRY;
2065
2066         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
2067                                    &RQF_MDS_HSM_STATE_SET);
2068         if (req == NULL)
2069                 RETURN(-ENOMEM);
2070
2071         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
2072
2073         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_SET);
2074         if (rc) {
2075                 ptlrpc_request_free(req);
2076                 RETURN(rc);
2077         }
2078
2079         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
2080                       OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
2081
2082         /* Copy states */
2083         req_hss = req_capsule_client_get(&req->rq_pill, &RMF_HSM_STATE_SET);
2084         if (req_hss == NULL)
2085                 GOTO(out, rc = -EPROTO);
2086         *req_hss = *hss;
2087
2088         ptlrpc_request_set_replen(req);
2089
2090         rc = mdc_queue_wait(req);
2091         GOTO(out, rc);
2092
2093         EXIT;
2094 out:
2095         ptlrpc_req_finished(req);
2096         return rc;
2097 }
2098
2099 static int mdc_ioc_hsm_request(struct obd_export *exp,
2100                                struct hsm_user_request *hur)
2101 {
2102         struct obd_import       *imp = class_exp2cliimp(exp);
2103         struct ptlrpc_request   *req;
2104         struct hsm_request      *req_hr;
2105         struct hsm_user_item    *req_hui;
2106         char                    *req_opaque;
2107         int                      rc;
2108         ENTRY;
2109
2110         req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_REQUEST);
2111         if (req == NULL)
2112                 GOTO(out, rc = -ENOMEM);
2113
2114         req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM, RCL_CLIENT,
2115                              hur->hur_request.hr_itemcount
2116                              * sizeof(struct hsm_user_item));
2117         req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA, RCL_CLIENT,
2118                              hur->hur_request.hr_data_len);
2119
2120         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_REQUEST);
2121         if (rc) {
2122                 ptlrpc_request_free(req);
2123                 RETURN(rc);
2124         }
2125
2126         mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
2127
2128         /* Copy hsm_request struct */
2129         req_hr = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_REQUEST);
2130         if (req_hr == NULL)
2131                 GOTO(out, rc = -EPROTO);
2132         *req_hr = hur->hur_request;
2133
2134         /* Copy hsm_user_item structs */
2135         req_hui = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM);
2136         if (req_hui == NULL)
2137                 GOTO(out, rc = -EPROTO);
2138         memcpy(req_hui, hur->hur_user_item,
2139                hur->hur_request.hr_itemcount * sizeof(struct hsm_user_item));
2140
2141         /* Copy opaque field */
2142         req_opaque = req_capsule_client_get(&req->rq_pill, &RMF_GENERIC_DATA);
2143         if (req_opaque == NULL)
2144                 GOTO(out, rc = -EPROTO);
2145         memcpy(req_opaque, hur_data(hur), hur->hur_request.hr_data_len);
2146
2147         ptlrpc_request_set_replen(req);
2148
2149         rc = mdc_queue_wait(req);
2150         GOTO(out, rc);
2151
2152 out:
2153         ptlrpc_req_finished(req);
2154         return rc;
2155 }
2156
2157 static struct kuc_hdr *changelog_kuc_hdr(char *buf, int len, int flags)
2158 {
2159         struct kuc_hdr *lh = (struct kuc_hdr *)buf;
2160
2161         LASSERT(len <= KUC_CHANGELOG_MSG_MAXSIZE);
2162
2163         lh->kuc_magic = KUC_MAGIC;
2164         lh->kuc_transport = KUC_TRANSPORT_CHANGELOG;
2165         lh->kuc_flags = flags;
2166         lh->kuc_msgtype = CL_RECORD;
2167         lh->kuc_msglen = len;
2168         return lh;
2169 }
2170
2171 #define D_CHANGELOG 0
2172
2173 struct changelog_show {
2174         __u64           cs_startrec;
2175         __u32           cs_flags;
2176         struct file     *cs_fp;
2177         char            *cs_buf;
2178         struct obd_device *cs_obd;
2179 };
2180
2181 static int changelog_kkuc_cb(const struct lu_env *env, struct llog_handle *llh,
2182                              struct llog_rec_hdr *hdr, void *data)
2183 {
2184         struct changelog_show *cs = data;
2185         struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
2186         struct kuc_hdr *lh;
2187         int len, rc;
2188         ENTRY;
2189
2190         if (rec->cr_hdr.lrh_type != CHANGELOG_REC) {
2191                 rc = -EINVAL;
2192                 CERROR("%s: not a changelog rec %x/%d: rc = %d\n",
2193                        cs->cs_obd->obd_name, rec->cr_hdr.lrh_type,
2194                        rec->cr.cr_type, rc);
2195                 RETURN(rc);
2196         }
2197
2198         if (rec->cr.cr_index < cs->cs_startrec) {
2199                 /* Skip entries earlier than what we are interested in */
2200                 CDEBUG(D_CHANGELOG, "rec="LPU64" start="LPU64"\n",
2201                        rec->cr.cr_index, cs->cs_startrec);
2202                 RETURN(0);
2203         }
2204
2205         CDEBUG(D_CHANGELOG, LPU64" %02d%-5s "LPU64" 0x%x t="DFID" p="DFID
2206                 " %.*s\n", rec->cr.cr_index, rec->cr.cr_type,
2207                 changelog_type2str(rec->cr.cr_type), rec->cr.cr_time,
2208                 rec->cr.cr_flags & CLF_FLAGMASK,
2209                 PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid),
2210                 rec->cr.cr_namelen, changelog_rec_name(&rec->cr));
2211
2212         len = sizeof(*lh) + changelog_rec_size(&rec->cr) + rec->cr.cr_namelen;
2213
2214         /* Set up the message */
2215         lh = changelog_kuc_hdr(cs->cs_buf, len, cs->cs_flags);
2216         memcpy(lh + 1, &rec->cr, len - sizeof(*lh));
2217
2218         rc = libcfs_kkuc_msg_put(cs->cs_fp, lh);
2219         CDEBUG(D_CHANGELOG, "kucmsg fp %p len %d rc %d\n", cs->cs_fp, len,rc);
2220
2221         RETURN(rc);
2222 }
2223
2224 static int mdc_changelog_send_thread(void *csdata)
2225 {
2226         struct changelog_show *cs = csdata;
2227         struct llog_ctxt *ctxt = NULL;
2228         struct llog_handle *llh = NULL;
2229         struct kuc_hdr *kuch;
2230         int rc;
2231
2232         CDEBUG(D_CHANGELOG, "changelog to fp=%p start "LPU64"\n",
2233                cs->cs_fp, cs->cs_startrec);
2234
2235         OBD_ALLOC(cs->cs_buf, KUC_CHANGELOG_MSG_MAXSIZE);
