lustre/mdc/mdc_request.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
  30  * Use is subject to license terms.
  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 #ifndef EXPORT_SYMTAB
  38 # define EXPORT_SYMTAB
  39 #endif
  40 #define DEBUG_SUBSYSTEM S_MDC
  41
  42 #ifdef __KERNEL__
  43 # include <linux/module.h>
  44 # include <linux/pagemap.h>
  45 # include <linux/miscdevice.h>
  46 # include <linux/init.h>
  47 #else
  48 # include <liblustre.h>
  49 #endif
  50
  51 #include <lustre_acl.h>
  52 #include <obd_class.h>
  53 #include <lustre_dlm.h>
  54 #include <lustre_fid.h>
  55 #include <md_object.h>
  56 #include <lprocfs_status.h>
  57 #include <lustre_param.h>
  58 #include "mdc_internal.h"
  59 #include <lustre/lustre_idl.h>
  60
  61 #define REQUEST_MINOR 244
  62
  63 struct mdc_renew_capa_args {
  64         struct obd_capa        *ra_oc;
  65         renew_capa_cb_t         ra_cb;
  66 };
  67
  68 static quota_interface_t *quota_interface;
  69 extern quota_interface_t mdc_quota_interface;
  70
  71 static int mdc_cleanup(struct obd_device *obd);
  72
  73 int mdc_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
  74                     const struct req_msg_field *field, struct obd_capa **oc)
  75 {
  76         struct lustre_capa *capa;
  77         struct obd_capa *c;
  78         ENTRY;
  79
  80         /* swabbed already in mdc_enqueue */
  81         capa = req_capsule_server_get(&req->rq_pill, field);
  82         if (capa == NULL)
  83                 RETURN(-EPROTO);
  84
  85         c = alloc_capa(CAPA_SITE_CLIENT);
  86         if (IS_ERR(c)) {
  87                 CDEBUG(D_INFO, "alloc capa failed!\n");
  88                 RETURN(PTR_ERR(c));
  89         } else {
  90                 c->c_capa = *capa;
  91                 *oc = c;
  92                 RETURN(0);
  93         }
  94 }
  95
  96 /* Helper that implements most of mdc_getstatus and signal_completed_replay. */
  97 /* XXX this should become mdc_get_info("key"), sending MDS_GET_INFO RPC */
  98 static int send_getstatus(struct obd_import *imp, struct lu_fid *rootfid,
  99                           struct obd_capa **pc, int level, int msg_flags)
 100 {
 101         struct ptlrpc_request *req;
 102         struct mdt_body       *body;
 103         int                    rc;
 104         ENTRY;
 105
 106         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_GETSTATUS,
 107                                         LUSTRE_MDS_VERSION, MDS_GETSTATUS);
 108         if (req == NULL)
 109                 RETURN(-ENOMEM);
 110
 111         mdc_pack_body(req, NULL, NULL, 0, 0, -1, 0);
 112         lustre_msg_add_flags(req->rq_reqmsg, msg_flags);
 113         req->rq_send_state = level;
 114
 115         ptlrpc_request_set_replen(req);
 116
 117         rc = ptlrpc_queue_wait(req);
 118         if (rc)
 119                 GOTO(out, rc);
 120
 121         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 122         if (body == NULL)
 123                 GOTO(out, rc = -EPROTO);
 124
 125         if (body->valid & OBD_MD_FLMDSCAPA) {
 126                 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA1, pc);
 127                 if (rc)
 128                         GOTO(out, rc);
 129         }
 130
 131         *rootfid = body->fid1;
 132         CDEBUG(D_NET,
 133                "root fid="DFID", last_committed="LPU64"\n",
 134                PFID(rootfid),
 135                lustre_msg_get_last_committed(req->rq_repmsg));
 136         EXIT;
 137 out:
 138         ptlrpc_req_finished(req);
 139         return rc;
 140 }
 141
 142 /* This should be mdc_get_info("rootfid") */
 143 int mdc_getstatus(struct obd_export *exp, struct lu_fid *rootfid,
 144                   struct obd_capa **pc)
 145 {
 146         return send_getstatus(class_exp2cliimp(exp), rootfid, pc,
 147                               LUSTRE_IMP_FULL, 0);
 148 }
 149
 150 /*
 151  * This function now is known to always saying that it will receive 4 buffers
 152  * from server. Even for cases when acl_size and md_size is zero, RPC header
 153  * will contain 4 fields and RPC itself will contain zero size fields. This is
 154  * because mdt_getattr*() _always_ returns 4 fields, but if acl is not needed
 155  * and thus zero, it shrinks it, making zero size. The same story about
 156  * md_size. And this is course of problem when client waits for smaller number
 157  * of fields. This issue will be fixed later when client gets aware of RPC
 158  * layouts.  --umka
 159  */
 160 static int mdc_getattr_common(struct obd_export *exp,
 161                               struct ptlrpc_request *req)
 162 {
 163         struct req_capsule *pill = &req->rq_pill;
 164         struct mdt_body    *body;
 165         void               *eadata;
 166         int                 rc;
 167         ENTRY;
 168
 169         /* Request message already built. */
 170         rc = ptlrpc_queue_wait(req);
 171         if (rc != 0)
 172                 RETURN(rc);
 173
 174         /* sanity check for the reply */
 175         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 176         if (body == NULL)
 177                 RETURN(-EPROTO);
 178
 179         CDEBUG(D_NET, "mode: %o\n", body->mode);
 180
 181         if (body->eadatasize != 0) {
 182                 mdc_update_max_ea_from_body(exp, body);
 183
 184                 eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
 185                                                       body->eadatasize);
 186                 if (eadata == NULL)
 187                         RETURN(-EPROTO);
 188         }
 189
 190         if (body->valid & OBD_MD_FLRMTPERM) {
 191                 struct mdt_remote_perm *perm;
 192
 193                 LASSERT(client_is_remote(exp));
 194                 perm = req_capsule_server_swab_get(pill, &RMF_ACL,
 195                                                 lustre_swab_mdt_remote_perm);
 196                 if (perm == NULL)
 197                         RETURN(-EPROTO);
 198         }
 199
 200         if (body->valid & OBD_MD_FLMDSCAPA) {
 201                 struct lustre_capa *capa;
 202                 capa = req_capsule_server_get(pill, &RMF_CAPA1);
 203                 if (capa == NULL)
 204                         RETURN(-EPROTO);
 205         }
 206
 207         RETURN(0);
 208 }
 209
 210 int mdc_getattr(struct obd_export *exp, struct md_op_data *op_data,
 211                 struct ptlrpc_request **request)
 212 {
 213         struct ptlrpc_request *req;
 214         int                    rc;
 215         ENTRY;
 216
 217         *request = NULL;
 218         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
 219         if (req == NULL)
 220                 RETURN(-ENOMEM);
 221
 222         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 223
 224         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
 225         if (rc) {
 226                 ptlrpc_request_free(req);
 227                 RETURN(rc);
 228         }
 229
 230         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
 231                       op_data->op_valid, op_data->op_mode, -1, 0);
 232
 233         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 234                              op_data->op_mode);
 235         if (op_data->op_valid & OBD_MD_FLRMTPERM) {
 236                 LASSERT(client_is_remote(exp));
 237                 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
 238                                      sizeof(struct mdt_remote_perm));
 239         }
 240         ptlrpc_request_set_replen(req);
 241
 242         rc = mdc_getattr_common(exp, req);
 243         if (rc)
 244                 ptlrpc_req_finished(req);
 245         else
 246                 *request = req;
 247         RETURN(rc);
 248 }
 249
 250 int mdc_getattr_name(struct obd_export *exp, struct md_op_data *op_data,
 251                      struct ptlrpc_request **request)
 252 {
 253         struct ptlrpc_request *req;
 254         int                    rc;
 255         ENTRY;
 256
 257         *request = NULL;
 258         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 259                                    &RQF_MDS_GETATTR_NAME);
 260         if (req == NULL)
 261                 RETURN(-ENOMEM);
 262
 263         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 264         req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 265                              op_data->op_namelen + 1);
 266
 267         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR_NAME);
 268         if (rc) {
 269                 ptlrpc_request_free(req);
 270                 RETURN(rc);
 271         }
 272
 273         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
 274                       op_data->op_valid, op_data->op_mode,
 275                       op_data->op_suppgids[0], 0);
 276
 277         if (op_data->op_name) {
 278                 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 279                 LASSERT(strnlen(op_data->op_name, op_data->op_namelen) ==
 280                                 op_data->op_namelen);
 281                 memcpy(name, op_data->op_name, op_data->op_namelen);
 282         }
 283
 284         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 285                              op_data->op_mode);
 286         ptlrpc_request_set_replen(req);
 287
 288         rc = mdc_getattr_common(exp, req);
 289         if (rc)
 290                 ptlrpc_req_finished(req);
 291         else
 292                 *request = req;
 293         RETURN(rc);
 294 }
 295
 296 static int mdc_is_subdir(struct obd_export *exp,
 297                          const struct lu_fid *pfid,
 298                          const struct lu_fid *cfid,
 299                          struct ptlrpc_request **request)
 300 {
 301         struct ptlrpc_request  *req;
 302         int                     rc;
 303
 304         ENTRY;
 305
 306         *request = NULL;
 307         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
 308                                         &RQF_MDS_IS_SUBDIR, LUSTRE_MDS_VERSION,
 309                                         MDS_IS_SUBDIR);
 310         if (req == NULL)
 311                 RETURN(-ENOMEM);
 312
 313         mdc_is_subdir_pack(req, pfid, cfid, 0);
 314         ptlrpc_request_set_replen(req);
 315
 316         rc = ptlrpc_queue_wait(req);
 317         if (rc && rc != -EREMOTE)
 318                 ptlrpc_req_finished(req);
 319         else
 320                 *request = req;
 321         RETURN(rc);
 322 }
 323
 324 static int mdc_xattr_common(struct obd_export *exp,const struct req_format *fmt,
 325                             const struct lu_fid *fid,
 326                             struct obd_capa *oc, int opcode, obd_valid valid,
 327                             const char *xattr_name, const char *input,
 328                             int input_size, int output_size, int flags,
 329                             __u32 suppgid, struct ptlrpc_request **request)
 330 {
 331         struct ptlrpc_request *req;
 332         int   xattr_namelen = 0;
 333         char *tmp;
 334         int   rc;
 335         ENTRY;
 336
 337         *request = NULL;
 338         req = ptlrpc_request_alloc(class_exp2cliimp(exp), fmt);
 339         if (req == NULL)
 340                 RETURN(-ENOMEM);
 341
 342         mdc_set_capa_size(req, &RMF_CAPA1, oc);
 343         if (xattr_name) {
 344                 xattr_namelen = strlen(xattr_name) + 1;
 345                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 346                                      xattr_namelen);
 347         }
 348         if (input_size) {
 349                 LASSERT(input);
 350                 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
 351                                      input_size);
 352         }
 353
 354         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode);
 355         if (rc) {
 356                 ptlrpc_request_free(req);
 357                 RETURN(rc);
 358         }
 359
 360         if (opcode == MDS_REINT) {
 361                 struct mdt_rec_setxattr *rec;
 362
 363                 CLASSERT(sizeof(struct mdt_rec_setxattr) ==
 364                          sizeof(struct mdt_rec_reint));
 365                 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
 366                 rec->sx_opcode = REINT_SETXATTR;
 367                 /* TODO:
 368                  *  cfs_curproc_fs{u,g}id() should replace
 369                  *  current->fs{u,g}id for portability.
