drivers/staging/lustre/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.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2015, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  */
  32
  33 #define DEBUG_SUBSYSTEM S_MDC
  34
  35 # include <linux/module.h>
  36 # include <linux/pagemap.h>
  37 # include <linux/miscdevice.h>
  38 # include <linux/init.h>
  39 # include <linux/utsname.h>
  40
  41 #include "../include/lustre_acl.h"
  42 #include "../include/lustre/lustre_ioctl.h"
  43 #include "../include/obd_class.h"
  44 #include "../include/lustre_lmv.h"
  45 #include "../include/lustre_fid.h"
  46 #include "../include/lprocfs_status.h"
  47 #include "../include/lustre_param.h"
  48 #include "../include/lustre_log.h"
  49 #include "../include/lustre_kernelcomm.h"
  50
  51 #include "mdc_internal.h"
  52
  53 #define REQUEST_MINOR 244
  54
  55 static int mdc_cleanup(struct obd_device *obd);
  56
  57 static inline int mdc_queue_wait(struct ptlrpc_request *req)
  58 {
  59         struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
  60         int rc;
  61
  62         /* obd_get_request_slot() ensures that this client has no more
  63          * than cl_max_rpcs_in_flight RPCs simultaneously inf light
  64          * against an MDT.
  65          */
  66         rc = obd_get_request_slot(cli);
  67         if (rc != 0)
  68                 return rc;
  69
  70         rc = ptlrpc_queue_wait(req);
  71         obd_put_request_slot(cli);
  72
  73         return rc;
  74 }
  75
  76 static int mdc_getstatus(struct obd_export *exp, struct lu_fid *rootfid)
  77 {
  78         struct ptlrpc_request *req;
  79         struct mdt_body       *body;
  80         int                 rc;
  81
  82         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
  83                                         &RQF_MDS_GETSTATUS,
  84                                         LUSTRE_MDS_VERSION, MDS_GETSTATUS);
  85         if (!req)
  86                 return -ENOMEM;
  87
  88         mdc_pack_body(req, NULL, 0, 0, -1, 0);
  89         req->rq_send_state = LUSTRE_IMP_FULL;
  90
  91         ptlrpc_request_set_replen(req);
  92
  93         rc = ptlrpc_queue_wait(req);
  94         if (rc)
  95                 goto out;
  96
  97         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
  98         if (!body) {
  99                 rc = -EPROTO;
 100                 goto out;
 101         }
 102
 103         *rootfid = body->mbo_fid1;
 104         CDEBUG(D_NET,
 105                "root fid="DFID", last_committed=%llu\n",
 106                PFID(rootfid),
 107                lustre_msg_get_last_committed(req->rq_repmsg));
 108 out:
 109         ptlrpc_req_finished(req);
 110         return rc;
 111 }
 112
 113 /*
 114  * This function now is known to always saying that it will receive 4 buffers
 115  * from server. Even for cases when acl_size and md_size is zero, RPC header
 116  * will contain 4 fields and RPC itself will contain zero size fields. This is
 117  * because mdt_getattr*() _always_ returns 4 fields, but if acl is not needed
 118  * and thus zero, it shrinks it, making zero size. The same story about
 119  * md_size. And this is course of problem when client waits for smaller number
 120  * of fields. This issue will be fixed later when client gets aware of RPC
 121  * layouts.  --umka
 122  */
 123 static int mdc_getattr_common(struct obd_export *exp,
 124                               struct ptlrpc_request *req)
 125 {
 126         struct req_capsule *pill = &req->rq_pill;
 127         struct mdt_body    *body;
 128         void           *eadata;
 129         int              rc;
 130
 131         /* Request message already built. */
 132         rc = ptlrpc_queue_wait(req);
 133         if (rc != 0)
 134                 return rc;
 135
 136         /* sanity check for the reply */
 137         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 138         if (!body)
 139                 return -EPROTO;
 140
 141         CDEBUG(D_NET, "mode: %o\n", body->mbo_mode);
 142
 143         mdc_update_max_ea_from_body(exp, body);
 144         if (body->mbo_eadatasize != 0) {
 145                 eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
 146                                                       body->mbo_eadatasize);
 147                 if (!eadata)
 148                         return -EPROTO;
 149         }
 150
 151         return 0;
 152 }
 153
 154 static int mdc_getattr(struct obd_export *exp, struct md_op_data *op_data,
 155                        struct ptlrpc_request **request)
 156 {
 157         struct ptlrpc_request *req;
 158         int                 rc;
 159
 160         /* Single MDS without an LMV case */
 161         if (op_data->op_flags & MF_GET_MDT_IDX) {
 162                 op_data->op_mds = 0;
 163                 return 0;
 164         }
 165         *request = NULL;
 166         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
 167         if (!req)
 168                 return -ENOMEM;
 169
 170         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
 171         if (rc) {
 172                 ptlrpc_request_free(req);
 173                 return rc;
 174         }
 175
 176         mdc_pack_body(req, &op_data->op_fid1, op_data->op_valid,
 177                       op_data->op_mode, -1, 0);
 178
 179         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 180                              op_data->op_mode);
 181         ptlrpc_request_set_replen(req);
 182
 183         rc = mdc_getattr_common(exp, req);
 184         if (rc)
 185                 ptlrpc_req_finished(req);
 186         else
 187                 *request = req;
 188         return rc;
 189 }
 190
 191 static int mdc_getattr_name(struct obd_export *exp, struct md_op_data *op_data,
 192                             struct ptlrpc_request **request)
 193 {
 194         struct ptlrpc_request *req;
 195         int                 rc;
 196
 197         *request = NULL;
 198         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 199                                    &RQF_MDS_GETATTR_NAME);
 200         if (!req)
 201                 return -ENOMEM;
 202
 203         req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 204                              op_data->op_namelen + 1);
 205
 206         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR_NAME);
 207         if (rc) {
 208                 ptlrpc_request_free(req);
 209                 return rc;
 210         }
 211
 212         mdc_pack_body(req, &op_data->op_fid1, op_data->op_valid,
 213                       op_data->op_mode, op_data->op_suppgids[0], 0);
 214
 215         if (op_data->op_name) {
 216                 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 217
 218                 LASSERT(strnlen(op_data->op_name, op_data->op_namelen) ==
 219                                 op_data->op_namelen);
 220                 memcpy(name, op_data->op_name, op_data->op_namelen);
 221         }
 222
 223         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 224                              op_data->op_mode);
 225         ptlrpc_request_set_replen(req);
 226
 227         rc = mdc_getattr_common(exp, req);
 228         if (rc)
 229                 ptlrpc_req_finished(req);
 230         else
 231                 *request = req;
 232         return rc;
 233 }
 234
 235 static int mdc_xattr_common(struct obd_export *exp,
 236                             const struct req_format *fmt,
 237                             const struct lu_fid *fid,
 238                             int opcode, u64 valid,
 239                             const char *xattr_name, const char *input,
 240                             int input_size, int output_size, int flags,
 241                             __u32 suppgid, struct ptlrpc_request **request)
 242 {
 243         struct ptlrpc_request *req;
 244         int   xattr_namelen = 0;
 245         char *tmp;
 246         int   rc;
 247
 248         *request = NULL;
 249         req = ptlrpc_request_alloc(class_exp2cliimp(exp), fmt);
 250         if (!req)
 251                 return -ENOMEM;
 252
 253         if (xattr_name) {
 254                 xattr_namelen = strlen(xattr_name) + 1;
 255                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 256                                      xattr_namelen);
 257         }
 258         if (input_size) {
 259                 LASSERT(input);
 260                 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
 261                                      input_size);
 262         }
 263
 264         /* Flush local XATTR locks to get rid of a possible cancel RPC */
 265         if (opcode == MDS_REINT && fid_is_sane(fid) &&
 266             exp->exp_connect_data.ocd_ibits_known & MDS_INODELOCK_XATTR) {
 267                 LIST_HEAD(cancels);
 268                 int count;
 269
 270                 /* Without that packing would fail */
 271                 if (input_size == 0)
 272                         req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
 273                                              RCL_CLIENT, 0);
 274
 275                 count = mdc_resource_get_unused(exp, fid,
 276                                                 &cancels, LCK_EX,
 277                                                 MDS_INODELOCK_XATTR);
 278
 279                 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
 280                 if (rc) {
 281                         ptlrpc_request_free(req);
 282                         return rc;
 283                 }
 284         } else {
 285                 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode);
 286                 if (rc) {
 287                         ptlrpc_request_free(req);
 288                         return rc;
 289                 }
 290         }
 291
 292         if (opcode == MDS_REINT) {
 293                 struct mdt_rec_setxattr *rec;
 294
 295                 CLASSERT(sizeof(struct mdt_rec_setxattr) ==
 296                          sizeof(struct mdt_rec_reint));
 297                 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
 298                 rec->sx_opcode = REINT_SETXATTR;
 299                 rec->sx_fsuid  = from_kuid(&init_user_ns, current_fsuid());
 300                 rec->sx_fsgid  = from_kgid(&init_user_ns, current_fsgid());
 301                 rec->sx_cap    = cfs_curproc_cap_pack();
 302                 rec->sx_suppgid1 = suppgid;
 303                 rec->sx_suppgid2 = -1;
 304                 rec->sx_fid    = *fid;
 305                 rec->sx_valid  = valid | OBD_MD_FLCTIME;
 306                 rec->sx_time   = ktime_get_real_seconds();
 307                 rec->sx_size   = output_size;
 308                 rec->sx_flags  = flags;
 309
 310         } else {
 311                 mdc_pack_body(req, fid, valid, output_size, suppgid, flags);
