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         kunmap(page);
1189         if (remove) {
1190                 lock_page(page);
1191                 if (likely(page->mapping != NULL))
1192                         truncate_complete_page(page->mapping, page);
1193                 unlock_page(page);
1194         }
1195         page_cache_release(page);
1196 }
1197
1198 static struct page *mdc_page_locate(struct address_space *mapping, __u64 *hash,
1199                                     __u64 *start, __u64 *end, int hash64)
1200 {
1201         /*
1202          * Complement of hash is used as an index so that
1203          * radix_tree_gang_lookup() can be used to find a page with starting
1204          * hash _smaller_ than one we are looking for.
1205          */
1206         unsigned long offset = hash_x_index(*hash, hash64);
1207         struct page *page;
1208         int found;
1209
1210         spin_lock_irq(&mapping->tree_lock);
1211         found = radix_tree_gang_lookup(&mapping->page_tree,
1212                                        (void **)&page, offset, 1);
1213         if (found > 0) {
1214                 struct lu_dirpage *dp;
1215
1216                 page_cache_get(page);
1217                 spin_unlock_irq(&mapping->tree_lock);
1218                 /*
1219                  * In contrast to find_lock_page() we are sure that directory
1220                  * page cannot be truncated (while DLM lock is held) and,
1221                  * hence, can avoid restart.
1222                  *
1223                  * In fact, page cannot be locked here at all, because
1224                  * mdc_read_page_remote does synchronous io.
1225                  */
1226                 wait_on_page_locked(page);
1227                 if (PageUptodate(page)) {
1228                         dp = kmap(page);
1229                         if (BITS_PER_LONG == 32 && hash64) {
1230                                 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1231                                 *end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
1232                                 *hash  = *hash >> 32;
1233                         } else {
1234                                 *start = le64_to_cpu(dp->ldp_hash_start);
1235                                 *end   = le64_to_cpu(dp->ldp_hash_end);
1236                         }
1237                         LASSERTF(*start <= *hash, "start = "LPX64",end = "
1238                                  LPX64",hash = "LPX64"\n", *start, *end, *hash);
1239                         CDEBUG(D_VFSTRACE, "page%lu [%llu %llu], hash"LPU64"\n",
1240                                offset, *start, *end, *hash);
1241                         if (*hash > *end) {
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                                 mdc_release_page(page,
1252                                     le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
1253                                 page = NULL;
1254                         }
1255                 } else {
1256                         page_cache_release(page);
1257                         page = ERR_PTR(-EIO);
1258                 }
1259         } else {
1260                 spin_unlock_irq(&mapping->tree_lock);
1261                 page = NULL;
1262         }
1263         return page;
1264 }
1265
1266 /*
1267  * Adjust a set of pages, each page containing an array of lu_dirpages,
1268  * so that each page can be used as a single logical lu_dirpage.
1269  *
1270  * A lu_dirpage is laid out as follows, where s = ldp_hash_start,
1271  * e = ldp_hash_end, f = ldp_flags, p = padding, and each "ent" is a
1272  * struct lu_dirent.  It has size up to LU_PAGE_SIZE. The ldp_hash_end
1273  * value is used as a cookie to request the next lu_dirpage in a
1274  * directory listing that spans multiple pages (two in this example):
1275  *   ________
1276  *  |        |
1277  * .|--------v-------   -----.
1278  * |s|e|f|p|ent|ent| ... |ent|
1279  * '--|--------------   -----'   Each CFS_PAGE contains a single
1280  *    '------.                   lu_dirpage.
1281  * .---------v-------   -----.
1282  * |s|e|f|p|ent| 0 | ... | 0 |
1283  * '-----------------   -----'
1284  *
1285  * However, on hosts where the native VM page size (PAGE_CACHE_SIZE) is
1286  * larger than LU_PAGE_SIZE, a single host page may contain multiple
1287  * lu_dirpages. After reading the lu_dirpages from the MDS, the
1288  * ldp_hash_end of the first lu_dirpage refers to the one immediately
1289  * after it in the same CFS_PAGE (arrows simplified for brevity, but
1290  * in general e0==s1, e1==s2, etc.):
1291  *
1292  * .--------------------   -----.
1293  * |s0|e0|f0|p|ent|ent| ... |ent|
1294  * |---v----------------   -----|
1295  * |s1|e1|f1|p|ent|ent| ... |ent|
1296  * |---v----------------   -----|  Here, each CFS_PAGE contains
1297  *             ...                 multiple lu_dirpages.
1298  * |---v----------------   -----|
1299  * |s'|e'|f'|p|ent|ent| ... |ent|
1300  * '---|----------------   -----'
1301  *     v
1302  * .----------------------------.
1303  * |        next CFS_PAGE       |
1304  *
1305  * This structure is transformed into a single logical lu_dirpage as follows:
1306  *
1307  * - Replace e0 with e' so the request for the next lu_dirpage gets the page
1308  *   labeled 'next CFS_PAGE'.
1309  *
1310  * - Copy the LDF_COLLIDE flag from f' to f0 to correctly reflect whether
1311  *   a hash collision with the next page exists.
1312  *
1313  * - Adjust the lde_reclen of the ending entry of each lu_dirpage to span
1314  *   to the first entry of the next lu_dirpage.
1315  */
1316 #if PAGE_CACHE_SIZE > LU_PAGE_SIZE
1317 static void mdc_adjust_dirpages(struct page **pages, int cfs_pgs, int lu_pgs)
1318 {
1319         int i;
1320
1321         for (i = 0; i < cfs_pgs; i++) {
1322                 struct lu_dirpage       *dp = kmap(pages[i]);
1323                 struct lu_dirpage       *first = dp;
1324                 struct lu_dirent        *end_dirent = NULL;
1325                 struct lu_dirent        *ent;
1326                 __u64                   hash_end = dp->ldp_hash_end;
1327                 __u32                   flags = dp->ldp_flags;
1328
1329                 while (--lu_pgs > 0) {
1330                         ent = lu_dirent_start(dp);
1331                         for (end_dirent = ent; ent != NULL;
1332                              end_dirent = ent, ent = lu_dirent_next(ent));
1333
1334                         /* Advance dp to next lu_dirpage. */
1335                         dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE);
1336
1337                         /* Check if we've reached the end of the CFS_PAGE. */
1338                         if (!((unsigned long)dp & ~CFS_PAGE_MASK))
1339                                 break;
1340
1341                         /* Save the hash and flags of this lu_dirpage. */
1342                         hash_end = dp->ldp_hash_end;
1343                         flags = dp->ldp_flags;
1344
1345                         /* Check if lu_dirpage contains no entries. */
1346                         if (end_dirent == NULL)
1347                                 break;
1348
1349                         /* Enlarge the end entry lde_reclen from 0 to
1350                          * first entry of next lu_dirpage. */
1351                         LASSERT(le16_to_cpu(end_dirent->lde_reclen) == 0);
1352                         end_dirent->lde_reclen =
1353                                 cpu_to_le16((char *)(dp->ldp_entries) -
1354                                             (char *)end_dirent);
1355                 }
1356
1357                 first->ldp_hash_end = hash_end;
1358                 first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE);
1359                 first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE);
1360
1361                 kunmap(pages[i]);
1362         }
1363         LASSERTF(lu_pgs == 0, "left = %d", lu_pgs);
1364 }
1365 #else
1366 #define mdc_adjust_dirpages(pages, cfs_pgs, lu_pgs) do {} while (0)
1367 #endif  /* PAGE_CACHE_SIZE > LU_PAGE_SIZE */
1368
1369 /* parameters for readdir page */
1370 struct readpage_param {
1371         struct md_op_data       *rp_mod;
1372         __u64                   rp_off;
1373         int                     rp_hash64;
1374         struct obd_export       *rp_exp;
1375         struct md_callback      *rp_cb;
1376 };
1377
1378 /**
1379  * Read pages from server.
1380  *
1381  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
1382  * a header lu_dirpage which describes the start/end hash, and whether this
1383  * page is empty (contains no dir entry) or hash collide with next page.
1384  * After client receives reply, several pages will be integrated into dir page
1385  * in CFS_PAGE_SIZE (if CFS_PAGE_SIZE greater than LU_PAGE_SIZE), and the
1386  * lu_dirpage for this integrated page will be adjusted.
1387  **/
1388 static int mdc_read_page_remote(void *data, struct page *page0)
1389 {
1390         struct readpage_param   *rp = data;
1391         struct page             **page_pool;
1392         struct page             *page;
1393         struct lu_dirpage       *dp;
1394         int                     rd_pgs = 0; /* number of pages read actually */
1395         int                     npages;
1396         struct md_op_data       *op_data = rp->rp_mod;
1397         struct ptlrpc_request   *req;
1398         int                     max_pages = op_data->op_max_pages;
1399         struct inode            *inode;
1400         struct lu_fid           *fid;
1401         int                     i;
1402         int                     rc;
1403         ENTRY;
1404
1405         LASSERT(max_pages > 0 && max_pages <= PTLRPC_MAX_BRW_PAGES);
1406         if (op_data->op_mea1 != NULL) {
1407                 __u32 index = op_data->op_stripe_offset;
1408
1409                 inode = op_data->op_mea1->lsm_md_oinfo[index].lmo_root;
1410                 fid = &op_data->op_mea1->lsm_md_oinfo[index].lmo_fid;
1411         } else {
1412                 inode = op_data->op_data;
1413                 fid = &op_data->op_fid1;
1414         }
1415         LASSERT(inode != NULL);
1416
1417         OBD_ALLOC(page_pool, sizeof(page_pool[0]) * max_pages);
1418         if (page_pool != NULL) {
1419                 page_pool[0] = page0;
1420         } else {
1421                 page_pool = &page0;
1422                 max_pages = 1;
1423         }
1424
1425         for (npages = 1; npages < max_pages; npages++) {
1426                 page = page_cache_alloc_cold(inode->i_mapping);
1427                 if (page == NULL)
1428                         break;
1429                 page_pool[npages] = page;
1430         }
1431
1432         rc = mdc_getpage(rp->rp_exp, fid, rp->rp_off, op_data->op_capa1,
1433                          page_pool, npages, &req);
1434         if (rc == 0) {
1435                 int lu_pgs;
1436
1437                 rd_pgs = (req->rq_bulk->bd_nob_transferred +
1438                             PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1439                 lu_pgs = req->rq_bulk->bd_nob_transferred >>
1440                                                         LU_PAGE_SHIFT;
1441                 LASSERT(!(req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK));
1442
1443                 CDEBUG(D_INODE, "read %d(%d)/%d pages\n", rd_pgs, lu_pgs,
1444                        op_data->op_npages);
1445
1446                 mdc_adjust_dirpages(page_pool, rd_pgs, lu_pgs);
1447
1448                 SetPageUptodate(page0);
1449         }
1450
1451         unlock_page(page0);
1452         ptlrpc_req_finished(req);
1453         CDEBUG(D_CACHE, "read %d/%d pages\n", rd_pgs, npages);
1454         for (i = 1; i < npages; i++) {
1455                 unsigned long   offset;
1456                 __u64           hash;
1457                 int ret;
1458
1459                 page = page_pool[i];
1460
1461                 if (rc < 0 || i >= rd_pgs) {
1462                         page_cache_release(page);
1463                         continue;
1464                 }
1465
1466                 SetPageUptodate(page);
1467
1468                 dp = kmap(page);
1469                 hash = le64_to_cpu(dp->ldp_hash_start);
1470                 kunmap(page);
1471
1472                 offset = hash_x_index(hash, rp->rp_hash64);
1473
1474                 prefetchw(&page->flags);
1475                 ret = add_to_page_cache_lru(page, inode->i_mapping, offset,
1476                                             GFP_KERNEL);
1477                 if (ret == 0)
1478                         unlock_page(page);
1479                 else
1480                         CDEBUG(D_VFSTRACE, "page %lu add to page cache failed:"
1481                                " rc = %d\n", offset, ret);
1482                 page_cache_release(page);
1483         }
1484
1485         if (page_pool != &page0)
1486                 OBD_FREE(page_pool, sizeof(page_pool[0]) * max_pages);
1487
1488         RETURN(rc);
1489 }
1490
1491 /**
1492  * Read dir page from cache first, if it can not find it, read it from
1493  * server and add into the cache.
