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, 2012, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36
  37 #define DEBUG_SUBSYSTEM S_MDC
  38
  39 #ifdef __KERNEL__
  40 # include <linux/module.h>
  41 # include <linux/pagemap.h>
  42 # include <linux/miscdevice.h>
  43 # include <linux/init.h>
  44 #else
  45 # include <liblustre.h>
  46 #endif
  47
  48 #include <lustre_acl.h>
  49 #include <obd_class.h>
  50 #include <lustre_fid.h>
  51 #include <md_object.h>
  52 #include <lprocfs_status.h>
  53 #include <lustre_param.h>
  54 #include <lustre_log.h>
  55
  56 #include "mdc_internal.h"
  57
  58 #define REQUEST_MINOR 244
  59
  60 struct mdc_renew_capa_args {
  61         struct obd_capa        *ra_oc;
  62         renew_capa_cb_t         ra_cb;
  63 };
  64
  65 static int mdc_cleanup(struct obd_device *obd);
  66
  67 int mdc_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
  68                     const struct req_msg_field *field, 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 /* Helper that implements most of mdc_getstatus and signal_completed_replay. */
  91 /* XXX this should become mdc_get_info("key"), sending MDS_GET_INFO RPC */
  92 static int send_getstatus(struct obd_import *imp, struct lu_fid *rootfid,
  93                           struct obd_capa **pc, int level, int msg_flags)
  94 {
  95         struct ptlrpc_request *req;
  96         struct mdt_body       *body;
  97         int                    rc;
  98         ENTRY;
  99
 100         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_GETSTATUS,
 101                                         LUSTRE_MDS_VERSION, MDS_GETSTATUS);
 102         if (req == NULL)
 103                 RETURN(-ENOMEM);
 104
 105         mdc_pack_body(req, NULL, NULL, 0, 0, -1, 0);
 106         lustre_msg_add_flags(req->rq_reqmsg, msg_flags);
 107         req->rq_send_state = level;
 108
 109         ptlrpc_request_set_replen(req);
 110
 111         rc = ptlrpc_queue_wait(req);
 112         if (rc)
 113                 GOTO(out, rc);
 114
 115         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 116         if (body == NULL)
 117                 GOTO(out, rc = -EPROTO);
 118
 119         if (body->valid & OBD_MD_FLMDSCAPA) {
 120                 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA1, pc);
 121                 if (rc)
 122                         GOTO(out, rc);
 123         }
 124
 125         *rootfid = body->fid1;
 126         CDEBUG(D_NET,
 127                "root fid="DFID", last_committed="LPU64"\n",
 128                PFID(rootfid),
 129                lustre_msg_get_last_committed(req->rq_repmsg));
 130         EXIT;
 131 out:
 132         ptlrpc_req_finished(req);
 133         return rc;
 134 }
 135
 136 /* This should be mdc_get_info("rootfid") */
 137 int mdc_getstatus(struct obd_export *exp, struct lu_fid *rootfid,
 138                   struct obd_capa **pc)
 139 {
 140         return send_getstatus(class_exp2cliimp(exp), rootfid, pc,
 141                               LUSTRE_IMP_FULL, 0);
 142 }
 143
 144 /*
 145  * This function now is known to always saying that it will receive 4 buffers
 146  * from server. Even for cases when acl_size and md_size is zero, RPC header
 147  * will contain 4 fields and RPC itself will contain zero size fields. This is
 148  * because mdt_getattr*() _always_ returns 4 fields, but if acl is not needed
 149  * and thus zero, it shrinks it, making zero size. The same story about
 150  * md_size. And this is course of problem when client waits for smaller number
 151  * of fields. This issue will be fixed later when client gets aware of RPC
 152  * layouts.  --umka
 153  */
 154 static int mdc_getattr_common(struct obd_export *exp,
 155                               struct ptlrpc_request *req)
 156 {
 157         struct req_capsule *pill = &req->rq_pill;
 158         struct mdt_body    *body;
 159         void               *eadata;
 160         int                 rc;
 161         ENTRY;
 162
 163         /* Request message already built. */
 164         rc = ptlrpc_queue_wait(req);
 165         if (rc != 0)
 166                 RETURN(rc);
 167
 168         /* sanity check for the reply */
 169         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 170         if (body == NULL)
 171                 RETURN(-EPROTO);
 172
 173         CDEBUG(D_NET, "mode: %o\n", body->mode);
 174
 175         if (body->eadatasize != 0) {
 176                 mdc_update_max_ea_from_body(exp, body);
 177
 178                 eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
 179                                                       body->eadatasize);
 180                 if (eadata == NULL)
 181                         RETURN(-EPROTO);
 182         }
 183
 184         if (body->valid & OBD_MD_FLRMTPERM) {
 185                 struct mdt_remote_perm *perm;
 186
 187                 LASSERT(client_is_remote(exp));
 188                 perm = req_capsule_server_swab_get(pill, &RMF_ACL,
 189                                                 lustre_swab_mdt_remote_perm);
 190                 if (perm == NULL)
 191                         RETURN(-EPROTO);
 192         }
 193
 194         if (body->valid & OBD_MD_FLMDSCAPA) {
 195                 struct lustre_capa *capa;
 196                 capa = req_capsule_server_get(pill, &RMF_CAPA1);
 197                 if (capa == NULL)
 198                         RETURN(-EPROTO);
 199         }
 200
 201         RETURN(0);
 202 }
 203
 204 int mdc_getattr(struct obd_export *exp, struct md_op_data *op_data,
 205                 struct ptlrpc_request **request)
 206 {
 207         struct ptlrpc_request *req;
 208         int                    rc;
 209         ENTRY;
 210
 211         *request = NULL;
 212         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
 213         if (req == NULL)
 214                 RETURN(-ENOMEM);
 215
 216         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 217
 218         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
 219         if (rc) {
 220                 ptlrpc_request_free(req);
 221                 RETURN(rc);
 222         }
 223
 224         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
 225                       op_data->op_valid, op_data->op_mode, -1, 0);
 226
 227         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 228                              op_data->op_mode);
 229         if (op_data->op_valid & OBD_MD_FLRMTPERM) {
 230                 LASSERT(client_is_remote(exp));
 231                 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
 232                                      sizeof(struct mdt_remote_perm));
 233         }
 234         ptlrpc_request_set_replen(req);
 235
 236         rc = mdc_getattr_common(exp, req);
 237         if (rc)
 238                 ptlrpc_req_finished(req);
 239         else
 240                 *request = req;
 241         RETURN(rc);
 242 }
 243
 244 int mdc_getattr_name(struct obd_export *exp, struct md_op_data *op_data,
 245                      struct ptlrpc_request **request)
 246 {
 247         struct ptlrpc_request *req;
 248         int                    rc;
 249         ENTRY;
 250
 251         *request = NULL;
 252         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 253                                    &RQF_MDS_GETATTR_NAME);
 254         if (req == NULL)
 255                 RETURN(-ENOMEM);
 256
 257         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 258         req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 259                              op_data->op_namelen + 1);
 260
 261         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR_NAME);
 262         if (rc) {
 263                 ptlrpc_request_free(req);
 264                 RETURN(rc);
 265         }
 266
 267         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
 268                       op_data->op_valid, op_data->op_mode,
 269                       op_data->op_suppgids[0], 0);
 270
 271         if (op_data->op_name) {
 272                 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 273                 LASSERT(strnlen(op_data->op_name, op_data->op_namelen) ==
 274                                 op_data->op_namelen);
 275                 memcpy(name, op_data->op_name, op_data->op_namelen);
 276         }
 277
 278         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 279                              op_data->op_mode);
 280         ptlrpc_request_set_replen(req);
 281
 282         rc = mdc_getattr_common(exp, req);
 283         if (rc)
 284                 ptlrpc_req_finished(req);
 285         else
 286                 *request = req;
 287         RETURN(rc);
 288 }
 289
 290 static int mdc_is_subdir(struct obd_export *exp,
 291                          const struct lu_fid *pfid,
 292                          const struct lu_fid *cfid,
 293                          struct ptlrpc_request **request)
 294 {
 295         struct ptlrpc_request  *req;
 296         int                     rc;
 297
 298         ENTRY;
 299
 300         *request = NULL;
 301         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
 302                                         &RQF_MDS_IS_SUBDIR, LUSTRE_MDS_VERSION,
 303                                         MDS_IS_SUBDIR);
 304         if (req == NULL)
 305                 RETURN(-ENOMEM);
 306
 307         mdc_is_subdir_pack(req, pfid, cfid, 0);
 308         ptlrpc_request_set_replen(req);
 309
 310         rc = ptlrpc_queue_wait(req);
 311         if (rc && rc != -EREMOTE)
 312                 ptlrpc_req_finished(req);
 313         else
 314                 *request = req;
 315         RETURN(rc);
 316 }
 317
 318 static int mdc_xattr_common(struct obd_export *exp,const struct req_format *fmt,
 319                             const struct lu_fid *fid,
 320                             struct obd_capa *oc, int opcode, obd_valid valid,
 321                             const char *xattr_name, const char *input,
 322                             int input_size, int output_size, int flags,
 323                             __u32 suppgid, struct ptlrpc_request **request)
 324 {
 325         struct ptlrpc_request *req;
 326         int   xattr_namelen = 0;
 327         char *tmp;
 328         int   rc;
 329         ENTRY;
 330
 331         *request = NULL;
 332         req = ptlrpc_request_alloc(class_exp2cliimp(exp), fmt);
 333         if (req == NULL)
 334                 RETURN(-ENOMEM);
 335
 336         mdc_set_capa_size(req, &RMF_CAPA1, oc);
 337         if (xattr_name) {
 338                 xattr_namelen = strlen(xattr_name) + 1;
 339                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 340                                      xattr_namelen);
 341         }
 342         if (input_size) {
 343                 LASSERT(input);
 344                 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
 345                                      input_size);
 346         }
 347
 348         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode);
 349         if (rc) {
 350                 ptlrpc_request_free(req);
 351                 RETURN(rc);
 352         }
 353
 354         if (opcode == MDS_REINT) {
 355                 struct mdt_rec_setxattr *rec;
 356
 357                 CLASSERT(sizeof(struct mdt_rec_setxattr) ==
 358                          sizeof(struct mdt_rec_reint));
 359                 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
 360                 rec->sx_opcode = REINT_SETXATTR;
 361                 /* TODO:
 362                  *  cfs_curproc_fs{u,g}id() should replace
 363                  *  current->fs{u,g}id for portability.