 370                  */
 371                 rec->sx_fsuid  = cfs_curproc_fsuid();
 372                 rec->sx_fsgid  = cfs_curproc_fsgid();
 373                 rec->sx_cap    = cfs_curproc_cap_pack();
 374                 rec->sx_suppgid1 = suppgid;
 375                 rec->sx_suppgid2 = -1;
 376                 rec->sx_fid    = *fid;
 377                 rec->sx_valid  = valid | OBD_MD_FLCTIME;
 378                 rec->sx_time   = cfs_time_current_sec();
 379                 rec->sx_size   = output_size;
 380                 rec->sx_flags  = flags;
 381
 382                 mdc_pack_capa(req, &RMF_CAPA1, oc);
 383         } else {
 384                 mdc_pack_body(req, fid, oc, valid, output_size, suppgid, flags);
 385         }
 386
 387         if (xattr_name) {
 388                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 389                 memcpy(tmp, xattr_name, xattr_namelen);
 390         }
 391         if (input_size) {
 392                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
 393                 memcpy(tmp, input, input_size);
 394         }
 395
 396         if (req_capsule_has_field(&req->rq_pill, &RMF_EADATA, RCL_SERVER))
 397                 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
 398                                      RCL_SERVER, output_size);
 399         ptlrpc_request_set_replen(req);
 400
 401         /* make rpc */
 402         if (opcode == MDS_REINT)
 403                 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 404
 405         rc = ptlrpc_queue_wait(req);
 406
 407         if (opcode == MDS_REINT)
 408                 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 409
 410         if (rc)
 411                 ptlrpc_req_finished(req);
 412         else
 413                 *request = req;
 414         RETURN(rc);
 415 }
 416
 417 int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
 418                  struct obd_capa *oc, obd_valid valid, const char *xattr_name,
 419                  const char *input, int input_size, int output_size,
 420                  int flags, __u32 suppgid, struct ptlrpc_request **request)
 421 {
 422         return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
 423                                 fid, oc, MDS_REINT, valid, xattr_name,
 424                                 input, input_size, output_size, flags,
 425                                 suppgid, request);
 426 }
 427
 428 int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
 429                  struct obd_capa *oc, obd_valid valid, const char *xattr_name,
 430                  const char *input, int input_size, int output_size,
 431                  int flags, struct ptlrpc_request **request)
 432 {
 433         return mdc_xattr_common(exp, &RQF_MDS_GETXATTR,
 434                                 fid, oc, MDS_GETXATTR, valid, xattr_name,
 435                                 input, input_size, output_size, flags,
 436                                 -1, request);
 437 }
 438
 439 #ifdef CONFIG_FS_POSIX_ACL
 440 static int mdc_unpack_acl(struct ptlrpc_request *req, struct lustre_md *md)
 441 {
 442         struct req_capsule     *pill = &req->rq_pill;
 443         struct mdt_body        *body = md->body;
 444         struct posix_acl       *acl;
 445         void                   *buf;
 446         int                     rc;
 447         ENTRY;
 448
 449         if (!body->aclsize)
 450                 RETURN(0);
 451
 452         buf = req_capsule_server_sized_get(pill, &RMF_ACL, body->aclsize);
 453
 454         if (!buf)
 455                 RETURN(-EPROTO);
 456
 457         acl = posix_acl_from_xattr(buf, body->aclsize);
 458         if (IS_ERR(acl)) {
 459                 rc = PTR_ERR(acl);
 460                 CERROR("convert xattr to acl: %d\n", rc);
 461                 RETURN(rc);
 462         }
 463
 464         rc = posix_acl_valid(acl);
 465         if (rc) {
 466                 CERROR("validate acl: %d\n", rc);
 467                 posix_acl_release(acl);
 468                 RETURN(rc);
 469         }
 470
 471         md->posix_acl = acl;
 472         RETURN(0);
 473 }
 474 #else
 475 #define mdc_unpack_acl(req, md) 0
 476 #endif
 477
 478 int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
 479                       struct obd_export *dt_exp, struct obd_export *md_exp,
 480                       struct lustre_md *md)
 481 {
 482         struct req_capsule *pill = &req->rq_pill;
 483         int rc;
 484         ENTRY;
 485
 486         LASSERT(md);
 487         memset(md, 0, sizeof(*md));
 488
 489         md->body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 490         LASSERT(md->body != NULL);
 491
 492         if (md->body->valid & OBD_MD_FLEASIZE) {
 493                 int lmmsize;
 494                 struct lov_mds_md *lmm;
 495
 496                 if (!S_ISREG(md->body->mode)) {
 497                         CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, should be a "
 498                                "regular file, but is not\n");
 499                         GOTO(out, rc = -EPROTO);
 500                 }
 501
 502                 if (md->body->eadatasize == 0) {
 503                         CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, "
 504                                "but eadatasize 0\n");
 505                         GOTO(out, rc = -EPROTO);
 506                 }
 507                 lmmsize = md->body->eadatasize;
 508                 lmm = req_capsule_server_sized_get(pill, &RMF_MDT_MD, lmmsize);
 509                 if (!lmm)
 510                         GOTO(out, rc = -EPROTO);
 511
 512                 rc = obd_unpackmd(dt_exp, &md->lsm, lmm, lmmsize);
 513                 if (rc < 0)
 514                         GOTO(out, rc);
 515
 516                 if (rc < sizeof(*md->lsm)) {
 517                         CDEBUG(D_INFO, "lsm size too small: "
 518                                "rc < sizeof (*md->lsm) (%d < %d)\n",
 519                                rc, (int)sizeof(*md->lsm));
 520                         GOTO(out, rc = -EPROTO);
 521                 }
 522
 523         } else if (md->body->valid & OBD_MD_FLDIREA) {
 524                 int lmvsize;
 525                 struct lov_mds_md *lmv;
 526
 527                 if(!S_ISDIR(md->body->mode)) {
 528                         CDEBUG(D_INFO, "OBD_MD_FLDIREA set, should be a "
 529                                "directory, but is not\n");
 530                         GOTO(out, rc = -EPROTO);
 531                 }
 532
 533                 if (md->body->eadatasize == 0) {
 534                         CDEBUG(D_INFO, "OBD_MD_FLDIREA is set, "
 535                                "but eadatasize 0\n");
 536                         RETURN(-EPROTO);
 537                 }
 538                 if (md->body->valid & OBD_MD_MEA) {
 539                         lmvsize = md->body->eadatasize;
 540                         lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
 541                                                            lmvsize);
 542                         if (!lmv)
 543                                 GOTO(out, rc = -EPROTO);
 544
 545                         rc = obd_unpackmd(md_exp, (void *)&md->mea, lmv,
 546                                           lmvsize);
 547                         if (rc < 0)
 548                                 GOTO(out, rc);
 549
 550                         if (rc < sizeof(*md->mea)) {
 551                                 CDEBUG(D_INFO, "size too small:  "
 552                                        "rc < sizeof(*md->mea) (%d < %d)\n",
 553                                         rc, (int)sizeof(*md->mea));
 554                                 GOTO(out, rc = -EPROTO);
 555                         }
 556                 }
 557         }
 558         rc = 0;
 559
 560         if (md->body->valid & OBD_MD_FLRMTPERM) {
 561                 /* remote permission */
 562                 LASSERT(client_is_remote(exp));
 563                 md->remote_perm = req_capsule_server_swab_get(pill, &RMF_ACL,
 564                                                 lustre_swab_mdt_remote_perm);
 565                 if (!md->remote_perm)
 566                         GOTO(out, rc = -EPROTO);
 567         }
 568         else if (md->body->valid & OBD_MD_FLACL) {
 569                 /* for ACL, it's possible that FLACL is set but aclsize is zero.
 570                  * only when aclsize != 0 there's an actual segment for ACL
 571                  * in reply buffer.
 572                  */
 573                 if (md->body->aclsize) {
 574                         rc = mdc_unpack_acl(req, md);
 575                         if (rc)
 576                                 GOTO(out, rc);
 577 #ifdef CONFIG_FS_POSIX_ACL
 578                 } else {
 579                         md->posix_acl = NULL;
 580 #endif
 581                 }
 582         }
 583         if (md->body->valid & OBD_MD_FLMDSCAPA) {
 584                 struct obd_capa *oc = NULL;
 585
 586                 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA1, &oc);
 587                 if (rc)
 588                         GOTO(out, rc);
 589                 md->mds_capa = oc;
 590         }
 591
 592         if (md->body->valid & OBD_MD_FLOSSCAPA) {
 593                 struct obd_capa *oc = NULL;
 594
 595                 rc = mdc_unpack_capa(NULL, req, &RMF_CAPA2, &oc);
 596                 if (rc)
 597                         GOTO(out, rc);
 598                 md->oss_capa = oc;
 599         }
 600
 601         EXIT;
 602 out:
 603         if (rc) {
 604                 if (md->oss_capa) {
 605                         capa_put(md->oss_capa);
 606                         md->oss_capa = NULL;
 607                 }
 608                 if (md->mds_capa) {
 609                         capa_put(md->mds_capa);
 610                         md->mds_capa = NULL;
 611                 }
 612 #ifdef CONFIG_FS_POSIX_ACL
 613                 posix_acl_release(md->posix_acl);
 614 #endif
 615                 if (md->lsm)
 616                         obd_free_memmd(dt_exp, &md->lsm);
 617         }
 618         return rc;
 619 }
 620
 621 int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
 622 {
 623         ENTRY;
 624         RETURN(0);
 625 }
 626
 627 /**
 628  * Handles both OPEN and SETATTR RPCs for OPEN-CLOSE and SETATTR-DONE_WRITING
 629  * RPC chains.