2236         if (cs->cs_buf == NULL)
2237                 GOTO(out, rc = -ENOMEM);
2238
2239         /* Set up the remote catalog handle */
2240         ctxt = llog_get_context(cs->cs_obd, LLOG_CHANGELOG_REPL_CTXT);
2241         if (ctxt == NULL)
2242                 GOTO(out, rc = -ENOENT);
2243         rc = llog_open(NULL, ctxt, &llh, NULL, CHANGELOG_CATALOG,
2244                        LLOG_OPEN_EXISTS);
2245         if (rc) {
2246                 CERROR("%s: fail to open changelog catalog: rc = %d\n",
2247                        cs->cs_obd->obd_name, rc);
2248                 GOTO(out, rc);
2249         }
2250         rc = llog_init_handle(NULL, llh, LLOG_F_IS_CAT, NULL);
2251         if (rc) {
2252                 CERROR("llog_init_handle failed %d\n", rc);
2253                 GOTO(out, rc);
2254         }
2255
2256         rc = llog_cat_process(NULL, llh, changelog_kkuc_cb, cs, 0, 0);
2257
2258         /* Send EOF no matter what our result */
2259         if ((kuch = changelog_kuc_hdr(cs->cs_buf, sizeof(*kuch),
2260                                       cs->cs_flags))) {
2261                 kuch->kuc_msgtype = CL_EOF;
2262                 libcfs_kkuc_msg_put(cs->cs_fp, kuch);
2263         }
2264
2265 out:
2266         fput(cs->cs_fp);
2267         if (llh)
2268                 llog_cat_close(NULL, llh);
2269         if (ctxt)
2270                 llog_ctxt_put(ctxt);
2271         if (cs->cs_buf)
2272                 OBD_FREE(cs->cs_buf, KUC_CHANGELOG_MSG_MAXSIZE);
2273         OBD_FREE_PTR(cs);
2274         return rc;
2275 }
2276
2277 static int mdc_ioc_changelog_send(struct obd_device *obd,
2278                                   struct ioc_changelog *icc)
2279 {
2280         struct changelog_show *cs;
2281         int rc;
2282
2283         /* Freed in mdc_changelog_send_thread */
2284         OBD_ALLOC_PTR(cs);
2285         if (!cs)
2286                 return -ENOMEM;
2287
2288         cs->cs_obd = obd;
2289         cs->cs_startrec = icc->icc_recno;
2290         /* matching fput in mdc_changelog_send_thread */
2291         cs->cs_fp = fget(icc->icc_id);
2292         cs->cs_flags = icc->icc_flags;
2293
2294         /*
2295          * New thread because we should return to user app before
2296          * writing into our pipe
2297          */
2298         rc = PTR_ERR(kthread_run(mdc_changelog_send_thread, cs,
2299                                  "mdc_clg_send_thread"));
2300         if (!IS_ERR_VALUE(rc)) {
2301                 CDEBUG(D_CHANGELOG, "start changelog thread\n");
2302                 return 0;
2303         }
2304
2305         CERROR("Failed to start changelog thread: %d\n", rc);
2306         OBD_FREE_PTR(cs);
2307         return rc;
2308 }
2309
2310 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2311                                 struct lustre_kernelcomm *lk);
2312
2313 static int mdc_quotacheck(struct obd_device *unused, struct obd_export *exp,
2314                           struct obd_quotactl *oqctl)
2315 {
2316         struct client_obd       *cli = &exp->exp_obd->u.cli;
2317         struct ptlrpc_request   *req;
2318         struct obd_quotactl     *body;
2319         int                      rc;
2320         ENTRY;
2321
2322         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
2323                                         &RQF_MDS_QUOTACHECK, LUSTRE_MDS_VERSION,
2324                                         MDS_QUOTACHECK);
2325         if (req == NULL)
2326                 RETURN(-ENOMEM);
2327
2328         body = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
2329         *body = *oqctl;
2330
2331         ptlrpc_request_set_replen(req);
2332
2333         /* the next poll will find -ENODATA, that means quotacheck is
2334          * going on */
2335         cli->cl_qchk_stat = -ENODATA;
2336         rc = ptlrpc_queue_wait(req);
2337         if (rc)
2338                 cli->cl_qchk_stat = rc;
2339         ptlrpc_req_finished(req);
2340         RETURN(rc);
2341 }
2342
2343 static int mdc_quota_poll_check(struct obd_export *exp,
2344                                 struct if_quotacheck *qchk)
2345 {
2346         struct client_obd *cli = &exp->exp_obd->u.cli;
2347         int rc;
2348         ENTRY;
2349
2350         qchk->obd_uuid = cli->cl_target_uuid;
2351         memcpy(qchk->obd_type, LUSTRE_MDS_NAME, strlen(LUSTRE_MDS_NAME));
2352
2353         rc = cli->cl_qchk_stat;
2354         /* the client is not the previous one */
2355         if (rc == CL_NOT_QUOTACHECKED)
2356                 rc = -EINTR;
2357         RETURN(rc);
2358 }
2359
2360 static int mdc_quotactl(struct obd_device *unused, struct obd_export *exp,
2361                         struct obd_quotactl *oqctl)
2362 {
2363         struct ptlrpc_request   *req;
2364         struct obd_quotactl     *oqc;
2365         int                      rc;
2366         ENTRY;
2367
2368         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
2369                                         &RQF_MDS_QUOTACTL, LUSTRE_MDS_VERSION,
2370                                         MDS_QUOTACTL);
2371         if (req == NULL)
2372                 RETURN(-ENOMEM);
2373
2374         oqc = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
2375         *oqc = *oqctl;
2376
2377         ptlrpc_request_set_replen(req);
2378         ptlrpc_at_set_req_timeout(req);
2379         req->rq_no_resend = 1;
2380
2381         rc = ptlrpc_queue_wait(req);
2382         if (rc)
2383                 CERROR("ptlrpc_queue_wait failed, rc: %d\n", rc);
2384
2385         if (req->rq_repmsg &&
2386             (oqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL))) {
2387                 *oqctl = *oqc;
2388         } else if (!rc) {
2389                 CERROR ("Can't unpack obd_quotactl\n");
2390                 rc = -EPROTO;
2391         }
2392         ptlrpc_req_finished(req);
2393
2394         RETURN(rc);
2395 }
2396
2397 static int mdc_ioc_swap_layouts(struct obd_export *exp,
2398                                 struct md_op_data *op_data)
2399 {
2400         CFS_LIST_HEAD(cancels);
2401         struct ptlrpc_request   *req;
2402         int                      rc, count;
2403         struct mdc_swap_layouts *msl, *payload;
2404         ENTRY;
2405
2406         msl = op_data->op_data;
2407
2408         /* When the MDT will get the MDS_SWAP_LAYOUTS RPC the
2409          * first thing it will do is to cancel the 2 layout
2410          * locks hold by this client.