 364                  */
 365                 rec->sx_fsuid  = cfs_curproc_fsuid();
 366                 rec->sx_fsgid  = cfs_curproc_fsgid();
 367                 rec->sx_cap    = cfs_curproc_cap_pack();
 368                 rec->sx_suppgid1 = suppgid;
 369                 rec->sx_suppgid2 = -1;
 370                 rec->sx_fid    = *fid;
 371                 rec->sx_valid  = valid | OBD_MD_FLCTIME;
 372                 rec->sx_time   = cfs_time_current_sec();
 373                 rec->sx_size   = output_size;
 374                 rec->sx_flags  = flags;
 375
 376                 mdc_pack_capa(req, &RMF_CAPA1, oc);
 377         } else {
 378                 mdc_pack_body(req, fid, oc, valid, output_size, suppgid, flags);
 379         }
 380
 381         if (xattr_name) {
 382                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 383                 memcpy(tmp, xattr_name, xattr_namelen);
 384         }
 385         if (input_size) {
 386                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
 387                 memcpy(tmp, input, input_size);
 388         }
 389
 390         if (req_capsule_has_field(&req->rq_pill, &RMF_EADATA, RCL_SERVER))
 391                 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
 392                                      RCL_SERVER, output_size);
 393         ptlrpc_request_set_replen(req);
 394
 395         /* make rpc */
 396         if (opcode == MDS_REINT)
 397                 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 398
 399         rc = ptlrpc_queue_wait(req);
 400
 401         if (opcode == MDS_REINT)
 402                 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 403
 404         if (rc)
 405                 ptlrpc_req_finished(req);
 406         else
 407                 *request = req;
 408         RETURN(rc);
 409 }
 410
 411 int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
 412                  struct obd_capa *oc, obd_valid valid, const char *xattr_name,
 413                  const char *input, int input_size, int output_size,
 414                  int flags, __u32 suppgid, struct ptlrpc_request **request)
 415 {
 416         return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
 417                                 fid, oc, MDS_REINT, valid, xattr_name,
 418                                 input, input_size, output_size, flags,
 419                                 suppgid, request);
 420 }
 421
 422 int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
 423                  struct obd_capa *oc, obd_valid valid, const char *xattr_name,
 424                  const char *input, int input_size, int output_size,
 425                  int flags, struct ptlrpc_request **request)
 426 {
 427         return mdc_xattr_common(exp, &RQF_MDS_GETXATTR,
 428                                 fid, oc, MDS_GETXATTR, valid, xattr_name,
 429                                 input, input_size, output_size, flags,
 430                                 -1, request);
 431 }
 432
 433 #ifdef CONFIG_FS_POSIX_ACL
 434 static int mdc_unpack_acl(struct ptlrpc_request *req, struct lustre_md *md)
 435 {
 436         struct req_capsule     *pill = &req->rq_pill;
 437         struct mdt_body        *body = md->body;
 438         struct posix_acl       *acl;
 439         void                   *buf;
 440         int                     rc;
 441         ENTRY;
 442
 443         if (!body->aclsize)
 444                 RETURN(0);
 445
 446         buf = req_capsule_server_sized_get(pill, &RMF_ACL, body->aclsize);
 447
 448         if (!buf)
 449                 RETURN(-EPROTO);
 450
 451         acl = posix_acl_from_xattr(buf, body->aclsize);
 452         if (IS_ERR(acl)) {
 453                 rc = PTR_ERR(acl);
 454                 CERROR("convert xattr to acl: %d\n", rc);
 455                 RETURN(rc);
 456         }
 457
 458         rc = posix_acl_valid(acl);
 459         if (rc) {
 460                 CERROR("validate acl: %d\n", rc);
 461                 posix_acl_release(acl);
 462                 RETURN(rc);
 463         }
 464
 465         md->posix_acl = acl;
 466         RETURN(0);
 467 }
 468 #else
 469 #define mdc_unpack_acl(req, md) 0
 470 #endif
 471
 472 int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
 473                       struct obd_export *dt_exp, struct obd_export *md_exp,
 474                       struct lustre_md *md)
 475 {
 476         struct req_capsule *pill = &req->rq_pill;
 477         int rc;
 478         ENTRY;
 479
 480         LASSERT(md);
 481         memset(md, 0, sizeof(*md));
 482
 483         md->body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 484         LASSERT(md->body != NULL);
 485
 486         if (md->body->valid & OBD_MD_FLEASIZE) {
 487                 int lmmsize;
 488                 struct lov_mds_md *lmm;
 489
 490                 if (!S_ISREG(md->body->mode)) {
 491                         CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, should be a "
 492                                "regular file, but is not\n");
 493                         GOTO(out, rc = -EPROTO);
 494                 }
 495
 496                 if (md->body->eadatasize == 0) {
 497                         CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, "
 498                                "but eadatasize 0\n");
 499                         GOTO(out, rc = -EPROTO);
 500                 }
 501                 lmmsize = md->body->eadatasize;
 502                 lmm = req_capsule_server_sized_get(pill, &RMF_MDT_MD, lmmsize);
 503                 if (!lmm)
 504                         GOTO(out, rc = -EPROTO);
 505
 506                 rc = obd_unpackmd(dt_exp, &md->lsm, lmm, lmmsize);
 507                 if (rc < 0)
 508                         GOTO(out, rc);
 509
 510                 if (rc < sizeof(*md->lsm)) {
 511                         CDEBUG(D_INFO, "lsm size too small: "
 512                                "rc < sizeof (*md->lsm) (%d < %d)\n",
 513                                rc, (int)sizeof(*md->lsm));
 514                         GOTO(out, rc = -EPROTO);
 515                 }
 516
 517         } else if (md->body->valid & OBD_MD_FLDIREA) {
 518                 int lmvsize;
 519                 struct lov_mds_md *lmv;
 520
 521                 if(!S_ISDIR(md->body->mode)) {
 522                         CDEBUG(D_INFO, "OBD_MD_FLDIREA set, should be a "
 523                                "directory, but is not\n");
 524                         GOTO(out, rc = -EPROTO);
 525                 }
 526
 527                 if (md->body->eadatasize == 0) {
 528                         CDEBUG(D_INFO, "OBD_MD_FLDIREA is set, "
 529                                "but eadatasize 0\n");
 530                         RETURN(-EPROTO);
 531                 }
 532                 if (md->body->valid & OBD_MD_MEA) {
 533                         lmvsize = md->body->eadatasize;
 534                         lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
 535                                                            lmvsize);
 536                         if (!lmv)
 537                                 GOTO(out, rc = -EPROTO);
 538
 539                         rc = obd_unpackmd(md_exp, (void *)&md->mea, lmv,
 540                                           lmvsize);
 541                         if (rc < 0)
 542                                 GOTO(out, rc);
 543
 544                         if (rc < sizeof(*md->mea)) {
 545                                 CDEBUG(D_INFO, "size too small:  "
 546                                        "rc < sizeof(*md->mea) (%d < %d)\n",
 547                                         rc, (int)sizeof(*md->mea));
 548                                 GOTO(out, rc = -EPROTO);
 549                         }
 550                 }
 551         }
 552         rc = 0;
 553
 554         if (md->body->valid & OBD_MD_FLRMTPERM) {
 555                 /* remote permission */
 556                 LASSERT(client_is_remote(exp));
 557                 md->remote_perm = req_capsule_server_swab_get(pill, &RMF_ACL,
 558                                                 lustre_swab_mdt_remote_perm);
 559                 if (!md->remote_perm)
 560                         GOTO(out, rc = -EPROTO);
 561         }
 562         else if (md->body->valid & OBD_MD_FLACL) {
 563                 /* for ACL, it's possible that FLACL is set but aclsize is zero.
 564                  * only when aclsize != 0 there's an actual segment for ACL
 565                  * in reply buffer.
 566                  */
 567                 if (md->body->aclsize) {
 568                         rc = mdc_unpack_acl(req, md);
 569                         if (rc)
 570                                 GOTO(out, rc);
 571 #ifdef CONFIG_FS_POSIX_ACL
 572                 } else {
 573                         md->posix_acl = NULL;
 574 #endif
 575                 }
 576         }
 577         if (md->body->valid & OBD_MD_FLMDSCAPA) {
 578                 struct obd_capa *oc = NULL;
 579
 580                 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA1, &oc);
 581                 if (rc)
 582                         GOTO(out, rc);
 583                 md->mds_capa = oc;
 584         }
 585
 586         if (md->body->valid & OBD_MD_FLOSSCAPA) {
 587                 struct obd_capa *oc = NULL;
 588
 589                 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA2, &oc);
 590                 if (rc)
 591                         GOTO(out, rc);
 592                 md->oss_capa = oc;
 593         }
 594
 595         EXIT;
 596 out:
 597         if (rc) {
 598                 if (md->oss_capa) {
 599                         capa_put(md->oss_capa);
 600                         md->oss_capa = NULL;
 601                 }
 602                 if (md->mds_capa) {
 603                         capa_put(md->mds_capa);
 604                         md->mds_capa = NULL;
 605                 }
 606 #ifdef CONFIG_FS_POSIX_ACL
 607                 posix_acl_release(md->posix_acl);
 608 #endif
 609                 if (md->lsm)
 610                         obd_free_memmd(dt_exp, &md->lsm);
 611         }
 612         return rc;
 613 }
 614
 615 int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
 616 {
 617         ENTRY;
 618         RETURN(0);
 619 }
 620
 621 /**
 622  * Handles both OPEN and SETATTR RPCs for OPEN-CLOSE and SETATTR-DONE_WRITING
 623  * RPC chains.
 624  */
 625 void mdc_replay_open(struct ptlrpc_request *req)
 626 {
 627         struct md_open_data *mod = req->rq_cb_data;
 628         struct ptlrpc_request *close_req;
 629         struct obd_client_handle *och;
 630         struct lustre_handle old;
 631         struct mdt_body *body;
 632         ENTRY;
 633
 634         if (mod == NULL) {
 635                 DEBUG_REQ(D_ERROR, req,
 636                           "Can't properly replay without open data.");
 637                 EXIT;
 638                 return;
 639         }
 640
 641         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 642         LASSERT(body != NULL);
 643
 644         och = mod->mod_och;
 645         if (och != NULL) {
 646                 struct lustre_handle *file_fh;
 647
 648                 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
 649
 650                 file_fh = &och->och_fh;
 651                 CDEBUG(D_HA, "updating handle from "LPX64" to "LPX64"\n",
 652                        file_fh->cookie, body->handle.cookie);
 653                 old = *file_fh;
 654                 *file_fh = body->handle;
 655         }
 656         close_req = mod->mod_close_req;
 657         if (close_req != NULL) {
 658                 __u32 opc = lustre_msg_get_opc(close_req->rq_reqmsg);
 659                 struct mdt_ioepoch *epoch;
 660
 661                 LASSERT(opc == MDS_CLOSE || opc == MDS_DONE_WRITING);
 662                 epoch = req_capsule_client_get(&close_req->rq_pill,
 663                                                &RMF_MDT_EPOCH);
 664                 LASSERT(epoch);
 665
 666                 if (och != NULL)
 667                         LASSERT(!memcmp(&old, &epoch->handle, sizeof(old)));
 668                 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
 669                 epoch->handle = body->handle;
 670         }
 671         EXIT;
 672 }
 673
 674 void mdc_commit_open(struct ptlrpc_request *req)
 675 {
 676         struct md_open_data *mod = req->rq_cb_data;
 677         if (mod == NULL)
 678                 return;
 679
 680         /**
 681          * No need to touch md_open_data::mod_och, it holds a reference on
 682          * \var mod and will zero references to each other, \var mod will be
 683          * freed after that when md_open_data::mod_och will put the reference.
 684          */
 685
 686         /**
 687          * Do not let open request to disappear as it still may be needed
 688          * for close rpc to happen (it may happen on evict only, otherwise
 689          * ptlrpc_request::rq_replay does not let mdc_commit_open() to be
 690          * called), just mark this rpc as committed to distinguish these 2
 691          * cases, see mdc_close() for details. The open request reference will
 692          * be put along with freeing \var mod.
 693          */
 694         ptlrpc_request_addref(req);
 695         spin_lock(&req->rq_lock);
 696         req->rq_committed = 1;
 697         spin_unlock(&req->rq_lock);
 698         req->rq_cb_data = NULL;
 699         obd_mod_put(mod);
 700 }
 701
 702 int mdc_set_open_replay_data(struct obd_export *exp,
 703                              struct obd_client_handle *och,
 704                              struct ptlrpc_request *open_req)
 705 {
 706         struct md_open_data   *mod;
 707         struct mdt_rec_create *rec;
 708         struct mdt_body       *body;
 709         struct obd_import     *imp = open_req->rq_import;
 710         ENTRY;
 711
 712         if (!open_req->rq_replay)
 713                 RETURN(0);
 714
 715         rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
 716         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
 717         LASSERT(rec != NULL);
 718         /* Incoming message in my byte order (it's been swabbed). */
 719         /* Outgoing messages always in my byte order. */
 720         LASSERT(body != NULL);
 721
 722         /* Only if the import is replayable, we set replay_open data */
 723         if (och && imp->imp_replayable) {
 724                 mod = obd_mod_alloc();
 725                 if (mod == NULL) {
 726                         DEBUG_REQ(D_ERROR, open_req,
 727                                   "Can't allocate md_open_data");
 728                         RETURN(0);
 729                 }
 730
 731                 /**
 732                  * Take a reference on \var mod, to be freed on mdc_close().
 733                  * It protects \var mod from being freed on eviction (commit
 734                  * callback is called despite rq_replay flag).
 735                  * Another reference for \var och.