 630  */
 631 void mdc_replay_open(struct ptlrpc_request *req)
 632 {
 633         struct md_open_data *mod = req->rq_cb_data;
 634         struct ptlrpc_request *close_req;
 635         struct obd_client_handle *och;
 636         struct lustre_handle old;
 637         struct mdt_body *body;
 638         ENTRY;
 639
 640         if (mod == NULL) {
 641                 DEBUG_REQ(D_ERROR, req,
 642                           "Can't properly replay without open data.");
 643                 EXIT;
 644                 return;
 645         }
 646
 647         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 648         LASSERT(body != NULL);
 649
 650         och = mod->mod_och;
 651         if (och != NULL) {
 652                 struct lustre_handle *file_fh;
 653
 654                 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
 655
 656                 file_fh = &och->och_fh;
 657                 CDEBUG(D_HA, "updating handle from "LPX64" to "LPX64"\n",
 658                        file_fh->cookie, body->handle.cookie);
 659                 old = *file_fh;
 660                 *file_fh = body->handle;
 661         }
 662         close_req = mod->mod_close_req;
 663         if (close_req != NULL) {
 664                 __u32 opc = lustre_msg_get_opc(close_req->rq_reqmsg);
 665                 struct mdt_ioepoch *epoch;
 666
 667                 LASSERT(opc == MDS_CLOSE || opc == MDS_DONE_WRITING);
 668                 epoch = req_capsule_client_get(&close_req->rq_pill,
 669                                                &RMF_MDT_EPOCH);
 670                 LASSERT(epoch);
 671
 672                 if (och != NULL)
 673                         LASSERT(!memcmp(&old, &epoch->handle, sizeof(old)));
 674                 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
 675                 epoch->handle = body->handle;
 676         }
 677         EXIT;
 678 }
 679
 680 void mdc_commit_open(struct ptlrpc_request *req)
 681 {
 682         struct md_open_data *mod = req->rq_cb_data;
 683         if (mod == NULL)
 684                 return;
 685
 686         /**
 687          * No need to touch md_open_data::mod_och, it holds a reference on
 688          * \var mod and will zero references to each other, \var mod will be
 689          * freed after that when md_open_data::mod_och will put the reference.
 690          */
 691
 692         /**
 693          * Do not let open request to disappear as it still may be needed
 694          * for close rpc to happen (it may happen on evict only, otherwise
 695          * ptlrpc_request::rq_replay does not let mdc_commit_open() to be
 696          * called), just mark this rpc as committed to distinguish these 2
 697          * cases, see mdc_close() for details. The open request reference will
 698          * be put along with freeing \var mod.
 699          */
 700         ptlrpc_request_addref(req);
 701         cfs_spin_lock(&req->rq_lock);
 702         req->rq_committed = 1;
 703         cfs_spin_unlock(&req->rq_lock);
 704         req->rq_cb_data = NULL;
 705         obd_mod_put(mod);
 706 }
 707
 708 int mdc_set_open_replay_data(struct obd_export *exp,
 709                              struct obd_client_handle *och,
 710                              struct ptlrpc_request *open_req)
 711 {
 712         struct md_open_data   *mod;
 713         struct mdt_rec_create *rec;
 714         struct mdt_body       *body;
 715         struct obd_import     *imp = open_req->rq_import;
 716         ENTRY;
 717
 718         if (!open_req->rq_replay)
 719                 RETURN(0);
 720
 721         rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
 722         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
 723         LASSERT(rec != NULL);
 724         /* Incoming message in my byte order (it's been swabbed). */
 725         /* Outgoing messages always in my byte order. */
 726         LASSERT(body != NULL);
 727
 728         /* Only if the import is replayable, we set replay_open data */
 729         if (och && imp->imp_replayable) {
 730                 mod = obd_mod_alloc();
 731                 if (mod == NULL) {
 732                         DEBUG_REQ(D_ERROR, open_req,
 733                                   "Can't allocate md_open_data");
 734                         RETURN(0);
 735                 }
 736
 737                 /**
 738                  * Take a reference on \var mod, to be freed on mdc_close().
 739                  * It protects \var mod from being freed on eviction (commit
 740                  * callback is called despite rq_replay flag).
 741                  * Another reference for \var och.
 742                  */
 743                 obd_mod_get(mod);
 744                 obd_mod_get(mod);
 745
 746                 cfs_spin_lock(&open_req->rq_lock);
 747                 och->och_mod = mod;
 748                 mod->mod_och = och;
 749                 mod->mod_open_req = open_req;
 750                 open_req->rq_cb_data = mod;
 751                 open_req->rq_commit_cb = mdc_commit_open;
 752                 cfs_spin_unlock(&open_req->rq_lock);
 753         }
 754
 755         rec->cr_fid2 = body->fid1;
 756         rec->cr_ioepoch = body->ioepoch;
 757         rec->cr_old_handle.cookie = body->handle.cookie;
 758         open_req->rq_replay_cb = mdc_replay_open;
 759         if (!fid_is_sane(&body->fid1)) {
 760                 DEBUG_REQ(D_ERROR, open_req, "Saving replay request with "
 761                           "insane fid");
 762                 LBUG();
 763         }
 764
 765         DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
 766         RETURN(0);
 767 }
 768
 769 int mdc_clear_open_replay_data(struct obd_export *exp,
 770                                struct obd_client_handle *och)
 771 {
 772         struct md_open_data *mod = och->och_mod;
 773         ENTRY;
 774
 775         /**
 776          * It is possible to not have \var mod in a case of eviction between
 777          * lookup and ll_file_open().
 778          **/
 779         if (mod == NULL)
 780                 RETURN(0);
 781
 782         LASSERT(mod != LP_POISON);
 783
 784         mod->mod_och = NULL;
 785         och->och_mod = NULL;
 786         obd_mod_put(mod);
 787
 788         RETURN(0);
 789 }
 790
 791 /* Prepares the request for the replay by the given reply */
 792 static void mdc_close_handle_reply(struct ptlrpc_request *req,
 793                                    struct md_op_data *op_data, int rc) {
 794         struct mdt_body  *repbody;
 795         struct mdt_ioepoch *epoch;
 796
 797         if (req && rc == -EAGAIN) {
 798                 repbody = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 799                 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
 800
 801                 epoch->flags |= MF_SOM_AU;
 802                 if (repbody->valid & OBD_MD_FLGETATTRLOCK)
 803                         op_data->op_flags |= MF_GETATTR_LOCK;
 804         }
 805 }
 806
 807 int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
 808               struct md_open_data *mod, struct ptlrpc_request **request)
 809 {
 810         struct obd_device     *obd = class_exp2obd(exp);
 811         struct ptlrpc_request *req;
 812         int                    rc;
 813         ENTRY;
 814
 815         *request = NULL;
 816         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_CLOSE);
 817         if (req == NULL)
 818                 RETURN(-ENOMEM);
 819
 820         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 821
 822         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
 823         if (rc) {
 824                 ptlrpc_request_free(req);
 825                 RETURN(rc);
 826         }
 827
 828         /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
 829          * portal whose threads are not taking any DLM locks and are therefore
 830          * always progressing */
 831         req->rq_request_portal = MDS_READPAGE_PORTAL;
 832         ptlrpc_at_set_req_timeout(req);
 833
 834         /* Ensure that this close's handle is fixed up during replay. */
 835         if (likely(mod != NULL)) {
 836                 LASSERTF(mod->mod_open_req != NULL &&
 837                          mod->mod_open_req->rq_type != LI_POISON,
 838                          "POISONED open %p!\n", mod->mod_open_req);
 839
 840                 mod->mod_close_req = req;
 841
 842                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
 843                 /* We no longer want to preserve this open for replay even
 844                  * though the open was committed. b=3632, b=3633 */
 845                 cfs_spin_lock(&mod->mod_open_req->rq_lock);
 846                 mod->mod_open_req->rq_replay = 0;
 847                 cfs_spin_unlock(&mod->mod_open_req->rq_lock);
 848         } else {
 849                  CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
 850         }
 851
 852         mdc_close_pack(req, op_data);
 853
 854         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 855                              obd->u.cli.cl_max_mds_easize);
 856         req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
 857                              obd->u.cli.cl_max_mds_cookiesize);
 858
 859         ptlrpc_request_set_replen(req);
 860
 861         mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 862         rc = ptlrpc_queue_wait(req);
 863         mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 864
 865         if (req->rq_repmsg == NULL) {
 866                 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
 867                        req->rq_status);
 868                 if (rc == 0)
 869                         rc = req->rq_status ?: -EIO;
 870         } else if (rc == 0 || rc == -EAGAIN) {
 871                 struct mdt_body *body;
 872
 873                 rc = lustre_msg_get_status(req->rq_repmsg);
 874                 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
 875                         DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
 876                                   "= %d", rc);
 877                         if (rc > 0)
 878                                 rc = -rc;
 879                 }
 880                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 881                 if (body == NULL)
 882                         rc = -EPROTO;
 883         } else if (rc == -ESTALE) {
 884                 /**
 885                  * it can be allowed error after 3633 if open was committed and
 886                  * server failed before close was sent. Let's check if mod
 887                  * exists and return no error in that case
 888                  */
 889                 if (mod) {
 890                         DEBUG_REQ(D_HA, req, "Reset ESTALE = %d", rc);
 891                         LASSERT(mod->mod_open_req != NULL);
 892                         if (mod->mod_open_req->rq_committed)
 893                                 rc = 0;
 894                 }
 895         }
 896
 897         if (mod) {
 898                 if (rc != 0)
 899                         mod->mod_close_req = NULL;
 900                 /* Since now, mod is accessed through open_req only,
 901                  * thus close req does not keep a reference on mod anymore. */
 902                 obd_mod_put(mod);
 903         }
 904         *request = req;
 905         mdc_close_handle_reply(req, op_data, rc);
 906         RETURN(rc);
 907 }
 908
 909 int mdc_done_writing(struct obd_export *exp, struct md_op_data *op_data,
 910                      struct md_open_data *mod)
 911 {
 912         struct obd_device     *obd = class_exp2obd(exp);
 913         struct ptlrpc_request *req;
 914         int                    rc;
 915         ENTRY;
 916
 917         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 918                                    &RQF_MDS_DONE_WRITING);
 919         if (req == NULL)
 920                 RETURN(-ENOMEM);
 921
 922         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 923         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_DONE_WRITING);
 924         if (rc) {
 925                 ptlrpc_request_free(req);
 926                 RETURN(rc);
 927         }
 928
 929         if (mod != NULL) {
 930                 LASSERTF(mod->mod_open_req != NULL &&
 931                          mod->mod_open_req->rq_type != LI_POISON,
 932                          "POISONED setattr %p!\n", mod->mod_open_req);
 933
 934                 mod->mod_close_req = req;
 935                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched setattr");
 936                 /* We no longer want to preserve this setattr for replay even
 937                  * though the open was committed. b=3632, b=3633 */
 938                 cfs_spin_lock(&mod->mod_open_req->rq_lock);
 939                 mod->mod_open_req->rq_replay = 0;
 940                 cfs_spin_unlock(&mod->mod_open_req->rq_lock);
 941         }
 942
 943         mdc_close_pack(req, op_data);
 944         ptlrpc_request_set_replen(req);
 945
 946         mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 947         rc = ptlrpc_queue_wait(req);
 948         mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 949
 950         if (rc == -ESTALE) {
 951                 /**
 952                  * it can be allowed error after 3633 if open or setattr were
 953                  * committed and server failed before close was sent.