2411          * So the client must cancel its layout locks on the 2 fids
2412          * with the request RPC to avoid extra RPC round trips
2413          */
2414         count = mdc_resource_get_unused(exp, &op_data->op_fid1, &cancels,
2415                                         LCK_CR, MDS_INODELOCK_LAYOUT);
2416         count += mdc_resource_get_unused(exp, &op_data->op_fid2, &cancels,
2417                                          LCK_CR, MDS_INODELOCK_LAYOUT);
2418
2419         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
2420                                    &RQF_MDS_SWAP_LAYOUTS);
2421         if (req == NULL) {
2422                 ldlm_lock_list_put(&cancels, l_bl_ast, count);
2423                 RETURN(-ENOMEM);
2424         }
2425
2426         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
2427         mdc_set_capa_size(req, &RMF_CAPA2, op_data->op_capa2);
2428
2429         rc = mdc_prep_elc_req(exp, req, MDS_SWAP_LAYOUTS, &cancels, count);
2430         if (rc) {
2431                 ptlrpc_request_free(req);
2432                 RETURN(rc);
2433         }
2434
2435         mdc_swap_layouts_pack(req, op_data);
2436
2437         payload = req_capsule_client_get(&req->rq_pill, &RMF_SWAP_LAYOUTS);
2438         LASSERT(payload);
2439
2440         *payload = *msl;
2441
2442         ptlrpc_request_set_replen(req);
2443
2444         rc = ptlrpc_queue_wait(req);
2445         if (rc)
2446                 GOTO(out, rc);
2447         EXIT;
2448
2449 out:
2450         ptlrpc_req_finished(req);
2451         return rc;
2452 }
2453
2454 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2455                          void *karg, void *uarg)
2456 {
2457         struct obd_device *obd = exp->exp_obd;
2458         struct obd_ioctl_data *data = karg;
2459         struct obd_import *imp = obd->u.cli.cl_import;
2460         int rc;
2461         ENTRY;
2462
2463         if (!try_module_get(THIS_MODULE)) {
2464                 CERROR("Can't get module. Is it alive?");
2465                 return -EINVAL;
2466         }
2467         switch (cmd) {
2468         case OBD_IOC_CHANGELOG_SEND:
2469                 rc = mdc_ioc_changelog_send(obd, karg);
2470                 GOTO(out, rc);
2471         case OBD_IOC_CHANGELOG_CLEAR: {
2472                 struct ioc_changelog *icc = karg;
2473                 struct changelog_setinfo cs =
2474                         {.cs_recno = icc->icc_recno, .cs_id = icc->icc_id};
2475                 rc = obd_set_info_async(NULL, exp, strlen(KEY_CHANGELOG_CLEAR),
2476                                         KEY_CHANGELOG_CLEAR, sizeof(cs), &cs,
2477                                         NULL);
2478                 GOTO(out, rc);
2479         }
2480         case OBD_IOC_FID2PATH:
2481                 rc = mdc_ioc_fid2path(exp, karg);
2482                 GOTO(out, rc);
2483         case LL_IOC_HSM_CT_START:
2484                 rc = mdc_ioc_hsm_ct_start(exp, karg);
2485                 /* ignore if it was already registered on this MDS. */
2486                 if (rc == -EEXIST)
2487                         rc = 0;
2488                 GOTO(out, rc);
2489         case LL_IOC_HSM_PROGRESS:
2490                 rc = mdc_ioc_hsm_progress(exp, karg);
2491                 GOTO(out, rc);
2492         case LL_IOC_HSM_STATE_GET:
2493                 rc = mdc_ioc_hsm_state_get(exp, karg);
2494                 GOTO(out, rc);
2495         case LL_IOC_HSM_STATE_SET:
2496                 rc = mdc_ioc_hsm_state_set(exp, karg);
2497                 GOTO(out, rc);
2498         case LL_IOC_HSM_ACTION:
2499                 rc = mdc_ioc_hsm_current_action(exp, karg);
2500                 GOTO(out, rc);
2501         case LL_IOC_HSM_REQUEST:
2502                 rc = mdc_ioc_hsm_request(exp, karg);
2503                 GOTO(out, rc);
2504         case OBD_IOC_CLIENT_RECOVER:
2505                 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1, 0);
2506                 if (rc < 0)
2507                         GOTO(out, rc);
2508                 GOTO(out, rc = 0);
2509         case IOC_OSC_SET_ACTIVE:
2510                 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
2511                 GOTO(out, rc);
2512         case OBD_IOC_POLL_QUOTACHECK:
2513                 rc = mdc_quota_poll_check(exp, (struct if_quotacheck *)karg);
2514                 GOTO(out, rc);
2515         case OBD_IOC_PING_TARGET:
2516                 rc = ptlrpc_obd_ping(obd);
2517                 GOTO(out, rc);
2518         /*
2519          * Normally IOC_OBD_STATFS, OBD_IOC_QUOTACTL iocontrol are handled by
2520          * LMV instead of MDC. But when the cluster is upgraded from 1.8,
2521          * there'd be no LMV layer thus we might be called here. Eventually
2522          * this code should be removed.
2523          * bz20731, LU-592.