 312         }
 313
 314         if (xattr_name) {
 315                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 316                 memcpy(tmp, xattr_name, xattr_namelen);
 317         }
 318         if (input_size) {
 319                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
 320                 memcpy(tmp, input, input_size);
 321         }
 322
 323         if (req_capsule_has_field(&req->rq_pill, &RMF_EADATA, RCL_SERVER))
 324                 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
 325                                      RCL_SERVER, output_size);
 326         ptlrpc_request_set_replen(req);
 327
 328         /* make rpc */
 329         if (opcode == MDS_REINT)
 330                 mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 331
 332         rc = ptlrpc_queue_wait(req);
 333
 334         if (opcode == MDS_REINT)
 335                 mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
 336
 337         if (rc)
 338                 ptlrpc_req_finished(req);
 339         else
 340                 *request = req;
 341         return rc;
 342 }
 343
 344 static int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
 345                         u64 valid, const char *xattr_name,
 346                         const char *input, int input_size, int output_size,
 347                         int flags, __u32 suppgid,
 348                         struct ptlrpc_request **request)
 349 {
 350         return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
 351                                 fid, MDS_REINT, valid, xattr_name,
 352                                 input, input_size, output_size, flags,
 353                                 suppgid, request);
 354 }
 355
 356 static int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
 357                         u64 valid, const char *xattr_name,
 358                         const char *input, int input_size, int output_size,
 359                         int flags, struct ptlrpc_request **request)
 360 {
 361         return mdc_xattr_common(exp, &RQF_MDS_GETXATTR,
 362                                 fid, MDS_GETXATTR, valid, xattr_name,
 363                                 input, input_size, output_size, flags,
 364                                 -1, request);
 365 }
 366
 367 #ifdef CONFIG_FS_POSIX_ACL
 368 static int mdc_unpack_acl(struct ptlrpc_request *req, struct lustre_md *md)
 369 {
 370         struct req_capsule     *pill = &req->rq_pill;
 371         struct mdt_body *body = md->body;
 372         struct posix_acl       *acl;
 373         void               *buf;
 374         int                  rc;
 375
 376         if (!body->mbo_aclsize)
 377                 return 0;
 378
 379         buf = req_capsule_server_sized_get(pill, &RMF_ACL, body->mbo_aclsize);
 380
 381         if (!buf)
 382                 return -EPROTO;
 383
 384         acl = posix_acl_from_xattr(&init_user_ns, buf, body->mbo_aclsize);
 385         if (!acl)
 386                 return 0;
 387
 388         if (IS_ERR(acl)) {
 389                 rc = PTR_ERR(acl);
 390                 CERROR("convert xattr to acl: %d\n", rc);
 391                 return rc;
 392         }
 393
 394         rc = posix_acl_valid(&init_user_ns, acl);
 395         if (rc) {
 396                 CERROR("validate acl: %d\n", rc);
 397                 posix_acl_release(acl);
 398                 return rc;
 399         }
 400
 401         md->posix_acl = acl;
 402         return 0;
 403 }
 404 #else
 405 #define mdc_unpack_acl(req, md) 0
 406 #endif
 407
 408 static int mdc_get_lustre_md(struct obd_export *exp,
 409                              struct ptlrpc_request *req,
 410                              struct obd_export *dt_exp,
 411                              struct obd_export *md_exp,
 412                              struct lustre_md *md)
 413 {
 414         struct req_capsule *pill = &req->rq_pill;
 415         int rc;
 416
 417         LASSERT(md);
 418         memset(md, 0, sizeof(*md));
 419
 420         md->body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 421
 422         if (md->body->mbo_valid & OBD_MD_FLEASIZE) {
 423                 int lmmsize;
 424                 struct lov_mds_md *lmm;
 425
 426                 if (!S_ISREG(md->body->mbo_mode)) {
 427                         CDEBUG(D_INFO,
 428                                "OBD_MD_FLEASIZE set, should be a regular file, but is not\n");
 429                         rc = -EPROTO;
 430                         goto out;
 431                 }
 432
 433                 if (md->body->mbo_eadatasize == 0) {
 434                         CDEBUG(D_INFO,
 435                                "OBD_MD_FLEASIZE set, but eadatasize 0\n");
 436                         rc = -EPROTO;
 437                         goto out;
 438                 }
 439                 lmmsize = md->body->mbo_eadatasize;
 440                 lmm = req_capsule_server_sized_get(pill, &RMF_MDT_MD, lmmsize);
 441                 if (!lmm) {
 442                         rc = -EPROTO;
 443                         goto out;
 444                 }
 445
 446                 rc = obd_unpackmd(dt_exp, &md->lsm, lmm, lmmsize);
 447                 if (rc < 0)
 448                         goto out;
 449
 450                 if (rc < (typeof(rc))sizeof(*md->lsm)) {
 451                         CDEBUG(D_INFO,
 452                                "lsm size too small: rc < sizeof (*md->lsm) (%d < %d)\n",
 453                                rc, (int)sizeof(*md->lsm));
 454                         rc = -EPROTO;
 455                         goto out;
 456                 }
 457
 458         } else if (md->body->mbo_valid & OBD_MD_FLDIREA) {
 459                 int lmvsize;
 460                 struct lov_mds_md *lmv;
 461
 462                 if (!S_ISDIR(md->body->mbo_mode)) {
 463                         CDEBUG(D_INFO,
 464                                "OBD_MD_FLDIREA set, should be a directory, but is not\n");
 465                         rc = -EPROTO;
 466                         goto out;
 467                 }
 468
 469                 if (md->body->mbo_eadatasize == 0) {
 470                         CDEBUG(D_INFO,
 471                                "OBD_MD_FLDIREA is set, but eadatasize 0\n");
 472                         return -EPROTO;
 473                 }
 474                 if (md->body->mbo_valid & OBD_MD_MEA) {
 475                         lmvsize = md->body->mbo_eadatasize;
 476                         lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
 477                                                            lmvsize);
 478                         if (!lmv) {
 479                                 rc = -EPROTO;
 480                                 goto out;
 481                         }
 482
 483                         rc = obd_unpackmd(md_exp, (void *)&md->lmv, lmv,
 484                                           lmvsize);
 485                         if (rc < 0)
 486                                 goto out;
 487
 488                         if (rc < (typeof(rc))sizeof(*md->lmv)) {
 489                                 CDEBUG(D_INFO,
 490                                        "size too small: rc < sizeof(*md->lmv) (%d < %d)\n",
 491                                         rc, (int)sizeof(*md->lmv));
 492                                 rc = -EPROTO;
 493                                 goto out;
 494                         }
 495                 }
 496         }
 497         rc = 0;
 498
 499         if (md->body->mbo_valid & OBD_MD_FLACL) {
 500                 /* for ACL, it's possible that FLACL is set but aclsize is zero.
 501                  * only when aclsize != 0 there's an actual segment for ACL
 502                  * in reply buffer.
 503                  */
 504                 if (md->body->mbo_aclsize) {
 505                         rc = mdc_unpack_acl(req, md);
 506                         if (rc)
 507                                 goto out;
 508 #ifdef CONFIG_FS_POSIX_ACL
 509                 } else {
 510                         md->posix_acl = NULL;
 511 #endif
 512                 }
 513         }
 514
 515 out:
 516         if (rc) {
 517 #ifdef CONFIG_FS_POSIX_ACL
 518                 posix_acl_release(md->posix_acl);
 519 #endif
 520                 if (md->lsm)
 521                         obd_free_memmd(dt_exp, &md->lsm);
 522         }
 523         return rc;
 524 }
 525
 526 static int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
 527 {
 528         return 0;
 529 }
 530
 531 /**
 532  * Handles both OPEN and SETATTR RPCs for OPEN-CLOSE and SETATTR-DONE_WRITING
 533  * RPC chains.
 534  */
 535 void mdc_replay_open(struct ptlrpc_request *req)
 536 {
 537         struct md_open_data *mod = req->rq_cb_data;
 538         struct ptlrpc_request *close_req;
 539         struct obd_client_handle *och;
 540         struct lustre_handle old;
 541         struct mdt_body *body;
 542
 543         if (!mod) {
 544                 DEBUG_REQ(D_ERROR, req,
 545                           "Can't properly replay without open data.");
 546                 return;
 547         }
 548
 549         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 550
 551         och = mod->mod_och;
 552         if (och) {
 553                 struct lustre_handle *file_fh;
 554
 555                 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
 556
 557                 file_fh = &och->och_fh;
 558                 CDEBUG(D_HA, "updating handle from %#llx to %#llx\n",
 559                        file_fh->cookie, body->mbo_handle.cookie);
 560                 old = *file_fh;
 561                 *file_fh = body->mbo_handle;
 562         }
 563         close_req = mod->mod_close_req;
 564         if (close_req) {
 565                 __u32 opc = lustre_msg_get_opc(close_req->rq_reqmsg);
 566                 struct mdt_ioepoch *epoch;
 567
 568                 LASSERT(opc == MDS_CLOSE || opc == MDS_DONE_WRITING);
 569                 epoch = req_capsule_client_get(&close_req->rq_pill,
 570                                                &RMF_MDT_EPOCH);
 571                 LASSERT(epoch);
 572
 573                 if (och)
 574                         LASSERT(!memcmp(&old, &epoch->handle, sizeof(old)));
 575                 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
 576                 epoch->handle = body->mbo_handle;
 577         }
 578 }
 579
 580 void mdc_commit_open(struct ptlrpc_request *req)
 581 {
 582         struct md_open_data *mod = req->rq_cb_data;
 583
 584         if (!mod)
 585                 return;
 586
 587         /**
 588          * No need to touch md_open_data::mod_och, it holds a reference on
 589          * \var mod and will zero references to each other, \var mod will be
 590          * freed after that when md_open_data::mod_och will put the reference.