1494  */
1495 static int mdc_read_page(struct obd_export *exp, struct md_op_data *op_data,
1496                          struct md_callback *cb_op, struct page **ppage)
1497 {
1498         struct lookup_intent    it = { .it_op = IT_READDIR };
1499         struct page             *page;
1500         struct inode            *dir = NULL;
1501         struct address_space    *mapping;
1502         struct lu_dirpage       *dp;
1503         __u64                   start = 0;
1504         __u64                   end = 0;
1505         struct lustre_handle    lockh;
1506         struct ptlrpc_request   *enq_req = NULL;
1507         struct readpage_param   rp_param;
1508         int rc;
1509
1510         ENTRY;
1511
1512         *ppage = NULL;
1513
1514         if (op_data->op_mea1 != NULL) {
1515                 __u32 index = op_data->op_stripe_offset;
1516
1517                 dir = op_data->op_mea1->lsm_md_oinfo[index].lmo_root;
1518         } else {
1519                 dir = op_data->op_data;
1520         }
1521         LASSERT(dir != NULL);
1522
1523         mapping = dir->i_mapping;
1524
1525         rc = mdc_intent_lock(exp, op_data, NULL, 0, &it, 0, &enq_req,
1526                              cb_op->md_blocking_ast, 0);
1527         if (enq_req != NULL)
1528                 ptlrpc_req_finished(enq_req);
1529
1530         if (rc < 0) {
1531                 CERROR("%s: "DFID" lock enqueue fails: rc = %d\n",
1532                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1), rc);
1533                 RETURN(rc);
1534         }
1535
1536         rc = 0;
1537         mdc_set_lock_data(exp, &it.d.lustre.it_lock_handle, dir, NULL);
1538
1539         rp_param.rp_off = op_data->op_hash_offset;
1540         rp_param.rp_hash64 = op_data->op_cli_flags & CLI_HASH64;
1541         page = mdc_page_locate(mapping, &rp_param.rp_off, &start, &end,
1542                                rp_param.rp_hash64);
1543         if (IS_ERR(page)) {
1544                 CERROR("%s: dir page locate: "DFID" at "LPU64": rc %ld\n",
1545                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1546                        rp_param.rp_off, PTR_ERR(page));
1547                 GOTO(out_unlock, rc = PTR_ERR(page));
1548         } else if (page != NULL) {
1549                 /*
1550                  * XXX nikita: not entirely correct handling of a corner case:
1551                  * suppose hash chain of entries with hash value HASH crosses
1552                  * border between pages P0 and P1. First both P0 and P1 are
1553                  * cached, seekdir() is called for some entry from the P0 part
1554                  * of the chain. Later P0 goes out of cache. telldir(HASH)
1555                  * happens and finds P1, as it starts with matching hash
1556                  * value. Remaining entries from P0 part of the chain are
1557                  * skipped. (Is that really a bug?)
1558                  *
1559                  * Possible solutions: 0. don't cache P1 is such case, handle
1560                  * it as an "overflow" page. 1. invalidate all pages at
1561                  * once. 2. use HASH|1 as an index for P1.
1562                  */
1563                 GOTO(hash_collision, page);
1564         }
1565
1566         rp_param.rp_exp = exp;
1567         rp_param.rp_mod = op_data;
1568         page = read_cache_page(mapping,
1569                                hash_x_index(rp_param.rp_off,
1570                                             rp_param.rp_hash64),
1571                                mdc_read_page_remote, &rp_param);
1572         if (IS_ERR(page)) {
1573                 CERROR("%s: read cache page: "DFID" at "LPU64": rc %ld\n",
1574                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1575                        rp_param.rp_off, PTR_ERR(page));
1576                 GOTO(out_unlock, rc = PTR_ERR(page));
1577         }
1578
1579         wait_on_page_locked(page);
1580         (void)kmap(page);
1581         if (!PageUptodate(page)) {
1582                 CERROR("%s: page not updated: "DFID" at "LPU64": rc %d\n",
1583                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1584                        rp_param.rp_off, -5);
1585                 goto fail;
1586         }
1587         if (!PageChecked(page))
1588                 SetPageChecked(page);
1589         if (PageError(page)) {
1590                 CERROR("%s: page error: "DFID" at "LPU64": rc %d\n",
1591                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1592                        rp_param.rp_off, -5);
1593                 goto fail;
1594         }
1595
1596 hash_collision:
1597         dp = page_address(page);
1598         if (BITS_PER_LONG == 32 && rp_param.rp_hash64) {
1599                 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1600                 end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
1601                 rp_param.rp_off = op_data->op_hash_offset >> 32;
1602         } else {
1603                 start = le64_to_cpu(dp->ldp_hash_start);
1604                 end   = le64_to_cpu(dp->ldp_hash_end);
1605                 rp_param.rp_off = op_data->op_hash_offset;
1606         }
1607         if (end == start) {
1608                 LASSERT(start == rp_param.rp_off);
1609                 CWARN("Page-wide hash collision: %#lx\n", (unsigned long)end);
1610 #if BITS_PER_LONG == 32
1611                 CWARN("Real page-wide hash collision at ["LPU64" "LPU64"] with "
1612                       "hash "LPU64"\n", le64_to_cpu(dp->ldp_hash_start),
1613                       le64_to_cpu(dp->ldp_hash_end), op_data->op_hash_offset);
1614 #endif
1615
1616                 /*
1617                  * Fetch whole overflow chain...
1618                  *
1619                  * XXX not yet.
1620                  */
1621                 goto fail;
1622         }
1623         *ppage = page;
1624 out_unlock:
1625         lockh.cookie = it.d.lustre.it_lock_handle;
1626         ldlm_lock_decref(&lockh, it.d.lustre.it_lock_mode);
1627         it.d.lustre.it_lock_handle = 0;
1628         return rc;
1629 fail:
1630         kunmap(page);
1631         mdc_release_page(page, 1);
1632         rc = -EIO;
1633         goto out_unlock;
1634 }
1635
1636 /**
1637  * Read one directory entry from the cache.