 736                  */
 737                 obd_mod_get(mod);
 738                 obd_mod_get(mod);
 739
 740                 spin_lock(&open_req->rq_lock);
 741                 och->och_mod = mod;
 742                 mod->mod_och = och;
 743                 mod->mod_open_req = open_req;
 744                 open_req->rq_cb_data = mod;
 745                 open_req->rq_commit_cb = mdc_commit_open;
 746                 spin_unlock(&open_req->rq_lock);
 747         }
 748
 749         rec->cr_fid2 = body->fid1;
 750         rec->cr_ioepoch = body->ioepoch;
 751         rec->cr_old_handle.cookie = body->handle.cookie;
 752         open_req->rq_replay_cb = mdc_replay_open;
 753         if (!fid_is_sane(&body->fid1)) {
 754                 DEBUG_REQ(D_ERROR, open_req, "Saving replay request with "
 755                           "insane fid");
 756                 LBUG();
 757         }
 758
 759         DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
 760         RETURN(0);
 761 }
 762
 763 int mdc_clear_open_replay_data(struct obd_export *exp,
 764                                struct obd_client_handle *och)
 765 {
 766         struct md_open_data *mod = och->och_mod;
 767         ENTRY;
 768
 769         /**
 770          * It is possible to not have \var mod in a case of eviction between
 771          * lookup and ll_file_open().
 772          **/
 773         if (mod == NULL)
 774                 RETURN(0);
 775
 776         LASSERT(mod != LP_POISON);
 777
 778         mod->mod_och = NULL;
 779         och->och_mod = NULL;
 780         obd_mod_put(mod);
 781
 782         RETURN(0);
 783 }
 784
 785 /* Prepares the request for the replay by the given reply */
 786 static void mdc_close_handle_reply(struct ptlrpc_request *req,
 787                                    struct md_op_data *op_data, int rc) {
 788         struct mdt_body  *repbody;
 789         struct mdt_ioepoch *epoch;
 790
 791         if (req && rc == -EAGAIN) {
 792                 repbody = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 793                 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
 794
 795                 epoch->flags |= MF_SOM_AU;
 796                 if (repbody->valid & OBD_MD_FLGETATTRLOCK)
 797                         op_data->op_flags |= MF_GETATTR_LOCK;
 798         }
 799 }
 800
 801 int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
 802               struct md_open_data *mod, struct ptlrpc_request **request)
 803 {
 804         struct obd_device     *obd = class_exp2obd(exp);
 805         struct ptlrpc_request *req;
 806         int                    rc;
 807         ENTRY;
 808
 809         *request = NULL;
 810         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_CLOSE);
 811         if (req == NULL)
 812                 RETURN(-ENOMEM);
 813
 814         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 815
 816         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
 817         if (rc) {
 818                 ptlrpc_request_free(req);
 819                 RETURN(rc);
 820         }
 821
 822         /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
 823          * portal whose threads are not taking any DLM locks and are therefore
 824          * always progressing */
 825         req->rq_request_portal = MDS_READPAGE_PORTAL;
 826         ptlrpc_at_set_req_timeout(req);
 827
 828         /* Ensure that this close's handle is fixed up during replay. */
 829         if (likely(mod != NULL)) {
 830                 LASSERTF(mod->mod_open_req != NULL &&
 831                          mod->mod_open_req->rq_type != LI_POISON,
 832                          "POISONED open %p!\n", mod->mod_open_req);
 833
 834                 mod->mod_close_req = req;
 835
 836                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
 837                 /* We no longer want to preserve this open for replay even
 838                  * though the open was committed. b=3632, b=3633 */
 839                 spin_lock(&mod->mod_open_req->rq_lock);
 840                 mod->mod_open_req->rq_replay = 0;
 841                 spin_unlock(&mod->mod_open_req->rq_lock);
 842         } else {
 843                  CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
 844         }
 845
 846         mdc_close_pack(req, op_data);
 847
 848         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 849                              obd->u.cli.cl_max_mds_easize);
 850         req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
 851                              obd->u.cli.cl_max_mds_cookiesize);
 852
 853         ptlrpc_request_set_replen(req);
 854
 855         mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 856         rc = ptlrpc_queue_wait(req);
 857         mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 858
 859         if (req->rq_repmsg == NULL) {
 860                 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
 861                        req->rq_status);
 862                 if (rc == 0)
 863                         rc = req->rq_status ?: -EIO;
 864         } else if (rc == 0 || rc == -EAGAIN) {
 865                 struct mdt_body *body;
 866
 867                 rc = lustre_msg_get_status(req->rq_repmsg);
 868                 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
 869                         DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
 870                                   "= %d", rc);
 871                         if (rc > 0)
 872                                 rc = -rc;
 873                 }
 874                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 875                 if (body == NULL)
 876                         rc = -EPROTO;
 877         } else if (rc == -ESTALE) {
 878                 /**
 879                  * it can be allowed error after 3633 if open was committed and
 880                  * server failed before close was sent. Let's check if mod
 881                  * exists and return no error in that case
 882                  */
 883                 if (mod) {
 884                         DEBUG_REQ(D_HA, req, "Reset ESTALE = %d", rc);
 885                         LASSERT(mod->mod_open_req != NULL);
 886                         if (mod->mod_open_req->rq_committed)
 887                                 rc = 0;
 888                 }
 889         }
 890
 891         if (mod) {
 892                 if (rc != 0)
 893                         mod->mod_close_req = NULL;
 894                 /* Since now, mod is accessed through open_req only,
 895                  * thus close req does not keep a reference on mod anymore. */
 896                 obd_mod_put(mod);
 897         }
 898         *request = req;
 899         mdc_close_handle_reply(req, op_data, rc);
 900         RETURN(rc);
 901 }
 902
 903 int mdc_done_writing(struct obd_export *exp, struct md_op_data *op_data,
 904                      struct md_open_data *mod)
 905 {
 906         struct obd_device     *obd = class_exp2obd(exp);
 907         struct ptlrpc_request *req;
 908         int                    rc;
 909         ENTRY;
 910
 911         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 912                                    &RQF_MDS_DONE_WRITING);
 913         if (req == NULL)
 914                 RETURN(-ENOMEM);
 915
 916         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 917         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_DONE_WRITING);
 918         if (rc) {
 919                 ptlrpc_request_free(req);
 920                 RETURN(rc);
 921         }
 922
 923         if (mod != NULL) {
 924                 LASSERTF(mod->mod_open_req != NULL &&
 925                          mod->mod_open_req->rq_type != LI_POISON,
 926                          "POISONED setattr %p!\n", mod->mod_open_req);
 927
 928                 mod->mod_close_req = req;
 929                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched setattr");
 930                 /* We no longer want to preserve this setattr for replay even
 931                  * though the open was committed. b=3632, b=3633 */
 932                 spin_lock(&mod->mod_open_req->rq_lock);
 933                 mod->mod_open_req->rq_replay = 0;
 934                 spin_unlock(&mod->mod_open_req->rq_lock);
 935         }
 936
 937         mdc_close_pack(req, op_data);
 938         ptlrpc_request_set_replen(req);
 939
 940         mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 941         rc = ptlrpc_queue_wait(req);
 942         mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 943
 944         if (rc == -ESTALE) {
 945                 /**
 946                  * it can be allowed error after 3633 if open or setattr were
 947                  * committed and server failed before close was sent.
 948                  * Let's check if mod exists and return no error in that case
 949                  */
 950                 if (mod) {
 951                         LASSERT(mod->mod_open_req != NULL);
 952                         if (mod->mod_open_req->rq_committed)
 953                                 rc = 0;
 954                 }
 955         }
 956
 957         if (mod) {
 958                 if (rc != 0)
 959                         mod->mod_close_req = NULL;
 960                 /* Since now, mod is accessed through setattr req only,
 961                  * thus DW req does not keep a reference on mod anymore. */
 962                 obd_mod_put(mod);
 963         }
 964
 965         mdc_close_handle_reply(req, op_data, rc);
 966         ptlrpc_req_finished(req);
 967         RETURN(rc);
 968 }
 969
 970 #ifdef HAVE_SPLIT_SUPPORT
 971 int mdc_sendpage(struct obd_export *exp, const struct lu_fid *fid,
 972                  const struct page *page, int offset)
 973 {
 974         struct ptlrpc_request   *req;
 975         struct ptlrpc_bulk_desc *desc;
 976         int                      rc;
 977         ENTRY;
 978
 979         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_WRITEPAGE);
 980         if (req == NULL)
 981                 RETURN(-ENOMEM);
 982
 983         /* FIXME: capa doesn't support split yet */
 984         mdc_set_capa_size(req, &RMF_CAPA1, NULL);
 985
 986         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_WRITEPAGE);
 987         if (rc) {
 988                 ptlrpc_request_free(req);
 989                 RETURN(rc);
 990         }
 991
 992         req->rq_request_portal = MDS_READPAGE_PORTAL;
 993         ptlrpc_at_set_req_timeout(req);
 994
 995         desc = ptlrpc_prep_bulk_imp(req, 1, BULK_GET_SOURCE, MDS_BULK_PORTAL);
 996         if (desc == NULL)
 997                 GOTO(out, rc = -ENOMEM);
 998
 999         /* NB req now owns desc and will free it when it gets freed. */
1000         ptlrpc_prep_bulk_page(desc, (struct page *)page, 0, offset);
1001         mdc_readdir_pack(req, 0, offset, fid, NULL);
1002
1003         ptlrpc_request_set_replen(req);
1004         rc = ptlrpc_queue_wait(req);
1005         if (rc)
1006                 GOTO(out, rc);
1007
1008         rc = sptlrpc_cli_unwrap_bulk_write(req, req->rq_bulk);
1009 out:
1010         ptlrpc_req_finished(req);
1011         return rc;
1012 }
1013 EXPORT_SYMBOL(mdc_sendpage);
1014 #endif
1015
1016 int mdc_readpage(struct obd_export *exp, struct md_op_data *op_data,
1017                  struct page **pages, struct ptlrpc_request **request)
1018 {
1019         struct ptlrpc_request   *req;
1020         struct ptlrpc_bulk_desc *desc;
1021         int                      i;
1022         cfs_waitq_t              waitq;
1023         int                      resends = 0;
1024         struct l_wait_info       lwi;
1025         int                      rc;
1026         ENTRY;
1027
1028         *request = NULL;
1029         cfs_waitq_init(&waitq);
1030
1031 restart_bulk:
1032         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
1033         if (req == NULL)
1034                 RETURN(-ENOMEM);
1035
1036         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1037
1038         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
1039         if (rc) {
1040                 ptlrpc_request_free(req);
1041                 RETURN(rc);
1042         }
1043
1044         req->rq_request_portal = MDS_READPAGE_PORTAL;
1045         ptlrpc_at_set_req_timeout(req);
1046
1047         desc = ptlrpc_prep_bulk_imp(req, op_data->op_npages, BULK_PUT_SINK,
1048                                     MDS_BULK_PORTAL);
1049         if (desc == NULL) {
1050                 ptlrpc_request_free(req);
1051                 RETURN(-ENOMEM);
1052         }
1053
1054         /* NB req now owns desc and will free it when it gets freed */
1055         for (i = 0; i < op_data->op_npages; i++)
1056                 ptlrpc_prep_bulk_page_pin(desc, pages[i], 0, CFS_PAGE_SIZE);
1057
1058         mdc_readdir_pack(req, op_data->op_offset,
1059                          CFS_PAGE_SIZE * op_data->op_npages,
1060                          &op_data->op_fid1, op_data->op_capa1);
1061
1062         ptlrpc_request_set_replen(req);
1063         rc = ptlrpc_queue_wait(req);
1064         if (rc) {
1065                 ptlrpc_req_finished(req);
1066                 if (rc != -ETIMEDOUT)
1067                         RETURN(rc);
1068
1069                 resends++;
1070                 if (!client_should_resend(resends, &exp->exp_obd->u.cli)) {
1071                         CERROR("too many resend retries, returning error\n");
1072                         RETURN(-EIO);
1073                 }
1074                 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(resends), NULL, NULL, NULL);
1075                 l_wait_event(waitq, 0, &lwi);
1076
1077                 goto restart_bulk;
1078         }
1079
1080         rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
1081                                           req->rq_bulk->bd_nob_transferred);
1082         if (rc < 0) {
1083                 ptlrpc_req_finished(req);
1084                 RETURN(rc);
1085         }
1086
1087         if (req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK) {
1088                 CERROR("Unexpected # bytes transferred: %d (%ld expected)\n",
1089                         req->rq_bulk->bd_nob_transferred,
1090                         CFS_PAGE_SIZE * op_data->op_npages);
1091                 ptlrpc_req_finished(req);
1092                 RETURN(-EPROTO);
1093         }
1094
1095         *request = req;
1096         RETURN(0);
1097 }
1098
1099 static int mdc_statfs(const struct lu_env *env,
1100                       struct obd_export *exp, struct obd_statfs *osfs,
1101                       __u64 max_age, __u32 flags)
1102 {
1103         struct obd_device     *obd = class_exp2obd(exp);
1104         struct ptlrpc_request *req;
1105         struct obd_statfs     *msfs;
1106         struct obd_import     *imp = NULL;
1107         int                    rc;
1108         ENTRY;
1109
1110         /*
1111          * Since the request might also come from lprocfs, so we need
1112          * sync this with client_disconnect_export Bug15684
1113          */
1114         down_read(&obd->u.cli.cl_sem);
1115         if (obd->u.cli.cl_import)
1116                 imp = class_import_get(obd->u.cli.cl_import);