 591          */
 592
 593         /**
 594          * Do not let open request to disappear as it still may be needed
 595          * for close rpc to happen (it may happen on evict only, otherwise
 596          * ptlrpc_request::rq_replay does not let mdc_commit_open() to be
 597          * called), just mark this rpc as committed to distinguish these 2
 598          * cases, see mdc_close() for details. The open request reference will
 599          * be put along with freeing \var mod.
 600          */
 601         ptlrpc_request_addref(req);
 602         spin_lock(&req->rq_lock);
 603         req->rq_committed = 1;
 604         spin_unlock(&req->rq_lock);
 605         req->rq_cb_data = NULL;
 606         obd_mod_put(mod);
 607 }
 608
 609 int mdc_set_open_replay_data(struct obd_export *exp,
 610                              struct obd_client_handle *och,
 611                              struct lookup_intent *it)
 612 {
 613         struct md_open_data   *mod;
 614         struct mdt_rec_create *rec;
 615         struct mdt_body       *body;
 616         struct ptlrpc_request *open_req = it->it_request;
 617         struct obd_import     *imp = open_req->rq_import;
 618
 619         if (!open_req->rq_replay)
 620                 return 0;
 621
 622         rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
 623         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
 624         LASSERT(rec);
 625         /* Incoming message in my byte order (it's been swabbed). */
 626         /* Outgoing messages always in my byte order. */
 627         LASSERT(body);
 628
 629         /* Only if the import is replayable, we set replay_open data */
 630         if (och && imp->imp_replayable) {
 631                 mod = obd_mod_alloc();
 632                 if (!mod) {
 633                         DEBUG_REQ(D_ERROR, open_req,
 634                                   "Can't allocate md_open_data");
 635                         return 0;
 636                 }
 637
 638                 /**
 639                  * Take a reference on \var mod, to be freed on mdc_close().
 640                  * It protects \var mod from being freed on eviction (commit
 641                  * callback is called despite rq_replay flag).
 642                  * Another reference for \var och.
 643                  */
 644                 obd_mod_get(mod);
 645                 obd_mod_get(mod);
 646
 647                 spin_lock(&open_req->rq_lock);
 648                 och->och_mod = mod;
 649                 mod->mod_och = och;
 650                 mod->mod_is_create = it_disposition(it, DISP_OPEN_CREATE) ||
 651                                      it_disposition(it, DISP_OPEN_STRIPE);
 652                 mod->mod_open_req = open_req;
 653                 open_req->rq_cb_data = mod;
 654                 open_req->rq_commit_cb = mdc_commit_open;
 655                 spin_unlock(&open_req->rq_lock);
 656         }
 657
 658         rec->cr_fid2 = body->mbo_fid1;
 659         rec->cr_ioepoch = body->mbo_ioepoch;
 660         rec->cr_old_handle.cookie = body->mbo_handle.cookie;
 661         open_req->rq_replay_cb = mdc_replay_open;
 662         if (!fid_is_sane(&body->mbo_fid1)) {
 663                 DEBUG_REQ(D_ERROR, open_req,
 664                           "Saving replay request with insane fid");
 665                 LBUG();
 666         }
 667
 668         DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
 669         return 0;
 670 }
 671
 672 static void mdc_free_open(struct md_open_data *mod)
 673 {
 674         int committed = 0;
 675
 676         if (mod->mod_is_create == 0 &&
 677             imp_connect_disp_stripe(mod->mod_open_req->rq_import))
 678                 committed = 1;
 679
 680         /*
 681          * No reason to asssert here if the open request has
 682          * rq_replay == 1. It means that mdc_close failed, and
 683          * close request wasn`t sent. It is not fatal to client.
 684          * The worst thing is eviction if the client gets open lock
 685          */
 686         DEBUG_REQ(D_RPCTRACE, mod->mod_open_req,
 687                   "free open request rq_replay = %d\n",
 688                    mod->mod_open_req->rq_replay);
 689
 690         ptlrpc_request_committed(mod->mod_open_req, committed);
 691         if (mod->mod_close_req)
 692                 ptlrpc_request_committed(mod->mod_close_req, committed);
 693 }
 694
 695 static int mdc_clear_open_replay_data(struct obd_export *exp,
 696                                       struct obd_client_handle *och)
 697 {
 698         struct md_open_data *mod = och->och_mod;
 699
 700         /**
 701          * It is possible to not have \var mod in a case of eviction between
 702          * lookup and ll_file_open().
 703          **/
 704         if (!mod)
 705                 return 0;
 706
 707         LASSERT(mod != LP_POISON);
 708         LASSERT(mod->mod_open_req);
 709         mdc_free_open(mod);
 710
 711         mod->mod_och = NULL;
 712         och->och_mod = NULL;
 713         obd_mod_put(mod);
 714
 715         return 0;
 716 }
 717
 718 /* Prepares the request for the replay by the given reply */
 719 static void mdc_close_handle_reply(struct ptlrpc_request *req,
 720                                    struct md_op_data *op_data, int rc) {
 721         struct mdt_body  *repbody;
 722         struct mdt_ioepoch *epoch;
 723
 724         if (req && rc == -EAGAIN) {
 725                 repbody = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 726                 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
 727
 728                 epoch->flags |= MF_SOM_AU;
 729                 if (repbody->mbo_valid & OBD_MD_FLGETATTRLOCK)
 730                         op_data->op_flags |= MF_GETATTR_LOCK;
 731         }
 732 }
 733
 734 static int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
 735                      struct md_open_data *mod, struct ptlrpc_request **request)
 736 {
 737         struct obd_device     *obd = class_exp2obd(exp);
 738         struct ptlrpc_request *req;
 739         struct req_format     *req_fmt;
 740         int                    rc;
 741         int                    saved_rc = 0;
 742
 743         req_fmt = &RQF_MDS_CLOSE;
 744         if (op_data->op_bias & MDS_HSM_RELEASE) {
 745                 req_fmt = &RQF_MDS_RELEASE_CLOSE;
 746
 747                 /* allocate a FID for volatile file */
 748                 rc = mdc_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
 749                 if (rc < 0) {
 750                         CERROR("%s: "DFID" failed to allocate FID: %d\n",
 751                                obd->obd_name, PFID(&op_data->op_fid1), rc);
 752                         /* save the errcode and proceed to close */
 753                         saved_rc = rc;
 754                 }
 755         }
 756
 757         *request = NULL;
 758         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_CLOSE))
 759                 req = NULL;
 760         else
 761                 req = ptlrpc_request_alloc(class_exp2cliimp(exp), req_fmt);
 762
 763         /* Ensure that this close's handle is fixed up during replay. */
 764         if (likely(mod)) {
 765                 LASSERTF(mod->mod_open_req &&
 766                          mod->mod_open_req->rq_type != LI_POISON,
 767                          "POISONED open %p!\n", mod->mod_open_req);
 768
 769                 mod->mod_close_req = req;
 770
 771                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
 772                 /* We no longer want to preserve this open for replay even
 773                  * though the open was committed. b=3632, b=3633
 774                  */
 775                 spin_lock(&mod->mod_open_req->rq_lock);
 776                 mod->mod_open_req->rq_replay = 0;
 777                 spin_unlock(&mod->mod_open_req->rq_lock);
 778         } else {
 779                  CDEBUG(D_HA,
 780                         "couldn't find open req; expecting close error\n");
 781         }
 782         if (!req) {
 783                 /*
 784                  * TODO: repeat close after errors
 785                  */
 786                 CWARN("%s: close of FID "DFID" failed, file reference will be dropped when this client unmounts or is evicted\n",
 787                       obd->obd_name, PFID(&op_data->op_fid1));
 788                 rc = -ENOMEM;
 789                 goto out;
 790         }
 791
 792         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
 793         if (rc) {
 794                 ptlrpc_request_free(req);
 795                 goto out;
 796         }
 797
 798         /*
 799          * To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
 800          * portal whose threads are not taking any DLM locks and are therefore
 801          * always progressing
 802          */
 803         req->rq_request_portal = MDS_READPAGE_PORTAL;
 804         ptlrpc_at_set_req_timeout(req);
 805
 806         mdc_close_pack(req, op_data);
 807
 808         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 809                              obd->u.cli.cl_default_mds_easize);
 810         req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
 811                              obd->u.cli.cl_default_mds_cookiesize);
 812
 813         ptlrpc_request_set_replen(req);
 814
 815         mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 816         rc = ptlrpc_queue_wait(req);
 817         mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 818
 819         if (!req->rq_repmsg) {
 820                 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
 821                        req->rq_status);
 822                 if (rc == 0)
 823                         rc = req->rq_status ?: -EIO;
 824         } else if (rc == 0 || rc == -EAGAIN) {
 825                 struct mdt_body *body;
 826
 827                 rc = lustre_msg_get_status(req->rq_repmsg);
 828                 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
 829                         DEBUG_REQ(D_ERROR, req,
 830                                   "type == PTL_RPC_MSG_ERR, err = %d", rc);
 831                         if (rc > 0)
 832                                 rc = -rc;
 833                 }
 834                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 835                 if (!body)
 836                         rc = -EPROTO;
 837         } else if (rc == -ESTALE) {
 838                 /**
 839                  * it can be allowed error after 3633 if open was committed and
 840                  * server failed before close was sent. Let's check if mod
 841                  * exists and return no error in that case
 842                  */
 843                 if (mod) {
 844                         DEBUG_REQ(D_HA, req, "Reset ESTALE = %d", rc);
 845                         if (mod->mod_open_req->rq_committed)
 846                                 rc = 0;
 847                 }
 848         }
 849
 850 out:
 851         if (mod) {
 852                 if (rc != 0)
 853                         mod->mod_close_req = NULL;
 854                 /* Since now, mod is accessed through open_req only,
 855                  * thus close req does not keep a reference on mod anymore.