1638  */
1639 int mdc_read_entry(struct obd_export *exp, struct md_op_data *op_data,
1640                    struct md_callback *cb_op, struct lu_dirent **entp)
1641 {
1642         struct page             *page = NULL;
1643         struct lu_dirpage       *dp;
1644         struct lu_dirent        *ent;
1645         int                     rc = 0;
1646         int                     index = 0;
1647         ENTRY;
1648
1649         if (op_data->op_hash_offset == MDS_DIR_END_OFF) {
1650                 *entp = NULL;
1651                 RETURN(0);
1652         }
1653
1654         rc = mdc_read_page(exp, op_data, cb_op, &page);
1655         if (rc != 0)
1656                 RETURN(rc);
1657
1658         if (op_data->op_cli_flags & CLI_READENT_END) {
1659                 mdc_release_page(page, 0);
1660                 RETURN(0);
1661         }
1662
1663         dp = kmap(page);
1664         for (ent = lu_dirent_start(dp); ent != NULL;
1665              ent = lu_dirent_next(ent)) {
1666                 index++;
1667                 if (ent->lde_hash > op_data->op_hash_offset)
1668                         break;
1669         }
1670         kunmap(page);
1671
1672         /* If it can not find entry in current page, try next page. */
1673         if (ent == NULL) {
1674                 __u64 orig_offset = op_data->op_hash_offset;
1675
1676                 if (dp->ldp_hash_end == MDS_DIR_END_OFF) {
1677                         mdc_release_page(page, 0);
1678                         RETURN(0);
1679                 }
1680
1681                 op_data->op_hash_offset = dp->ldp_hash_end;
1682                 mdc_release_page(page,
1683                                  le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
1684                 rc = mdc_read_page(exp, op_data, cb_op, &page);
1685                 if (rc != 0)
1686                         RETURN(rc);
1687
1688                 if (page != NULL) {
1689                         dp = kmap(page);
1690                         ent = lu_dirent_start(dp);
1691                         kunmap(page);
1692                 }
1693
1694                 op_data->op_hash_offset = orig_offset;
1695         }
1696
1697         *entp = ent;
1698
1699         RETURN(rc);
1700 }
1701
1702 #else /* __KERNEL__ */
1703
1704 static struct page
1705 *mdc_read_page_remote(struct obd_export *exp, const struct lmv_oinfo *lmo,
1706                       const __u64 hash, struct obd_capa *oc)
1707 {
1708         struct ptlrpc_request *req = NULL;
1709         struct page *page;
1710         int rc;
1711
1712         OBD_PAGE_ALLOC(page, 0);
1713         if (page == NULL)
1714                 return ERR_PTR(-ENOMEM);
1715
1716         rc = mdc_getpage(exp, &lmo->lmo_fid, hash, oc, &page, 1, &req);
1717         if (req != NULL)
1718                 ptlrpc_req_finished(req);
1719
1720         if (unlikely(rc)) {
1721                 OBD_PAGE_FREE(page);
1722                 return ERR_PTR(rc);
1723         }
1724         return page;
1725 }
1726
1727
1728 static int mdc_read_page(struct obd_export *exp, struct md_op_data *op_data,
1729                         struct md_callback *cb_op,
1730                         struct page **ppage)
1731 {
1732         struct page *page;
1733         struct lmv_oinfo *lmo;
1734         int rc = 0;
1735
1736         /* No local cache for liblustre, always read entry remotely */
1737         lmo = &op_data->op_mea1->lsm_md_oinfo[op_data->op_stripe_offset];
1738         page = mdc_read_page_remote(exp, lmo, op_data->op_hash_offset,
1739                                     op_data->op_capa1);
1740         if (IS_ERR(page))
1741                 return PTR_ERR(page);
1742
1743         *ppage = page;
1744
1745         return rc;
1746 }
1747
1748 int mdc_read_entry(struct obd_export *exp, struct md_op_data *op_data,
1749                    struct md_callback *cb_op, struct lu_dirent **entp)
1750 {
1751         struct page             *page = NULL;
1752         struct lu_dirpage       *dp;
1753         struct lu_dirent        *ent;
1754         int                     rc;
1755         ENTRY;
1756
1757         rc = mdc_read_page(exp, op_data, cb_op, &page);
1758         if (rc != 0)
1759                 RETURN(rc);
1760
1761         dp = page_address(page);
1762         if (dp->ldp_hash_end < op_data->op_hash_offset)
1763                 GOTO(out, *entp = NULL);
1764
1765         for (ent = lu_dirent_start(dp); ent != NULL;
1766              ent = lu_dirent_next(ent))
1767                 if (ent->lde_hash >= op_data->op_hash_offset)
1768                         break;
1769         *entp = ent;
1770 out:
1771
1772         OBD_PAGE_FREE(page);
1773         RETURN(rc);
1774 }
1775
1776 #endif
1777
1778 static int mdc_statfs(const struct lu_env *env,
1779                       struct obd_export *exp, struct obd_statfs *osfs,
1780                       __u64 max_age, __u32 flags)
1781 {
1782         struct obd_device     *obd = class_exp2obd(exp);
1783         struct ptlrpc_request *req;
1784         struct obd_statfs     *msfs;
1785         struct obd_import     *imp = NULL;
1786         int                    rc;
1787         ENTRY;
1788
1789         /*
1790          * Since the request might also come from lprocfs, so we need
1791          * sync this with client_disconnect_export Bug15684
1792          */
1793         down_read(&obd->u.cli.cl_sem);
1794         if (obd->u.cli.cl_import)
1795                 imp = class_import_get(obd->u.cli.cl_import);
1796         up_read(&obd->u.cli.cl_sem);
1797         if (!imp)
1798                 RETURN(-ENODEV);
1799
1800         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_STATFS,
1801                                         LUSTRE_MDS_VERSION, MDS_STATFS);
1802         if (req == NULL)
1803                 GOTO(output, rc = -ENOMEM);
1804
1805         ptlrpc_request_set_replen(req);
1806
1807         if (flags & OBD_STATFS_NODELAY) {
1808                 /* procfs requests not want stay in wait for avoid deadlock */
1809                 req->rq_no_resend = 1;
1810                 req->rq_no_delay = 1;
1811         }
1812
1813         rc = ptlrpc_queue_wait(req);
1814         if (rc) {
1815                 /* check connection error first */
1816                 if (imp->imp_connect_error)
1817                         rc = imp->imp_connect_error;
1818                 GOTO(out, rc);
1819         }
1820
1821         msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1822         if (msfs == NULL)
1823                 GOTO(out, rc = -EPROTO);
1824
1825         *osfs = *msfs;
1826         EXIT;
1827 out:
1828         ptlrpc_req_finished(req);
1829 output:
1830         class_import_put(imp);
1831         return rc;
1832 }
1833
1834 static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
1835 {
1836         __u32 keylen, vallen;
1837         void *key;
1838         int rc;
1839
1840         if (gf->gf_pathlen > PATH_MAX)
1841                 RETURN(-ENAMETOOLONG);
1842         if (gf->gf_pathlen < 2)
1843                 RETURN(-EOVERFLOW);
1844
1845         /* Key is KEY_FID2PATH + getinfo_fid2path description */
1846         keylen = cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf);
1847         OBD_ALLOC(key, keylen);
1848         if (key == NULL)
1849                 RETURN(-ENOMEM);
1850         memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
1851         memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf));
1852
1853         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
1854                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
1855
1856         if (!fid_is_sane(&gf->gf_fid))
1857                 GOTO(out, rc = -EINVAL);
1858
1859         /* Val is struct getinfo_fid2path result plus path */
1860         vallen = sizeof(*gf) + gf->gf_pathlen;
1861
1862         rc = obd_get_info(NULL, exp, keylen, key, &vallen, gf, NULL);
1863         if (rc != 0 && rc != -EREMOTE)
1864                 GOTO(out, rc);
1865
1866         if (vallen <= sizeof(*gf))
1867                 GOTO(out, rc = -EPROTO);
1868         else if (vallen > sizeof(*gf) + gf->gf_pathlen)
1869                 GOTO(out, rc = -EOVERFLOW);
1870
1871         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n%s\n",
1872                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno, gf->gf_path);
1873
1874 out:
1875         OBD_FREE(key, keylen);
1876         return rc;
1877 }
1878
1879 static int mdc_ioc_hsm_progress(struct obd_export *exp,
1880                                 struct hsm_progress_kernel *hpk)
1881 {
1882         struct obd_import               *imp = class_exp2cliimp(exp);
1883         struct hsm_progress_kernel      *req_hpk;
1884         struct ptlrpc_request           *req;
1885         int                              rc;
1886         ENTRY;
1887
1888         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_PROGRESS,
1889                                         LUSTRE_MDS_VERSION, MDS_HSM_PROGRESS);
1890         if (req == NULL)
1891                 GOTO(out, rc = -ENOMEM);
1892
1893         mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
1894
1895         /* Copy hsm_progress struct */
1896         req_hpk = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_PROGRESS);
1897         if (req_hpk == NULL)
1898                 GOTO(out, rc = -EPROTO);
1899
1900         *req_hpk = *hpk;
1901         req_hpk->hpk_errval = lustre_errno_hton(hpk->hpk_errval);
1902
1903         ptlrpc_request_set_replen(req);
1904
1905         rc = mdc_queue_wait(req);
1906         GOTO(out, rc);
1907 out:
1908         ptlrpc_req_finished(req);
1909         return rc;
1910 }
1911
1912 static int mdc_ioc_hsm_ct_register(struct obd_import *imp, __u32 archives)
1913 {
1914         __u32                   *archive_mask;
1915         struct ptlrpc_request   *req;
1916         int                      rc;
1917         ENTRY;
1918
1919         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_REGISTER,
1920                                         LUSTRE_MDS_VERSION,
1921                                         MDS_HSM_CT_REGISTER);
1922         if (req == NULL)
1923                 GOTO(out, rc = -ENOMEM);
1924
1925         mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
1926
1927         /* Copy hsm_progress struct */
1928         archive_mask = req_capsule_client_get(&req->rq_pill,
1929                                               &RMF_MDS_HSM_ARCHIVE);
1930         if (archive_mask == NULL)
1931                 GOTO(out, rc = -EPROTO);
1932
1933         *archive_mask = archives;
1934
1935         ptlrpc_request_set_replen(req);
1936
1937         rc = mdc_queue_wait(req);
1938         GOTO(out, rc);
1939 out:
1940         ptlrpc_req_finished(req);
1941         return rc;
1942 }
1943
1944 static int mdc_ioc_hsm_current_action(struct obd_export *exp,
