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