 856                  */
 857                 obd_mod_put(mod);
 858         }
 859         *request = req;
 860         mdc_close_handle_reply(req, op_data, rc);
 861         return rc < 0 ? rc : saved_rc;
 862 }
 863
 864 static int mdc_done_writing(struct obd_export *exp, struct md_op_data *op_data,
 865                             struct md_open_data *mod)
 866 {
 867         struct obd_device     *obd = class_exp2obd(exp);
 868         struct ptlrpc_request *req;
 869         int                 rc;
 870
 871         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 872                                    &RQF_MDS_DONE_WRITING);
 873         if (!req)
 874                 return -ENOMEM;
 875
 876         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_DONE_WRITING);
 877         if (rc) {
 878                 ptlrpc_request_free(req);
 879                 return rc;
 880         }
 881
 882         if (mod) {
 883                 LASSERTF(mod->mod_open_req &&
 884                          mod->mod_open_req->rq_type != LI_POISON,
 885                          "POISONED setattr %p!\n", mod->mod_open_req);
 886
 887                 mod->mod_close_req = req;
 888                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched setattr");
 889                 /* We no longer want to preserve this setattr for replay even
 890                  * though the open was committed. b=3632, b=3633
 891                  */
 892                 spin_lock(&mod->mod_open_req->rq_lock);
 893                 mod->mod_open_req->rq_replay = 0;
 894                 spin_unlock(&mod->mod_open_req->rq_lock);
 895         }
 896
 897         mdc_close_pack(req, op_data);
 898         ptlrpc_request_set_replen(req);
 899
 900         mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 901         rc = ptlrpc_queue_wait(req);
 902         mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
 903
 904         if (rc == -ESTALE) {
 905                 /**
 906                  * it can be allowed error after 3633 if open or setattr were
 907                  * committed and server failed before close was sent.
 908                  * Let's check if mod exists and return no error in that case
 909                  */
 910                 if (mod) {
 911                         if (mod->mod_open_req->rq_committed)
 912                                 rc = 0;
 913                 }
 914         }
 915
 916         if (mod) {
 917                 if (rc != 0)
 918                         mod->mod_close_req = NULL;
 919                 LASSERT(mod->mod_open_req);
 920                 mdc_free_open(mod);
 921
 922                 /* Since now, mod is accessed through setattr req only,
 923                  * thus DW req does not keep a reference on mod anymore.
 924                  */
 925                 obd_mod_put(mod);
 926         }
 927
 928         mdc_close_handle_reply(req, op_data, rc);
 929         ptlrpc_req_finished(req);
 930         return rc;
 931 }
 932
 933 static int mdc_getpage(struct obd_export *exp, const struct lu_fid *fid,
 934                        u64 offset, struct page **pages, int npages,
 935                        struct ptlrpc_request **request)
 936 {
 937         struct ptlrpc_bulk_desc *desc;
 938         struct ptlrpc_request *req;
 939         wait_queue_head_t waitq;
 940         struct l_wait_info lwi;
 941         int resends = 0;
 942         int rc;
 943         int i;
 944
 945         *request = NULL;
 946         init_waitqueue_head(&waitq);
 947
 948 restart_bulk:
 949         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
 950         if (!req)
 951                 return -ENOMEM;
 952
 953         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
 954         if (rc) {
 955                 ptlrpc_request_free(req);
 956                 return rc;
 957         }
 958
 959         req->rq_request_portal = MDS_READPAGE_PORTAL;
 960         ptlrpc_at_set_req_timeout(req);
 961
 962         desc = ptlrpc_prep_bulk_imp(req, npages, 1, BULK_PUT_SINK,
 963                                     MDS_BULK_PORTAL);
 964         if (!desc) {
 965                 ptlrpc_request_free(req);
 966                 return -ENOMEM;
 967         }
 968
 969         /* NB req now owns desc and will free it when it gets freed */
 970         for (i = 0; i < npages; i++)
 971                 ptlrpc_prep_bulk_page_pin(desc, pages[i], 0, PAGE_SIZE);
 972
 973         mdc_readdir_pack(req, offset, PAGE_SIZE * npages, fid);
 974
 975         ptlrpc_request_set_replen(req);
 976         rc = ptlrpc_queue_wait(req);
 977         if (rc) {
 978                 ptlrpc_req_finished(req);
 979                 if (rc != -ETIMEDOUT)
 980                         return rc;
 981
 982                 resends++;
 983                 if (!client_should_resend(resends, &exp->exp_obd->u.cli)) {
 984                         CERROR("%s: too many resend retries: rc = %d\n",
 985                                exp->exp_obd->obd_name, -EIO);
 986                         return -EIO;
 987                 }
 988                 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(resends), NULL, NULL,
 989                                        NULL);
 990                 l_wait_event(waitq, 0, &lwi);
 991
 992                 goto restart_bulk;
 993         }
 994
 995         rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
 996                                           req->rq_bulk->bd_nob_transferred);
 997         if (rc < 0) {
 998                 ptlrpc_req_finished(req);
 999                 return rc;
1000         }
1001
1002         if (req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK) {
1003                 CERROR("%s: unexpected bytes transferred: %d (%ld expected)\n",
1004                        exp->exp_obd->obd_name, req->rq_bulk->bd_nob_transferred,
1005                        PAGE_SIZE * npages);
1006                 ptlrpc_req_finished(req);
1007                 return -EPROTO;
1008         }
1009
1010         *request = req;
1011         return 0;
1012 }
1013
1014 static void mdc_release_page(struct page *page, int remove)
1015 {
1016         if (remove) {
1017                 lock_page(page);
1018                 if (likely(page->mapping))
1019                         truncate_complete_page(page->mapping, page);
1020                 unlock_page(page);
1021         }
1022         put_page(page);
1023 }
1024
1025 static struct page *mdc_page_locate(struct address_space *mapping, __u64 *hash,
1026                                     __u64 *start, __u64 *end, int hash64)
1027 {
1028         /*
1029          * Complement of hash is used as an index so that
1030          * radix_tree_gang_lookup() can be used to find a page with starting
1031          * hash _smaller_ than one we are looking for.
1032          */
1033         unsigned long offset = hash_x_index(*hash, hash64);
1034         struct page *page;
1035         int found;
1036
1037         spin_lock_irq(&mapping->tree_lock);
1038         found = radix_tree_gang_lookup(&mapping->page_tree,
1039                                        (void **)&page, offset, 1);
1040         if (found > 0 && !radix_tree_exceptional_entry(page)) {
1041                 struct lu_dirpage *dp;
1042
1043                 get_page(page);
1044                 spin_unlock_irq(&mapping->tree_lock);
1045                 /*
1046                  * In contrast to find_lock_page() we are sure that directory
1047                  * page cannot be truncated (while DLM lock is held) and,
1048                  * hence, can avoid restart.
1049                  *
1050                  * In fact, page cannot be locked here at all, because
1051                  * mdc_read_page_remote does synchronous io.
1052                  */
1053                 wait_on_page_locked(page);
1054                 if (PageUptodate(page)) {
1055                         dp = kmap(page);
1056                         if (BITS_PER_LONG == 32 && hash64) {
1057                                 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1058                                 *end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
1059                                 *hash  = *hash >> 32;
1060                         } else {
1061                                 *start = le64_to_cpu(dp->ldp_hash_start);
1062                                 *end   = le64_to_cpu(dp->ldp_hash_end);
1063                         }
1064                         if (unlikely(*start == 1 && *hash == 0))
1065                                 *hash = *start;
1066                         else
1067                                 LASSERTF(*start <= *hash, "start = %#llx,end = %#llx,hash = %#llx\n",
1068                                          *start, *end, *hash);
1069                         CDEBUG(D_VFSTRACE, "offset %lx [%#llx %#llx], hash %#llx\n",
1070                                offset, *start, *end, *hash);
1071                         if (*hash > *end) {
1072                                 kunmap(page);
1073                                 mdc_release_page(page, 0);
1074                                 page = NULL;
1075                         } else if (*end != *start && *hash == *end) {
1076                                 /*
1077                                  * upon hash collision, remove this page,
1078                                  * otherwise put page reference, and
1079                                  * mdc_read_page_remote() will issue RPC to
1080                                  * fetch the page we want.