2524          */
2525         case IOC_OBD_STATFS: {
2526                 struct obd_statfs stat_buf = {0};
2527
2528                 if (*((__u32 *) data->ioc_inlbuf2) != 0)
2529                         GOTO(out, rc = -ENODEV);
2530
2531                 /* copy UUID */
2532                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd),
2533                                  min((int)data->ioc_plen2,
2534                                      (int)sizeof(struct obd_uuid))))
2535                         GOTO(out, rc = -EFAULT);
2536
2537                 rc = mdc_statfs(NULL, obd->obd_self_export, &stat_buf,
2538                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
2539                                 0);
2540                 if (rc != 0)
2541                         GOTO(out, rc);
2542
2543                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
2544                                      min((int) data->ioc_plen1,
2545                                          (int) sizeof(stat_buf))))
2546                         GOTO(out, rc = -EFAULT);
2547
2548                 GOTO(out, rc = 0);
2549         }
2550         case OBD_IOC_QUOTACTL: {
2551                 struct if_quotactl *qctl = karg;
2552                 struct obd_quotactl *oqctl;
2553
2554                 OBD_ALLOC_PTR(oqctl);
2555                 if (oqctl == NULL)
2556                         GOTO(out, rc = -ENOMEM);
2557
2558                 QCTL_COPY(oqctl, qctl);
2559                 rc = obd_quotactl(exp, oqctl);
2560                 if (rc == 0) {
2561                         QCTL_COPY(qctl, oqctl);
2562                         qctl->qc_valid = QC_MDTIDX;
2563                         qctl->obd_uuid = obd->u.cli.cl_target_uuid;
2564                 }
2565
2566                 OBD_FREE_PTR(oqctl);
2567                 GOTO(out, rc);
2568         }
2569         case LL_IOC_GET_CONNECT_FLAGS:
2570                 if (copy_to_user(uarg, exp_connect_flags_ptr(exp),
2571                                  sizeof(*exp_connect_flags_ptr(exp))))
2572                         GOTO(out, rc = -EFAULT);
2573
2574                 GOTO(out, rc = 0);
2575         case LL_IOC_LOV_SWAP_LAYOUTS:
2576                 rc = mdc_ioc_swap_layouts(exp, karg);
2577                 GOTO(out, rc);
2578         default:
2579                 CERROR("unrecognised ioctl: cmd = %#x\n", cmd);
2580                 GOTO(out, rc = -ENOTTY);
2581         }
2582 out:
2583         module_put(THIS_MODULE);
2584
2585         return rc;
2586 }
2587
2588 int mdc_get_info_rpc(struct obd_export *exp,
2589                      obd_count keylen, void *key,
2590                      int vallen, void *val)
2591 {
2592         struct obd_import      *imp = class_exp2cliimp(exp);
2593         struct ptlrpc_request  *req;
2594         char                   *tmp;
2595         int                     rc = -EINVAL;
2596         ENTRY;
2597
2598         req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
2599         if (req == NULL)
2600                 RETURN(-ENOMEM);
2601
2602         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
2603                              RCL_CLIENT, keylen);
2604         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
2605                              RCL_CLIENT, sizeof(__u32));
2606
2607         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
2608         if (rc) {
2609                 ptlrpc_request_free(req);
2610                 RETURN(rc);
2611         }
2612
2613         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
2614         memcpy(tmp, key, keylen);
2615         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
2616         memcpy(tmp, &vallen, sizeof(__u32));
2617
2618         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
2619                              RCL_SERVER, vallen);
2620         ptlrpc_request_set_replen(req);
2621
2622         rc = ptlrpc_queue_wait(req);
2623         /* -EREMOTE means the get_info result is partial, and it needs to
2624          * continue on another MDT, see fid2path part in lmv_iocontrol */
2625         if (rc == 0 || rc == -EREMOTE) {
2626                 tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
2627                 memcpy(val, tmp, vallen);
2628                 if (ptlrpc_rep_need_swab(req)) {
2629                         if (KEY_IS(KEY_FID2PATH))
2630                                 lustre_swab_fid2path(val);
2631                 }
2632         }
2633         ptlrpc_req_finished(req);
2634
2635         RETURN(rc);
2636 }
2637
2638 static void lustre_swab_hai(struct hsm_action_item *h)
2639 {
2640         __swab32s(&h->hai_len);
2641         __swab32s(&h->hai_action);
2642         lustre_swab_lu_fid(&h->hai_fid);
2643         lustre_swab_lu_fid(&h->hai_dfid);
2644         __swab64s(&h->hai_cookie);
2645         __swab64s(&h->hai_extent.offset);
2646         __swab64s(&h->hai_extent.length);
2647         __swab64s(&h->hai_gid);
2648 }
2649
2650 static void lustre_swab_hal(struct hsm_action_list *h)
2651 {
2652         struct hsm_action_item  *hai;
2653         int                      i;
2654
2655         __swab32s(&h->hal_version);
2656         __swab32s(&h->hal_count);
2657         __swab32s(&h->hal_archive_id);
2658         __swab64s(&h->hal_flags);
2659         hai = hai_first(h);
2660         for (i = 0; i < h->hal_count; i++, hai = hai_next(hai))
2661                 lustre_swab_hai(hai);
2662 }
2663
2664 static void lustre_swab_kuch(struct kuc_hdr *l)
2665 {
2666         __swab16s(&l->kuc_magic);
2667         /* __u8 l->kuc_transport */
2668         __swab16s(&l->kuc_msgtype);
2669         __swab16s(&l->kuc_msglen);
2670 }
2671
2672 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2673                                 struct lustre_kernelcomm *lk)
2674 {
2675         struct obd_import  *imp = class_exp2cliimp(exp);
2676         __u32               archive = lk->lk_data;
2677         int                 rc = 0;
2678
2679         if (lk->lk_group != KUC_GRP_HSM) {
2680                 CERROR("Bad copytool group %d\n", lk->lk_group);
2681                 return -EINVAL;
2682         }
2683
2684         CDEBUG(D_HSM, "CT start r%d w%d u%d g%d f%#x\n", lk->lk_rfd, lk->lk_wfd,
2685                lk->lk_uid, lk->lk_group, lk->lk_flags);
2686
2687         if (lk->lk_flags & LK_FLG_STOP) {
2688                 /* Unregister with the coordinator */
2689                 rc = mdc_ioc_hsm_ct_unregister(imp);
2690         } else {
2691                 rc = mdc_ioc_hsm_ct_register(imp, archive);
2692         }
2693
2694         return rc;
2695 }
2696
2697 /**
2698  * Send a message to any listening copytools
2699  * @param val KUC message (kuc_hdr + hsm_action_list)
2700  * @param len total length of message
2701  */
2702 static int mdc_hsm_copytool_send(int len, void *val)
2703 {
2704         struct kuc_hdr          *lh = (struct kuc_hdr *)val;
2705         struct hsm_action_list  *hal = (struct hsm_action_list *)(lh + 1);
2706         int                      rc;
2707         ENTRY;
2708
2709         if (len < sizeof(*lh) + sizeof(*hal)) {
2710                 CERROR("Short HSM message %d < %d\n", len,
2711                        (int) (sizeof(*lh) + sizeof(*hal)));
2712                 RETURN(-EPROTO);
2713         }
2714         if (lh->kuc_magic == __swab16(KUC_MAGIC)) {
2715                 lustre_swab_kuch(lh);
2716                 lustre_swab_hal(hal);
2717         } else if (lh->kuc_magic != KUC_MAGIC) {
2718                 CERROR("Bad magic %x!=%x\n", lh->kuc_magic, KUC_MAGIC);
2719                 RETURN(-EPROTO);
2720         }
2721
2722         CDEBUG(D_HSM, " Received message mg=%x t=%d m=%d l=%d actions=%d "
2723                "on %s\n",
2724                lh->kuc_magic, lh->kuc_transport, lh->kuc_msgtype,
2725                lh->kuc_msglen, hal->hal_count, hal->hal_fsname);
2726
2727         /* Broadcast to HSM listeners */
2728         rc = libcfs_kkuc_group_put(KUC_GRP_HSM, lh);
2729
2730         RETURN(rc);
2731 }
2732
2733 /**
2734  * callback function passed to kuc for re-registering each HSM copytool
2735  * running on MDC, after MDT shutdown/recovery.