1117         up_read(&obd->u.cli.cl_sem);
1118         if (!imp)
1119                 RETURN(-ENODEV);
1120
1121         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_STATFS,
1122                                         LUSTRE_MDS_VERSION, MDS_STATFS);
1123         if (req == NULL)
1124                 GOTO(output, rc = -ENOMEM);
1125
1126         ptlrpc_request_set_replen(req);
1127
1128         if (flags & OBD_STATFS_NODELAY) {
1129                 /* procfs requests not want stay in wait for avoid deadlock */
1130                 req->rq_no_resend = 1;
1131                 req->rq_no_delay = 1;
1132         }
1133
1134         rc = ptlrpc_queue_wait(req);
1135         if (rc) {
1136                 /* check connection error first */
1137                 if (imp->imp_connect_error)
1138                         rc = imp->imp_connect_error;
1139                 GOTO(out, rc);
1140         }
1141
1142         msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1143         if (msfs == NULL)
1144                 GOTO(out, rc = -EPROTO);
1145
1146         *osfs = *msfs;
1147         EXIT;
1148 out:
1149         ptlrpc_req_finished(req);
1150 output:
1151         class_import_put(imp);
1152         return rc;
1153 }
1154
1155 static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
1156 {
1157         __u32 keylen, vallen;
1158         void *key;
1159         int rc;
1160
1161         if (gf->gf_pathlen > PATH_MAX)
1162                 RETURN(-ENAMETOOLONG);
1163         if (gf->gf_pathlen < 2)
1164                 RETURN(-EOVERFLOW);
1165
1166         /* Key is KEY_FID2PATH + getinfo_fid2path description */
1167         keylen = cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf);
1168         OBD_ALLOC(key, keylen);
1169         if (key == NULL)
1170                 RETURN(-ENOMEM);
1171         memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
1172         memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf));
1173
1174         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
1175                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
1176
1177         if (!fid_is_sane(&gf->gf_fid))
1178                 GOTO(out, rc = -EINVAL);
1179
1180         /* Val is struct getinfo_fid2path result plus path */
1181         vallen = sizeof(*gf) + gf->gf_pathlen;
1182
1183         rc = obd_get_info(NULL, exp, keylen, key, &vallen, gf, NULL);
1184         if (rc)
1185                 GOTO(out, rc);
1186
1187         if (vallen <= sizeof(*gf))
1188                 GOTO(out, rc = -EPROTO);
1189         else if (vallen > sizeof(*gf) + gf->gf_pathlen)
1190                 GOTO(out, rc = -EOVERFLOW);
1191
1192         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n%s\n",
1193                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno, gf->gf_path);
1194
1195 out:
1196         OBD_FREE(key, keylen);
1197         return rc;
1198 }
1199
1200 static int mdc_ioc_hsm_progress(struct obd_export *exp,
1201                                 struct hsm_progress_kernel *hpk)
1202 {
1203         struct obd_import               *imp = class_exp2cliimp(exp);
1204         struct hsm_progress_kernel      *req_hpk;
1205         struct ptlrpc_request           *req;
1206         int                              rc;
1207         ENTRY;
1208
1209         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_PROGRESS,
1210                                         LUSTRE_MDS_VERSION, MDS_HSM_PROGRESS);
1211         if (req == NULL)
1212                 GOTO(out, rc = -ENOMEM);
1213
1214         /* Copy hsm_progress struct */
1215         req_hpk = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_PROGRESS);
1216         LASSERT(req_hpk);
1217         *req_hpk = *hpk;
1218
1219         ptlrpc_request_set_replen(req);
1220
1221         rc = ptlrpc_queue_wait(req);
1222         GOTO(out, rc);
1223 out:
1224         ptlrpc_req_finished(req);
1225         return rc;
1226 }
1227
1228 static int mdc_ioc_hsm_ct_register(struct obd_import *imp, __u32 archives)
1229 {
1230         __u32                   *archive_mask;
1231         struct ptlrpc_request   *req;
1232         int                      rc;
1233         ENTRY;
1234
1235         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_REGISTER,
1236                                         LUSTRE_MDS_VERSION,
1237                                         MDS_HSM_CT_REGISTER);
1238         if (req == NULL)
1239                 GOTO(out, rc = -ENOMEM);
1240
1241         /* Copy hsm_progress struct */
1242         archive_mask = req_capsule_client_get(&req->rq_pill,
1243                                               &RMF_MDS_HSM_ARCHIVE);
1244         LASSERT(archive_mask);
1245         *archive_mask = archives;
1246
1247         ptlrpc_request_set_replen(req);
1248
1249         rc = ptlrpc_queue_wait(req);
1250         GOTO(out, rc);
1251 out:
1252         ptlrpc_req_finished(req);
1253         return rc;
1254 }
1255
1256 static int mdc_ioc_hsm_current_action(struct obd_export *exp,
1257                                       struct md_op_data *op_data)
1258 {
1259         struct hsm_current_action       *hca = op_data->op_data;
1260         struct hsm_current_action       *req_hca;
1261         struct ptlrpc_request           *req;
1262         int                              rc;
1263         ENTRY;
1264
1265         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1266                                    &RQF_MDS_HSM_ACTION);
1267         if (req == NULL)
1268                 RETURN(-ENOMEM);
1269
1270         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1271
1272         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_ACTION);
1273         if (rc) {
1274                 ptlrpc_request_free(req);
1275                 RETURN(rc);
1276         }
1277
1278         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
1279                       OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
1280
1281         ptlrpc_request_set_replen(req);
1282
1283         rc = ptlrpc_queue_wait(req);
1284         if (rc)
1285                 GOTO(out, rc);
1286
1287         req_hca = req_capsule_server_get(&req->rq_pill,
1288                                          &RMF_MDS_HSM_CURRENT_ACTION);
1289         if (req_hca == NULL)
1290                 GOTO(out, rc = -EPROTO);
1291
1292         *hca = *req_hca;
1293
1294         EXIT;
1295 out:
1296         ptlrpc_req_finished(req);
1297         return rc;
1298 }
1299
1300 static int mdc_ioc_hsm_ct_unregister(struct obd_import *imp)
1301 {
1302         struct ptlrpc_request   *req;
1303         int                      rc;
1304         ENTRY;
1305
1306         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_UNREGISTER,
1307                                         LUSTRE_MDS_VERSION,
1308                                         MDS_HSM_CT_UNREGISTER);
1309         if (req == NULL)
1310                 GOTO(out, rc = -ENOMEM);
1311
1312         ptlrpc_request_set_replen(req);
1313
1314         rc = ptlrpc_queue_wait(req);
1315         GOTO(out, rc);
1316 out:
1317         ptlrpc_req_finished(req);
1318         return rc;
1319 }
1320
1321 static int mdc_ioc_hsm_state_get(struct obd_export *exp,
1322                                  struct md_op_data *op_data)
1323 {
1324         struct hsm_user_state   *hus = op_data->op_data;
1325         struct hsm_user_state   *req_hus;
1326         struct ptlrpc_request   *req;
1327         int                      rc;
1328         ENTRY;
1329
1330         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1331                                    &RQF_MDS_HSM_STATE_GET);
1332         if (req == NULL)
1333                 RETURN(-ENOMEM);
1334
1335         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1336
1337         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_GET);
1338         if (rc != 0) {
1339                 ptlrpc_request_free(req);
1340                 RETURN(rc);
1341         }
1342
1343         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
1344                       OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
1345
1346         ptlrpc_request_set_replen(req);
1347
1348         rc = ptlrpc_queue_wait(req);
1349         if (rc)
1350                 GOTO(out, rc);
1351
1352         req_hus = req_capsule_server_get(&req->rq_pill, &RMF_HSM_USER_STATE);
1353         if (req_hus == NULL)
1354                 GOTO(out, rc = -EPROTO);
1355
1356         *hus = *req_hus;
1357
1358         EXIT;
1359 out:
1360         ptlrpc_req_finished(req);
1361         return rc;
1362 }
1363
1364 static int mdc_ioc_hsm_state_set(struct obd_export *exp,
1365                                  struct md_op_data *op_data)
1366 {
1367         struct hsm_state_set    *hss = op_data->op_data;
1368         struct hsm_state_set    *req_hss;
1369         struct ptlrpc_request   *req;
1370         int                      rc;
1371         ENTRY;
1372
1373         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1374                                    &RQF_MDS_HSM_STATE_SET);
1375         if (req == NULL)
1376                 RETURN(-ENOMEM);
1377
1378         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1379
1380         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_SET);
1381         if (rc) {
1382                 ptlrpc_request_free(req);
1383                 RETURN(rc);
1384         }
1385
1386         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
1387                       OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
1388
1389         /* Copy states */
1390         req_hss = req_capsule_client_get(&req->rq_pill, &RMF_HSM_STATE_SET);
1391         LASSERT(req_hss);
1392         *req_hss = *hss;
1393
1394         ptlrpc_request_set_replen(req);
1395
1396         rc = ptlrpc_queue_wait(req);
1397         GOTO(out, rc);
1398
1399         EXIT;
1400 out:
1401         ptlrpc_req_finished(req);
1402         return rc;
1403 }
1404
1405 static int mdc_ioc_hsm_request(struct obd_export *exp,
1406                                struct hsm_user_request *hur)
1407 {
1408         struct obd_import       *imp = class_exp2cliimp(exp);
1409         struct ptlrpc_request   *req;
1410         struct hsm_request      *req_hr;
1411         struct hsm_user_item    *req_hui;
1412         char                    *req_opaque;
1413         int                      rc;
1414         ENTRY;
1415
1416         req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_REQUEST);
1417         if (req == NULL)
1418                 GOTO(out, rc = -ENOMEM);
1419
1420         req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM, RCL_CLIENT,
1421                              hur->hur_request.hr_itemcount
1422                              * sizeof(struct hsm_user_item));
1423         req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA, RCL_CLIENT,
1424                              hur->hur_request.hr_data_len);
1425
1426         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_REQUEST);
1427         if (rc) {
1428                 ptlrpc_request_free(req);
1429                 RETURN(rc);
1430         }
1431
1432         mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, 0, 0);
1433
1434         /* Copy hsm_request struct */
1435         req_hr = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_REQUEST);
1436         LASSERT(req_hr);
1437         *req_hr = hur->hur_request;
1438
1439         /* Copy hsm_user_item structs */
1440         req_hui = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM);
1441         LASSERT(req_hui);
1442         memcpy(req_hui, hur->hur_user_item,
1443                hur->hur_request.hr_itemcount * sizeof(struct hsm_user_item));
1444
1445         /* Copy opaque field */
1446         req_opaque = req_capsule_client_get(&req->rq_pill, &RMF_GENERIC_DATA);
1447         LASSERT(req_opaque);
1448         memcpy(req_opaque, hur_data(hur), hur->hur_request.hr_data_len);
1449
1450         ptlrpc_request_set_replen(req);
1451
1452         rc = ptlrpc_queue_wait(req);
1453         GOTO(out, rc);
1454
1455 out:
1456         ptlrpc_req_finished(req);
1457         return rc;
1458 }
1459
1460 static struct kuc_hdr *changelog_kuc_hdr(char *buf, int len, int flags)
1461 {
1462         struct kuc_hdr *lh = (struct kuc_hdr *)buf;
1463
1464         LASSERT(len <= CR_MAXSIZE);
1465
1466         lh->kuc_magic = KUC_MAGIC;
1467         lh->kuc_transport = KUC_TRANSPORT_CHANGELOG;
1468         lh->kuc_flags = flags;
1469         lh->kuc_msgtype = CL_RECORD;
1470         lh->kuc_msglen = len;
1471         return lh;
1472 }
1473
1474 #define D_CHANGELOG 0
1475
1476 struct changelog_show {
1477         __u64       cs_startrec;
1478         __u32       cs_flags;
1479         cfs_file_t *cs_fp;
1480         char       *cs_buf;
1481         struct obd_device *cs_obd;
1482 };
1483
1484 static int changelog_show_cb(const struct lu_env *env, struct llog_handle *llh,
1485                              struct llog_rec_hdr *hdr, void *data)
1486 {
1487         struct changelog_show *cs = data;
1488         struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
1489         struct kuc_hdr *lh;
1490         int len, rc;
1491         ENTRY;
1492
1493         if ((rec->cr_hdr.lrh_type != CHANGELOG_REC) ||
1494             (rec->cr.cr_type >= CL_LAST)) {
1495                 CERROR("Not a changelog rec %d/%d\n", rec->cr_hdr.lrh_type,
1496                        rec->cr.cr_type);
1497                 RETURN(-EINVAL);
1498         }
1499
1500         if (rec->cr.cr_index < cs->cs_startrec) {
1501                 /* Skip entries earlier than what we are interested in */
1502                 CDEBUG(D_CHANGELOG, "rec="LPU64" start="LPU64"\n",
1503                        rec->cr.cr_index, cs->cs_startrec);
1504                 RETURN(0);
1505         }
1506
1507         CDEBUG(D_CHANGELOG, LPU64" %02d%-5s "LPU64" 0x%x t="DFID" p="DFID
1508                 " %.*s\n", rec->cr.cr_index, rec->cr.cr_type,
1509                 changelog_type2str(rec->cr.cr_type), rec->cr.cr_time,
1510                 rec->cr.cr_flags & CLF_FLAGMASK,
1511                 PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid),
1512                 rec->cr.cr_namelen, changelog_rec_name(&rec->cr));
1513
1514         len = sizeof(*lh) + changelog_rec_size(&rec->cr) + rec->cr.cr_namelen;
1515
1516         /* Set up the message */
1517         lh = changelog_kuc_hdr(cs->cs_buf, len, cs->cs_flags);
1518         memcpy(lh + 1, &rec->cr, len - sizeof(*lh));
1519
1520         rc = libcfs_kkuc_msg_put(cs->cs_fp, lh);
1521         CDEBUG(D_CHANGELOG, "kucmsg fp %p len %d rc %d\n", cs->cs_fp, len,rc);
1522
1523         RETURN(rc);
1524 }
1525
1526 static int mdc_changelog_send_thread(void *csdata)
1527 {
1528         struct changelog_show *cs = csdata;
1529         struct llog_ctxt *ctxt = NULL;
1530         struct llog_handle *llh = NULL;
1531         struct kuc_hdr *kuch;
1532         int rc;
1533
1534         CDEBUG(D_CHANGELOG, "changelog to fp=%p start "LPU64"\n",
1535                cs->cs_fp, cs->cs_startrec);
1536
1537         /*
1538          * It's important to daemonize here to close unused FDs.