1081                                  */
1082                                 kunmap(page);
1083                                 mdc_release_page(page,
1084                                                  le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
1085                                 page = NULL;
1086                         }
1087                 } else {
1088                         put_page(page);
1089                         page = ERR_PTR(-EIO);
1090                 }
1091         } else {
1092                 spin_unlock_irq(&mapping->tree_lock);
1093                 page = NULL;
1094         }
1095         return page;
1096 }
1097
1098 /*
1099  * Adjust a set of pages, each page containing an array of lu_dirpages,
1100  * so that each page can be used as a single logical lu_dirpage.
1101  *
1102  * A lu_dirpage is laid out as follows, where s = ldp_hash_start,
1103  * e = ldp_hash_end, f = ldp_flags, p = padding, and each "ent" is a
1104  * struct lu_dirent.  It has size up to LU_PAGE_SIZE. The ldp_hash_end
1105  * value is used as a cookie to request the next lu_dirpage in a
1106  * directory listing that spans multiple pages (two in this example):
1107  *   ________
1108  *  |        |
1109  * .|--------v-------   -----.
1110  * |s|e|f|p|ent|ent| ... |ent|
1111  * '--|--------------   -----'   Each PAGE contains a single
1112  *    '------.                   lu_dirpage.
1113  * .---------v-------   -----.
1114  * |s|e|f|p|ent| 0 | ... | 0 |
1115  * '-----------------   -----'
1116  *
1117  * However, on hosts where the native VM page size (PAGE_SIZE) is
1118  * larger than LU_PAGE_SIZE, a single host page may contain multiple
1119  * lu_dirpages. After reading the lu_dirpages from the MDS, the
1120  * ldp_hash_end of the first lu_dirpage refers to the one immediately
1121  * after it in the same PAGE (arrows simplified for brevity, but
1122  * in general e0==s1, e1==s2, etc.):
1123  *
1124  * .--------------------   -----.
1125  * |s0|e0|f0|p|ent|ent| ... |ent|
1126  * |---v----------------   -----|
1127  * |s1|e1|f1|p|ent|ent| ... |ent|
1128  * |---v----------------   -----|  Here, each PAGE contains
1129  *             ...                 multiple lu_dirpages.
1130  * |---v----------------   -----|
1131  * |s'|e'|f'|p|ent|ent| ... |ent|
1132  * '---|----------------   -----'
1133  *     v
1134  * .----------------------------.
1135  * |        next PAGE           |
1136  *
1137  * This structure is transformed into a single logical lu_dirpage as follows:
1138  *
1139  * - Replace e0 with e' so the request for the next lu_dirpage gets the page
1140  *   labeled 'next PAGE'.
1141  *
1142  * - Copy the LDF_COLLIDE flag from f' to f0 to correctly reflect whether
1143  *   a hash collision with the next page exists.
1144  *
1145  * - Adjust the lde_reclen of the ending entry of each lu_dirpage to span
1146  *   to the first entry of the next lu_dirpage.
1147  */
1148 #if PAGE_SIZE > LU_PAGE_SIZE
1149 static void mdc_adjust_dirpages(struct page **pages, int cfs_pgs, int lu_pgs)
1150 {
1151         int i;
1152
1153         for (i = 0; i < cfs_pgs; i++) {
1154                 struct lu_dirpage *dp = kmap(pages[i]);
1155                 __u64 hash_end = le64_to_cpu(dp->ldp_hash_end);
1156                 __u32 flags = le32_to_cpu(dp->ldp_flags);
1157                 struct lu_dirpage *first = dp;
1158                 struct lu_dirent *end_dirent = NULL;
1159                 struct lu_dirent *ent;
1160
1161                 while (--lu_pgs > 0) {
1162                         ent = lu_dirent_start(dp);
1163                         for (end_dirent = ent; ent;
1164                              end_dirent = ent, ent = lu_dirent_next(ent));
1165
1166                         /* Advance dp to next lu_dirpage. */
1167                         dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE);
1168
1169                         /* Check if we've reached the end of the CFS_PAGE. */
1170                         if (!((unsigned long)dp & ~PAGE_MASK))
1171                                 break;
1172
1173                         /* Save the hash and flags of this lu_dirpage. */
1174                         hash_end = le64_to_cpu(dp->ldp_hash_end);
1175                         flags = le32_to_cpu(dp->ldp_flags);
1176
1177                         /* Check if lu_dirpage contains no entries. */
1178                         if (!end_dirent)
1179                                 break;
1180
1181                         /*
1182                          * Enlarge the end entry lde_reclen from 0 to
1183                          * first entry of next lu_dirpage.
1184                          */
1185                         LASSERT(!le16_to_cpu(end_dirent->lde_reclen));
1186                         end_dirent->lde_reclen =
1187                                 cpu_to_le16((char *)(dp->ldp_entries) -
1188                                             (char *)end_dirent);
1189                 }
1190
1191                 first->ldp_hash_end = hash_end;
1192                 first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE);
1193                 first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE);
1194
1195                 kunmap(pages[i]);
1196         }
1197         LASSERTF(lu_pgs == 0, "left = %d", lu_pgs);
1198 }
1199 #else
1200 #define mdc_adjust_dirpages(pages, cfs_pgs, lu_pgs) do {} while (0)
1201 #endif  /* PAGE_SIZE > LU_PAGE_SIZE */
1202
1203 /* parameters for readdir page */
1204 struct readpage_param {
1205         struct md_op_data       *rp_mod;
1206         __u64                   rp_off;
1207         int                     rp_hash64;
1208         struct obd_export       *rp_exp;
1209         struct md_callback      *rp_cb;
1210 };
1211
1212 /**
1213  * Read pages from server.
1214  *
1215  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
1216  * a header lu_dirpage which describes the start/end hash, and whether this
1217  * page is empty (contains no dir entry) or hash collide with next page.
1218  * After client receives reply, several pages will be integrated into dir page
1219  * in PAGE_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the
1220  * lu_dirpage for this integrated page will be adjusted.
1221  **/
1222 static int mdc_read_page_remote(void *data, struct page *page0)
1223 {
1224         struct readpage_param *rp = data;
1225         struct page **page_pool;
1226         struct page *page;
1227         struct lu_dirpage *dp;
1228         int rd_pgs = 0; /* number of pages read actually */
1229         int npages;
1230         struct md_op_data *op_data = rp->rp_mod;
1231         struct ptlrpc_request *req;
1232         int max_pages = op_data->op_max_pages;
1233         struct inode *inode;
1234         struct lu_fid *fid;
1235         int i;
1236         int rc;
1237
1238         LASSERT(max_pages > 0 && max_pages <= PTLRPC_MAX_BRW_PAGES);
1239         inode = op_data->op_data;
1240         fid = &op_data->op_fid1;
1241         LASSERT(inode);
1242
1243         page_pool = kcalloc(max_pages, sizeof(page), GFP_NOFS);
1244         if (page_pool) {
1245                 page_pool[0] = page0;
1246         } else {
1247                 page_pool = &page0;
1248                 max_pages = 1;
1249         }
1250
1251         for (npages = 1; npages < max_pages; npages++) {
1252                 page = page_cache_alloc_cold(inode->i_mapping);
1253                 if (!page)
1254                         break;
1255                 page_pool[npages] = page;
1256         }
1257
1258         rc = mdc_getpage(rp->rp_exp, fid, rp->rp_off, page_pool, npages, &req);
1259         if (!rc) {
1260                 int lu_pgs = req->rq_bulk->bd_nob_transferred;
1261
1262                 rd_pgs = (req->rq_bulk->bd_nob_transferred +
1263                           PAGE_SIZE - 1) >> PAGE_SHIFT;
1264                 lu_pgs >>= LU_PAGE_SHIFT;
1265                 LASSERT(!(req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK));
1266
1267                 CDEBUG(D_INODE, "read %d(%d) pages\n", rd_pgs, lu_pgs);
1268
1269                 mdc_adjust_dirpages(page_pool, rd_pgs, lu_pgs);
1270
1271                 SetPageUptodate(page0);
1272         }
1273
1274         unlock_page(page0);
1275         ptlrpc_req_finished(req);
1276         CDEBUG(D_CACHE, "read %d/%d pages\n", rd_pgs, npages);
1277         for (i = 1; i < npages; i++) {
1278                 unsigned long offset;
1279                 __u64 hash;
1280                 int ret;
1281
1282                 page = page_pool[i];
1283
1284                 if (rc < 0 || i >= rd_pgs) {
1285                         put_page(page);
1286                         continue;
1287                 }
1288
1289                 SetPageUptodate(page);
1290
1291                 dp = kmap(page);
1292                 hash = le64_to_cpu(dp->ldp_hash_start);
1293                 kunmap(page);
1294
1295                 offset = hash_x_index(hash, rp->rp_hash64);
1296
1297                 prefetchw(&page->flags);
1298                 ret = add_to_page_cache_lru(page, inode->i_mapping, offset,
1299                                             GFP_KERNEL);
1300                 if (!ret)
1301                         unlock_page(page);
1302                 else
1303                         CDEBUG(D_VFSTRACE, "page %lu add to page cache failed: rc = %d\n",
1304                                offset, ret);
1305                 put_page(page);
1306         }
1307
1308         if (page_pool != &page0)
1309                 kfree(page_pool);
1310
1311         return rc;
1312 }
1313
1314 /**
1315  * Read dir page from cache first, if it can not find it, read it from
1316  * server and add into the cache.