2736  * @param data copytool registration data
2737  * @param cb_arg callback argument (obd_import)
2738  */
2739 static int mdc_hsm_ct_reregister(void *data, void *cb_arg)
2740 {
2741         struct kkuc_ct_data     *kcd = data;
2742         struct obd_import       *imp = (struct obd_import *)cb_arg;
2743         int                      rc;
2744
2745         if (kcd == NULL || kcd->kcd_magic != KKUC_CT_DATA_MAGIC)
2746                 return -EPROTO;
2747
2748         if (!obd_uuid_equals(&kcd->kcd_uuid, &imp->imp_obd->obd_uuid))
2749                 return 0;
2750
2751         CDEBUG(D_HA, "%s: recover copytool registration to MDT (archive=%#x)\n",
2752                imp->imp_obd->obd_name, kcd->kcd_archive);
2753         rc = mdc_ioc_hsm_ct_register(imp, kcd->kcd_archive);
2754
2755         /* ignore error if the copytool is already registered */
2756         return (rc == -EEXIST) ? 0 : rc;
2757 }
2758
2759 /**
2760  * Re-establish all kuc contexts with MDT
2761  * after MDT shutdown/recovery.
2762  */
2763 static int mdc_kuc_reregister(struct obd_import *imp)
2764 {
2765         /* re-register HSM agents */
2766         return libcfs_kkuc_group_foreach(KUC_GRP_HSM, mdc_hsm_ct_reregister,
2767                                          (void *)imp);
2768 }
2769
2770 int mdc_set_info_async(const struct lu_env *env,
2771                        struct obd_export *exp,
2772                        obd_count keylen, void *key,
2773                        obd_count vallen, void *val,
2774                        struct ptlrpc_request_set *set)
2775 {
2776         struct obd_import       *imp = class_exp2cliimp(exp);
2777         int                      rc;
2778         ENTRY;
2779
2780         if (KEY_IS(KEY_READ_ONLY)) {
2781                 if (vallen != sizeof(int))
2782                         RETURN(-EINVAL);
2783
2784                 spin_lock(&imp->imp_lock);
2785                 if (*((int *)val)) {
2786                         imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
2787                         imp->imp_connect_data.ocd_connect_flags |=
2788                                                         OBD_CONNECT_RDONLY;
2789                 } else {
2790                         imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
2791                         imp->imp_connect_data.ocd_connect_flags &=
2792                                                         ~OBD_CONNECT_RDONLY;
2793                 }
2794                 spin_unlock(&imp->imp_lock);
2795
2796                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2797                                        keylen, key, vallen, val, set);
2798                 RETURN(rc);
2799         }
2800         if (KEY_IS(KEY_SPTLRPC_CONF)) {
2801                 sptlrpc_conf_client_adapt(exp->exp_obd);
2802                 RETURN(0);
2803         }
2804         if (KEY_IS(KEY_FLUSH_CTX)) {
2805                 sptlrpc_import_flush_my_ctx(imp);
2806                 RETURN(0);
2807         }
2808         if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2809                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2810                                        keylen, key, vallen, val, set);
2811                 RETURN(rc);
2812         }
2813         if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) {
2814                 rc = mdc_hsm_copytool_send(vallen, val);
2815                 RETURN(rc);
2816         }
2817
2818         CERROR("Unknown key %s\n", (char *)key);
2819         RETURN(-EINVAL);
2820 }
2821
2822 int mdc_get_info(const struct lu_env *env, struct obd_export *exp,
2823                  __u32 keylen, void *key, __u32 *vallen, void *val,
2824                  struct lov_stripe_md *lsm)
2825 {
2826         int rc = -EINVAL;
2827
2828         if (KEY_IS(KEY_MAX_EASIZE)) {
2829                 int mdsize, *max_easize;
2830
2831                 if (*vallen != sizeof(int))
2832                         RETURN(-EINVAL);
2833                 mdsize = *(int*)val;
2834                 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
2835                         exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
2836                 max_easize = val;
2837                 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
2838                 RETURN(0);
2839         } else if (KEY_IS(KEY_CONN_DATA)) {
2840                 struct obd_import *imp = class_exp2cliimp(exp);
2841                 struct obd_connect_data *data = val;
2842
2843                 if (*vallen != sizeof(*data))
2844                         RETURN(-EINVAL);
2845
2846                 *data = imp->imp_connect_data;
2847                 RETURN(0);
2848         } else if (KEY_IS(KEY_TGT_COUNT)) {
2849                 *((int *)val) = 1;
2850                 RETURN(0);
2851         }
2852
2853         rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
2854
2855         RETURN(rc);
2856 }
2857
2858 static int mdc_pin(struct obd_export *exp, const struct lu_fid *fid,
2859                    struct obd_capa *oc, struct obd_client_handle *handle,
2860                    int flags)
2861 {
2862         struct ptlrpc_request *req;
2863         struct mdt_body       *body;
2864         int                    rc;
2865         ENTRY;
2866
2867         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_PIN);
2868         if (req == NULL)
2869                 RETURN(-ENOMEM);
2870
2871         mdc_set_capa_size(req, &RMF_CAPA1, oc);
2872
2873         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_PIN);
2874         if (rc) {
2875                 ptlrpc_request_free(req);
2876                 RETURN(rc);
2877         }
2878
2879         mdc_pack_body(req, fid, oc, 0, 0, -1, flags);
2880
2881         ptlrpc_request_set_replen(req);
2882
2883         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
2884         rc = ptlrpc_queue_wait(req);
2885         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
2886         if (rc) {
2887                 CERROR("Pin failed: %d\n", rc);
2888                 GOTO(err_out, rc);
2889         }
2890
2891         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2892         if (body == NULL)
2893                 GOTO(err_out, rc = -EPROTO);
2894
2895         handle->och_fh = body->handle;
2896         handle->och_magic = OBD_CLIENT_HANDLE_MAGIC;
2897
2898         handle->och_mod = obd_mod_alloc();
2899         if (handle->och_mod == NULL) {
2900                 DEBUG_REQ(D_ERROR, req, "can't allocate md_open_data");
2901                 GOTO(err_out, rc = -ENOMEM);
2902         }
2903         handle->och_mod->mod_open_req = req; /* will be dropped by unpin */
2904
2905         RETURN(0);
2906
2907 err_out:
2908         ptlrpc_req_finished(req);
2909         RETURN(rc);
2910 }
2911
2912 static int mdc_unpin(struct obd_export *exp, struct obd_client_handle *handle,