 954                  * Let's check if mod exists and return no error in that case
 955                  */
 956                 if (mod) {
 957                         LASSERT(mod->mod_open_req != NULL);
 958                         if (mod->mod_open_req->rq_committed)
 959                                 rc = 0;
 960                 }
 961         }
 962
 963         if (mod) {
 964                 if (rc != 0)
 965                         mod->mod_close_req = NULL;
 966                 /* Since now, mod is accessed through setattr req only,
 967                  * thus DW req does not keep a reference on mod anymore. */
 968                 obd_mod_put(mod);
 969         }
 970
 971         mdc_close_handle_reply(req, op_data, rc);
 972         ptlrpc_req_finished(req);
 973         RETURN(rc);
 974 }
 975
 976 #ifdef HAVE_SPLIT_SUPPORT
 977 int mdc_sendpage(struct obd_export *exp, const struct lu_fid *fid,
 978                  const struct page *page, int offset)
 979 {
 980         struct ptlrpc_request   *req;
 981         struct ptlrpc_bulk_desc *desc;
 982         int                      rc;
 983         ENTRY;
 984
 985         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_WRITEPAGE);
 986         if (req == NULL)
 987                 RETURN(-ENOMEM);
 988
 989         /* FIXME: capa doesn't support split yet */
 990         mdc_set_capa_size(req, &RMF_CAPA1, NULL);
 991
 992         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_WRITEPAGE);
 993         if (rc) {
 994                 ptlrpc_request_free(req);
 995                 RETURN(rc);
 996         }
 997
 998         req->rq_request_portal = MDS_READPAGE_PORTAL;
 999         ptlrpc_at_set_req_timeout(req);
1000
1001         desc = ptlrpc_prep_bulk_imp(req, 1, BULK_GET_SOURCE, MDS_BULK_PORTAL);
1002         if (desc == NULL)
1003                 GOTO(out, rc = -ENOMEM);
1004
1005         /* NB req now owns desc and will free it when it gets freed. */
1006         ptlrpc_prep_bulk_page(desc, (struct page *)page, 0, offset);
1007         mdc_readdir_pack(req, 0, offset, fid, NULL);
1008
1009         ptlrpc_request_set_replen(req);
1010         rc = ptlrpc_queue_wait(req);
1011         if (rc)
1012                 GOTO(out, rc);
1013
1014         rc = sptlrpc_cli_unwrap_bulk_write(req, req->rq_bulk);
1015 out:
1016         ptlrpc_req_finished(req);
1017         return rc;
1018 }
1019 EXPORT_SYMBOL(mdc_sendpage);
1020 #endif
1021
1022 int mdc_readpage(struct obd_export *exp, const struct lu_fid *fid,
1023                  struct obd_capa *oc, __u64 offset, struct page *page,
1024                  struct ptlrpc_request **request)
1025 {
1026         struct ptlrpc_request   *req;
1027         struct ptlrpc_bulk_desc *desc;
1028         int                      rc;
1029         ENTRY;
1030
1031         *request = NULL;
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, oc);
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, 1, BULK_PUT_SINK, MDS_BULK_PORTAL);
1048         if (desc == NULL) {
1049                 ptlrpc_request_free(req);
1050                 RETURN(-ENOMEM);
1051         }
1052
1053         /* NB req now owns desc and will free it when it gets freed */
1054         ptlrpc_prep_bulk_page(desc, page, 0, CFS_PAGE_SIZE);
1055         mdc_readdir_pack(req, offset, CFS_PAGE_SIZE, fid, oc);
1056
1057         ptlrpc_request_set_replen(req);
1058         rc = ptlrpc_queue_wait(req);
1059         if (rc) {
1060                 ptlrpc_req_finished(req);
1061                 RETURN(rc);
1062         }
1063
1064         rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
1065                                           req->rq_bulk->bd_nob_transferred);
1066         if (rc < 0) {
1067                 ptlrpc_req_finished(req);
1068                 RETURN(rc);
1069         }
1070
1071         if (req->rq_bulk->bd_nob_transferred != CFS_PAGE_SIZE) {
1072                 CERROR("Unexpected # bytes transferred: %d (%ld expected)\n",
1073                         req->rq_bulk->bd_nob_transferred, CFS_PAGE_SIZE);
1074                 ptlrpc_req_finished(req);
1075                 RETURN(-EPROTO);
1076         }
1077
1078         *request = req;
1079         RETURN(0);
1080 }
1081
1082 static int mdc_statfs(struct obd_device *obd, struct obd_statfs *osfs,
1083                       __u64 max_age, __u32 flags)
1084 {
1085         struct ptlrpc_request *req;
1086         struct obd_statfs     *msfs;
1087         struct obd_import     *imp = NULL;
1088         int                    rc;
1089         ENTRY;
1090
1091         /*
1092          * Since the request might also come from lprocfs, so we need
1093          * sync this with client_disconnect_export Bug15684
1094          */
1095         cfs_down_read(&obd->u.cli.cl_sem);
1096         if (obd->u.cli.cl_import)
1097                 imp = class_import_get(obd->u.cli.cl_import);
1098         cfs_up_read(&obd->u.cli.cl_sem);
1099         if (!imp)
1100                 RETURN(-ENODEV);
1101
1102         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_STATFS,
1103                                         LUSTRE_MDS_VERSION, MDS_STATFS);
1104         if (req == NULL)
1105                 GOTO(output, rc = -ENOMEM);
1106
1107         ptlrpc_request_set_replen(req);
1108
1109         if (flags & OBD_STATFS_NODELAY) {
1110                 /* procfs requests not want stay in wait for avoid deadlock */
1111                 req->rq_no_resend = 1;
1112                 req->rq_no_delay = 1;
1113         }
1114
1115         rc = ptlrpc_queue_wait(req);
1116         if (rc) {
1117                 /* check connection error first */
1118                 if (imp->imp_connect_error)
1119                         rc = imp->imp_connect_error;
1120                 GOTO(out, rc);
1121         }
1122
1123         msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1124         if (msfs == NULL)
1125                 GOTO(out, rc = -EPROTO);
1126
1127         *osfs = *msfs;
1128         EXIT;
1129 out:
1130         ptlrpc_req_finished(req);
1131 output:
1132         class_import_put(imp);
1133         return rc;
1134 }
1135
1136 static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
1137 {
1138         __u32 keylen, vallen;
1139         void *key;
1140         int rc;
1141
1142         if (gf->gf_pathlen > PATH_MAX)
1143                 RETURN(-ENAMETOOLONG);
1144         if (gf->gf_pathlen < 2)
1145                 RETURN(-EOVERFLOW);
1146
1147         /* Key is KEY_FID2PATH + getinfo_fid2path description */
1148         keylen = cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf);
1149         OBD_ALLOC(key, keylen);
1150         if (key == NULL)
1151                 RETURN(-ENOMEM);
1152         memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
1153         memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf));
1154
1155         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
1156                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
1157
1158         if (!fid_is_sane(&gf->gf_fid))
1159                 GOTO(out, rc = -EINVAL);
1160
1161         /* Val is struct getinfo_fid2path result plus path */
1162         vallen = sizeof(*gf) + gf->gf_pathlen;
1163
1164         rc = obd_get_info(exp, keylen, key, &vallen, gf, NULL);
1165         if (rc)
1166                 GOTO(out, rc);
1167
1168         if (vallen <= sizeof(*gf))
1169                 GOTO(out, rc = -EPROTO);
1170         else if (vallen > sizeof(*gf) + gf->gf_pathlen)
1171                 GOTO(out, rc = -EOVERFLOW);
1172
1173         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n%s\n",
1174                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno, gf->gf_path);
1175
1176 out:
1177         OBD_FREE(key, keylen);
1178         return rc;
1179 }
1180
1181 static struct kuc_hdr *changelog_kuc_hdr(char *buf, int len, int flags)
1182 {
1183         struct kuc_hdr *lh = (struct kuc_hdr *)buf;
1184
1185         LASSERT(len <= CR_MAXSIZE);
1186
1187         lh->kuc_magic = KUC_MAGIC;
1188         lh->kuc_transport = KUC_TRANSPORT_CHANGELOG;
1189         lh->kuc_flags = flags;
1190         lh->kuc_msgtype = CL_RECORD;
1191         lh->kuc_msglen = len;
1192         return lh;
1193 }
1194
1195 #define D_CHANGELOG 0
1196
1197 struct changelog_show {
1198         __u64       cs_startrec;
1199         __u32       cs_flags;