1945                                       struct md_op_data *op_data)
1946 {
1947         struct hsm_current_action       *hca = op_data->op_data;
1948         struct hsm_current_action       *req_hca;
1949         struct ptlrpc_request           *req;
1950         int                              rc;
1951         ENTRY;
1952
1953         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1954                                    &RQF_MDS_HSM_ACTION);
1955         if (req == NULL)
1956                 RETURN(-ENOMEM);
1957
1958         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1959
1960         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_ACTION);
1961         if (rc) {
1962                 ptlrpc_request_free(req);
1963                 RETURN(rc);
1964         }
1965
1966         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
1967                       OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
1968
1969         ptlrpc_request_set_replen(req);
1970
1971         rc = mdc_queue_wait(req);
1972         if (rc)
1973                 GOTO(out, rc);
1974
1975         req_hca = req_capsule_server_get(&req->rq_pill,
1976                                          &RMF_MDS_HSM_CURRENT_ACTION);
1977         if (req_hca == NULL)
1978                 GOTO(out, rc = -EPROTO);
1979
1980         *hca = *req_hca;
1981
1982         EXIT;
1983 out:
1984         ptlrpc_req_finished(req);
1985         return rc;
1986 }
1987
1988 static int mdc_ioc_hsm_ct_unregister(struct obd_import *imp)
1989 {
1990         struct ptlrpc_request   *req;
1991         int                      rc;
1992         ENTRY;
1993
1994         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_UNREGISTER,
1995                                         LUSTRE_MDS_VERSION,
1996                                         MDS_HSM_CT_UNREGISTER);
1997         if (req == NULL)
1998                 GOTO(out, rc = -ENOMEM);
1999
2000         mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
2001
2002         ptlrpc_request_set_replen(req);
2003
2004         rc = mdc_queue_wait(req);
2005         GOTO(out, rc);
2006 out:
2007         ptlrpc_req_finished(req);
2008         return rc;
2009 }
2010
2011 static int mdc_ioc_hsm_state_get(struct obd_export *exp,
2012                                  struct md_op_data *op_data)
2013 {
2014         struct hsm_user_state   *hus = op_data->op_data;
2015         struct hsm_user_state   *req_hus;
2016         struct ptlrpc_request   *req;
2017         int                      rc;
2018         ENTRY;
2019
2020         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
2021                                    &RQF_MDS_HSM_STATE_GET);
2022         if (req == NULL)
2023                 RETURN(-ENOMEM);
2024
2025         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
2026
2027         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_GET);
2028         if (rc != 0) {
2029                 ptlrpc_request_free(req);
2030                 RETURN(rc);
2031         }
2032
2033         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
2034                       OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
2035
2036         ptlrpc_request_set_replen(req);
2037
2038         rc = mdc_queue_wait(req);
2039         if (rc)
2040                 GOTO(out, rc);
2041
2042         req_hus = req_capsule_server_get(&req->rq_pill, &RMF_HSM_USER_STATE);
2043         if (req_hus == NULL)
2044                 GOTO(out, rc = -EPROTO);
2045
2046         *hus = *req_hus;
2047
2048         EXIT;
2049 out:
2050         ptlrpc_req_finished(req);
2051         return rc;
2052 }
2053
2054 static int mdc_ioc_hsm_state_set(struct obd_export *exp,
2055                                  struct md_op_data *op_data)
2056 {
2057         struct hsm_state_set    *hss = op_data->op_data;
2058         struct hsm_state_set    *req_hss;
2059         struct ptlrpc_request   *req;
2060         int                      rc;
2061         ENTRY;
2062
2063         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
2064                                    &RQF_MDS_HSM_STATE_SET);
2065         if (req == NULL)
2066                 RETURN(-ENOMEM);
2067
2068         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
2069
2070         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_SET);
2071         if (rc) {
2072                 ptlrpc_request_free(req);
2073                 RETURN(rc);
2074         }
2075
2076         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
2077                       OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
2078
2079         /* Copy states */
2080         req_hss = req_capsule_client_get(&req->rq_pill, &RMF_HSM_STATE_SET);
2081         if (req_hss == NULL)
2082                 GOTO(out, rc = -EPROTO);
2083         *req_hss = *hss;
2084
2085         ptlrpc_request_set_replen(req);
2086
2087         rc = mdc_queue_wait(req);
2088         GOTO(out, rc);
2089
2090         EXIT;
2091 out:
2092         ptlrpc_req_finished(req);
2093         return rc;
2094 }
2095
2096 static int mdc_ioc_hsm_request(struct obd_export *exp,
2097                                struct hsm_user_request *hur)
2098 {
2099         struct obd_import       *imp = class_exp2cliimp(exp);
2100         struct ptlrpc_request   *req;
2101         struct hsm_request      *req_hr;
2102         struct hsm_user_item    *req_hui;
2103         char                    *req_opaque;
2104         int                      rc;
2105         ENTRY;
2106
2107         req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_REQUEST);
2108         if (req == NULL)
2109                 GOTO(out, rc = -ENOMEM);
2110
2111         req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM, RCL_CLIENT,
2112                              hur->hur_request.hr_itemcount
2113                              * sizeof(struct hsm_user_item));
2114         req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA, RCL_CLIENT,
2115                              hur->hur_request.hr_data_len);
2116
2117         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_REQUEST);
2118         if (rc) {
2119                 ptlrpc_request_free(req);
2120                 RETURN(rc);
2121         }
2122
2123         mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
2124
2125         /* Copy hsm_request struct */
2126         req_hr = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_REQUEST);
2127         if (req_hr == NULL)
2128                 GOTO(out, rc = -EPROTO);
2129         *req_hr = hur->hur_request;
2130
2131         /* Copy hsm_user_item structs */
2132         req_hui = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM);
2133         if (req_hui == NULL)
2134                 GOTO(out, rc = -EPROTO);
2135         memcpy(req_hui, hur->hur_user_item,
2136                hur->hur_request.hr_itemcount * sizeof(struct hsm_user_item));
2137
2138         /* Copy opaque field */
2139         req_opaque = req_capsule_client_get(&req->rq_pill, &RMF_GENERIC_DATA);
2140         if (req_opaque == NULL)
2141                 GOTO(out, rc = -EPROTO);
2142         memcpy(req_opaque, hur_data(hur), hur->hur_request.hr_data_len);
2143
2144         ptlrpc_request_set_replen(req);
2145
2146         rc = mdc_queue_wait(req);
2147         GOTO(out, rc);
2148
2149 out:
2150         ptlrpc_req_finished(req);
2151         return rc;
2152 }
2153
2154 static struct kuc_hdr *changelog_kuc_hdr(char *buf, int len, int flags)
2155 {
2156         struct kuc_hdr *lh = (struct kuc_hdr *)buf;
2157
2158         LASSERT(len <= KUC_CHANGELOG_MSG_MAXSIZE);
2159
2160         lh->kuc_magic = KUC_MAGIC;
2161         lh->kuc_transport = KUC_TRANSPORT_CHANGELOG;
2162         lh->kuc_flags = flags;
2163         lh->kuc_msgtype = CL_RECORD;
2164         lh->kuc_msglen = len;
2165         return lh;
2166 }
2167
2168 #define D_CHANGELOG 0
2169
2170 struct changelog_show {
2171         __u64           cs_startrec;
2172         __u32           cs_flags;
2173         struct file     *cs_fp;
2174         char            *cs_buf;
2175         struct obd_device *cs_obd;
2176 };
2177
2178 static int changelog_kkuc_cb(const struct lu_env *env, struct llog_handle *llh,
2179                              struct llog_rec_hdr *hdr, void *data)
2180 {
2181         struct changelog_show *cs = data;
2182         struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
2183         struct kuc_hdr *lh;
2184         int len, rc;
2185         ENTRY;
2186
2187         if (rec->cr_hdr.lrh_type != CHANGELOG_REC) {
2188                 rc = -EINVAL;
2189                 CERROR("%s: not a changelog rec %x/%d: rc = %d\n",
2190                        cs->cs_obd->obd_name, rec->cr_hdr.lrh_type,
2191                        rec->cr.cr_type, rc);
2192                 RETURN(rc);
2193         }
2194
2195         if (rec->cr.cr_index < cs->cs_startrec) {
2196                 /* Skip entries earlier than what we are interested in */
2197                 CDEBUG(D_CHANGELOG, "rec="LPU64" start="LPU64"\n",
2198                        rec->cr.cr_index, cs->cs_startrec);
2199                 RETURN(0);
2200         }
2201
2202         CDEBUG(D_CHANGELOG, LPU64" %02d%-5s "LPU64" 0x%x t="DFID" p="DFID
2203                 " %.*s\n", rec->cr.cr_index, rec->cr.cr_type,
2204                 changelog_type2str(rec->cr.cr_type), rec->cr.cr_time,
2205                 rec->cr.cr_flags & CLF_FLAGMASK,
2206                 PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid),
2207                 rec->cr.cr_namelen, changelog_rec_name(&rec->cr));
2208
2209         len = sizeof(*lh) + changelog_rec_size(&rec->cr) + rec->cr.cr_namelen;
2210
2211         /* Set up the message */
2212         lh = changelog_kuc_hdr(cs->cs_buf, len, cs->cs_flags);
2213         memcpy(lh + 1, &rec->cr, len - sizeof(*lh));
2214
2215         rc = libcfs_kkuc_msg_put(cs->cs_fp, lh);
2216         CDEBUG(D_CHANGELOG, "kucmsg fp %p len %d rc %d\n", cs->cs_fp, len,rc);
2217
2218         RETURN(rc);
2219 }
2220
2221 static int mdc_changelog_send_thread(void *csdata)
2222 {
2223         struct changelog_show *cs = csdata;
2224         struct llog_ctxt *ctxt = NULL;
2225         struct llog_handle *llh = NULL;
2226         struct kuc_hdr *kuch;
2227         int rc;
2228
2229         CDEBUG(D_CHANGELOG, "changelog to fp=%p start "LPU64"\n",
2230                cs->cs_fp, cs->cs_startrec);
2231
2232         OBD_ALLOC(cs->cs_buf, KUC_CHANGELOG_MSG_MAXSIZE);
2233         if (cs->cs_buf == NULL)
2234                 GOTO(out, rc = -ENOMEM);
2235
2236         /* Set up the remote catalog handle */
2237         ctxt = llog_get_context(cs->cs_obd, LLOG_CHANGELOG_REPL_CTXT);
2238         if (ctxt == NULL)
2239                 GOTO(out, rc = -ENOENT);
2240         rc = llog_open(NULL, ctxt, &llh, NULL, CHANGELOG_CATALOG,
2241                        LLOG_OPEN_EXISTS);