2913                      int flag)
2914 {
2915         struct ptlrpc_request *req;
2916         struct mdt_body       *body;
2917         int                    rc;
2918         ENTRY;
2919
2920         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_UNPIN,
2921                                         LUSTRE_MDS_VERSION, MDS_UNPIN);
2922         if (req == NULL)
2923                 RETURN(-ENOMEM);
2924
2925         body = req_capsule_client_get(&req->rq_pill, &RMF_MDT_BODY);
2926         body->handle = handle->och_fh;
2927         body->flags = flag;
2928
2929         ptlrpc_request_set_replen(req);
2930
2931         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
2932         rc = ptlrpc_queue_wait(req);
2933         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
2934
2935         if (rc != 0)
2936                 CERROR("Unpin failed: %d\n", rc);
2937
2938         ptlrpc_req_finished(req);
2939         ptlrpc_req_finished(handle->och_mod->mod_open_req);
2940
2941         obd_mod_put(handle->och_mod);
2942         RETURN(rc);
2943 }
2944
2945 int mdc_fsync(struct obd_export *exp, const struct lu_fid *fid,
2946               struct obd_capa *oc, struct ptlrpc_request **request)
2947 {
2948         struct ptlrpc_request *req;
2949         int                    rc;
2950         ENTRY;
2951
2952         *request = NULL;
2953         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
2954         if (req == NULL)
2955                 RETURN(-ENOMEM);
2956
2957         mdc_set_capa_size(req, &RMF_CAPA1, oc);
2958
2959         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
2960         if (rc) {
2961                 ptlrpc_request_free(req);
2962                 RETURN(rc);
2963         }
2964
2965         mdc_pack_body(req, fid, oc, 0, 0, -1, 0);
2966
2967         ptlrpc_request_set_replen(req);
2968
2969         rc = ptlrpc_queue_wait(req);
2970         if (rc)
2971                 ptlrpc_req_finished(req);
2972         else
2973                 *request = req;
2974         RETURN(rc);
2975 }
2976
2977 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
2978                             enum obd_import_event event)
2979 {
2980         int rc = 0;
2981
2982         LASSERT(imp->imp_obd == obd);
2983
2984         switch (event) {
2985         case IMP_EVENT_DISCON: {
2986 #if 0
2987                 /* XXX Pass event up to OBDs stack. used only for FLD now */
2988                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DISCON, NULL);
2989 #endif
2990                 break;
2991         }
2992         case IMP_EVENT_INACTIVE: {
2993                 struct client_obd *cli = &obd->u.cli;
2994                 /*
2995                  * Flush current sequence to make client obtain new one
2996                  * from server in case of disconnect/reconnect.
2997                  */
2998                 if (cli->cl_seq != NULL)
2999                         seq_client_flush(cli->cl_seq);
3000
3001                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
3002                 break;
3003         }
3004         case IMP_EVENT_INVALIDATE: {
3005                 struct ldlm_namespace *ns = obd->obd_namespace;
3006
3007                 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
3008
3009                 break;
3010         }
3011         case IMP_EVENT_ACTIVE:
3012                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
3013                 /* redo the kuc registration after reconnecting */
3014                 if (rc == 0)
3015                         rc = mdc_kuc_reregister(imp);
3016                 break;
3017         case IMP_EVENT_OCD:
3018                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
3019                 break;
3020         case IMP_EVENT_DEACTIVATE:
3021         case IMP_EVENT_ACTIVATE:
3022                 break;
3023         default:
3024                 CERROR("Unknown import event %x\n", event);
3025                 LBUG();
3026         }
3027         RETURN(rc);
3028 }
3029
3030 int mdc_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
3031                   struct md_op_data *op_data)
3032 {
3033         struct client_obd *cli = &exp->exp_obd->u.cli;
3034         struct lu_client_seq *seq = cli->cl_seq;
3035         ENTRY;
3036         RETURN(seq_client_alloc_fid(NULL, seq, fid));
3037 }
3038
3039 struct obd_uuid *mdc_get_uuid(struct obd_export *exp) {
3040         struct client_obd *cli = &exp->exp_obd->u.cli;
3041         return &cli->cl_target_uuid;
3042 }
3043
3044 /**
3045  * Determine whether the lock can be canceled before replaying it during
3046  * recovery, non zero value will be return if the lock can be canceled,
3047  * or zero returned for not
3048  */
3049 static int mdc_cancel_for_recovery(struct ldlm_lock *lock)
3050 {
3051         if (lock->l_resource->lr_type != LDLM_IBITS)
3052                 RETURN(0);
3053
3054         /* FIXME: if we ever get into a situation where there are too many
3055          * opened files with open locks on a single node, then we really
3056          * should replay these open locks to reget it */
3057         if (lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_OPEN)
3058                 RETURN(0);
3059
3060         RETURN(1);
3061 }
3062
3063 static int mdc_resource_inode_free(struct ldlm_resource *res)
3064 {
3065         if (res->lr_lvb_inode)
3066                 res->lr_lvb_inode = NULL;
3067
3068         return 0;
3069 }
3070
3071 struct ldlm_valblock_ops inode_lvbo = {
3072         lvbo_free: mdc_resource_inode_free
3073 };
3074
3075 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
3076 {
3077         struct client_obd               *cli = &obd->u.cli;
3078         int                             rc;
3079         ENTRY;
3080
3081         OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
3082         if (!cli->cl_rpc_lock)
3083                 RETURN(-ENOMEM);
3084         mdc_init_rpc_lock(cli->cl_rpc_lock);
3085
3086         rc = ptlrpcd_addref();
3087         if (rc < 0)
3088                 GOTO(err_rpc_lock, rc);
3089
3090         OBD_ALLOC(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
3091         if (!cli->cl_close_lock)
3092                 GOTO(err_ptlrpcd_decref, rc = -ENOMEM);
3093         mdc_init_rpc_lock(cli->cl_close_lock);
3094
3095         rc = client_obd_setup(obd, cfg);
3096         if (rc)
3097                 GOTO(err_close_lock, rc);
3098 #ifdef LPROCFS
3099         obd->obd_vars = lprocfs_mdc_obd_vars;
3100         lprocfs_seq_obd_setup(obd);
3101         lprocfs_alloc_md_stats(obd, 0);
3102 #endif
3103         sptlrpc_lprocfs_cliobd_attach(obd);
3104         ptlrpc_lprocfs_register_obd(obd);
3105
3106         ns_register_cancel(obd->obd_namespace, mdc_cancel_for_recovery);
3107
3108         obd->obd_namespace->ns_lvbo = &inode_lvbo;
3109