1200         cfs_file_t *cs_fp;
1201         char       *cs_buf;
1202         struct obd_device *cs_obd;
1203 };
1204
1205 static int changelog_show_cb(struct llog_handle *llh, struct llog_rec_hdr *hdr,
1206                              void *data)
1207 {
1208         struct changelog_show *cs = data;
1209         struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
1210         struct kuc_hdr *lh;
1211         int len, rc;
1212         ENTRY;
1213
1214         if ((rec->cr_hdr.lrh_type != CHANGELOG_REC) ||
1215             (rec->cr.cr_type >= CL_LAST)) {
1216                 CERROR("Not a changelog rec %d/%d\n", rec->cr_hdr.lrh_type,
1217                        rec->cr.cr_type);
1218                 RETURN(-EINVAL);
1219         }
1220
1221         if (rec->cr.cr_index < cs->cs_startrec) {
1222                 /* Skip entries earlier than what we are interested in */
1223                 CDEBUG(D_CHANGELOG, "rec="LPU64" start="LPU64"\n",
1224                        rec->cr.cr_index, cs->cs_startrec);
1225                 RETURN(0);
1226         }
1227
1228         CDEBUG(D_CHANGELOG, LPU64" %02d%-5s "LPU64" 0x%x t="DFID" p="DFID
1229                " %.*s\n", rec->cr.cr_index, rec->cr.cr_type,
1230                changelog_type2str(rec->cr.cr_type), rec->cr.cr_time,
1231                rec->cr.cr_flags & CLF_FLAGMASK,
1232                PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid),
1233                rec->cr.cr_namelen, rec->cr.cr_name);
1234
1235         len = sizeof(*lh) + sizeof(rec->cr) + rec->cr.cr_namelen;
1236
1237         /* Set up the message */
1238         lh = changelog_kuc_hdr(cs->cs_buf, len, cs->cs_flags);
1239         memcpy(lh + 1, &rec->cr, len - sizeof(*lh));
1240
1241         rc = libcfs_kkuc_msg_put(cs->cs_fp, lh);
1242         CDEBUG(D_CHANGELOG, "kucmsg fp %p len %d rc %d\n", cs->cs_fp, len,rc);
1243
1244         RETURN(rc);
1245 }
1246
1247 static int mdc_changelog_send_thread(void *csdata)
1248 {
1249         struct changelog_show *cs = csdata;
1250         struct llog_ctxt *ctxt = NULL;
1251         struct llog_handle *llh = NULL;
1252         struct kuc_hdr *kuch;
1253         int rc;
1254
1255         CDEBUG(D_CHANGELOG, "changelog to fp=%p start "LPU64"\n",
1256                cs->cs_fp, cs->cs_startrec);
1257
1258         OBD_ALLOC(cs->cs_buf, CR_MAXSIZE);
1259         if (cs->cs_buf == NULL)
1260                 GOTO(out, rc = -ENOMEM);
1261
1262         /* Set up the remote catalog handle */
1263         ctxt = llog_get_context(cs->cs_obd, LLOG_CHANGELOG_REPL_CTXT);
1264         if (ctxt == NULL)
1265                 GOTO(out, rc = -ENOENT);
1266         rc = llog_create(ctxt, &llh, NULL, CHANGELOG_CATALOG);
1267         if (rc) {
1268                 CERROR("llog_create() failed %d\n", rc);
1269                 GOTO(out, rc);
1270         }
1271         rc = llog_init_handle(llh, LLOG_F_IS_CAT, NULL);
1272         if (rc) {
1273                 CERROR("llog_init_handle failed %d\n", rc);
1274                 GOTO(out, rc);
1275         }
1276
1277         /* We need the pipe fd open, so llog_process can't daemonize */
1278         rc = llog_cat_process_flags(llh, changelog_show_cb, cs,
1279                                     LLOG_FLAG_NODEAMON, 0, 0);
1280
1281         /* Send EOF no matter what our result */
1282         if ((kuch = changelog_kuc_hdr(cs->cs_buf, sizeof(*kuch),
1283                                       cs->cs_flags))) {
1284                 kuch->kuc_msgtype = CL_EOF;
1285                 libcfs_kkuc_msg_put(cs->cs_fp, kuch);
1286         }
1287
1288 out:
1289         cfs_put_file(cs->cs_fp);
1290         if (llh)
1291                 llog_cat_put(llh);
1292         if (ctxt)
1293                 llog_ctxt_put(ctxt);
1294         if (cs->cs_buf)
1295                 OBD_FREE(cs->cs_buf, CR_MAXSIZE);
1296         OBD_FREE_PTR(cs);
1297         /* detach from parent process so we get cleaned up */
1298         cfs_daemonize("cl_send");
1299         return rc;
1300 }
1301
1302 static int mdc_ioc_changelog_send(struct obd_device *obd,
1303                                   struct ioc_changelog *icc)
1304 {
1305         struct changelog_show *cs;
1306         int rc;
1307
1308         /* Freed in mdc_changelog_send_thread */
1309         OBD_ALLOC_PTR(cs);
1310         if (!cs)
1311                 return -ENOMEM;
1312
1313         cs->cs_obd = obd;
1314         cs->cs_startrec = icc->icc_recno;
1315         /* matching cfs_put_file in mdc_changelog_send_thread */
1316         cs->cs_fp = cfs_get_fd(icc->icc_id);
1317         cs->cs_flags = icc->icc_flags;
1318
1319         /* New thread because we should return to user app before
1320            writing into our pipe */
1321         rc = cfs_kernel_thread(mdc_changelog_send_thread, cs,
1322                                CLONE_VM | CLONE_FILES);
1323         if (rc >= 0) {
1324                 CDEBUG(D_CHANGELOG, "start changelog thread: %d\n", rc);
1325                 return 0;
1326         }
1327
1328         CERROR("Failed to start changelog thread: %d\n", rc);
1329         OBD_FREE_PTR(cs);
1330         return rc;
1331 }
1332
1333 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
1334                                 struct lustre_kernelcomm *lk);
1335
1336 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1337                          void *karg, void *uarg)
1338 {
1339         struct obd_device *obd = exp->exp_obd;
1340         struct obd_ioctl_data *data = karg;
1341         struct obd_import *imp = obd->u.cli.cl_import;
1342         struct llog_ctxt *ctxt;
1343         int rc;
1344         ENTRY;
1345
1346         if (!cfs_try_module_get(THIS_MODULE)) {
1347                 CERROR("Can't get module. Is it alive?");
1348                 return -EINVAL;
1349         }
1350         switch (cmd) {
1351         case LL_IOC_HSM_CT_START:
1352                 rc = mdc_ioc_hsm_ct_start(exp, karg);
1353                 GOTO(out, rc);
1354         case OBD_IOC_CHANGELOG_SEND:
1355                 rc = mdc_ioc_changelog_send(obd, karg);
1356                 GOTO(out, rc);
1357         case OBD_IOC_CHANGELOG_CLEAR: {
1358                 struct ioc_changelog *icc = karg;
1359                 struct changelog_setinfo cs =
1360                         {.cs_recno = icc->icc_recno, .cs_id = icc->icc_id};
1361                 rc = obd_set_info_async(exp, strlen(KEY_CHANGELOG_CLEAR),
1362                                         KEY_CHANGELOG_CLEAR, sizeof(cs), &cs,
1363                                         NULL);
1364                 GOTO(out, rc);
1365         }
1366         case OBD_IOC_FID2PATH: {
1367                 rc = mdc_ioc_fid2path(exp, karg);
1368                 GOTO(out, rc);
1369         }
1370         case OBD_IOC_CLIENT_RECOVER:
1371                 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1);
1372                 if (rc < 0)
1373                         GOTO(out, rc);
1374                 GOTO(out, rc = 0);
1375         case IOC_OSC_SET_ACTIVE:
1376                 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
1377                 GOTO(out, rc);
1378         case OBD_IOC_PARSE: {
1379                 ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
1380                 rc = class_config_parse_llog(ctxt, data->ioc_inlbuf1, NULL);
1381                 llog_ctxt_put(ctxt);
1382                 GOTO(out, rc);
1383         }
1384 #ifdef __KERNEL__
1385         case OBD_IOC_LLOG_INFO:
1386         case OBD_IOC_LLOG_PRINT: {
1387                 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
1388                 rc = llog_ioctl(ctxt, cmd, data);
1389                 llog_ctxt_put(ctxt);
1390                 GOTO(out, rc);
1391         }
1392 #endif
1393         case OBD_IOC_POLL_QUOTACHECK:
1394                 rc = lquota_poll_check(quota_interface, exp,
1395                                        (struct if_quotacheck *)karg);
1396                 GOTO(out, rc);
1397         case OBD_IOC_PING_TARGET:
1398                 rc = ptlrpc_obd_ping(obd);
1399                 GOTO(out, rc);
1400         /*
1401          * Normally IOC_OBD_STATFS iocontrol is handled by LMV instead of MDC.
1402          * But when the cluster is upgraded from 1.8, there'd be no LMV layer
1403          * thus we might be called here. Eventually this code should be removed.
1404          * bz20731.