1539          * The write fd from pipe is already opened by the caller,
1540          * so it's fine to clear all files here
1541          */
1542         cfs_daemonize("mdc_clg_send_thread");
1543
1544         OBD_ALLOC(cs->cs_buf, CR_MAXSIZE);
1545         if (cs->cs_buf == NULL)
1546                 GOTO(out, rc = -ENOMEM);
1547
1548         /* Set up the remote catalog handle */
1549         ctxt = llog_get_context(cs->cs_obd, LLOG_CHANGELOG_REPL_CTXT);
1550         if (ctxt == NULL)
1551                 GOTO(out, rc = -ENOENT);
1552         rc = llog_open(NULL, ctxt, &llh, NULL, CHANGELOG_CATALOG,
1553                        LLOG_OPEN_EXISTS);
1554         if (rc) {
1555                 CERROR("%s: fail to open changelog catalog: rc = %d\n",
1556                        cs->cs_obd->obd_name, rc);
1557                 GOTO(out, rc);
1558         }
1559         rc = llog_init_handle(NULL, llh, LLOG_F_IS_CAT, NULL);
1560         if (rc) {
1561                 CERROR("llog_init_handle failed %d\n", rc);
1562                 GOTO(out, rc);
1563         }
1564
1565         rc = llog_cat_process(NULL, llh, changelog_show_cb, cs, 0, 0);
1566
1567         /* Send EOF no matter what our result */
1568         if ((kuch = changelog_kuc_hdr(cs->cs_buf, sizeof(*kuch),
1569                                       cs->cs_flags))) {
1570                 kuch->kuc_msgtype = CL_EOF;
1571                 libcfs_kkuc_msg_put(cs->cs_fp, kuch);
1572         }
1573
1574 out:
1575         cfs_put_file(cs->cs_fp);
1576         if (llh)
1577                 llog_cat_close(NULL, llh);
1578         if (ctxt)
1579                 llog_ctxt_put(ctxt);
1580         if (cs->cs_buf)
1581                 OBD_FREE(cs->cs_buf, CR_MAXSIZE);
1582         OBD_FREE_PTR(cs);
1583         /* detach from parent process so we get cleaned up */
1584         cfs_daemonize("cl_send");
1585         return rc;
1586 }
1587
1588 static int mdc_ioc_changelog_send(struct obd_device *obd,
1589                                   struct ioc_changelog *icc)
1590 {
1591         struct changelog_show *cs;
1592         int rc;
1593
1594         /* Freed in mdc_changelog_send_thread */
1595         OBD_ALLOC_PTR(cs);
1596         if (!cs)
1597                 return -ENOMEM;
1598
1599         cs->cs_obd = obd;
1600         cs->cs_startrec = icc->icc_recno;
1601         /* matching cfs_put_file in mdc_changelog_send_thread */
1602         cs->cs_fp = cfs_get_fd(icc->icc_id);
1603         cs->cs_flags = icc->icc_flags;
1604
1605         /* New thread because we should return to user app before
1606            writing into our pipe */
1607         rc = cfs_create_thread(mdc_changelog_send_thread, cs, CFS_DAEMON_FLAGS);
1608         if (rc >= 0) {
1609                 CDEBUG(D_CHANGELOG, "start changelog thread: %d\n", rc);
1610                 return 0;
1611         }
1612
1613         CERROR("Failed to start changelog thread: %d\n", rc);
1614         OBD_FREE_PTR(cs);
1615         return rc;
1616 }
1617
1618 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
1619                                 struct lustre_kernelcomm *lk);
1620
1621 static int mdc_quotacheck(struct obd_device *unused, struct obd_export *exp,
1622                           struct obd_quotactl *oqctl)
1623 {
1624         struct client_obd       *cli = &exp->exp_obd->u.cli;
1625         struct ptlrpc_request   *req;
1626         struct obd_quotactl     *body;
1627         int                      rc;
1628         ENTRY;
1629
1630         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
1631                                         &RQF_MDS_QUOTACHECK, LUSTRE_MDS_VERSION,
1632                                         MDS_QUOTACHECK);
1633         if (req == NULL)
1634                 RETURN(-ENOMEM);
1635
1636         body = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1637         *body = *oqctl;
1638
1639         ptlrpc_request_set_replen(req);
1640
1641         /* the next poll will find -ENODATA, that means quotacheck is
1642          * going on */
1643         cli->cl_qchk_stat = -ENODATA;
1644         rc = ptlrpc_queue_wait(req);
1645         if (rc)
1646                 cli->cl_qchk_stat = rc;
1647         ptlrpc_req_finished(req);
1648         RETURN(rc);
1649 }
1650
1651 static int mdc_quota_poll_check(struct obd_export *exp,
1652                                 struct if_quotacheck *qchk)
1653 {
1654         struct client_obd *cli = &exp->exp_obd->u.cli;
1655         int rc;
1656         ENTRY;
1657
1658         qchk->obd_uuid = cli->cl_target_uuid;
1659         memcpy(qchk->obd_type, LUSTRE_MDS_NAME, strlen(LUSTRE_MDS_NAME));
1660
1661         rc = cli->cl_qchk_stat;
1662         /* the client is not the previous one */
1663         if (rc == CL_NOT_QUOTACHECKED)
1664                 rc = -EINTR;
1665         RETURN(rc);
1666 }
1667
1668 static int mdc_quotactl(struct obd_device *unused, struct obd_export *exp,
1669                         struct obd_quotactl *oqctl)
1670 {
1671         struct ptlrpc_request   *req;
1672         struct obd_quotactl     *oqc;
1673         int                      rc;
1674         ENTRY;
1675
1676         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
1677                                         &RQF_MDS_QUOTACTL, LUSTRE_MDS_VERSION,
1678                                         MDS_QUOTACTL);
1679         if (req == NULL)
1680                 RETURN(-ENOMEM);
1681
1682         oqc = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1683         *oqc = *oqctl;
1684
1685         ptlrpc_request_set_replen(req);
1686         ptlrpc_at_set_req_timeout(req);
1687         req->rq_no_resend = 1;
1688
1689         rc = ptlrpc_queue_wait(req);
1690         if (rc)
1691                 CERROR("ptlrpc_queue_wait failed, rc: %d\n", rc);
1692
1693         if (req->rq_repmsg &&
1694             (oqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL))) {
1695                 *oqctl = *oqc;
1696         } else if (!rc) {
1697                 CERROR ("Can't unpack obd_quotactl\n");
1698                 rc = -EPROTO;
1699         }
1700         ptlrpc_req_finished(req);
1701
1702         RETURN(rc);
1703 }
1704
1705 static int mdc_ioc_swap_layouts(struct obd_export *exp,
1706                                 struct md_op_data *op_data)
1707 {
1708         CFS_LIST_HEAD(cancels);
1709         struct ptlrpc_request   *req;
1710         int                      rc, count;
1711         struct mdc_swap_layouts *msl, *payload;
1712         ENTRY;
1713
1714         msl = op_data->op_data;
1715
1716         /* When the MDT will get the MDS_SWAP_LAYOUTS RPC the
1717          * first thing it will do is to cancel the 2 layout
1718          * locks hold by this client.