3110         rc = obd_llog_init(obd, &obd->obd_olg, obd, NULL);
3111         if (rc) {
3112                 mdc_cleanup(obd);
3113                 CERROR("failed to setup llogging subsystems\n");
3114         }
3115
3116         RETURN(rc);
3117
3118 err_close_lock:
3119         OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
3120 err_ptlrpcd_decref:
3121         ptlrpcd_decref();
3122 err_rpc_lock:
3123         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
3124         RETURN(rc);
3125 }
3126
3127 /* Initialize the default and maximum LOV EA and cookie sizes.  This allows
3128  * us to make MDS RPCs with large enough reply buffers to hold the
3129  * maximum-sized (= maximum striped) EA and cookie without having to
3130  * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
3131 static int mdc_init_ea_size(struct obd_export *exp, int easize,
3132                      int def_easize, int cookiesize)
3133 {
3134         struct obd_device *obd = exp->exp_obd;
3135         struct client_obd *cli = &obd->u.cli;
3136         ENTRY;
3137
3138         if (cli->cl_max_mds_easize < easize)
3139                 cli->cl_max_mds_easize = easize;
3140
3141         if (cli->cl_default_mds_easize < def_easize)
3142                 cli->cl_default_mds_easize = def_easize;
3143
3144         if (cli->cl_max_mds_cookiesize < cookiesize)
3145                 cli->cl_max_mds_cookiesize = cookiesize;
3146
3147         RETURN(0);
3148 }
3149
3150 static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
3151 {
3152         int rc = 0;
3153         ENTRY;
3154
3155         switch (stage) {
3156         case OBD_CLEANUP_EARLY:
3157                 break;
3158         case OBD_CLEANUP_EXPORTS:
3159                 /* Failsafe, ok if racy */
3160                 if (obd->obd_type->typ_refcnt <= 1)
3161                         libcfs_kkuc_group_rem(0, KUC_GRP_HSM, NULL);
3162
3163                 obd_cleanup_client_import(obd);
3164                 ptlrpc_lprocfs_unregister_obd(obd);
3165                 lprocfs_obd_cleanup(obd);
3166                 lprocfs_free_md_stats(obd);
3167
3168                 rc = obd_llog_finish(obd, 0);
3169                 if (rc != 0)
3170                         CERROR("failed to cleanup llogging subsystems\n");
3171                 break;
3172         }
3173         RETURN(rc);
3174 }
3175
3176 static int mdc_cleanup(struct obd_device *obd)
3177 {
3178         struct client_obd *cli = &obd->u.cli;
3179
3180         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
3181         OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
3182
3183         ptlrpcd_decref();
3184
3185         return client_obd_cleanup(obd);
3186 }
3187
3188
3189 static int mdc_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
3190                          struct obd_device *tgt, int *index)
3191 {
3192         struct llog_ctxt        *ctxt;
3193         int                      rc;
3194
3195         ENTRY;
3196
3197         LASSERT(olg == &obd->obd_olg);
3198
3199         rc = llog_setup(NULL, obd, olg, LLOG_CHANGELOG_REPL_CTXT, tgt,
3200                         &llog_client_ops);
3201         if (rc)
3202                 RETURN(rc);
3203
3204         ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
3205         llog_initiator_connect(ctxt);
3206         llog_ctxt_put(ctxt);
3207
3208         RETURN(0);
3209 }
3210
3211 static int mdc_llog_finish(struct obd_device *obd, int count)
3212 {
3213         struct llog_ctxt *ctxt;
3214
3215         ENTRY;
3216
3217         ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
3218         if (ctxt)
3219                 llog_cleanup(NULL, ctxt);
3220
3221         RETURN(0);
3222 }
3223
3224 static int mdc_process_config(struct obd_device *obd, obd_count len, void *buf)
3225 {
3226         struct lustre_cfg *lcfg = buf;
3227         int rc = class_process_proc_seq_param(PARAM_MDC, obd->obd_vars,
3228                                               lcfg, obd);
3229         return (rc > 0 ? 0: rc);
3230 }
3231
3232
3233 /* get remote permission for current user on fid */
3234 int mdc_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
3235                         struct obd_capa *oc, __u32 suppgid,
3236                         struct ptlrpc_request **request)
3237 {
3238         struct ptlrpc_request  *req;
3239         int                    rc;
3240         ENTRY;
3241
3242         LASSERT(client_is_remote(exp));
3243
3244         *request = NULL;
3245         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
3246         if (req == NULL)
3247                 RETURN(-ENOMEM);
3248
3249         mdc_set_capa_size(req, &RMF_CAPA1, oc);
3250
3251         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
3252         if (rc) {
3253                 ptlrpc_request_free(req);
3254                 RETURN(rc);
3255         }
3256
3257         mdc_pack_body(req, fid, oc, OBD_MD_FLRMTPERM, 0, suppgid, 0);
3258
3259         req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
3260                              sizeof(struct mdt_remote_perm));
3261
3262         ptlrpc_request_set_replen(req);
3263
3264         rc = ptlrpc_queue_wait(req);
3265         if (rc)
3266                 ptlrpc_req_finished(req);
3267         else
3268                 *request = req;
3269         RETURN(rc);
3270 }
3271
3272 static int mdc_interpret_renew_capa(const struct lu_env *env,
3273                                     struct ptlrpc_request *req, void *args,
3274                                     int status)
3275 {
3276         struct mdc_renew_capa_args *ra = args;
3277         struct mdt_body *body = NULL;
3278         struct lustre_capa *capa;
3279         ENTRY;
3280
3281         if (status)
3282                 GOTO(out, capa = ERR_PTR(status));
3283
3284         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
3285         if (body == NULL)
3286                 GOTO(out, capa = ERR_PTR(-EFAULT));
3287
3288         if ((body->valid & OBD_MD_FLOSSCAPA) == 0)
3289                 GOTO(out, capa = ERR_PTR(-ENOENT));
3290
3291         capa = req_capsule_server_get(&req->rq_pill, &RMF_CAPA2);
3292         if (!capa)
3293                 GOTO(out, capa = ERR_PTR(-EFAULT));
3294         EXIT;
3295 out:
3296         ra->ra_cb(ra->ra_oc, capa);
3297         return 0;
3298 }
3299
3300 static int mdc_renew_capa(struct obd_export *exp, struct obd_capa *oc,
3301                           renew_capa_cb_t cb)
3302 {
3303         struct ptlrpc_request *req;
3304         struct mdc_renew_capa_args *ra;
3305         ENTRY;
3306
3307         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_GETATTR,
3308                                         LUSTRE_MDS_VERSION, MDS_GETATTR);
3309         if (req == NULL)
3310                 RETURN(-ENOMEM);
3311
3312         /* NB, OBD_MD_FLOSSCAPA is set here, but it doesn't necessarily mean the
3313          * capa to renew is oss capa.