1405          */
1406         case IOC_OBD_STATFS: {
1407                 struct obd_statfs stat_buf = {0};
1408
1409                 if (*((__u32 *) data->ioc_inlbuf2) != 0)
1410                         GOTO(out, rc = -ENODEV);
1411
1412                 /* copy UUID */
1413                 if (cfs_copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd),
1414                                      min((int) data->ioc_plen2,
1415                                          (int) sizeof(struct obd_uuid))))
1416                         GOTO(out, rc = -EFAULT);
1417
1418                 rc = mdc_statfs(obd, &stat_buf,
1419                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1420                                 0);
1421                 if (rc != 0)
1422                         GOTO(out, rc);
1423
1424                 if (cfs_copy_to_user(data->ioc_pbuf1, &stat_buf,
1425                                      min((int) data->ioc_plen1,
1426                                          (int) sizeof(stat_buf))))
1427                         GOTO(out, rc = -EFAULT);
1428
1429                 GOTO(out, rc = 0);
1430         }
1431         case LL_IOC_GET_CONNECT_FLAGS: {
1432                 if (cfs_copy_to_user(uarg, &exp->exp_connect_flags,
1433                                      sizeof(__u64)))
1434                         GOTO(out, rc = -EFAULT);
1435                 else
1436                         GOTO(out, rc = 0);
1437         }
1438         default:
1439                 CERROR("mdc_ioctl(): unrecognised ioctl %#x\n", cmd);
1440                 GOTO(out, rc = -ENOTTY);
1441         }
1442 out:
1443         cfs_module_put(THIS_MODULE);
1444
1445         return rc;
1446 }
1447
1448 int mdc_get_info_rpc(struct obd_export *exp,
1449                      obd_count keylen, void *key,
1450                      int vallen, void *val)
1451 {
1452         struct obd_import      *imp = class_exp2cliimp(exp);
1453         struct ptlrpc_request  *req;
1454         char                   *tmp;
1455         int                     rc = -EINVAL;
1456         ENTRY;
1457
1458         req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
1459         if (req == NULL)
1460                 RETURN(-ENOMEM);
1461
1462         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
1463                              RCL_CLIENT, keylen);
1464         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
1465                              RCL_CLIENT, sizeof(__u32));
1466
1467         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
1468         if (rc) {
1469                 ptlrpc_request_free(req);
1470                 RETURN(rc);
1471         }
1472
1473         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
1474         memcpy(tmp, key, keylen);
1475         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
1476         memcpy(tmp, &vallen, sizeof(__u32));
1477
1478         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
1479                              RCL_SERVER, vallen);
1480         ptlrpc_request_set_replen(req);
1481
1482         rc = ptlrpc_queue_wait(req);
1483         if (rc == 0) {
1484                 tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
1485                 memcpy(val, tmp, vallen);
1486                 if (ptlrpc_rep_need_swab(req)) {
1487                         if (KEY_IS(KEY_FID2PATH)) {
1488                                 lustre_swab_fid2path(val);
1489                         }
1490                 }
1491         }
1492         ptlrpc_req_finished(req);
1493
1494         RETURN(rc);
1495 }
1496
1497 static void lustre_swab_hai(struct hsm_action_item *h)
1498 {
1499         __swab32s(&h->hai_len);
1500         __swab32s(&h->hai_action);
1501         lustre_swab_lu_fid(&h->hai_fid);
1502         __swab64s(&h->hai_cookie);
1503         __swab64s(&h->hai_extent.offset);
1504         __swab64s(&h->hai_extent.length);
1505         __swab64s(&h->hai_gid);
1506 }
1507
1508 static void lustre_swab_hal(struct hsm_action_list *h)
1509 {
1510         struct hsm_action_item *hai;
1511         int i;
1512
1513         __swab32s(&h->hal_version);
1514         __swab32s(&h->hal_count);
1515         __swab32s(&h->hal_archive_num);
1516         hai = hai_zero(h);
1517         for (i = 0; i < h->hal_count; i++) {
1518                 lustre_swab_hai(hai);
1519                 hai = hai_next(hai);
1520         }
1521 }
1522
1523 static void lustre_swab_kuch(struct kuc_hdr *l)
1524 {
1525         __swab16s(&l->kuc_magic);
1526         /* __u8 l->kuc_transport */
1527         __swab16s(&l->kuc_msgtype);
1528         __swab16s(&l->kuc_msglen);
1529 }
1530
1531 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
1532                                 struct lustre_kernelcomm *lk)
1533 {
1534         int rc = 0;
1535
1536         if (lk->lk_group != KUC_GRP_HSM) {
1537                 CERROR("Bad copytool group %d\n", lk->lk_group);
1538                 return -EINVAL;
1539         }
1540
1541         CDEBUG(D_HSM, "CT start r%d w%d u%d g%d f%#x\n", lk->lk_rfd, lk->lk_wfd,
1542                lk->lk_uid, lk->lk_group, lk->lk_flags);
1543
1544         if (lk->lk_flags & LK_FLG_STOP)
1545                 rc = libcfs_kkuc_group_rem(lk->lk_uid,lk->lk_group);
1546         else {
1547                 cfs_file_t *fp = cfs_get_fd(lk->lk_wfd);
1548                 rc = libcfs_kkuc_group_add(fp, lk->lk_uid,lk->lk_group,
1549                                            lk->lk_data);
1550                 if (rc && fp)
1551                         cfs_put_file(fp);
1552         }
1553
1554         /* lk_data is archive number mask */
1555         /* TODO: register archive num with mdt so coordinator can choose
1556            correct agent. */
1557
1558         return rc;
1559 }
1560
1561 /**
1562  * Send a message to any listening copytools
1563  * @param val KUC message (kuc_hdr + hsm_action_list)
1564  * @param len total length of message
1565  */
1566 static int mdc_hsm_copytool_send(int len, void *val)
1567 {
1568         struct kuc_hdr *lh = (struct kuc_hdr *)val;
1569         struct hsm_action_list *hal = (struct hsm_action_list *)(lh + 1);
1570         int rc;
1571         ENTRY;
1572
1573         if (len < sizeof(*lh) + sizeof(*hal)) {
1574                 CERROR("Short HSM message %d < %d\n", len,
1575                       (int) (sizeof(*lh) + sizeof(*hal)));
1576                 RETURN(-EPROTO);
1577         }
1578         if (lh->kuc_magic == __swab16(KUC_MAGIC)) {
1579                 lustre_swab_kuch(lh);
1580                 lustre_swab_hal(hal);
1581         } else if (lh->kuc_magic != KUC_MAGIC) {
1582                 CERROR("Bad magic %x!=%x\n", lh->kuc_magic, KUC_MAGIC);
1583                 RETURN(-EPROTO);
1584         }
1585
1586         CDEBUG(D_HSM, " Received message mg=%x t=%d m=%d l=%d actions=%d\n",
1587                lh->kuc_magic, lh->kuc_transport, lh->kuc_msgtype,
1588                lh->kuc_msglen, hal->hal_count);
1589
1590         /* Broadcast to HSM listeners */
1591         rc = libcfs_kkuc_group_put(KUC_GRP_HSM, lh);
1592
1593         RETURN(rc);
1594 }
1595
1596 int mdc_set_info_async(struct obd_export *exp,
1597                        obd_count keylen, void *key,
1598                        obd_count vallen, void *val,
1599                        struct ptlrpc_request_set *set)
1600 {
1601         struct obd_import *imp = class_exp2cliimp(exp);
1602         int                rc = -EINVAL;
1603         ENTRY;
1604
1605         if (KEY_IS(KEY_READ_ONLY)) {
1606                 if (vallen != sizeof(int))
1607                         RETURN(-EINVAL);
1608
1609                 cfs_spin_lock(&imp->imp_lock);
1610                 if (*((int *)val)) {
1611                         imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
1612                         imp->imp_connect_data.ocd_connect_flags |= OBD_CONNECT_RDONLY;
1613                 } else {
1614                         imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
1615                         imp->imp_connect_data.ocd_connect_flags &= ~OBD_CONNECT_RDONLY;
1616                 }
1617                 cfs_spin_unlock(&imp->imp_lock);
1618
1619                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
1620                                        keylen, key, vallen, val, set);
1621                 RETURN(rc);
1622         }
1623         if (KEY_IS(KEY_SPTLRPC_CONF)) {
1624                 sptlrpc_conf_client_adapt(exp->exp_obd);
1625                 RETURN(0);
1626         }
1627         if (KEY_IS(KEY_FLUSH_CTX)) {
1628                 sptlrpc_import_flush_my_ctx(imp);
1629                 RETURN(0);
1630         }
1631         if (KEY_IS(KEY_MDS_CONN)) {
1632                 /* mds-mds import */
1633                 cfs_spin_lock(&imp->imp_lock);
1634                 imp->imp_server_timeout = 1;
1635                 cfs_spin_unlock(&imp->imp_lock);
1636                 imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
1637                 CDEBUG(D_OTHER, "%s: timeout / 2\n", exp->exp_obd->obd_name);
1638                 RETURN(0);
1639         }
1640         if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1641                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
1642                                        keylen, key, vallen, val, set);
1643                 RETURN(rc);
1644         }
1645         if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) {
1646                 rc = mdc_hsm_copytool_send(vallen, val);
1647                 RETURN(rc);
1648         }
1649
1650         RETURN(rc);
1651 }
1652
1653 int mdc_get_info(struct obd_export *exp, __u32 keylen, void *key,
1654                  __u32 *vallen, void *val, struct lov_stripe_md *lsm)
1655 {
1656         int rc = -EINVAL;
1657
1658         if (KEY_IS(KEY_MAX_EASIZE)) {
1659                 int mdsize, *max_easize;
1660
1661                 if (*vallen != sizeof(int))
1662                         RETURN(-EINVAL);
1663                 mdsize = *(int*)val;
1664                 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
1665                         exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
1666                 max_easize = val;
1667                 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
1668                 RETURN(0);
1669         }
1670         if (KEY_IS(KEY_CONN_DATA)) {
1671                 struct obd_import *imp = class_exp2cliimp(exp);
1672                 struct obd_connect_data *data = val;
1673
1674                 if (*vallen != sizeof(*data))
1675                         RETURN(-EINVAL);
1676
1677                 *data = imp->imp_connect_data;
1678                 RETURN(0);
1679         }
1680
1681         rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
1682
1683         RETURN(rc);
1684 }
1685
1686 static int mdc_pin(struct obd_export *exp, const struct lu_fid *fid,
1687                    struct obd_capa *oc, struct obd_client_handle *handle,
1688                    int flags)
1689 {
1690         struct ptlrpc_request *req;
1691         struct mdt_body       *body;
1692         int                    rc;
1693         ENTRY;
1694
1695         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_PIN);
1696         if (req == NULL)
1697                 RETURN(-ENOMEM);
1698
1699         mdc_set_capa_size(req, &RMF_CAPA1, oc);
1700
1701         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_PIN);
1702         if (rc) {
1703                 ptlrpc_request_free(req);
1704                 RETURN(rc);
1705         }
1706
1707         mdc_pack_body(req, fid, oc, 0, 0, -1, flags);
1708
1709         ptlrpc_request_set_replen(req);
1710
1711         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1712         rc = ptlrpc_queue_wait(req);
1713         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1714         if (rc) {
1715                 CERROR("Pin failed: %d\n", rc);
1716                 GOTO(err_out, rc);
1717         }
1718
1719         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1720         if (body == NULL)
1721                 GOTO(err_out, rc = -EPROTO);
1722
1723         handle->och_fh = body->handle;
1724         handle->och_magic = OBD_CLIENT_HANDLE_MAGIC;
1725
1726         handle->och_mod = obd_mod_alloc();
1727         if (handle->och_mod == NULL) {
1728                 DEBUG_REQ(D_ERROR, req, "can't allocate md_open_data");
1729                 GOTO(err_out, rc = -ENOMEM);
1730         }
1731         handle->och_mod->mod_open_req = req; /* will be dropped by unpin */
1732
1733         RETURN(0);
1734
1735 err_out:
1736         ptlrpc_req_finished(req);
1737         RETURN(rc);
1738 }
1739
1740 static int mdc_unpin(struct obd_export *exp, struct obd_client_handle *handle,
1741                      int flag)
1742 {
1743         struct ptlrpc_request *req;
1744         struct mdt_body       *body;
1745         int                    rc;
1746         ENTRY;
1747
1748         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_UNPIN,
1749                                         LUSTRE_MDS_VERSION, MDS_UNPIN);
1750         if (req == NULL)
1751                 RETURN(-ENOMEM);
1752
1753         body = req_capsule_client_get(&req->rq_pill, &RMF_MDT_BODY);
1754         body->handle = handle->och_fh;
1755         body->flags = flag;
1756
1757         ptlrpc_request_set_replen(req);
1758
1759         mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1760         rc = ptlrpc_queue_wait(req);
1761         mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
1762
1763         if (rc != 0)
1764                 CERROR("Unpin failed: %d\n", rc);
1765
1766         ptlrpc_req_finished(req);
1767         ptlrpc_req_finished(handle->och_mod->mod_open_req);
1768
1769         obd_mod_put(handle->och_mod);
1770         RETURN(rc);
1771 }
1772
1773 int mdc_sync(struct obd_export *exp, const struct lu_fid *fid,
1774              struct obd_capa *oc, struct ptlrpc_request **request)
1775 {
1776         struct ptlrpc_request *req;
1777         int                    rc;
1778         ENTRY;
1779
1780         *request = NULL;
1781         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
1782         if (req == NULL)
1783                 RETURN(-ENOMEM);
1784
1785         mdc_set_capa_size(req, &RMF_CAPA1, oc);
1786
1787         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
1788         if (rc) {
1789                 ptlrpc_request_free(req);
1790                 RETURN(rc);
1791         }
1792
1793         mdc_pack_body(req, fid, oc, 0, 0, -1, 0);
1794
1795         ptlrpc_request_set_replen(req);
1796
1797         rc = ptlrpc_queue_wait(req);
1798         if (rc)
1799                 ptlrpc_req_finished(req);
1800         else
1801                 *request = req;
1802         RETURN(rc);
1803 }
1804
1805 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
1806                             enum obd_import_event event)
1807 {
1808         int rc = 0;
1809
1810         LASSERT(imp->imp_obd == obd);
1811
1812         switch (event) {
1813         case IMP_EVENT_DISCON: {
1814 #if 0
1815                 /* XXX Pass event up to OBDs stack. used only for FLD now */
1816                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DISCON, NULL);
1817 #endif
1818                 break;
1819         }
1820         case IMP_EVENT_INACTIVE: {
1821                 struct client_obd *cli = &obd->u.cli;
1822                 /*
1823                  * Flush current sequence to make client obtain new one
1824                  * from server in case of disconnect/reconnect.