2242         if (rc) {
2243                 CERROR("%s: fail to open changelog catalog: rc = %d\n",
2244                        cs->cs_obd->obd_name, rc);
2245                 GOTO(out, rc);
2246         }
2247         rc = llog_init_handle(NULL, llh, LLOG_F_IS_CAT, NULL);
2248         if (rc) {
2249                 CERROR("llog_init_handle failed %d\n", rc);
2250                 GOTO(out, rc);
2251         }
2252
2253         rc = llog_cat_process(NULL, llh, changelog_kkuc_cb, cs, 0, 0);
2254
2255         /* Send EOF no matter what our result */
2256         if ((kuch = changelog_kuc_hdr(cs->cs_buf, sizeof(*kuch),
2257                                       cs->cs_flags))) {
2258                 kuch->kuc_msgtype = CL_EOF;
2259                 libcfs_kkuc_msg_put(cs->cs_fp, kuch);
2260         }
2261
2262 out:
2263         fput(cs->cs_fp);
2264         if (llh)
2265                 llog_cat_close(NULL, llh);
2266         if (ctxt)
2267                 llog_ctxt_put(ctxt);
2268         if (cs->cs_buf)
2269                 OBD_FREE(cs->cs_buf, KUC_CHANGELOG_MSG_MAXSIZE);
2270         OBD_FREE_PTR(cs);
2271         return rc;
2272 }
2273
2274 static int mdc_ioc_changelog_send(struct obd_device *obd,
2275                                   struct ioc_changelog *icc)
2276 {
2277         struct changelog_show *cs;
2278         int rc;
2279
2280         /* Freed in mdc_changelog_send_thread */
2281         OBD_ALLOC_PTR(cs);
2282         if (!cs)
2283                 return -ENOMEM;
2284
2285         cs->cs_obd = obd;
2286         cs->cs_startrec = icc->icc_recno;
2287         /* matching fput in mdc_changelog_send_thread */
2288         cs->cs_fp = fget(icc->icc_id);
2289         cs->cs_flags = icc->icc_flags;
2290
2291         /*
2292          * New thread because we should return to user app before
2293          * writing into our pipe
2294          */
2295         rc = PTR_ERR(kthread_run(mdc_changelog_send_thread, cs,
2296                                  "mdc_clg_send_thread"));
2297         if (!IS_ERR_VALUE(rc)) {
2298                 CDEBUG(D_CHANGELOG, "start changelog thread\n");
2299                 return 0;
2300         }
2301
2302         CERROR("Failed to start changelog thread: %d\n", rc);
2303         OBD_FREE_PTR(cs);
2304         return rc;
2305 }
2306
2307 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2308                                 struct lustre_kernelcomm *lk);
2309
2310 static int mdc_quotacheck(struct obd_device *unused, struct obd_export *exp,
2311                           struct obd_quotactl *oqctl)
2312 {
2313         struct client_obd       *cli = &exp->exp_obd->u.cli;
2314         struct ptlrpc_request   *req;
2315         struct obd_quotactl     *body;
2316         int                      rc;
2317         ENTRY;
2318
2319         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
2320                                         &RQF_MDS_QUOTACHECK, LUSTRE_MDS_VERSION,
2321                                         MDS_QUOTACHECK);
2322         if (req == NULL)
2323                 RETURN(-ENOMEM);
2324
2325         body = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
2326         *body = *oqctl;
2327
2328         ptlrpc_request_set_replen(req);
2329
2330         /* the next poll will find -ENODATA, that means quotacheck is
2331          * going on */
2332         cli->cl_qchk_stat = -ENODATA;
2333         rc = ptlrpc_queue_wait(req);
2334         if (rc)
2335                 cli->cl_qchk_stat = rc;
2336         ptlrpc_req_finished(req);
2337         RETURN(rc);
2338 }
2339
2340 static int mdc_quota_poll_check(struct obd_export *exp,
2341                                 struct if_quotacheck *qchk)
2342 {
2343         struct client_obd *cli = &exp->exp_obd->u.cli;
2344         int rc;
2345         ENTRY;
2346
2347         qchk->obd_uuid = cli->cl_target_uuid;
2348         memcpy(qchk->obd_type, LUSTRE_MDS_NAME, strlen(LUSTRE_MDS_NAME));
2349
2350         rc = cli->cl_qchk_stat;
2351         /* the client is not the previous one */
2352         if (rc == CL_NOT_QUOTACHECKED)
2353                 rc = -EINTR;
2354         RETURN(rc);
2355 }
2356
2357 static int mdc_quotactl(struct obd_device *unused, struct obd_export *exp,
2358                         struct obd_quotactl *oqctl)
2359 {
2360         struct ptlrpc_request   *req;
2361         struct obd_quotactl     *oqc;
2362         int                      rc;
2363         ENTRY;
2364
2365         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
2366                                         &RQF_MDS_QUOTACTL, LUSTRE_MDS_VERSION,
2367                                         MDS_QUOTACTL);
2368         if (req == NULL)
2369                 RETURN(-ENOMEM);
2370
2371         oqc = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
2372         *oqc = *oqctl;
2373
2374         ptlrpc_request_set_replen(req);
2375         ptlrpc_at_set_req_timeout(req);
2376         req->rq_no_resend = 1;
2377
2378         rc = ptlrpc_queue_wait(req);
2379         if (rc)
2380                 CERROR("ptlrpc_queue_wait failed, rc: %d\n", rc);
2381
2382         if (req->rq_repmsg &&
2383             (oqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL))) {
2384                 *oqctl = *oqc;
2385         } else if (!rc) {
2386                 CERROR ("Can't unpack obd_quotactl\n");
2387                 rc = -EPROTO;
2388         }
2389         ptlrpc_req_finished(req);
2390
2391         RETURN(rc);
2392 }
2393
2394 static int mdc_ioc_swap_layouts(struct obd_export *exp,
2395                                 struct md_op_data *op_data)
2396 {
2397         CFS_LIST_HEAD(cancels);
2398         struct ptlrpc_request   *req;
2399         int                      rc, count;
2400         struct mdc_swap_layouts *msl, *payload;
2401         ENTRY;
2402
2403         msl = op_data->op_data;
2404
2405         /* When the MDT will get the MDS_SWAP_LAYOUTS RPC the
2406          * first thing it will do is to cancel the 2 layout
2407          * locks hold by this client.
2408          * So the client must cancel its layout locks on the 2 fids
2409          * with the request RPC to avoid extra RPC round trips
2410          */
2411         count = mdc_resource_get_unused(exp, &op_data->op_fid1, &cancels,
2412                                         LCK_CR, MDS_INODELOCK_LAYOUT);
2413         count += mdc_resource_get_unused(exp, &op_data->op_fid2, &cancels,
2414                                          LCK_CR, MDS_INODELOCK_LAYOUT);
2415
2416         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
2417                                    &RQF_MDS_SWAP_LAYOUTS);
2418         if (req == NULL) {
2419                 ldlm_lock_list_put(&cancels, l_bl_ast, count);
2420                 RETURN(-ENOMEM);
2421         }
2422
2423         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
2424         mdc_set_capa_size(req, &RMF_CAPA2, op_data->op_capa2);
2425
2426         rc = mdc_prep_elc_req(exp, req, MDS_SWAP_LAYOUTS, &cancels, count);
2427         if (rc) {
2428                 ptlrpc_request_free(req);
2429                 RETURN(rc);
2430         }
2431
2432         mdc_swap_layouts_pack(req, op_data);
2433
2434         payload = req_capsule_client_get(&req->rq_pill, &RMF_SWAP_LAYOUTS);
2435         LASSERT(payload);
2436
2437         *payload = *msl;
2438
2439         ptlrpc_request_set_replen(req);
2440
2441         rc = ptlrpc_queue_wait(req);
2442         if (rc)
2443                 GOTO(out, rc);
2444         EXIT;
2445
2446 out:
2447         ptlrpc_req_finished(req);
2448         return rc;
2449 }
2450
2451 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2452                          void *karg, void *uarg)
2453 {
2454         struct obd_device *obd = exp->exp_obd;
2455         struct obd_ioctl_data *data = karg;
2456         struct obd_import *imp = obd->u.cli.cl_import;
2457         int rc;
2458         ENTRY;
2459
2460         if (!try_module_get(THIS_MODULE)) {
2461                 CERROR("Can't get module. Is it alive?");
2462                 return -EINVAL;
2463         }
2464         switch (cmd) {
2465         case OBD_IOC_CHANGELOG_SEND:
2466                 rc = mdc_ioc_changelog_send(obd, karg);
2467                 GOTO(out, rc);
2468         case OBD_IOC_CHANGELOG_CLEAR: {
2469                 struct ioc_changelog *icc = karg;
2470                 struct changelog_setinfo cs =
2471                         {.cs_recno = icc->icc_recno, .cs_id = icc->icc_id};
2472                 rc = obd_set_info_async(NULL, exp, strlen(KEY_CHANGELOG_CLEAR),
2473                                         KEY_CHANGELOG_CLEAR, sizeof(cs), &cs,
2474                                         NULL);
2475                 GOTO(out, rc);
2476         }
2477         case OBD_IOC_FID2PATH:
2478                 rc = mdc_ioc_fid2path(exp, karg);
2479                 GOTO(out, rc);
2480         case LL_IOC_HSM_CT_START:
2481                 rc = mdc_ioc_hsm_ct_start(exp, karg);
2482                 /* ignore if it was already registered on this MDS. */
2483                 if (rc == -EEXIST)
2484                         rc = 0;
2485                 GOTO(out, rc);
2486         case LL_IOC_HSM_PROGRESS:
2487                 rc = mdc_ioc_hsm_progress(exp, karg);
2488                 GOTO(out, rc);
2489         case LL_IOC_HSM_STATE_GET:
2490                 rc = mdc_ioc_hsm_state_get(exp, karg);
2491                 GOTO(out, rc);
2492         case LL_IOC_HSM_STATE_SET:
2493                 rc = mdc_ioc_hsm_state_set(exp, karg);
2494                 GOTO(out, rc);
2495         case LL_IOC_HSM_ACTION:
2496                 rc = mdc_ioc_hsm_current_action(exp, karg);
2497                 GOTO(out, rc);
2498         case LL_IOC_HSM_REQUEST:
2499                 rc = mdc_ioc_hsm_request(exp, karg);
2500                 GOTO(out, rc);
2501         case OBD_IOC_CLIENT_RECOVER:
2502                 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1, 0);
2503                 if (rc < 0)
2504                         GOTO(out, rc);
2505                 GOTO(out, rc = 0);
2506         case IOC_OSC_SET_ACTIVE:
2507                 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
2508                 GOTO(out, rc);
2509         case OBD_IOC_POLL_QUOTACHECK:
2510                 rc = mdc_quota_poll_check(exp, (struct if_quotacheck *)karg);
2511                 GOTO(out, rc);
2512         case OBD_IOC_PING_TARGET:
2513                 rc = ptlrpc_obd_ping(obd);
2514                 GOTO(out, rc);
2515         /*
2516          * Normally IOC_OBD_STATFS, OBD_IOC_QUOTACTL iocontrol are handled by
2517          * LMV instead of MDC. But when the cluster is upgraded from 1.8,
2518          * there'd be no LMV layer thus we might be called here. Eventually
2519          * this code should be removed.
2520          * bz20731, LU-592.