1317  *
1318  * \param[in] exp       MDC export
1319  * \param[in] op_data   client MD stack parameters, transferring parameters
1320  *                      between different layers on client MD stack.
1321  * \param[in] cb_op     callback required for ldlm lock enqueue during
1322  *                      read page
1323  * \param[in] hash_offset the hash offset of the page to be read
1324  * \param[in] ppage     the page to be read
1325  *
1326  * retval               = 0 get the page successfully
1327  *                      errno(<0) get the page failed
1328  */
1329 static int mdc_read_page(struct obd_export *exp, struct md_op_data *op_data,
1330                          struct md_callback *cb_op, __u64 hash_offset,
1331                          struct page **ppage)
1332 {
1333         struct lookup_intent it = { .it_op = IT_READDIR };
1334         struct page *page;
1335         struct inode *dir = op_data->op_data;
1336         struct address_space *mapping;
1337         struct lu_dirpage *dp;
1338         __u64 start = 0;
1339         __u64 end = 0;
1340         struct lustre_handle lockh;
1341         struct ptlrpc_request *enq_req = NULL;
1342         struct readpage_param rp_param;
1343         int rc;
1344
1345         *ppage = NULL;
1346
1347         LASSERT(dir);
1348         mapping = dir->i_mapping;
1349
1350         rc = mdc_intent_lock(exp, op_data, &it, &enq_req,
1351                              cb_op->md_blocking_ast, 0);
1352         if (enq_req)
1353                 ptlrpc_req_finished(enq_req);
1354
1355         if (rc < 0) {
1356                 CERROR("%s: "DFID" lock enqueue fails: rc = %d\n",
1357                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1), rc);
1358                 return rc;
1359         }
1360
1361         rc = 0;
1362         lockh.cookie = it.it_lock_handle;
1363         mdc_set_lock_data(exp, &lockh, dir, NULL);
1364
1365         rp_param.rp_off = hash_offset;
1366         rp_param.rp_hash64 = op_data->op_cli_flags & CLI_HASH64;
1367         page = mdc_page_locate(mapping, &rp_param.rp_off, &start, &end,
1368                                rp_param.rp_hash64);
1369         if (IS_ERR(page)) {
1370                 CDEBUG(D_INFO, "%s: dir page locate: " DFID " at %llu: rc %ld\n",
1371                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1372                        rp_param.rp_off, PTR_ERR(page));
1373                 rc = PTR_ERR(page);
1374                 goto out_unlock;
1375         } else if (page) {
1376                 /*
1377                  * XXX nikita: not entirely correct handling of a corner case:
1378                  * suppose hash chain of entries with hash value HASH crosses
1379                  * border between pages P0 and P1. First both P0 and P1 are
1380                  * cached, seekdir() is called for some entry from the P0 part
1381                  * of the chain. Later P0 goes out of cache. telldir(HASH)
1382                  * happens and finds P1, as it starts with matching hash
1383                  * value. Remaining entries from P0 part of the chain are
1384                  * skipped. (Is that really a bug?)
1385                  *
1386                  * Possible solutions: 0. don't cache P1 is such case, handle
1387                  * it as an "overflow" page. 1. invalidate all pages at
1388                  * once. 2. use HASH|1 as an index for P1.
1389                  */
1390                 goto hash_collision;
1391         }
1392
1393         rp_param.rp_exp = exp;
1394         rp_param.rp_mod = op_data;
1395         page = read_cache_page(mapping,
1396                                hash_x_index(rp_param.rp_off,
1397                                             rp_param.rp_hash64),
1398                                mdc_read_page_remote, &rp_param);
1399         if (IS_ERR(page)) {
1400                 CERROR("%s: read cache page: "DFID" at %llu: rc %ld\n",
1401                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1402                        rp_param.rp_off, PTR_ERR(page));
1403                 rc = PTR_ERR(page);
1404                 goto out_unlock;
1405         }
1406
1407         wait_on_page_locked(page);
1408         (void)kmap(page);
1409         if (!PageUptodate(page)) {
1410                 CERROR("%s: page not updated: "DFID" at %llu: rc %d\n",
1411                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1412                        rp_param.rp_off, -5);
1413                 goto fail;
1414         }
1415         if (!PageChecked(page))
1416                 SetPageChecked(page);
1417         if (PageError(page)) {
1418                 CERROR("%s: page error: "DFID" at %llu: rc %d\n",
1419                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1420                        rp_param.rp_off, -5);
1421                 goto fail;
1422         }
1423
1424 hash_collision:
1425         dp = page_address(page);
1426         if (BITS_PER_LONG == 32 && rp_param.rp_hash64) {
1427                 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1428                 end = le64_to_cpu(dp->ldp_hash_end) >> 32;
1429                 rp_param.rp_off = hash_offset >> 32;
1430         } else {
1431                 start = le64_to_cpu(dp->ldp_hash_start);
1432                 end = le64_to_cpu(dp->ldp_hash_end);
1433                 rp_param.rp_off = hash_offset;
1434         }
1435         if (end == start) {
1436                 LASSERT(start == rp_param.rp_off);
1437                 CWARN("Page-wide hash collision: %#lx\n", (unsigned long)end);
1438 #if BITS_PER_LONG == 32
1439                 CWARN("Real page-wide hash collision at [%llu %llu] with hash %llu\n",
1440                       le64_to_cpu(dp->ldp_hash_start),
1441                       le64_to_cpu(dp->ldp_hash_end), hash_offset);
1442 #endif
1443                 /*
1444                  * Fetch whole overflow chain...
1445                  *
1446                  * XXX not yet.
1447                  */
1448                 goto fail;
1449         }
1450         *ppage = page;
1451 out_unlock:
1452         ldlm_lock_decref(&lockh, it.it_lock_mode);
1453         return rc;
1454 fail:
1455         kunmap(page);
1456         mdc_release_page(page, 1);
1457         rc = -EIO;
1458         goto out_unlock;
1459 }
1460
1461 static int mdc_statfs(const struct lu_env *env,
1462                       struct obd_export *exp, struct obd_statfs *osfs,
1463                       __u64 max_age, __u32 flags)
1464 {
1465         struct obd_device     *obd = class_exp2obd(exp);
1466         struct ptlrpc_request *req;
1467         struct obd_statfs     *msfs;
1468         struct obd_import     *imp = NULL;
1469         int                 rc;
1470
1471         /*
1472          * Since the request might also come from lprocfs, so we need
1473          * sync this with client_disconnect_export Bug15684
1474          */
1475         down_read(&obd->u.cli.cl_sem);
1476         if (obd->u.cli.cl_import)
1477                 imp = class_import_get(obd->u.cli.cl_import);
1478         up_read(&obd->u.cli.cl_sem);
1479         if (!imp)
1480                 return -ENODEV;
1481
1482         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_STATFS,
1483                                         LUSTRE_MDS_VERSION, MDS_STATFS);
1484         if (!req) {
1485                 rc = -ENOMEM;
1486                 goto output;
1487         }
1488
1489         ptlrpc_request_set_replen(req);
1490
1491         if (flags & OBD_STATFS_NODELAY) {
1492                 /* procfs requests not want stay in wait for avoid deadlock */
1493                 req->rq_no_resend = 1;
1494                 req->rq_no_delay = 1;
1495         }
1496
1497         rc = ptlrpc_queue_wait(req);
1498         if (rc) {
1499                 /* check connection error first */
1500                 if (imp->imp_connect_error)
1501                         rc = imp->imp_connect_error;
1502                 goto out;
1503         }
1504
1505         msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1506         if (!msfs) {
1507                 rc = -EPROTO;
1508                 goto out;
1509         }
1510
1511         *osfs = *msfs;
1512 out:
1513         ptlrpc_req_finished(req);
1514 output:
1515         class_import_put(imp);
1516         return rc;
1517 }
1518
1519 static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
1520 {
1521         __u32 keylen, vallen;
1522         void *key;
1523         int rc;
1524
1525         if (gf->gf_pathlen > PATH_MAX)
1526                 return -ENAMETOOLONG;
1527         if (gf->gf_pathlen < 2)
1528                 return -EOVERFLOW;
1529
1530         /* Key is KEY_FID2PATH + getinfo_fid2path description */
1531         keylen = cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf);
1532         key = kzalloc(keylen, GFP_NOFS);
1533         if (!key)
1534                 return -ENOMEM;
1535         memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
1536         memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf));
1537
1538         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
1539                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
1540
1541         if (!fid_is_sane(&gf->gf_fid)) {
1542                 rc = -EINVAL;
1543                 goto out;
1544         }
1545
1546         /* Val is struct getinfo_fid2path result plus path */
1547         vallen = sizeof(*gf) + gf->gf_pathlen;
1548
1549         rc = obd_get_info(NULL, exp, keylen, key, &vallen, gf, NULL);
1550         if (rc != 0 && rc != -EREMOTE)