1825                  */
1826                 if (cli->cl_seq != NULL)
1827                         seq_client_flush(cli->cl_seq);
1828
1829                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
1830                 break;
1831         }
1832         case IMP_EVENT_INVALIDATE: {
1833                 struct ldlm_namespace *ns = obd->obd_namespace;
1834
1835                 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
1836
1837                 break;
1838         }
1839         case IMP_EVENT_ACTIVE: {
1840                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
1841                 break;
1842         }
1843         case IMP_EVENT_OCD:
1844                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
1845                 break;
1846
1847         default:
1848                 CERROR("Unknown import event %x\n", event);
1849                 LBUG();
1850         }
1851         RETURN(rc);
1852 }
1853
1854 static int mdc_fid_init(struct obd_export *exp)
1855 {
1856         struct client_obd *cli = &exp->exp_obd->u.cli;
1857         char *prefix;
1858         int rc;
1859         ENTRY;
1860
1861         OBD_ALLOC_PTR(cli->cl_seq);
1862         if (cli->cl_seq == NULL)
1863                 RETURN(-ENOMEM);
1864
1865         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
1866         if (prefix == NULL)
1867                 GOTO(out_free_seq, rc = -ENOMEM);
1868
1869         snprintf(prefix, MAX_OBD_NAME + 5, "srv-%s",
1870                  exp->exp_obd->obd_name);
1871
1872         /* Init client side sequence-manager */
1873         rc = seq_client_init(cli->cl_seq, exp,
1874                              LUSTRE_SEQ_METADATA,
1875                              prefix, NULL);
1876         OBD_FREE(prefix, MAX_OBD_NAME + 5);
1877         if (rc)
1878                 GOTO(out_free_seq, rc);
1879
1880         RETURN(rc);
1881 out_free_seq:
1882         OBD_FREE_PTR(cli->cl_seq);
1883         cli->cl_seq = NULL;
1884         return rc;
1885 }
1886
1887 static int mdc_fid_fini(struct obd_export *exp)
1888 {
1889         struct client_obd *cli = &exp->exp_obd->u.cli;
1890         ENTRY;
1891
1892         if (cli->cl_seq != NULL) {
1893                 seq_client_fini(cli->cl_seq);
1894                 OBD_FREE_PTR(cli->cl_seq);
1895                 cli->cl_seq = NULL;
1896         }
1897
1898         RETURN(0);
1899 }
1900
1901 int mdc_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
1902                   struct md_op_data *op_data)
1903 {
1904         struct client_obd *cli = &exp->exp_obd->u.cli;
1905         struct lu_client_seq *seq = cli->cl_seq;
1906         ENTRY;
1907         RETURN(seq_client_alloc_fid(seq, fid));
1908 }
1909
1910 /* XXX This method is used only to clear current fid seq
1911  * once fld/mds insert failed */
1912 static int mdc_fid_delete(struct obd_export *exp, const struct lu_fid *fid)
1913 {
1914         struct client_obd *cli = &exp->exp_obd->u.cli;
1915
1916         seq_client_flush(cli->cl_seq);
1917         return 0;
1918 }
1919
1920 struct obd_uuid *mdc_get_uuid(struct obd_export *exp) {
1921         struct client_obd *cli = &exp->exp_obd->u.cli;
1922         return &cli->cl_target_uuid;
1923 }
1924
1925 /**
1926  * Determine whether the lock can be canceled before replaying it during
1927  * recovery, non zero value will be return if the lock can be canceled,
1928  * or zero returned for not
1929  */
1930 static int mdc_cancel_for_recovery(struct ldlm_lock *lock)
1931 {
1932         if (lock->l_resource->lr_type != LDLM_IBITS)
1933                 RETURN(0);
1934
1935         /* FIXME: if we ever get into a situation where there are too many
1936          * opened files with open locks on a single node, then we really
1937          * should replay these open locks to reget it */
1938         if (lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_OPEN)
1939                 RETURN(0);
1940
1941         RETURN(1);
1942 }
1943
1944 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
1945 {
1946         struct client_obd *cli = &obd->u.cli;
1947         struct lprocfs_static_vars lvars = { 0 };
1948         int rc;
1949         ENTRY;
1950
1951         OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1952         if (!cli->cl_rpc_lock)
1953                 RETURN(-ENOMEM);
1954         mdc_init_rpc_lock(cli->cl_rpc_lock);
1955
1956         ptlrpcd_addref();
1957
1958         OBD_ALLOC(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1959         if (!cli->cl_setattr_lock)
1960                 GOTO(err_rpc_lock, rc = -ENOMEM);
1961         mdc_init_rpc_lock(cli->cl_setattr_lock);
1962
1963         OBD_ALLOC(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1964         if (!cli->cl_close_lock)
1965                 GOTO(err_setattr_lock, rc = -ENOMEM);
1966         mdc_init_rpc_lock(cli->cl_close_lock);
1967
1968         rc = client_obd_setup(obd, cfg);
1969         if (rc)
1970                 GOTO(err_close_lock, rc);
1971         lprocfs_mdc_init_vars(&lvars);
1972         lprocfs_obd_setup(obd, lvars.obd_vars);
1973         sptlrpc_lprocfs_cliobd_attach(obd);
1974         ptlrpc_lprocfs_register_obd(obd);
1975
1976         ns_register_cancel(obd->obd_namespace, mdc_cancel_for_recovery);
1977
1978         rc = obd_llog_init(obd, &obd->obd_olg, obd, NULL);
1979         if (rc) {
1980                 mdc_cleanup(obd);
1981                 CERROR("failed to setup llogging subsystems\n");
1982         }
1983
1984         RETURN(rc);
1985
1986 err_close_lock:
1987         OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
1988 err_setattr_lock:
1989         OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
1990 err_rpc_lock:
1991         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
1992         ptlrpcd_decref();
1993         RETURN(rc);
1994 }
1995
1996 /* Initialize the default and maximum LOV EA and cookie sizes.  This allows
1997  * us to make MDS RPCs with large enough reply buffers to hold the
1998  * maximum-sized (= maximum striped) EA and cookie without having to
1999  * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
2000 static int mdc_init_ea_size(struct obd_export *exp, int easize,
2001                      int def_easize, int cookiesize)
2002 {
2003         struct obd_device *obd = exp->exp_obd;
2004         struct client_obd *cli = &obd->u.cli;
2005         ENTRY;
2006
2007         if (cli->cl_max_mds_easize < easize)
2008                 cli->cl_max_mds_easize = easize;
2009
2010         if (cli->cl_default_mds_easize < def_easize)
2011                 cli->cl_default_mds_easize = def_easize;
2012
2013         if (cli->cl_max_mds_cookiesize < cookiesize)
2014                 cli->cl_max_mds_cookiesize = cookiesize;
2015
2016         RETURN(0);
2017 }
2018
2019 static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2020 {
2021         int rc = 0;
2022         ENTRY;
2023
2024         switch (stage) {
2025         case OBD_CLEANUP_EARLY:
2026                 break;
2027         case OBD_CLEANUP_EXPORTS:
2028                 /* Failsafe, ok if racy */
2029                 if (obd->obd_type->typ_refcnt <= 1)
2030                         libcfs_kkuc_group_rem(0, KUC_GRP_HSM);
2031
2032                 /* If we set up but never connected, the
2033                    client import will not have been cleaned. */
2034                 if (obd->u.cli.cl_import) {
2035                         struct obd_import *imp;
2036                         cfs_down_write(&obd->u.cli.cl_sem);
2037                         imp = obd->u.cli.cl_import;
2038                         CERROR("client import never connected\n");
2039                         ptlrpc_invalidate_import(imp);
2040                         class_destroy_import(imp);
2041                         cfs_up_write(&obd->u.cli.cl_sem);
2042                         obd->u.cli.cl_import = NULL;
2043                 }
2044                 rc = obd_llog_finish(obd, 0);
2045                 if (rc != 0)
2046                         CERROR("failed to cleanup llogging subsystems\n");
2047                 break;
2048         }
2049         RETURN(rc);
2050 }
2051
2052 static int mdc_cleanup(struct obd_device *obd)
2053 {
2054         struct client_obd *cli = &obd->u.cli;
2055
2056         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
2057         OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
2058         OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
2059
2060         ptlrpc_lprocfs_unregister_obd(obd);
2061         lprocfs_obd_cleanup(obd);
2062         ptlrpcd_decref();
2063
2064         return client_obd_cleanup(obd);
2065 }
2066
2067
2068 static int mdc_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
2069                          struct obd_device *tgt, int *index)
2070 {
2071         struct llog_ctxt *ctxt;
2072         int rc;
2073         ENTRY;
2074
2075         LASSERT(olg == &obd->obd_olg);
2076
2077         rc = llog_setup(obd, olg, LLOG_LOVEA_REPL_CTXT, tgt, 0, NULL,
2078                         &llog_client_ops);
2079         if (rc)
2080                 RETURN(rc);
2081
2082         ctxt = llog_get_context(obd, LLOG_LOVEA_REPL_CTXT);
2083         llog_initiator_connect(ctxt);
2084         llog_ctxt_put(ctxt);
2085
2086         rc = llog_setup(obd, olg, LLOG_CHANGELOG_REPL_CTXT, tgt, 0, NULL,
2087                         &llog_client_ops);
2088         if (rc == 0) {
2089                 ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
2090                 llog_initiator_connect(ctxt);