2521          */
2522         case IOC_OBD_STATFS: {
2523                 struct obd_statfs stat_buf = {0};
2524
2525                 if (*((__u32 *) data->ioc_inlbuf2) != 0)
2526                         GOTO(out, rc = -ENODEV);
2527
2528                 /* copy UUID */
2529                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd),
2530                                  min((int)data->ioc_plen2,
2531                                      (int)sizeof(struct obd_uuid))))
2532                         GOTO(out, rc = -EFAULT);
2533
2534                 rc = mdc_statfs(NULL, obd->obd_self_export, &stat_buf,
2535                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
2536                                 0);
2537                 if (rc != 0)
2538                         GOTO(out, rc);
2539
2540                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
2541                                      min((int) data->ioc_plen1,
2542                                          (int) sizeof(stat_buf))))
2543                         GOTO(out, rc = -EFAULT);
2544
2545                 GOTO(out, rc = 0);
2546         }
2547         case OBD_IOC_QUOTACTL: {
2548                 struct if_quotactl *qctl = karg;
2549                 struct obd_quotactl *oqctl;
2550
2551                 OBD_ALLOC_PTR(oqctl);
2552                 if (oqctl == NULL)
2553                         GOTO(out, rc = -ENOMEM);
2554
2555                 QCTL_COPY(oqctl, qctl);
2556                 rc = obd_quotactl(exp, oqctl);
2557                 if (rc == 0) {
2558                         QCTL_COPY(qctl, oqctl);
2559                         qctl->qc_valid = QC_MDTIDX;
2560                         qctl->obd_uuid = obd->u.cli.cl_target_uuid;
2561                 }
2562
2563                 OBD_FREE_PTR(oqctl);
2564                 GOTO(out, rc);
2565         }
2566         case LL_IOC_GET_CONNECT_FLAGS:
2567                 if (copy_to_user(uarg, exp_connect_flags_ptr(exp),
2568                                  sizeof(*exp_connect_flags_ptr(exp))))
2569                         GOTO(out, rc = -EFAULT);
2570
2571                 GOTO(out, rc = 0);
2572         case LL_IOC_LOV_SWAP_LAYOUTS:
2573                 rc = mdc_ioc_swap_layouts(exp, karg);
2574                 GOTO(out, rc);
2575         default:
2576                 CERROR("unrecognised ioctl: cmd = %#x\n", cmd);
2577                 GOTO(out, rc = -ENOTTY);
2578         }
2579 out:
2580         module_put(THIS_MODULE);
2581
2582         return rc;
2583 }
2584
2585 int mdc_get_info_rpc(struct obd_export *exp,
2586                      obd_count keylen, void *key,
2587                      int vallen, void *val)
2588 {
2589         struct obd_import      *imp = class_exp2cliimp(exp);
2590         struct ptlrpc_request  *req;
2591         char                   *tmp;
2592         int                     rc = -EINVAL;
2593         ENTRY;
2594
2595         req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
2596         if (req == NULL)
2597                 RETURN(-ENOMEM);
2598
2599         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
2600                              RCL_CLIENT, keylen);
2601         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
2602                              RCL_CLIENT, sizeof(__u32));
2603
2604         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
2605         if (rc) {
2606                 ptlrpc_request_free(req);
2607                 RETURN(rc);
2608         }
2609
2610         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
2611         memcpy(tmp, key, keylen);
2612         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
2613         memcpy(tmp, &vallen, sizeof(__u32));
2614
2615         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
2616                              RCL_SERVER, vallen);
2617         ptlrpc_request_set_replen(req);
2618
2619         rc = ptlrpc_queue_wait(req);
2620         /* -EREMOTE means the get_info result is partial, and it needs to
2621          * continue on another MDT, see fid2path part in lmv_iocontrol */
2622         if (rc == 0 || rc == -EREMOTE) {
2623                 tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
2624                 memcpy(val, tmp, vallen);
2625                 if (ptlrpc_rep_need_swab(req)) {
2626                         if (KEY_IS(KEY_FID2PATH))
2627                                 lustre_swab_fid2path(val);
2628                 }
2629         }
2630         ptlrpc_req_finished(req);
2631
2632         RETURN(rc);
2633 }
2634
2635 static void lustre_swab_hai(struct hsm_action_item *h)
2636 {
2637         __swab32s(&h->hai_len);
2638         __swab32s(&h->hai_action);
2639         lustre_swab_lu_fid(&h->hai_fid);
2640         lustre_swab_lu_fid(&h->hai_dfid);
2641         __swab64s(&h->hai_cookie);
2642         __swab64s(&h->hai_extent.offset);
2643         __swab64s(&h->hai_extent.length);
2644         __swab64s(&h->hai_gid);
2645 }
2646
2647 static void lustre_swab_hal(struct hsm_action_list *h)
2648 {
2649         struct hsm_action_item  *hai;
2650         int                      i;
2651
2652         __swab32s(&h->hal_version);
2653         __swab32s(&h->hal_count);
2654         __swab32s(&h->hal_archive_id);
2655         __swab64s(&h->hal_flags);
2656         hai = hai_first(h);
2657         for (i = 0; i < h->hal_count; i++, hai = hai_next(hai))
2658                 lustre_swab_hai(hai);
2659 }
2660
2661 static void lustre_swab_kuch(struct kuc_hdr *l)
2662 {
2663         __swab16s(&l->kuc_magic);
2664         /* __u8 l->kuc_transport */
2665         __swab16s(&l->kuc_msgtype);
2666         __swab16s(&l->kuc_msglen);
2667 }
2668
2669 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2670                                 struct lustre_kernelcomm *lk)
2671 {
2672         struct obd_import  *imp = class_exp2cliimp(exp);
2673         __u32               archive = lk->lk_data;
2674         int                 rc = 0;
2675
2676         if (lk->lk_group != KUC_GRP_HSM) {
2677                 CERROR("Bad copytool group %d\n", lk->lk_group);
2678                 return -EINVAL;
2679         }
2680
2681         CDEBUG(D_HSM, "CT start r%d w%d u%d g%d f%#x\n", lk->lk_rfd, lk->lk_wfd,
2682                lk->lk_uid, lk->lk_group, lk->lk_flags);
2683
2684         if (lk->lk_flags & LK_FLG_STOP) {
2685                 /* Unregister with the coordinator */
2686                 rc = mdc_ioc_hsm_ct_unregister(imp);
2687         } else {
2688                 rc = mdc_ioc_hsm_ct_register(imp, archive);
2689         }
2690
2691         return rc;
2692 }
2693
2694 /**
2695  * Send a message to any listening copytools
2696  * @param val KUC message (kuc_hdr + hsm_action_list)
2697  * @param len total length of message
2698  */
2699 static int mdc_hsm_copytool_send(int len, void *val)
2700 {
2701         struct kuc_hdr          *lh = (struct kuc_hdr *)val;
2702         struct hsm_action_list  *hal = (struct hsm_action_list *)(lh + 1);
2703         int                      rc;
2704         ENTRY;
2705
2706         if (len < sizeof(*lh) + sizeof(*hal)) {
2707                 CERROR("Short HSM message %d < %d\n", len,
2708                        (int) (sizeof(*lh) + sizeof(*hal)));
2709                 RETURN(-EPROTO);
2710         }
2711         if (lh->kuc_magic == __swab16(KUC_MAGIC)) {
2712                 lustre_swab_kuch(lh);
2713                 lustre_swab_hal(hal);
2714         } else if (lh->kuc_magic != KUC_MAGIC) {
2715                 CERROR("Bad magic %x!=%x\n", lh->kuc_magic, KUC_MAGIC);
2716                 RETURN(-EPROTO);
2717         }
2718
2719         CDEBUG(D_HSM, " Received message mg=%x t=%d m=%d l=%d actions=%d "
2720                "on %s\n",
2721                lh->kuc_magic, lh->kuc_transport, lh->kuc_msgtype,
2722                lh->kuc_msglen, hal->hal_count, hal->hal_fsname);
2723
2724         /* Broadcast to HSM listeners */
2725         rc = libcfs_kkuc_group_put(KUC_GRP_HSM, lh);
2726
2727         RETURN(rc);
2728 }
2729
2730 /**
2731  * callback function passed to kuc for re-registering each HSM copytool
2732  * running on MDC, after MDT shutdown/recovery.
2733  * @param data copytool registration data
2734  * @param cb_arg callback argument (obd_import)
2735  */
2736 static int mdc_hsm_ct_reregister(void *data, void *cb_arg)
2737 {
2738         struct kkuc_ct_data     *kcd = data;
2739         struct obd_import       *imp = (struct obd_import *)cb_arg;
2740         int                      rc;
2741
2742         if (kcd == NULL || kcd->kcd_magic != KKUC_CT_DATA_MAGIC)
2743                 return -EPROTO;
2744
2745         if (!obd_uuid_equals(&kcd->kcd_uuid, &imp->imp_obd->obd_uuid))
2746                 return 0;
2747
2748         CDEBUG(D_HA, "%s: recover copytool registration to MDT (archive=%#x)\n",
2749                imp->imp_obd->obd_name, kcd->kcd_archive);
2750         rc = mdc_ioc_hsm_ct_register(imp, kcd->kcd_archive);
2751
2752         /* ignore error if the copytool is already registered */
2753         return (rc == -EEXIST) ? 0 : rc;
2754 }
2755
2756 /**
2757  * Re-establish all kuc contexts with MDT
2758  * after MDT shutdown/recovery.