1551                 goto out;
1552
1553         if (vallen <= sizeof(*gf)) {
1554                 rc = -EPROTO;
1555                 goto out;
1556         } else if (vallen > sizeof(*gf) + gf->gf_pathlen) {
1557                 rc = -EOVERFLOW;
1558                 goto out;
1559         }
1560
1561         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n%s\n",
1562                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno, gf->gf_path);
1563
1564 out:
1565         kfree(key);
1566         return rc;
1567 }
1568
1569 static int mdc_ioc_hsm_progress(struct obd_export *exp,
1570                                 struct hsm_progress_kernel *hpk)
1571 {
1572         struct obd_import               *imp = class_exp2cliimp(exp);
1573         struct hsm_progress_kernel      *req_hpk;
1574         struct ptlrpc_request           *req;
1575         int                              rc;
1576
1577         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_PROGRESS,
1578                                         LUSTRE_MDS_VERSION, MDS_HSM_PROGRESS);
1579         if (!req) {
1580                 rc = -ENOMEM;
1581                 goto out;
1582         }
1583
1584         mdc_pack_body(req, NULL, 0, 0, -1, 0);
1585
1586         /* Copy hsm_progress struct */
1587         req_hpk = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_PROGRESS);
1588         if (!req_hpk) {
1589                 rc = -EPROTO;
1590                 goto out;
1591         }
1592
1593         *req_hpk = *hpk;
1594         req_hpk->hpk_errval = lustre_errno_hton(hpk->hpk_errval);
1595
1596         ptlrpc_request_set_replen(req);
1597
1598         rc = mdc_queue_wait(req);
1599 out:
1600         ptlrpc_req_finished(req);
1601         return rc;
1602 }
1603
1604 static int mdc_ioc_hsm_ct_register(struct obd_import *imp, __u32 archives)
1605 {
1606         __u32                   *archive_mask;
1607         struct ptlrpc_request   *req;
1608         int                      rc;
1609
1610         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_REGISTER,
1611                                         LUSTRE_MDS_VERSION,
1612                                         MDS_HSM_CT_REGISTER);
1613         if (!req) {
1614                 rc = -ENOMEM;
1615                 goto out;
1616         }
1617
1618         mdc_pack_body(req, NULL, 0, 0, -1, 0);
1619
1620         /* Copy hsm_progress struct */
1621         archive_mask = req_capsule_client_get(&req->rq_pill,
1622                                               &RMF_MDS_HSM_ARCHIVE);
1623         if (!archive_mask) {
1624                 rc = -EPROTO;
1625                 goto out;
1626         }
1627
1628         *archive_mask = archives;
1629
1630         ptlrpc_request_set_replen(req);
1631
1632         rc = mdc_queue_wait(req);
1633 out:
1634         ptlrpc_req_finished(req);
1635         return rc;
1636 }
1637
1638 static int mdc_ioc_hsm_current_action(struct obd_export *exp,
1639                                       struct md_op_data *op_data)
1640 {
1641         struct hsm_current_action       *hca = op_data->op_data;
1642         struct hsm_current_action       *req_hca;
1643         struct ptlrpc_request           *req;
1644         int                              rc;
1645
1646         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1647                                    &RQF_MDS_HSM_ACTION);
1648         if (!req)
1649                 return -ENOMEM;
1650
1651         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_ACTION);
1652         if (rc) {
1653                 ptlrpc_request_free(req);
1654                 return rc;
1655         }
1656
1657         mdc_pack_body(req, &op_data->op_fid1, 0, 0,
1658                       op_data->op_suppgids[0], 0);
1659
1660         ptlrpc_request_set_replen(req);
1661
1662         rc = mdc_queue_wait(req);
1663         if (rc)
1664                 goto out;
1665
1666         req_hca = req_capsule_server_get(&req->rq_pill,
1667                                          &RMF_MDS_HSM_CURRENT_ACTION);
1668         if (!req_hca) {
1669                 rc = -EPROTO;
1670                 goto out;
1671         }
1672
1673         *hca = *req_hca;
1674
1675 out:
1676         ptlrpc_req_finished(req);
1677         return rc;
1678 }
1679
1680 static int mdc_ioc_hsm_ct_unregister(struct obd_import *imp)
1681 {
1682         struct ptlrpc_request   *req;
1683         int                      rc;
1684
1685         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_UNREGISTER,
1686                                         LUSTRE_MDS_VERSION,
1687                                         MDS_HSM_CT_UNREGISTER);
1688         if (!req) {
1689                 rc = -ENOMEM;
1690                 goto out;
1691         }
1692
1693         mdc_pack_body(req, NULL, 0, 0, -1, 0);
1694
1695         ptlrpc_request_set_replen(req);
1696
1697         rc = mdc_queue_wait(req);
1698 out:
1699         ptlrpc_req_finished(req);
1700         return rc;
1701 }
1702
1703 static int mdc_ioc_hsm_state_get(struct obd_export *exp,
1704                                  struct md_op_data *op_data)
1705 {
1706         struct hsm_user_state   *hus = op_data->op_data;
1707         struct hsm_user_state   *req_hus;
1708         struct ptlrpc_request   *req;
1709         int                      rc;
1710
1711         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1712                                    &RQF_MDS_HSM_STATE_GET);
1713         if (!req)
1714                 return -ENOMEM;
1715
1716         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_GET);
1717         if (rc != 0) {
1718                 ptlrpc_request_free(req);
1719                 return rc;
1720         }
1721
1722         mdc_pack_body(req, &op_data->op_fid1, 0, 0,
1723                       op_data->op_suppgids[0], 0);
1724
1725         ptlrpc_request_set_replen(req);
1726
1727         rc = mdc_queue_wait(req);
1728         if (rc)
1729                 goto out;
1730
1731         req_hus = req_capsule_server_get(&req->rq_pill, &RMF_HSM_USER_STATE);
1732         if (!req_hus) {
1733                 rc = -EPROTO;
1734                 goto out;
1735         }
1736
1737         *hus = *req_hus;
1738
1739 out:
1740         ptlrpc_req_finished(req);
1741         return rc;
1742 }
1743
1744 static int mdc_ioc_hsm_state_set(struct obd_export *exp,
1745                                  struct md_op_data *op_data)
1746 {
1747         struct hsm_state_set    *hss = op_data->op_data;
1748         struct hsm_state_set    *req_hss;
1749         struct ptlrpc_request   *req;
1750         int                      rc;
1751
1752         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1753                                    &RQF_MDS_HSM_STATE_SET);
1754         if (!req)
1755                 return -ENOMEM;
1756
1757         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_SET);
1758         if (rc) {
1759                 ptlrpc_request_free(req);
1760                 return rc;
1761         }
1762
1763         mdc_pack_body(req, &op_data->op_fid1, 0, 0,
1764                       op_data->op_suppgids[0], 0);
1765
1766         /* Copy states */
1767         req_hss = req_capsule_client_get(&req->rq_pill, &RMF_HSM_STATE_SET);
1768         if (!req_hss) {
1769                 rc = -EPROTO;
1770                 goto out;
1771         }
1772         *req_hss = *hss;
1773
1774         ptlrpc_request_set_replen(req);
1775
1776         rc = mdc_queue_wait(req);
1777 out:
1778         ptlrpc_req_finished(req);
1779         return rc;
1780 }
1781
1782 static int mdc_ioc_hsm_request(struct obd_export *exp,
1783                                struct hsm_user_request *hur)
1784 {
1785         struct obd_import       *imp = class_exp2cliimp(exp);
1786         struct ptlrpc_request   *req;
1787         struct hsm_request      *req_hr;
1788         struct hsm_user_item    *req_hui;
1789         char                    *req_opaque;
1790         int                      rc;
1791
1792         req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_REQUEST);
1793         if (!req) {
1794                 rc = -ENOMEM;
1795                 goto out;
1796         }
1797
1798         req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM, RCL_CLIENT,
1799                              hur->hur_request.hr_itemcount
1800                              * sizeof(struct hsm_user_item));
1801         req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA, RCL_CLIENT,
1802                              hur->hur_request.hr_data_len);
1803
1804         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_REQUEST);
1805         if (rc) {
1806                 ptlrpc_request_free(req);
1807                 return rc;
1808         }
1809
1810         mdc_pack_body(req, NULL, 0, 0, -1, 0);
1811
1812         /* Copy hsm_request struct */
1813         req_hr = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_REQUEST);
1814         if (!req_hr) {
1815                 rc = -EPROTO;
1816                 goto out;
1817         }
1818         *req_hr = hur->hur_request;
1819
1820         /* Copy hsm_user_item structs */
1821         req_hui = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM);
1822         if (!req_hui) {
1823                 rc = -EPROTO;
1824                 goto out;
1825         }
1826         memcpy(req_hui, hur->hur_user_item,
1827                hur->hur_request.hr_itemcount * sizeof(struct hsm_user_item));
1828
1829         /* Copy opaque field */
1830         req_opaque = req_capsule_client_get(&req->rq_pill, &RMF_GENERIC_DATA);