1719          * So the client must cancel its layout locks on the 2 fids
1720          * with the request RPC to avoid extra RPC round trips
1721          */
1722         count = mdc_resource_get_unused(exp, &op_data->op_fid1, &cancels,
1723                                         LCK_CR, MDS_INODELOCK_LAYOUT);
1724         count += mdc_resource_get_unused(exp, &op_data->op_fid2, &cancels,
1725                                          LCK_CR, MDS_INODELOCK_LAYOUT);
1726
1727         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1728                                    &RQF_MDS_SWAP_LAYOUTS);
1729         if (req == NULL) {
1730                 ldlm_lock_list_put(&cancels, l_bl_ast, count);
1731                 RETURN(-ENOMEM);
1732         }
1733
1734         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
1735         mdc_set_capa_size(req, &RMF_CAPA2, op_data->op_capa2);
1736
1737         rc = mdc_prep_elc_req(exp, req, MDS_SWAP_LAYOUTS, &cancels, count);
1738         if (rc) {
1739                 ptlrpc_request_free(req);
1740                 RETURN(rc);
1741         }
1742
1743         mdc_swap_layouts_pack(req, op_data);
1744
1745         payload = req_capsule_client_get(&req->rq_pill, &RMF_SWAP_LAYOUTS);
1746         LASSERT(payload);
1747
1748         *payload = *msl;
1749
1750         ptlrpc_request_set_replen(req);
1751
1752         rc = ptlrpc_queue_wait(req);
1753         if (rc)
1754                 GOTO(out, rc);
1755         EXIT;
1756
1757 out:
1758         ptlrpc_req_finished(req);
1759         return rc;
1760 }
1761
1762 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1763                          void *karg, void *uarg)
1764 {
1765         struct obd_device *obd = exp->exp_obd;
1766         struct obd_ioctl_data *data = karg;
1767         struct obd_import *imp = obd->u.cli.cl_import;
1768         struct llog_ctxt *ctxt;
1769         int rc;
1770         ENTRY;
1771
1772         if (!cfs_try_module_get(THIS_MODULE)) {
1773                 CERROR("Can't get module. Is it alive?");
1774                 return -EINVAL;
1775         }
1776         switch (cmd) {
1777         case OBD_IOC_CHANGELOG_SEND:
1778                 rc = mdc_ioc_changelog_send(obd, karg);
1779                 GOTO(out, rc);
1780         case OBD_IOC_CHANGELOG_CLEAR: {
1781                 struct ioc_changelog *icc = karg;
1782                 struct changelog_setinfo cs =
1783                         {.cs_recno = icc->icc_recno, .cs_id = icc->icc_id};
1784                 rc = obd_set_info_async(NULL, exp, strlen(KEY_CHANGELOG_CLEAR),
1785                                         KEY_CHANGELOG_CLEAR, sizeof(cs), &cs,
1786                                         NULL);
1787                 GOTO(out, rc);
1788         }
1789         case OBD_IOC_FID2PATH:
1790                 rc = mdc_ioc_fid2path(exp, karg);
1791                 GOTO(out, rc);
1792         case LL_IOC_HSM_CT_START:
1793                 rc = mdc_ioc_hsm_ct_start(exp, karg);
1794                 GOTO(out, rc);
1795         case LL_IOC_HSM_PROGRESS:
1796                 rc = mdc_ioc_hsm_progress(exp, karg);
1797                 GOTO(out, rc);
1798         case LL_IOC_HSM_STATE_GET:
1799                 rc = mdc_ioc_hsm_state_get(exp, karg);
1800                 GOTO(out, rc);
1801         case LL_IOC_HSM_STATE_SET:
1802                 rc = mdc_ioc_hsm_state_set(exp, karg);
1803         case LL_IOC_HSM_ACTION:
1804                 rc = mdc_ioc_hsm_current_action(exp, karg);
1805                 GOTO(out, rc);
1806         case LL_IOC_HSM_REQUEST:
1807                 rc = mdc_ioc_hsm_request(exp, karg);
1808                 GOTO(out, rc);
1809         case OBD_IOC_CLIENT_RECOVER:
1810                 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1, 0);
1811                 if (rc < 0)
1812                         GOTO(out, rc);
1813                 GOTO(out, rc = 0);
1814         case IOC_OSC_SET_ACTIVE:
1815                 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
1816                 GOTO(out, rc);
1817         case OBD_IOC_PARSE: {
1818                 ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
1819                 rc = class_config_parse_llog(NULL, ctxt, data->ioc_inlbuf1,
1820                                              NULL);
1821                 llog_ctxt_put(ctxt);
1822                 GOTO(out, rc);
1823         }
1824 #ifdef __KERNEL__
1825         case OBD_IOC_LLOG_INFO:
1826         case OBD_IOC_LLOG_PRINT: {
1827                 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
1828                 rc = llog_ioctl(NULL, ctxt, cmd, data);
1829                 llog_ctxt_put(ctxt);
1830                 GOTO(out, rc);
1831         }
1832 #endif
1833         case OBD_IOC_POLL_QUOTACHECK:
1834                 rc = mdc_quota_poll_check(exp, (struct if_quotacheck *)karg);
1835                 GOTO(out, rc);
1836         case OBD_IOC_PING_TARGET:
1837                 rc = ptlrpc_obd_ping(obd);
1838                 GOTO(out, rc);
1839         /*
1840          * Normally IOC_OBD_STATFS, OBD_IOC_QUOTACTL iocontrol are handled by
1841          * LMV instead of MDC. But when the cluster is upgraded from 1.8,
1842          * there'd be no LMV layer thus we might be called here. Eventually
1843          * this code should be removed.
1844          * bz20731, LU-592.
1845          */
1846         case IOC_OBD_STATFS: {
1847                 struct obd_statfs stat_buf = {0};
1848
1849                 if (*((__u32 *) data->ioc_inlbuf2) != 0)
1850                         GOTO(out, rc = -ENODEV);
1851
1852                 /* copy UUID */
1853                 if (cfs_copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd),
1854                                      min((int) data->ioc_plen2,
1855                                          (int) sizeof(struct obd_uuid))))
1856                         GOTO(out, rc = -EFAULT);
1857
1858                 rc = mdc_statfs(NULL, obd->obd_self_export, &stat_buf,
1859                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1860                                 0);
1861                 if (rc != 0)
1862                         GOTO(out, rc);
1863
1864                 if (cfs_copy_to_user(data->ioc_pbuf1, &stat_buf,
1865                                      min((int) data->ioc_plen1,
1866                                          (int) sizeof(stat_buf))))
1867                         GOTO(out, rc = -EFAULT);
1868
1869                 GOTO(out, rc = 0);
1870         }
1871         case OBD_IOC_QUOTACTL: {
1872                 struct if_quotactl *qctl = karg;
1873                 struct obd_quotactl *oqctl;
1874
1875                 OBD_ALLOC_PTR(oqctl);
1876                 if (!oqctl)
1877                         RETURN(-ENOMEM);
1878
1879                 QCTL_COPY(oqctl, qctl);
1880                 rc = obd_quotactl(exp, oqctl);
1881                 if (rc == 0) {
1882                         QCTL_COPY(qctl, oqctl);
1883                         qctl->qc_valid = QC_MDTIDX;
1884                         qctl->obd_uuid = obd->u.cli.cl_target_uuid;
1885                 }
1886                 OBD_FREE_PTR(oqctl);
1887                 break;
1888         }
1889         case LL_IOC_GET_CONNECT_FLAGS: {
1890                 if (cfs_copy_to_user(uarg,
1891                                      exp_connect_flags_ptr(exp),
1892                                      sizeof(__u64)))
1893                         GOTO(out, rc = -EFAULT);
1894                 else
1895                         GOTO(out, rc = 0);
1896         }
1897         case LL_IOC_LOV_SWAP_LAYOUTS: {
1898                 rc = mdc_ioc_swap_layouts(exp, karg);
1899                 break;
1900         }
1901         default:
1902                 CERROR("mdc_ioctl(): unrecognised ioctl %#x\n", cmd);
1903                 GOTO(out, rc = -ENOTTY);
1904         }
1905 out:
1906         cfs_module_put(THIS_MODULE);
1907
1908         return rc;
1909 }
1910
1911 int mdc_get_info_rpc(struct obd_export *exp,
1912                      obd_count keylen, void *key,
1913                      int vallen, void *val)
1914 {
1915         struct obd_import      *imp = class_exp2cliimp(exp);
1916         struct ptlrpc_request  *req;
1917         char                   *tmp;
1918         int                     rc = -EINVAL;
1919         ENTRY;
1920
1921         req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
1922         if (req == NULL)
1923                 RETURN(-ENOMEM);
1924
1925         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
1926                              RCL_CLIENT, keylen);
1927         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
1928                              RCL_CLIENT, sizeof(__u32));
1929
1930         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
1931         if (rc) {
1932                 ptlrpc_request_free(req);
1933                 RETURN(rc);
1934         }
1935
1936         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
1937         memcpy(tmp, key, keylen);
1938         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
1939         memcpy(tmp, &vallen, sizeof(__u32));
1940
1941         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
1942                              RCL_SERVER, vallen);
1943         ptlrpc_request_set_replen(req);
1944
1945         rc = ptlrpc_queue_wait(req);
1946         if (rc == 0) {
1947                 tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
1948                 memcpy(val, tmp, vallen);
1949                 if (ptlrpc_rep_need_swab(req)) {
1950                         if (KEY_IS(KEY_FID2PATH)) {
1951                                 lustre_swab_fid2path(val);
1952                         }
1953                 }
1954         }
1955         ptlrpc_req_finished(req);
1956
1957         RETURN(rc);
1958 }
1959
1960 static void lustre_swab_hai(struct hsm_action_item *h)
1961 {
1962         __swab32s(&h->hai_len);
1963         __swab32s(&h->hai_action);
1964         lustre_swab_lu_fid(&h->hai_fid);
1965         lustre_swab_lu_fid(&h->hai_dfid);
1966         __swab64s(&h->hai_cookie);
1967         __swab64s(&h->hai_extent.offset);
1968         __swab64s(&h->hai_extent.length);
1969         __swab64s(&h->hai_gid);
1970 }
1971
1972 static void lustre_swab_hal(struct hsm_action_list *h)
1973 {
1974         struct hsm_action_item  *hai;
1975         int                      i;
1976
1977         __swab32s(&h->hal_version);
1978         __swab32s(&h->hal_count);
1979         __swab32s(&h->hal_archive_id);
1980         __swab64s(&h->hal_flags);
1981         hai = hai_zero(h);
1982         for (i = 0; i < h->hal_count; i++) {
1983                 lustre_swab_hai(hai);
1984                 hai = hai_next(hai);
1985         }
1986 }
1987
1988 static void lustre_swab_kuch(struct kuc_hdr *l)
1989 {
1990         __swab16s(&l->kuc_magic);
1991         /* __u8 l->kuc_transport */
1992         __swab16s(&l->kuc_msgtype);
1993         __swab16s(&l->kuc_msglen);
1994 }
1995
1996 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
1997                                 struct lustre_kernelcomm *lk)
1998 {
1999         struct obd_import  *imp = class_exp2cliimp(exp);
2000         __u32               archive = lk->lk_data;
2001         int                 rc = 0;
2002
2003         if (lk->lk_group != KUC_GRP_HSM) {
2004                 CERROR("Bad copytool group %d\n", lk->lk_group);
2005                 return -EINVAL;
2006         }
2007
2008         CDEBUG(D_HSM, "CT start r%d w%d u%d g%d f%#x\n", lk->lk_rfd, lk->lk_wfd,
2009                lk->lk_uid, lk->lk_group, lk->lk_flags);
2010
2011         if (lk->lk_flags & LK_FLG_STOP) {
2012                 rc = libcfs_kkuc_group_rem(lk->lk_uid, lk->lk_group);
2013                 /* Unregister with the coordinator */
2014                 if (rc == 0)
2015                         rc = mdc_ioc_hsm_ct_unregister(imp);
2016         } else {
2017                 cfs_file_t *fp = cfs_get_fd(lk->lk_wfd);
2018                 rc = libcfs_kkuc_group_add(fp, lk->lk_uid, lk->lk_group,
2019                                            lk->lk_data);
2020                 if (rc && fp)
2021                         cfs_put_file(fp);
2022                 if (rc == 0)
2023                         rc = mdc_ioc_hsm_ct_register(imp, archive);
2024         }
2025
2026         return rc;
2027 }
2028
2029 /**
2030  * Send a message to any listening copytools
2031  * @param val KUC message (kuc_hdr + hsm_action_list)
2032  * @param len total length of message
2033  */
2034 static int mdc_hsm_copytool_send(int len, void *val)
2035 {
2036         struct kuc_hdr          *lh = (struct kuc_hdr *)val;
2037         struct hsm_action_list  *hal = (struct hsm_action_list *)(lh + 1);
2038         int                      rc;
2039         ENTRY;
2040
2041         if (len < sizeof(*lh) + sizeof(*hal)) {
2042                 CERROR("Short HSM message %d < %d\n", len,
2043                        (int) (sizeof(*lh) + sizeof(*hal)));
2044                 RETURN(-EPROTO);
2045         }
2046         if (lh->kuc_magic == __swab16(KUC_MAGIC)) {
2047                 lustre_swab_kuch(lh);
2048                 lustre_swab_hal(hal);
2049         } else if (lh->kuc_magic != KUC_MAGIC) {
2050                 CERROR("Bad magic %x!=%x\n", lh->kuc_magic, KUC_MAGIC);
2051                 RETURN(-EPROTO);
2052         }
2053
2054         CDEBUG(D_HSM, " Received message mg=%x t=%d m=%d l=%d actions=%d "
2055                "on %s\n",
2056                lh->kuc_magic, lh->kuc_transport, lh->kuc_msgtype,
2057                lh->kuc_msglen, hal->hal_count, hal->hal_fsname);
2058
2059         /* Broadcast to HSM listeners */
2060         rc = libcfs_kkuc_group_put(KUC_GRP_HSM, lh);
2061
2062         RETURN(rc);
2063 }
2064
2065 /**
2066  * callback function passed to kuc for re-registering each HSM copytool
2067  * running on MDC, after MDT shutdown/recovery.