2759  */
2760 static int mdc_kuc_reregister(struct obd_import *imp)
2761 {
2762         /* re-register HSM agents */
2763         return libcfs_kkuc_group_foreach(KUC_GRP_HSM, mdc_hsm_ct_reregister,
2764                                          (void *)imp);
2765 }
2766
2767 int mdc_set_info_async(const struct lu_env *env,
2768                        struct obd_export *exp,
2769                        obd_count keylen, void *key,
2770                        obd_count vallen, void *val,
2771                        struct ptlrpc_request_set *set)
2772 {
2773         struct obd_import       *imp = class_exp2cliimp(exp);
2774         int                      rc;
2775         ENTRY;
2776
2777         if (KEY_IS(KEY_READ_ONLY)) {
2778                 if (vallen != sizeof(int))
2779                         RETURN(-EINVAL);
2780
2781                 spin_lock(&imp->imp_lock);
2782                 if (*((int *)val)) {
2783                         imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
2784                         imp->imp_connect_data.ocd_connect_flags |=
2785                                                         OBD_CONNECT_RDONLY;
2786                 } else {
2787                         imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
2788                         imp->imp_connect_data.ocd_connect_flags &=
2789                                                         ~OBD_CONNECT_RDONLY;
2790                 }
2791                 spin_unlock(&imp->imp_lock);
2792
2793                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2794                                        keylen, key, vallen, val, set);
2795                 RETURN(rc);
2796         }
2797         if (KEY_IS(KEY_SPTLRPC_CONF)) {
2798                 sptlrpc_conf_client_adapt(exp->exp_obd);
2799                 RETURN(0);
2800         }
2801         if (KEY_IS(KEY_FLUSH_CTX)) {
2802                 sptlrpc_import_flush_my_ctx(imp);
2803                 RETURN(0);
2804         }
2805         if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2806                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2807                                        keylen, key, vallen, val, set);
2808                 RETURN(rc);
2809         }
2810         if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) {
2811                 rc = mdc_hsm_copytool_send(vallen, val);
2812                 RETURN(rc);
2813         }
2814
2815         CERROR("Unknown key %s\n", (char *)key);
2816         RETURN(-EINVAL);
2817 }
2818
2819 int mdc_get_info(const struct lu_env *env, struct obd_export *exp,
2820                  __u32 keylen, void *key, __u32 *vallen, void *val,
2821                  struct lov_stripe_md *lsm)
2822 {
2823         int rc = -EINVAL;
2824
2825         if (KEY_IS(KEY_MAX_EASIZE)) {
2826                 int mdsize, *max_easize;
2827
2828                 if (*vallen != sizeof(int))
2829                         RETURN(-EINVAL);
2830                 mdsize = *(int*)val;
2831                 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
2832                         exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
2833                 max_easize = val;
2834                 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
2835                 RETURN(0);
2836         } else if (KEY_IS(KEY_CONN_DATA)) {
2837                 struct obd_import *imp = class_exp2cliimp(exp);
2838                 struct obd_connect_data *data = val;
2839
2840                 if (*vallen != sizeof(*data))
2841                         RETURN(-EINVAL);
2842
2843                 *data = imp->imp_connect_data;
2844                 RETURN(0);
2845         } else if (KEY_IS(KEY_TGT_COUNT)) {
2846                 *((int *)val) = 1;
2847                 RETURN(0);
2848         }
2849
2850         rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
2851
2852         RETURN(rc);
2853 }
2854
2855 static int mdc_pin(struct obd_export *exp, const struct lu_fid *fid,
2856                    struct obd_capa *oc, struct obd_client_handle *handle,
2857                    int flags)
2858 {
2859         struct ptlrpc_request *req;
2860         struct mdt_body       *body;
2861         int                    rc;
2862         ENTRY;
2863
2864         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_PIN);
2865         if (req == NULL)
2866                 RETURN(-ENOMEM);
2867
2868         mdc_set_capa_size(req, &RMF_CAPA1, oc);
2869
2870         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_PIN);
2871         if (rc) {
2872                 ptlrpc_request_free(req);
2873                 RETURN(rc);
2874         }
2875
2876         mdc_pack_body(req, fid, oc, 0, 0, -1, flags);
2877
2878         ptlrpc_request_set_replen(req);
2879
2880         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
2881         rc = ptlrpc_queue_wait(req);
2882         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
2883         if (rc) {
2884                 CERROR("Pin failed: %d\n", rc);
2885                 GOTO(err_out, rc);
2886         }
2887
2888         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2889         if (body == NULL)
2890                 GOTO(err_out, rc = -EPROTO);
2891
2892         handle->och_fh = body->handle;
2893         handle->och_magic = OBD_CLIENT_HANDLE_MAGIC;
2894
2895         handle->och_mod = obd_mod_alloc();
2896         if (handle->och_mod == NULL) {
2897                 DEBUG_REQ(D_ERROR, req, "can't allocate md_open_data");
2898                 GOTO(err_out, rc = -ENOMEM);
2899         }
2900         handle->och_mod->mod_open_req = req; /* will be dropped by unpin */
2901
2902         RETURN(0);
2903
2904 err_out:
2905         ptlrpc_req_finished(req);
2906         RETURN(rc);
2907 }
2908
2909 static int mdc_unpin(struct obd_export *exp, struct obd_client_handle *handle,
2910                      int flag)
2911 {
2912         struct ptlrpc_request *req;
2913         struct mdt_body       *body;
2914         int                    rc;
2915         ENTRY;
2916
2917         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_UNPIN,
2918                                         LUSTRE_MDS_VERSION, MDS_UNPIN);
2919         if (req == NULL)
2920                 RETURN(-ENOMEM);
2921
2922         body = req_capsule_client_get(&req->rq_pill, &RMF_MDT_BODY);
2923         body->handle = handle->och_fh;
2924         body->flags = flag;
2925
2926         ptlrpc_request_set_replen(req);
2927
2928         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
2929         rc = ptlrpc_queue_wait(req);
2930         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
2931
2932         if (rc != 0)
2933                 CERROR("Unpin failed: %d\n", rc);
2934
2935         ptlrpc_req_finished(req);
2936         ptlrpc_req_finished(handle->och_mod->mod_open_req);
2937
2938         obd_mod_put(handle->och_mod);
2939         RETURN(rc);
2940 }
2941
2942 int mdc_fsync(struct obd_export *exp, const struct lu_fid *fid,
2943               struct obd_capa *oc, struct ptlrpc_request **request)
2944 {
2945         struct ptlrpc_request *req;
2946         int                    rc;
2947         ENTRY;
2948
2949         *request = NULL;
2950         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
2951         if (req == NULL)
2952                 RETURN(-ENOMEM);
2953
2954         mdc_set_capa_size(req, &RMF_CAPA1, oc);
2955
2956         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
2957         if (rc) {
2958                 ptlrpc_request_free(req);
2959                 RETURN(rc);
2960         }
2961
2962         mdc_pack_body(req, fid, oc, 0, 0, -1, 0);
2963
2964         ptlrpc_request_set_replen(req);
2965
2966         rc = ptlrpc_queue_wait(req);
2967         if (rc)
2968                 ptlrpc_req_finished(req);
2969         else
2970                 *request = req;
2971         RETURN(rc);
2972 }
2973
2974 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
2975                             enum obd_import_event event)
2976 {
2977         int rc = 0;
2978
2979         LASSERT(imp->imp_obd == obd);
2980
2981         switch (event) {
2982         case IMP_EVENT_DISCON: {
2983 #if 0
2984                 /* XXX Pass event up to OBDs stack. used only for FLD now */
2985                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DISCON, NULL);
2986 #endif
2987                 break;
2988         }
2989         case IMP_EVENT_INACTIVE: {
2990                 struct client_obd *cli = &obd->u.cli;
2991                 /*
2992                  * Flush current sequence to make client obtain new one
2993                  * from server in case of disconnect/reconnect.
2994                  */
2995                 if (cli->cl_seq != NULL)
2996                         seq_client_flush(cli->cl_seq);
2997
2998                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
2999                 break;
3000         }
3001         case IMP_EVENT_INVALIDATE: {
3002                 struct ldlm_namespace *ns = obd->obd_namespace;
3003
3004                 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
3005
3006                 break;
3007         }
3008         case IMP_EVENT_ACTIVE:
3009                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
3010                 /* redo the kuc registration after reconnecting */
3011                 if (rc == 0)
3012                         rc = mdc_kuc_reregister(imp);
3013                 break;
3014         case IMP_EVENT_OCD:
3015                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
3016                 break;
3017         case IMP_EVENT_DEACTIVATE:
3018         case IMP_EVENT_ACTIVATE:
3019                 break;
3020         default:
3021                 CERROR("Unknown import event %x\n", event);
3022                 LBUG();
3023         }
3024         RETURN(rc);
3025 }
3026
3027 int mdc_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
3028                   struct md_op_data *op_data)
3029 {
3030         struct client_obd *cli = &exp->exp_obd->u.cli;
3031         struct lu_client_seq *seq = cli->cl_seq;
3032         ENTRY;
3033         RETURN(seq_client_alloc_fid(NULL, seq, fid));
3034 }
3035
3036 struct obd_uuid *mdc_get_uuid(struct obd_export *exp) {
3037         struct client_obd *cli = &exp->exp_obd->u.cli;
3038         return &cli->cl_target_uuid;
3039 }
3040
3041 /**
3042  * Determine whether the lock can be canceled before replaying it during
3043  * recovery, non zero value will be return if the lock can be canceled,
3044  * or zero returned for not
3045  */
3046 static int mdc_cancel_for_recovery(struct ldlm_lock *lock)
3047 {
3048         if (lock->l_resource->lr_type != LDLM_IBITS)
3049                 RETURN(0);
3050
3051         /* FIXME: if we ever get into a situation where there are too many
3052          * opened files with open locks on a single node, then we really
3053          * should replay these open locks to reget it */
3054         if (lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_OPEN)
3055                 RETURN(0);
3056
3057         RETURN(1);
3058 }
3059
3060 static int mdc_resource_inode_free(struct ldlm_resource *res)
3061 {
3062         if (res->lr_lvb_inode)
3063                 res->lr_lvb_inode = NULL;
3064
3065         return 0;
3066 }
3067
3068 struct ldlm_valblock_ops inode_lvbo = {
3069         lvbo_free: mdc_resource_inode_free
3070 };
3071
3072 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
3073 {
3074         struct client_obd               *cli = &obd->u.cli;
3075         int                             rc;
3076         ENTRY;
3077
3078         OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
3079         if (!cli->cl_rpc_lock)
3080                 RETURN(-ENOMEM);
3081         mdc_init_rpc_lock(cli->cl_rpc_lock);
3082
3083         rc = ptlrpcd_addref();