2091                 llog_ctxt_put(ctxt);
2092         }
2093
2094         RETURN(rc);
2095 }
2096
2097 static int mdc_llog_finish(struct obd_device *obd, int count)
2098 {
2099         struct llog_ctxt *ctxt;
2100         int rc = 0;
2101         ENTRY;
2102
2103         ctxt = llog_get_context(obd, LLOG_LOVEA_REPL_CTXT);
2104         if (ctxt)
2105                 rc = llog_cleanup(ctxt);
2106
2107         ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
2108         if (ctxt)
2109                 rc = llog_cleanup(ctxt);
2110
2111         RETURN(rc);
2112 }
2113
2114 static int mdc_process_config(struct obd_device *obd, obd_count len, void *buf)
2115 {
2116         struct lustre_cfg *lcfg = buf;
2117         struct lprocfs_static_vars lvars = { 0 };
2118         int rc = 0;
2119
2120         lprocfs_mdc_init_vars(&lvars);
2121         switch (lcfg->lcfg_command) {
2122         default:
2123                 rc = class_process_proc_param(PARAM_MDC, lvars.obd_vars,
2124                                               lcfg, obd);
2125                 if (rc > 0)
2126                         rc = 0;
2127                 break;
2128         }
2129         return(rc);
2130 }
2131
2132
2133 /* get remote permission for current user on fid */
2134 int mdc_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
2135                         struct obd_capa *oc, __u32 suppgid,
2136                         struct ptlrpc_request **request)
2137 {
2138         struct ptlrpc_request  *req;
2139         int                    rc;
2140         ENTRY;
2141
2142         LASSERT(client_is_remote(exp));
2143
2144         *request = NULL;
2145         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
2146         if (req == NULL)
2147                 RETURN(-ENOMEM);
2148
2149         mdc_set_capa_size(req, &RMF_CAPA1, oc);
2150
2151         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
2152         if (rc) {
2153                 ptlrpc_request_free(req);
2154                 RETURN(rc);
2155         }
2156
2157         mdc_pack_body(req, fid, oc, OBD_MD_FLRMTPERM, 0, suppgid, 0);
2158
2159         req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
2160                              sizeof(struct mdt_remote_perm));
2161
2162         ptlrpc_request_set_replen(req);
2163
2164         rc = ptlrpc_queue_wait(req);
2165         if (rc)
2166                 ptlrpc_req_finished(req);
2167         else
2168                 *request = req;
2169         RETURN(rc);
2170 }
2171
2172 static int mdc_interpret_renew_capa(const struct lu_env *env,
2173                                     struct ptlrpc_request *req, void *args,
2174                                     int status)
2175 {
2176         struct mdc_renew_capa_args *ra = args;
2177         struct mdt_body *body = NULL;
2178         struct lustre_capa *capa;
2179         ENTRY;
2180
2181         if (status)
2182                 GOTO(out, capa = ERR_PTR(status));
2183
2184         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2185         if (body == NULL)
2186                 GOTO(out, capa = ERR_PTR(-EFAULT));
2187
2188         if ((body->valid & OBD_MD_FLOSSCAPA) == 0)
2189                 GOTO(out, capa = ERR_PTR(-ENOENT));
2190
2191         capa = req_capsule_server_get(&req->rq_pill, &RMF_CAPA2);
2192         if (!capa)
2193                 GOTO(out, capa = ERR_PTR(-EFAULT));
2194         EXIT;
2195 out:
2196         ra->ra_cb(ra->ra_oc, capa);
2197         return 0;
2198 }
2199
2200 static int mdc_renew_capa(struct obd_export *exp, struct obd_capa *oc,
2201                           renew_capa_cb_t cb)
2202 {
2203         struct ptlrpc_request *req;
2204         struct mdc_renew_capa_args *ra;
2205         ENTRY;
2206
2207         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_GETATTR,
2208                                         LUSTRE_MDS_VERSION, MDS_GETATTR);
2209         if (req == NULL)
2210                 RETURN(-ENOMEM);
2211
2212         /* NB, OBD_MD_FLOSSCAPA is set here, but it doesn't necessarily mean the
2213          * capa to renew is oss capa.
2214          */
2215         mdc_pack_body(req, &oc->c_capa.lc_fid, oc, OBD_MD_FLOSSCAPA, 0, -1, 0);
2216         ptlrpc_request_set_replen(req);
2217
2218         CLASSERT(sizeof(*ra) <= sizeof(req->rq_async_args));
2219         ra = ptlrpc_req_async_args(req);
2220         ra->ra_oc = oc;
2221         ra->ra_cb = cb;
2222         req->rq_interpret_reply = mdc_interpret_renew_capa;
2223         ptlrpcd_add_req(req, PSCOPE_OTHER);
2224         RETURN(0);
2225 }
2226
2227 static int mdc_connect(const struct lu_env *env,
2228                        struct obd_export **exp,
2229                        struct obd_device *obd, struct obd_uuid *cluuid,
2230                        struct obd_connect_data *data,
2231                        void *localdata)
2232 {
2233         struct obd_import *imp = obd->u.cli.cl_import;
2234
2235         /* mds-mds import features */
2236         if (data && (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
2237                 cfs_spin_lock(&imp->imp_lock);
2238                 imp->imp_server_timeout = 1;
2239                 cfs_spin_unlock(&imp->imp_lock);
2240                 imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
2241                 CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
2242                        obd->obd_name);
2243         }
2244
2245         return client_connect_import(env, exp, obd, cluuid, data, NULL);
2246 }
2247
2248 struct obd_ops mdc_obd_ops = {
2249         .o_owner            = THIS_MODULE,
2250         .o_setup            = mdc_setup,
2251         .o_precleanup       = mdc_precleanup,
2252         .o_cleanup          = mdc_cleanup,
2253         .o_add_conn         = client_import_add_conn,
2254         .o_del_conn         = client_import_del_conn,
2255         .o_connect          = mdc_connect,
2256         .o_disconnect       = client_disconnect_export,
2257         .o_iocontrol        = mdc_iocontrol,
2258         .o_set_info_async   = mdc_set_info_async,
2259         .o_statfs           = mdc_statfs,
2260         .o_pin              = mdc_pin,
2261         .o_unpin            = mdc_unpin,
2262         .o_fid_init         = mdc_fid_init,
2263         .o_fid_fini         = mdc_fid_fini,
2264         .o_fid_alloc        = mdc_fid_alloc,
2265         .o_fid_delete       = mdc_fid_delete,
2266         .o_import_event     = mdc_import_event,
2267         .o_llog_init        = mdc_llog_init,
2268         .o_llog_finish      = mdc_llog_finish,
2269         .o_get_info         = mdc_get_info,
2270         .o_process_config   = mdc_process_config,
2271         .o_get_uuid         = mdc_get_uuid,
2272 };
2273
2274 struct md_ops mdc_md_ops = {
2275         .m_getstatus        = mdc_getstatus,
2276         .m_change_cbdata    = mdc_change_cbdata,
2277         .m_find_cbdata      = mdc_find_cbdata,
2278         .m_close            = mdc_close,
2279         .m_create           = mdc_create,
2280         .m_done_writing     = mdc_done_writing,
2281         .m_enqueue          = mdc_enqueue,
2282         .m_getattr          = mdc_getattr,
2283         .m_getattr_name     = mdc_getattr_name,
2284         .m_intent_lock      = mdc_intent_lock,
2285         .m_link             = mdc_link,
2286         .m_is_subdir        = mdc_is_subdir,
2287         .m_rename           = mdc_rename,
2288         .m_setattr          = mdc_setattr,
2289         .m_setxattr         = mdc_setxattr,
2290         .m_getxattr         = mdc_getxattr,
2291         .m_sync             = mdc_sync,
2292         .m_readpage         = mdc_readpage,
2293         .m_unlink           = mdc_unlink,
2294         .m_cancel_unused    = mdc_cancel_unused,
2295         .m_init_ea_size     = mdc_init_ea_size,
2296         .m_set_lock_data    = mdc_set_lock_data,
2297         .m_lock_match       = mdc_lock_match,
2298         .m_get_lustre_md    = mdc_get_lustre_md,
2299         .m_free_lustre_md   = mdc_free_lustre_md,
2300         .m_set_open_replay_data = mdc_set_open_replay_data,
2301         .m_clear_open_replay_data = mdc_clear_open_replay_data,
2302         .m_renew_capa       = mdc_renew_capa,
2303         .m_unpack_capa      = mdc_unpack_capa,
2304         .m_get_remote_perm  = mdc_get_remote_perm,
2305         .m_intent_getattr_async = mdc_intent_getattr_async,
2306         .m_revalidate_lock      = mdc_revalidate_lock
2307 };
2308
2309 int __init mdc_init(void)
2310 {
2311         int rc;
2312         struct lprocfs_static_vars lvars = { 0 };
2313         lprocfs_mdc_init_vars(&lvars);
2314
2315         cfs_request_module("lquota");
2316         quota_interface = PORTAL_SYMBOL_GET(mdc_quota_interface);
2317         init_obd_quota_ops(quota_interface, &mdc_obd_ops);
2318
2319         rc = class_register_type(&mdc_obd_ops, &mdc_md_ops, lvars.module_vars,
2320                                  LUSTRE_MDC_NAME, NULL);
2321         if (rc && quota_interface)
2322                 PORTAL_SYMBOL_PUT(mdc_quota_interface);
2323
2324         RETURN(rc);
2325 }
2326
2327 #ifdef __KERNEL__
2328 static void /*__exit*/ mdc_exit(void)
2329 {
2330         if (quota_interface)
2331                 PORTAL_SYMBOL_PUT(mdc_quota_interface);
2332
2333         class_unregister_type(LUSTRE_MDC_NAME);
2334 }
2335
2336 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2337 MODULE_DESCRIPTION("Lustre Metadata Client");
2338 MODULE_LICENSE("GPL");
2339
2340 module_init(mdc_init);
2341 module_exit(mdc_exit);
2342 #endif