2068  * @param data archive id served by the copytool
2069  * @param cb_arg callback argument (obd_import)
2070  */
2071 static int mdc_hsm_ct_reregister(__u32 data, void *cb_arg)
2072 {
2073         struct obd_import       *imp = (struct obd_import *)cb_arg;
2074         __u32                    archive = data;
2075         int                      rc;
2076
2077         CDEBUG(D_HA, "recover copytool registration to MDT (archive=%#x)\n",
2078                archive);
2079         rc = mdc_ioc_hsm_ct_register(imp, archive);
2080
2081         /* ignore error if the copytool is already registered */
2082         return ((rc != 0) && (rc != -EEXIST)) ? rc : 0;
2083 }
2084
2085 /**
2086  * Re-establish all kuc contexts with MDT
2087  * after MDT shutdown/recovery.
2088  */
2089 static int mdc_kuc_reregister(struct obd_import *imp)
2090 {
2091         /* re-register HSM agents */
2092         return libcfs_kkuc_group_foreach(KUC_GRP_HSM, mdc_hsm_ct_reregister,
2093                                          (void *)imp);
2094 }
2095
2096 int mdc_set_info_async(const struct lu_env *env,
2097                        struct obd_export *exp,
2098                        obd_count keylen, void *key,
2099                        obd_count vallen, void *val,
2100                        struct ptlrpc_request_set *set)
2101 {
2102         struct obd_import       *imp = class_exp2cliimp(exp);
2103         int                      rc;
2104         ENTRY;
2105
2106         if (KEY_IS(KEY_READ_ONLY)) {
2107                 if (vallen != sizeof(int))
2108                         RETURN(-EINVAL);
2109
2110                 spin_lock(&imp->imp_lock);
2111                 if (*((int *)val)) {
2112                         imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
2113                         imp->imp_connect_data.ocd_connect_flags |=
2114                                                         OBD_CONNECT_RDONLY;
2115                 } else {
2116                         imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
2117                         imp->imp_connect_data.ocd_connect_flags &=
2118                                                         ~OBD_CONNECT_RDONLY;
2119                 }
2120                 spin_unlock(&imp->imp_lock);
2121
2122                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2123                                        keylen, key, vallen, val, set);
2124                 RETURN(rc);
2125         }
2126         if (KEY_IS(KEY_SPTLRPC_CONF)) {
2127                 sptlrpc_conf_client_adapt(exp->exp_obd);
2128                 RETURN(0);
2129         }
2130         if (KEY_IS(KEY_FLUSH_CTX)) {
2131                 sptlrpc_import_flush_my_ctx(imp);
2132                 RETURN(0);
2133         }
2134         if (KEY_IS(KEY_MDS_CONN)) {
2135                 /* mds-mds import */
2136                 spin_lock(&imp->imp_lock);
2137                 imp->imp_server_timeout = 1;
2138                 spin_unlock(&imp->imp_lock);
2139                 imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
2140                 CDEBUG(D_OTHER, "%s: timeout / 2\n", exp->exp_obd->obd_name);
2141                 RETURN(0);
2142         }
2143         if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2144                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2145                                        keylen, key, vallen, val, set);
2146                 RETURN(rc);
2147         }
2148         if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) {
2149                 rc = mdc_hsm_copytool_send(vallen, val);
2150                 RETURN(rc);
2151         }
2152
2153         CERROR("Unknown key %s\n", (char *)key);
2154         RETURN(-EINVAL);
2155 }
2156
2157 int mdc_get_info(const struct lu_env *env, struct obd_export *exp,
2158                  __u32 keylen, void *key, __u32 *vallen, void *val,
2159                  struct lov_stripe_md *lsm)
2160 {
2161         int rc = -EINVAL;
2162
2163         if (KEY_IS(KEY_MAX_EASIZE)) {
2164                 int mdsize, *max_easize;
2165
2166                 if (*vallen != sizeof(int))
2167                         RETURN(-EINVAL);
2168                 mdsize = *(int*)val;
2169                 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
2170                         exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
2171                 max_easize = val;
2172                 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
2173                 RETURN(0);
2174         } else if (KEY_IS(KEY_CONN_DATA)) {
2175                 struct obd_import *imp = class_exp2cliimp(exp);
2176                 struct obd_connect_data *data = val;
2177
2178                 if (*vallen != sizeof(*data))
2179                         RETURN(-EINVAL);
2180
2181                 *data = imp->imp_connect_data;
2182                 RETURN(0);
2183         } else if (KEY_IS(KEY_TGT_COUNT)) {
2184                 *((int *)val) = 1;
2185                 RETURN(0);
2186         }
2187
2188         rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
2189
2190         RETURN(rc);
2191 }
2192
2193 static int mdc_pin(struct obd_export *exp, const struct lu_fid *fid,
2194                    struct obd_capa *oc, struct obd_client_handle *handle,
2195                    int flags)
2196 {
2197         struct ptlrpc_request *req;
2198         struct mdt_body       *body;
2199         int                    rc;
2200         ENTRY;
2201
2202         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_PIN);
2203         if (req == NULL)
2204                 RETURN(-ENOMEM);
2205
2206         mdc_set_capa_size(req, &RMF_CAPA1, oc);
2207
2208         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_PIN);
2209         if (rc) {
2210                 ptlrpc_request_free(req);
2211                 RETURN(rc);
2212         }
2213
2214         mdc_pack_body(req, fid, oc, 0, 0, -1, flags);
2215
2216         ptlrpc_request_set_replen(req);
2217
2218         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
2219         rc = ptlrpc_queue_wait(req);
2220         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
2221         if (rc) {
2222                 CERROR("Pin failed: %d\n", rc);
2223                 GOTO(err_out, rc);
2224         }
2225
2226         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2227         if (body == NULL)
2228                 GOTO(err_out, rc = -EPROTO);
2229
2230         handle->och_fh = body->handle;
2231         handle->och_magic = OBD_CLIENT_HANDLE_MAGIC;
2232
2233         handle->och_mod = obd_mod_alloc();
2234         if (handle->och_mod == NULL) {
2235                 DEBUG_REQ(D_ERROR, req, "can't allocate md_open_data");
2236                 GOTO(err_out, rc = -ENOMEM);
2237         }
2238         handle->och_mod->mod_open_req = req; /* will be dropped by unpin */
2239
2240         RETURN(0);
2241
2242 err_out:
2243         ptlrpc_req_finished(req);
2244         RETURN(rc);
2245 }
2246
2247 static int mdc_unpin(struct obd_export *exp, struct obd_client_handle *handle,
2248                      int flag)
2249 {
2250         struct ptlrpc_request *req;
2251         struct mdt_body       *body;
2252         int                    rc;
2253         ENTRY;
2254
2255         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_UNPIN,
2256                                         LUSTRE_MDS_VERSION, MDS_UNPIN);
2257         if (req == NULL)
2258                 RETURN(-ENOMEM);
2259
2260         body = req_capsule_client_get(&req->rq_pill, &RMF_MDT_BODY);
2261         body->handle = handle->och_fh;
2262         body->flags = flag;
2263
2264         ptlrpc_request_set_replen(req);
2265
2266         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
2267         rc = ptlrpc_queue_wait(req);
2268         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
2269
2270         if (rc != 0)
2271                 CERROR("Unpin failed: %d\n", rc);
2272
2273         ptlrpc_req_finished(req);
2274         ptlrpc_req_finished(handle->och_mod->mod_open_req);
2275
2276         obd_mod_put(handle->och_mod);
2277         RETURN(rc);
2278 }
2279
2280 int mdc_sync(struct obd_export *exp, const struct lu_fid *fid,
2281              struct obd_capa *oc, struct ptlrpc_request **request)
2282 {
2283         struct ptlrpc_request *req;
2284         int                    rc;
2285         ENTRY;
2286
2287         *request = NULL;
2288         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
2289         if (req == NULL)
2290                 RETURN(-ENOMEM);
2291
2292         mdc_set_capa_size(req, &RMF_CAPA1, oc);
2293
2294         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
2295         if (rc) {
2296                 ptlrpc_request_free(req);
2297                 RETURN(rc);
2298         }
2299
2300         mdc_pack_body(req, fid, oc, 0, 0, -1, 0);
2301
2302         ptlrpc_request_set_replen(req);
2303
2304         rc = ptlrpc_queue_wait(req);
2305         if (rc)
2306                 ptlrpc_req_finished(req);
2307         else
2308                 *request = req;
2309         RETURN(rc);
2310 }
2311
2312 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
2313                             enum obd_import_event event)
2314 {
2315         int rc = 0;
2316
2317         LASSERT(imp->imp_obd == obd);
2318
2319         switch (event) {
2320         case IMP_EVENT_DISCON: {
2321 #if 0
2322                 /* XXX Pass event up to OBDs stack. used only for FLD now */
2323                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DISCON, NULL);
2324 #endif
2325                 break;
2326         }
2327         case IMP_EVENT_INACTIVE: {
2328                 struct client_obd *cli = &obd->u.cli;
2329                 /*
2330                  * Flush current sequence to make client obtain new one
2331                  * from server in case of disconnect/reconnect.