3084         if (rc < 0)
3085                 GOTO(err_rpc_lock, rc);
3086
3087         OBD_ALLOC(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
3088         if (!cli->cl_close_lock)
3089                 GOTO(err_ptlrpcd_decref, rc = -ENOMEM);
3090         mdc_init_rpc_lock(cli->cl_close_lock);
3091
3092         rc = client_obd_setup(obd, cfg);
3093         if (rc)
3094                 GOTO(err_close_lock, rc);
3095 #ifdef LPROCFS
3096         obd->obd_vars = lprocfs_mdc_obd_vars;
3097         lprocfs_seq_obd_setup(obd);
3098         lprocfs_alloc_md_stats(obd, 0);
3099 #endif
3100         sptlrpc_lprocfs_cliobd_attach(obd);
3101         ptlrpc_lprocfs_register_obd(obd);
3102
3103         ns_register_cancel(obd->obd_namespace, mdc_cancel_for_recovery);
3104
3105         obd->obd_namespace->ns_lvbo = &inode_lvbo;
3106
3107         rc = obd_llog_init(obd, &obd->obd_olg, obd, NULL);
3108         if (rc) {
3109                 mdc_cleanup(obd);
3110                 CERROR("failed to setup llogging subsystems\n");
3111         }
3112
3113         RETURN(rc);
3114
3115 err_close_lock:
3116         OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
3117 err_ptlrpcd_decref:
3118         ptlrpcd_decref();
3119 err_rpc_lock:
3120         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
3121         RETURN(rc);
3122 }
3123
3124 /* Initialize the default and maximum LOV EA and cookie sizes.  This allows
3125  * us to make MDS RPCs with large enough reply buffers to hold the
3126  * maximum-sized (= maximum striped) EA and cookie without having to
3127  * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
3128 static int mdc_init_ea_size(struct obd_export *exp, int easize,
3129                      int def_easize, int cookiesize)
3130 {
3131         struct obd_device *obd = exp->exp_obd;
3132         struct client_obd *cli = &obd->u.cli;
3133         ENTRY;
3134
3135         if (cli->cl_max_mds_easize < easize)
3136                 cli->cl_max_mds_easize = easize;
3137
3138         if (cli->cl_default_mds_easize < def_easize)
3139                 cli->cl_default_mds_easize = def_easize;
3140
3141         if (cli->cl_max_mds_cookiesize < cookiesize)
3142                 cli->cl_max_mds_cookiesize = cookiesize;
3143
3144         RETURN(0);
3145 }
3146
3147 static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
3148 {
3149         int rc = 0;
3150         ENTRY;
3151
3152         switch (stage) {
3153         case OBD_CLEANUP_EARLY:
3154                 break;
3155         case OBD_CLEANUP_EXPORTS:
3156                 /* Failsafe, ok if racy */
3157                 if (obd->obd_type->typ_refcnt <= 1)
3158                         libcfs_kkuc_group_rem(0, KUC_GRP_HSM, NULL);
3159
3160                 obd_cleanup_client_import(obd);
3161                 ptlrpc_lprocfs_unregister_obd(obd);
3162                 lprocfs_obd_cleanup(obd);
3163                 lprocfs_free_md_stats(obd);
3164
3165                 rc = obd_llog_finish(obd, 0);
3166                 if (rc != 0)
3167                         CERROR("failed to cleanup llogging subsystems\n");
3168                 break;
3169         }
3170         RETURN(rc);
3171 }
3172
3173 static int mdc_cleanup(struct obd_device *obd)
3174 {
3175         struct client_obd *cli = &obd->u.cli;
3176
3177         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
3178         OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
3179
3180         ptlrpcd_decref();
3181
3182         return client_obd_cleanup(obd);
3183 }
3184
3185
3186 static int mdc_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
3187                          struct obd_device *tgt, int *index)
3188 {
3189         struct llog_ctxt        *ctxt;
3190         int                      rc;
3191
3192         ENTRY;
3193
3194         LASSERT(olg == &obd->obd_olg);
3195
3196         rc = llog_setup(NULL, obd, olg, LLOG_CHANGELOG_REPL_CTXT, tgt,
3197                         &llog_client_ops);
3198         if (rc)
3199                 RETURN(rc);
3200
3201         ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
3202         llog_initiator_connect(ctxt);
3203         llog_ctxt_put(ctxt);
3204
3205         RETURN(0);
3206 }
3207
3208 static int mdc_llog_finish(struct obd_device *obd, int count)
3209 {
3210         struct llog_ctxt *ctxt;
3211
3212         ENTRY;
3213
3214         ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
3215         if (ctxt)
3216                 llog_cleanup(NULL, ctxt);
3217
3218         RETURN(0);
3219 }
3220
3221 static int mdc_process_config(struct obd_device *obd, obd_count len, void *buf)
3222 {
3223         struct lustre_cfg *lcfg = buf;
3224         int rc = class_process_proc_seq_param(PARAM_MDC, obd->obd_vars,
3225                                               lcfg, obd);
3226         return (rc > 0 ? 0: rc);
3227 }
3228
3229
3230 /* get remote permission for current user on fid */
3231 int mdc_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
3232                         struct obd_capa *oc, __u32 suppgid,
3233                         struct ptlrpc_request **request)
3234 {
3235         struct ptlrpc_request  *req;
3236         int                    rc;
3237         ENTRY;
3238
3239         LASSERT(client_is_remote(exp));
3240
3241         *request = NULL;
3242         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
3243         if (req == NULL)
3244                 RETURN(-ENOMEM);
3245
3246         mdc_set_capa_size(req, &RMF_CAPA1, oc);
3247
3248         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
3249         if (rc) {
3250                 ptlrpc_request_free(req);
3251                 RETURN(rc);
3252         }
3253
3254         mdc_pack_body(req, fid, oc, OBD_MD_FLRMTPERM, 0, suppgid, 0);
3255
3256         req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
3257                              sizeof(struct mdt_remote_perm));
3258
3259         ptlrpc_request_set_replen(req);
3260
3261         rc = ptlrpc_queue_wait(req);
3262         if (rc)
3263                 ptlrpc_req_finished(req);
3264         else
3265                 *request = req;
3266         RETURN(rc);
3267 }
3268
3269 static int mdc_interpret_renew_capa(const struct lu_env *env,
3270                                     struct ptlrpc_request *req, void *args,
3271                                     int status)
3272 {
3273         struct mdc_renew_capa_args *ra = args;
3274         struct mdt_body *body = NULL;
3275         struct lustre_capa *capa;
3276         ENTRY;
3277
3278         if (status)
3279                 GOTO(out, capa = ERR_PTR(status));
3280
3281         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
3282         if (body == NULL)
3283                 GOTO(out, capa = ERR_PTR(-EFAULT));
3284
3285         if ((body->valid & OBD_MD_FLOSSCAPA) == 0)
3286                 GOTO(out, capa = ERR_PTR(-ENOENT));
3287
3288         capa = req_capsule_server_get(&req->rq_pill, &RMF_CAPA2);
3289         if (!capa)
3290                 GOTO(out, capa = ERR_PTR(-EFAULT));
3291         EXIT;
3292 out:
3293         ra->ra_cb(ra->ra_oc, capa);
3294         return 0;
3295 }
3296
3297 static int mdc_renew_capa(struct obd_export *exp, struct obd_capa *oc,
3298                           renew_capa_cb_t cb)
3299 {
3300         struct ptlrpc_request *req;
3301         struct mdc_renew_capa_args *ra;
3302         ENTRY;
3303
3304         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_GETATTR,
3305                                         LUSTRE_MDS_VERSION, MDS_GETATTR);
3306         if (req == NULL)
3307                 RETURN(-ENOMEM);
3308
3309         /* NB, OBD_MD_FLOSSCAPA is set here, but it doesn't necessarily mean the
3310          * capa to renew is oss capa.
3311          */
3312         mdc_pack_body(req, &oc->c_capa.lc_fid, oc, OBD_MD_FLOSSCAPA, 0, -1, 0);
3313         ptlrpc_request_set_replen(req);
3314
3315         CLASSERT(sizeof(*ra) <= sizeof(req->rq_async_args));
3316         ra = ptlrpc_req_async_args(req);
3317         ra->ra_oc = oc;
3318         ra->ra_cb = cb;
3319         req->rq_interpret_reply = mdc_interpret_renew_capa;
3320         ptlrpcd_add_req(req, PDL_POLICY_LOCAL, -1);
3321         RETURN(0);
3322 }
3323
3324 struct obd_ops mdc_obd_ops = {
3325         .o_owner            = THIS_MODULE,
3326         .o_setup            = mdc_setup,
3327         .o_precleanup       = mdc_precleanup,
3328         .o_cleanup          = mdc_cleanup,
3329         .o_add_conn         = client_import_add_conn,
3330         .o_del_conn         = client_import_del_conn,
3331         .o_connect          = client_connect_import,
3332         .o_disconnect       = client_disconnect_export,
3333         .o_iocontrol        = mdc_iocontrol,
3334         .o_set_info_async   = mdc_set_info_async,
3335         .o_statfs           = mdc_statfs,
3336         .o_pin              = mdc_pin,
3337         .o_unpin            = mdc_unpin,
3338         .o_fid_init         = client_fid_init,
3339         .o_fid_fini         = client_fid_fini,
3340         .o_fid_alloc        = mdc_fid_alloc,
3341         .o_import_event     = mdc_import_event,
3342         .o_llog_init        = mdc_llog_init,
3343         .o_llog_finish      = mdc_llog_finish,
3344         .o_get_info         = mdc_get_info,
3345         .o_process_config   = mdc_process_config,
3346         .o_get_uuid         = mdc_get_uuid,
3347         .o_quotactl         = mdc_quotactl,
3348         .o_quotacheck       = mdc_quotacheck
3349 };
3350
3351 struct md_ops mdc_md_ops = {
3352         .m_getstatus        = mdc_getstatus,
3353         .m_null_inode       = mdc_null_inode,
3354         .m_find_cbdata      = mdc_find_cbdata,
3355         .m_close            = mdc_close,
3356         .m_create           = mdc_create,
3357         .m_done_writing     = mdc_done_writing,
3358         .m_enqueue          = mdc_enqueue,
3359         .m_getattr          = mdc_getattr,
3360         .m_getattr_name     = mdc_getattr_name,
3361         .m_intent_lock      = mdc_intent_lock,
3362         .m_link             = mdc_link,
3363         .m_is_subdir        = mdc_is_subdir,
3364         .m_rename           = mdc_rename,
3365         .m_setattr          = mdc_setattr,
3366         .m_setxattr         = mdc_setxattr,
3367         .m_getxattr         = mdc_getxattr,
3368         .m_fsync                = mdc_fsync,
3369         .m_read_entry           = mdc_read_entry,
3370         .m_unlink           = mdc_unlink,
3371         .m_cancel_unused    = mdc_cancel_unused,
3372         .m_init_ea_size     = mdc_init_ea_size,
3373         .m_set_lock_data    = mdc_set_lock_data,
3374         .m_lock_match       = mdc_lock_match,
3375         .m_get_lustre_md    = mdc_get_lustre_md,
3376         .m_free_lustre_md   = mdc_free_lustre_md,
3377         .m_set_open_replay_data = mdc_set_open_replay_data,
3378         .m_clear_open_replay_data = mdc_clear_open_replay_data,
3379         .m_renew_capa       = mdc_renew_capa,
3380         .m_unpack_capa      = mdc_unpack_capa,
3381         .m_get_remote_perm  = mdc_get_remote_perm,
3382         .m_intent_getattr_async = mdc_intent_getattr_async,
3383         .m_revalidate_lock      = mdc_revalidate_lock
3384 };
3385
3386 int __init mdc_init(void)
3387 {
3388         return class_register_type(&mdc_obd_ops, &mdc_md_ops, NULL,
3389 #ifndef HAVE_ONLY_PROCFS_SEQ
3390                                         NULL,
3391 #endif
3392                                         LUSTRE_MDC_NAME, NULL);
3393 }
3394
3395 #ifdef __KERNEL__
3396 static void /*__exit*/ mdc_exit(void)
3397 {
3398         class_unregister_type(LUSTRE_MDC_NAME);
3399 }
3400
3401 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3402 MODULE_DESCRIPTION("Lustre Metadata Client");
3403 MODULE_LICENSE("GPL");
3404
3405 module_init(mdc_init);
3406 module_exit(mdc_exit);
3407 #endif