3314          */
3315         mdc_pack_body(req, &oc->c_capa.lc_fid, oc, OBD_MD_FLOSSCAPA, 0, -1, 0);
3316         ptlrpc_request_set_replen(req);
3317
3318         CLASSERT(sizeof(*ra) <= sizeof(req->rq_async_args));
3319         ra = ptlrpc_req_async_args(req);
3320         ra->ra_oc = oc;
3321         ra->ra_cb = cb;
3322         req->rq_interpret_reply = mdc_interpret_renew_capa;
3323         ptlrpcd_add_req(req, PDL_POLICY_LOCAL, -1);
3324         RETURN(0);
3325 }
3326
3327 struct obd_ops mdc_obd_ops = {
3328         .o_owner            = THIS_MODULE,
3329         .o_setup            = mdc_setup,
3330         .o_precleanup       = mdc_precleanup,
3331         .o_cleanup          = mdc_cleanup,
3332         .o_add_conn         = client_import_add_conn,
3333         .o_del_conn         = client_import_del_conn,
3334         .o_connect          = client_connect_import,
3335         .o_disconnect       = client_disconnect_export,
3336         .o_iocontrol        = mdc_iocontrol,
3337         .o_set_info_async   = mdc_set_info_async,
3338         .o_statfs           = mdc_statfs,
3339         .o_pin              = mdc_pin,
3340         .o_unpin            = mdc_unpin,
3341         .o_fid_init         = client_fid_init,
3342         .o_fid_fini         = client_fid_fini,
3343         .o_fid_alloc        = mdc_fid_alloc,
3344         .o_import_event     = mdc_import_event,
3345         .o_llog_init        = mdc_llog_init,
3346         .o_llog_finish      = mdc_llog_finish,
3347         .o_get_info         = mdc_get_info,
3348         .o_process_config   = mdc_process_config,
3349         .o_get_uuid         = mdc_get_uuid,
3350         .o_quotactl         = mdc_quotactl,
3351         .o_quotacheck       = mdc_quotacheck
3352 };
3353
3354 struct md_ops mdc_md_ops = {
3355         .m_getstatus        = mdc_getstatus,
3356         .m_null_inode       = mdc_null_inode,
3357         .m_find_cbdata      = mdc_find_cbdata,
3358         .m_close            = mdc_close,
3359         .m_create           = mdc_create,
3360         .m_done_writing     = mdc_done_writing,
3361         .m_enqueue          = mdc_enqueue,
3362         .m_getattr          = mdc_getattr,
3363         .m_getattr_name     = mdc_getattr_name,
3364         .m_intent_lock      = mdc_intent_lock,
3365         .m_link             = mdc_link,
3366         .m_is_subdir        = mdc_is_subdir,
3367         .m_rename           = mdc_rename,
3368         .m_setattr          = mdc_setattr,
3369         .m_setxattr         = mdc_setxattr,
3370         .m_getxattr         = mdc_getxattr,
3371         .m_fsync                = mdc_fsync,
3372         .m_read_entry           = mdc_read_entry,
3373         .m_unlink           = mdc_unlink,
3374         .m_cancel_unused    = mdc_cancel_unused,
3375         .m_init_ea_size     = mdc_init_ea_size,
3376         .m_set_lock_data    = mdc_set_lock_data,
3377         .m_lock_match       = mdc_lock_match,
3378         .m_get_lustre_md    = mdc_get_lustre_md,
3379         .m_free_lustre_md   = mdc_free_lustre_md,
3380         .m_set_open_replay_data = mdc_set_open_replay_data,
3381         .m_clear_open_replay_data = mdc_clear_open_replay_data,
3382         .m_renew_capa       = mdc_renew_capa,
3383         .m_unpack_capa      = mdc_unpack_capa,
3384         .m_get_remote_perm  = mdc_get_remote_perm,
3385         .m_intent_getattr_async = mdc_intent_getattr_async,
3386         .m_revalidate_lock      = mdc_revalidate_lock
3387 };
3388
3389 int __init mdc_init(void)
3390 {
3391         return class_register_type(&mdc_obd_ops, &mdc_md_ops, NULL,
3392 #ifndef HAVE_ONLY_PROCFS_SEQ
3393                                         NULL,
3394 #endif
3395                                         LUSTRE_MDC_NAME, NULL);
3396 }
3397
3398 #ifdef __KERNEL__
3399 static void /*__exit*/ mdc_exit(void)
3400 {
3401         class_unregister_type(LUSTRE_MDC_NAME);
3402 }
3403
3404 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3405 MODULE_DESCRIPTION("Lustre Metadata Client");
3406 MODULE_LICENSE("GPL");
3407
3408 module_init(mdc_init);
3409 module_exit(mdc_exit);
3410 #endif