2332                  */
2333                 if (cli->cl_seq != NULL)
2334                         seq_client_flush(cli->cl_seq);
2335
2336                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
2337                 break;
2338         }
2339         case IMP_EVENT_INVALIDATE: {
2340                 struct ldlm_namespace *ns = obd->obd_namespace;
2341
2342                 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2343
2344                 break;
2345         }
2346         case IMP_EVENT_ACTIVE:
2347                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
2348                 /* restore re-establish kuc registration after reconnecting */
2349                 if (rc == 0)
2350                         rc = mdc_kuc_reregister(imp);
2351                 break;
2352         case IMP_EVENT_OCD:
2353                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
2354                 break;
2355         case IMP_EVENT_DEACTIVATE:
2356         case IMP_EVENT_ACTIVATE:
2357                 break;
2358         default:
2359                 CERROR("Unknown import event %x\n", event);
2360                 LBUG();
2361         }
2362         RETURN(rc);
2363 }
2364
2365 int mdc_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
2366                   struct md_op_data *op_data)
2367 {
2368         struct client_obd *cli = &exp->exp_obd->u.cli;
2369         struct lu_client_seq *seq = cli->cl_seq;
2370         ENTRY;
2371         RETURN(seq_client_alloc_fid(NULL, seq, fid));
2372 }
2373
2374 struct obd_uuid *mdc_get_uuid(struct obd_export *exp) {
2375         struct client_obd *cli = &exp->exp_obd->u.cli;
2376         return &cli->cl_target_uuid;
2377 }
2378
2379 /**
2380  * Determine whether the lock can be canceled before replaying it during
2381  * recovery, non zero value will be return if the lock can be canceled,
2382  * or zero returned for not
2383  */
2384 static int mdc_cancel_for_recovery(struct ldlm_lock *lock)
2385 {
2386         if (lock->l_resource->lr_type != LDLM_IBITS)
2387                 RETURN(0);
2388
2389         /* FIXME: if we ever get into a situation where there are too many
2390          * opened files with open locks on a single node, then we really
2391          * should replay these open locks to reget it */
2392         if (lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_OPEN)
2393                 RETURN(0);
2394
2395         RETURN(1);
2396 }
2397
2398 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
2399 {
2400         struct client_obd *cli = &obd->u.cli;
2401         struct lprocfs_static_vars lvars = { 0 };
2402         int rc;
2403         ENTRY;
2404
2405         OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
2406         if (!cli->cl_rpc_lock)
2407                 RETURN(-ENOMEM);
2408         mdc_init_rpc_lock(cli->cl_rpc_lock);
2409
2410         ptlrpcd_addref();
2411
2412         OBD_ALLOC(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
2413         if (!cli->cl_close_lock)
2414                 GOTO(err_rpc_lock, rc = -ENOMEM);
2415         mdc_init_rpc_lock(cli->cl_close_lock);
2416
2417         rc = client_obd_setup(obd, cfg);
2418         if (rc)
2419                 GOTO(err_close_lock, rc);
2420         lprocfs_mdc_init_vars(&lvars);
2421         lprocfs_obd_setup(obd, lvars.obd_vars);
2422         sptlrpc_lprocfs_cliobd_attach(obd);
2423         ptlrpc_lprocfs_register_obd(obd);
2424
2425         ns_register_cancel(obd->obd_namespace, mdc_cancel_for_recovery);
2426
2427         rc = obd_llog_init(obd, &obd->obd_olg, obd, NULL);
2428         if (rc) {
2429                 mdc_cleanup(obd);
2430                 CERROR("failed to setup llogging subsystems\n");
2431         }
2432
2433         RETURN(rc);
2434
2435 err_close_lock:
2436         OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
2437 err_rpc_lock:
2438         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
2439         ptlrpcd_decref();
2440         RETURN(rc);
2441 }
2442
2443 /* Initialize the default and maximum LOV EA and cookie sizes.  This allows
2444  * us to make MDS RPCs with large enough reply buffers to hold the
2445  * maximum-sized (= maximum striped) EA and cookie without having to
2446  * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
2447 static int mdc_init_ea_size(struct obd_export *exp, int easize,
2448                      int def_easize, int cookiesize)
2449 {
2450         struct obd_device *obd = exp->exp_obd;
2451         struct client_obd *cli = &obd->u.cli;
2452         ENTRY;
2453
2454         if (cli->cl_max_mds_easize < easize)
2455                 cli->cl_max_mds_easize = easize;
2456
2457         if (cli->cl_default_mds_easize < def_easize)
2458                 cli->cl_default_mds_easize = def_easize;
2459
2460         if (cli->cl_max_mds_cookiesize < cookiesize)
2461                 cli->cl_max_mds_cookiesize = cookiesize;
2462
2463         RETURN(0);
2464 }
2465
2466 static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2467 {
2468         int rc = 0;
2469         ENTRY;
2470
2471         switch (stage) {
2472         case OBD_CLEANUP_EARLY:
2473                 break;
2474         case OBD_CLEANUP_EXPORTS:
2475                 /* Failsafe, ok if racy */
2476                 if (obd->obd_type->typ_refcnt <= 1)
2477                         libcfs_kkuc_group_rem(0, KUC_GRP_HSM);
2478
2479                 obd_cleanup_client_import(obd);
2480                 ptlrpc_lprocfs_unregister_obd(obd);
2481                 lprocfs_obd_cleanup(obd);
2482
2483                 rc = obd_llog_finish(obd, 0);
2484                 if (rc != 0)
2485                         CERROR("failed to cleanup llogging subsystems\n");
2486                 break;
2487         }
2488         RETURN(rc);
2489 }
2490
2491 static int mdc_cleanup(struct obd_device *obd)
2492 {
2493         struct client_obd *cli = &obd->u.cli;
2494
2495         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
2496         OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
2497
2498         ptlrpcd_decref();
2499
2500         return client_obd_cleanup(obd);
2501 }
2502
2503
2504 static int mdc_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
2505                          struct obd_device *tgt, int *index)
2506 {
2507         struct llog_ctxt        *ctxt;
2508         int                      rc;
2509
2510         ENTRY;
2511
2512         LASSERT(olg == &obd->obd_olg);
2513
2514         rc = llog_setup(NULL, obd, olg, LLOG_CHANGELOG_REPL_CTXT, tgt,
2515                         &llog_client_ops);
2516         if (rc)
2517                 RETURN(rc);
2518
2519         ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
2520         llog_initiator_connect(ctxt);
2521         llog_ctxt_put(ctxt);
2522
2523         RETURN(0);
2524 }
2525
2526 static int mdc_llog_finish(struct obd_device *obd, int count)
2527 {
2528         struct llog_ctxt *ctxt;
2529
2530         ENTRY;
2531
2532         ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
2533         if (ctxt)
2534                 llog_cleanup(NULL, ctxt);
2535
2536         RETURN(0);
2537 }
2538
2539 static int mdc_process_config(struct obd_device *obd, obd_count len, void *buf)
2540 {
2541         struct lustre_cfg *lcfg = buf;
2542         struct lprocfs_static_vars lvars = { 0 };
2543         int rc = 0;
2544
2545         lprocfs_mdc_init_vars(&lvars);
2546         switch (lcfg->lcfg_command) {
2547         default:
2548                 rc = class_process_proc_param(PARAM_MDC, lvars.obd_vars,
2549                                               lcfg, obd);
2550                 if (rc > 0)
2551                         rc = 0;
2552                 break;
2553         }
2554         return(rc);
2555 }
2556
2557
2558 /* get remote permission for current user on fid */
2559 int mdc_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
2560                         struct obd_capa *oc, __u32 suppgid,
2561                         struct ptlrpc_request **request)
2562 {
2563         struct ptlrpc_request  *req;
2564         int                    rc;
2565         ENTRY;
2566
2567         LASSERT(client_is_remote(exp));
2568
2569         *request = NULL;
2570         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
2571         if (req == NULL)
2572                 RETURN(-ENOMEM);
2573
2574         mdc_set_capa_size(req, &RMF_CAPA1, oc);
2575
2576         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
2577         if (rc) {
2578                 ptlrpc_request_free(req);
2579                 RETURN(rc);
2580         }
2581
2582         mdc_pack_body(req, fid, oc, OBD_MD_FLRMTPERM, 0, suppgid, 0);
2583
2584         req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
2585                              sizeof(struct mdt_remote_perm));
2586
2587         ptlrpc_request_set_replen(req);
2588
2589         rc = ptlrpc_queue_wait(req);
2590         if (rc)
2591                 ptlrpc_req_finished(req);
2592         else
2593                 *request = req;
2594         RETURN(rc);
2595 }
2596
2597 static int mdc_interpret_renew_capa(const struct lu_env *env,
2598                                     struct ptlrpc_request *req, void *args,
2599                                     int status)
2600 {
2601         struct mdc_renew_capa_args *ra = args;
2602         struct mdt_body *body = NULL;
2603         struct lustre_capa *capa;
2604         ENTRY;
2605
2606         if (status)
2607                 GOTO(out, capa = ERR_PTR(status));
2608
2609         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2610         if (body == NULL)
2611                 GOTO(out, capa = ERR_PTR(-EFAULT));
2612
2613         if ((body->valid & OBD_MD_FLOSSCAPA) == 0)
2614                 GOTO(out, capa = ERR_PTR(-ENOENT));
2615
2616         capa = req_capsule_server_get(&req->rq_pill, &RMF_CAPA2);
2617         if (!capa)
2618                 GOTO(out, capa = ERR_PTR(-EFAULT));
2619         EXIT;
2620 out:
2621         ra->ra_cb(ra->ra_oc, capa);
2622         return 0;
2623 }
2624
2625 static int mdc_renew_capa(struct obd_export *exp, struct obd_capa *oc,
2626                           renew_capa_cb_t cb)
2627 {
2628         struct ptlrpc_request *req;
2629         struct mdc_renew_capa_args *ra;
2630         ENTRY;
2631
2632         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_GETATTR,
2633                                         LUSTRE_MDS_VERSION, MDS_GETATTR);
2634         if (req == NULL)
2635                 RETURN(-ENOMEM);
2636
2637         /* NB, OBD_MD_FLOSSCAPA is set here, but it doesn't necessarily mean the
2638          * capa to renew is oss capa.
2639          */
2640         mdc_pack_body(req, &oc->c_capa.lc_fid, oc, OBD_MD_FLOSSCAPA, 0, -1, 0);
2641         ptlrpc_request_set_replen(req);
2642
2643         CLASSERT(sizeof(*ra) <= sizeof(req->rq_async_args));
2644         ra = ptlrpc_req_async_args(req);
2645         ra->ra_oc = oc;
2646         ra->ra_cb = cb;
2647         req->rq_interpret_reply = mdc_interpret_renew_capa;
2648         ptlrpcd_add_req(req, PDL_POLICY_LOCAL, -1);
2649         RETURN(0);
2650 }
2651
2652 static int mdc_connect(const struct lu_env *env,
2653                        struct obd_export **exp,
2654                        struct obd_device *obd, struct obd_uuid *cluuid,
2655                        struct obd_connect_data *data,
2656                        void *localdata)
2657 {
2658         struct obd_import *imp = obd->u.cli.cl_import;
2659
2660         /* mds-mds import features */
2661         if (data && (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
2662                 spin_lock(&imp->imp_lock);
2663                 imp->imp_server_timeout = 1;
2664                 spin_unlock(&imp->imp_lock);
2665                 imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
2666                 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
2667                        obd->obd_name);
2668         }
2669
2670         return client_connect_import(env, exp, obd, cluuid, data, NULL);
2671 }
2672
2673 struct obd_ops mdc_obd_ops = {
2674         .o_owner            = THIS_MODULE,
2675         .o_setup            = mdc_setup,
2676         .o_precleanup       = mdc_precleanup,
2677         .o_cleanup          = mdc_cleanup,
2678         .o_add_conn         = client_import_add_conn,
2679         .o_del_conn         = client_import_del_conn,
2680         .o_connect          = mdc_connect,
2681         .o_disconnect       = client_disconnect_export,
2682         .o_iocontrol        = mdc_iocontrol,
2683         .o_set_info_async   = mdc_set_info_async,
2684         .o_statfs           = mdc_statfs,
2685         .o_pin              = mdc_pin,
2686         .o_unpin            = mdc_unpin,
2687         .o_fid_init         = client_fid_init,
2688         .o_fid_fini         = client_fid_fini,
2689         .o_fid_alloc        = mdc_fid_alloc,
2690         .o_import_event     = mdc_import_event,
2691         .o_llog_init        = mdc_llog_init,
2692         .o_llog_finish      = mdc_llog_finish,
2693         .o_get_info         = mdc_get_info,
2694         .o_process_config   = mdc_process_config,
2695         .o_get_uuid         = mdc_get_uuid,
2696         .o_quotactl         = mdc_quotactl,
2697         .o_quotacheck       = mdc_quotacheck
2698 };
2699
2700 struct md_ops mdc_md_ops = {
2701         .m_getstatus        = mdc_getstatus,
2702         .m_change_cbdata    = mdc_change_cbdata,
2703         .m_find_cbdata      = mdc_find_cbdata,
2704         .m_close            = mdc_close,
2705         .m_create           = mdc_create,
2706         .m_done_writing     = mdc_done_writing,
2707         .m_enqueue          = mdc_enqueue,
2708         .m_getattr          = mdc_getattr,
2709         .m_getattr_name     = mdc_getattr_name,
2710         .m_intent_lock      = mdc_intent_lock,
2711         .m_link             = mdc_link,
2712         .m_is_subdir        = mdc_is_subdir,
2713         .m_rename           = mdc_rename,
2714         .m_setattr          = mdc_setattr,
2715         .m_setxattr         = mdc_setxattr,
2716         .m_getxattr         = mdc_getxattr,
2717         .m_sync             = mdc_sync,
2718         .m_readpage         = mdc_readpage,
2719         .m_unlink           = mdc_unlink,
2720         .m_cancel_unused    = mdc_cancel_unused,
2721         .m_init_ea_size     = mdc_init_ea_size,
2722         .m_set_lock_data    = mdc_set_lock_data,
2723         .m_lock_match       = mdc_lock_match,
2724         .m_get_lustre_md    = mdc_get_lustre_md,
2725         .m_free_lustre_md   = mdc_free_lustre_md,
2726         .m_set_open_replay_data = mdc_set_open_replay_data,
2727         .m_clear_open_replay_data = mdc_clear_open_replay_data,
2728         .m_renew_capa       = mdc_renew_capa,
2729         .m_unpack_capa      = mdc_unpack_capa,
2730         .m_get_remote_perm  = mdc_get_remote_perm,
2731         .m_intent_getattr_async = mdc_intent_getattr_async,
2732         .m_revalidate_lock      = mdc_revalidate_lock
2733 };
2734
2735 int __init mdc_init(void)
2736 {
2737         int rc;
2738         struct lprocfs_static_vars lvars = { 0 };
2739         lprocfs_mdc_init_vars(&lvars);
2740
2741         rc = class_register_type(&mdc_obd_ops, &mdc_md_ops, lvars.module_vars,
2742                                  LUSTRE_MDC_NAME, NULL);
2743         RETURN(rc);
2744 }
2745
2746 #ifdef __KERNEL__
2747 static void /*__exit*/ mdc_exit(void)
2748 {
2749         class_unregister_type(LUSTRE_MDC_NAME);
2750 }
2751
2752 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2753 MODULE_DESCRIPTION("Lustre Metadata Client");
2754 MODULE_LICENSE("GPL");
2755
2756 module_init(mdc_init);
2757 module_exit(mdc_exit);
2758 #endif