1831         if (!req_opaque) {
1832                 rc = -EPROTO;
1833                 goto out;
1834         }
1835         memcpy(req_opaque, hur_data(hur), hur->hur_request.hr_data_len);
1836
1837         ptlrpc_request_set_replen(req);
1838
1839         rc = mdc_queue_wait(req);
1840 out:
1841         ptlrpc_req_finished(req);
1842         return rc;
1843 }
1844
1845 static struct kuc_hdr *changelog_kuc_hdr(char *buf, size_t len, u32 flags)
1846 {
1847         struct kuc_hdr *lh = (struct kuc_hdr *)buf;
1848
1849         LASSERT(len <= KUC_CHANGELOG_MSG_MAXSIZE);
1850
1851         lh->kuc_magic = KUC_MAGIC;
1852         lh->kuc_transport = KUC_TRANSPORT_CHANGELOG;
1853         lh->kuc_flags = flags;
1854         lh->kuc_msgtype = CL_RECORD;
1855         lh->kuc_msglen = len;
1856         return lh;
1857 }
1858
1859 struct changelog_show {
1860         __u64           cs_startrec;
1861         enum changelog_send_flag        cs_flags;
1862         struct file     *cs_fp;
1863         char            *cs_buf;
1864         struct obd_device *cs_obd;
1865 };
1866
1867 static inline char *cs_obd_name(struct changelog_show *cs)
1868 {
1869         return cs->cs_obd->obd_name;
1870 }
1871
1872 static int changelog_kkuc_cb(const struct lu_env *env, struct llog_handle *llh,
1873                              struct llog_rec_hdr *hdr, void *data)
1874 {
1875         struct changelog_show *cs = data;
1876         struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
1877         struct kuc_hdr *lh;
1878         size_t len;
1879         int rc;
1880
1881         if (rec->cr_hdr.lrh_type != CHANGELOG_REC) {
1882                 rc = -EINVAL;
1883                 CERROR("%s: not a changelog rec %x/%d: rc = %d\n",
1884                        cs_obd_name(cs), rec->cr_hdr.lrh_type,
1885                        rec->cr.cr_type, rc);
1886                 return rc;
1887         }
1888
1889         if (rec->cr.cr_index < cs->cs_startrec) {
1890                 /* Skip entries earlier than what we are interested in */
1891                 CDEBUG(D_HSM, "rec=%llu start=%llu\n",
1892                        rec->cr.cr_index, cs->cs_startrec);
1893                 return 0;
1894         }
1895
1896         CDEBUG(D_HSM, "%llu %02d%-5s %llu 0x%x t=" DFID " p=" DFID
1897                 " %.*s\n", rec->cr.cr_index, rec->cr.cr_type,
1898                 changelog_type2str(rec->cr.cr_type), rec->cr.cr_time,
1899                 rec->cr.cr_flags & CLF_FLAGMASK,
1900                 PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid),
1901                 rec->cr.cr_namelen, changelog_rec_name(&rec->cr));
1902
1903         len = sizeof(*lh) + changelog_rec_size(&rec->cr) + rec->cr.cr_namelen;
1904
1905         /* Set up the message */
1906         lh = changelog_kuc_hdr(cs->cs_buf, len, cs->cs_flags);
1907         memcpy(lh + 1, &rec->cr, len - sizeof(*lh));
1908
1909         rc = libcfs_kkuc_msg_put(cs->cs_fp, lh);
1910         CDEBUG(D_HSM, "kucmsg fp %p len %zu rc %d\n", cs->cs_fp, len, rc);
1911
1912         return rc;
1913 }
1914
1915 static int mdc_changelog_send_thread(void *csdata)
1916 {
1917         enum llog_flag flags = LLOG_F_IS_CAT;
1918         struct changelog_show *cs = csdata;
1919         struct llog_ctxt *ctxt = NULL;
1920         struct llog_handle *llh = NULL;
1921         struct kuc_hdr *kuch;
1922         int rc;
1923
1924         CDEBUG(D_HSM, "changelog to fp=%p start %llu\n",
1925                cs->cs_fp, cs->cs_startrec);
1926
1927         cs->cs_buf = kzalloc(KUC_CHANGELOG_MSG_MAXSIZE, GFP_NOFS);
1928         if (!cs->cs_buf) {
1929                 rc = -ENOMEM;
1930                 goto out;
1931         }
1932
1933         /* Set up the remote catalog handle */
1934         ctxt = llog_get_context(cs->cs_obd, LLOG_CHANGELOG_REPL_CTXT);
1935         if (!ctxt) {
1936                 rc = -ENOENT;
1937                 goto out;
1938         }
1939         rc = llog_open(NULL, ctxt, &llh, NULL, CHANGELOG_CATALOG,
1940                        LLOG_OPEN_EXISTS);
1941         if (rc) {
1942                 CERROR("%s: fail to open changelog catalog: rc = %d\n",
1943                        cs_obd_name(cs), rc);
1944                 goto out;
1945         }
1946
1947         if (cs->cs_flags & CHANGELOG_FLAG_JOBID)
1948                 flags |= LLOG_F_EXT_JOBID;
1949
1950         rc = llog_init_handle(NULL, llh, flags, NULL);
1951         if (rc) {
1952                 CERROR("llog_init_handle failed %d\n", rc);
1953                 goto out;
1954         }
1955
1956         rc = llog_cat_process(NULL, llh, changelog_kkuc_cb, cs, 0, 0);
1957
1958         /* Send EOF no matter what our result */
1959         kuch = changelog_kuc_hdr(cs->cs_buf, sizeof(*kuch), cs->cs_flags);
1960         if (kuch) {
1961                 kuch->kuc_msgtype = CL_EOF;
1962                 libcfs_kkuc_msg_put(cs->cs_fp, kuch);
1963         }
1964
1965 out:
1966         fput(cs->cs_fp);
1967         if (llh)
1968                 llog_cat_close(NULL, llh);
1969         if (ctxt)
1970                 llog_ctxt_put(ctxt);
1971         kfree(cs->cs_buf);
1972         kfree(cs);
1973         return rc;
1974 }
1975
1976 static int mdc_ioc_changelog_send(struct obd_device *obd,
1977                                   struct ioc_changelog *icc)
1978 {
1979         struct changelog_show *cs;
1980         struct task_struct *task;
1981         int rc;
1982
1983         /* Freed in mdc_changelog_send_thread */
1984         cs = kzalloc(sizeof(*cs), GFP_NOFS);
1985         if (!cs)
1986                 return -ENOMEM;
1987
1988         cs->cs_obd = obd;
1989         cs->cs_startrec = icc->icc_recno;
1990         /* matching fput in mdc_changelog_send_thread */
1991         cs->cs_fp = fget(icc->icc_id);
1992         cs->cs_flags = icc->icc_flags;
1993
1994         /*
1995          * New thread because we should return to user app before
1996          * writing into our pipe
1997          */
1998         task = kthread_run(mdc_changelog_send_thread, cs,
1999                            "mdc_clg_send_thread");
2000         if (IS_ERR(task)) {
2001                 rc = PTR_ERR(task);
2002                 CERROR("%s: can't start changelog thread: rc = %d\n",
2003                        cs_obd_name(cs), rc);
2004                 kfree(cs);
2005         } else {
2006                 rc = 0;
2007                 CDEBUG(D_HSM, "%s: started changelog thread\n",
2008                        cs_obd_name(cs));
2009         }
2010
2011         CERROR("Failed to start changelog thread: %d\n", rc);
2012         return rc;
2013 }
2014
2015 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2016                                 struct lustre_kernelcomm *lk);
2017
2018 static int mdc_quotacheck(struct obd_device *unused, struct obd_export *exp,
2019                           struct obd_quotactl *oqctl)
2020 {
2021         struct client_obd       *cli = &exp->exp_obd->u.cli;
2022         struct ptlrpc_request   *req;
2023         struct obd_quotactl     *body;
2024         int                   rc;
2025
2026         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
2027                                         &RQF_MDS_QUOTACHECK, LUSTRE_MDS_VERSION,
2028                                         MDS_QUOTACHECK);
2029         if (!req)
2030                 return -ENOMEM;
2031
2032         body = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
2033         *body = *oqctl;
2034
2035         ptlrpc_request_set_replen(req);
2036
2037         /* the next poll will find -ENODATA, that means quotacheck is
2038          * going on
2039          */
2040         cli->cl_qchk_stat = -ENODATA;
2041         rc = ptlrpc_queue_wait(req);
2042         if (rc)
2043                 cli->cl_qchk_stat = rc;
2044         ptlrpc_req_finished(req);
2045         return rc;
2046 }
2047
2048 static int mdc_quota_poll_check(struct obd_export *exp,
2049                                 struct if_quotacheck *qchk)
2050 {
2051         struct client_obd *cli = &exp->exp_obd->u.cli;
2052         int rc;
2053
2054         qchk->obd_uuid = cli->cl_target_uuid;
2055         memcpy(qchk->obd_type, LUSTRE_MDS_NAME, strlen(LUSTRE_MDS_NAME));
2056
2057         rc = cli->cl_qchk_stat;
2058         /* the client is not the previous one */
2059         if (rc == CL_NOT_QUOTACHECKED)
2060                 rc = -EINTR;
2061         return rc;
2062 }
2063
2064 static int mdc_quotactl(struct obd_device *unused, struct obd_export *exp,
2065                         struct obd_quotactl *oqctl)
2066 {
2067         struct ptlrpc_request   *req;
2068         struct obd_quotactl     *oqc;
2069         int                   rc;
2070
2071         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
2072                                         &RQF_MDS_QUOTACTL, LUSTRE_MDS_VERSION,
2073                                         MDS_QUOTACTL);
2074         if (!req)
2075                 return -ENOMEM;
2076
2077         oqc = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
2078         *oqc = *oqctl;
2079
2080         ptlrpc_request_set_replen(req);
2081         ptlrpc_at_set_req_timeout(req);
2082         req->rq_no_resend = 1;
2083
2084         rc = ptlrpc_queue_wait(req);
2085         if (rc)
2086                 CERROR("ptlrpc_queue_wait failed, rc: %d\n", rc);
2087
2088         if (req->rq_repmsg) {
2089                 oqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
2090                 if (oqc) {
2091                         *oqctl = *oqc;
2092                 } else if (!rc) {
2093                         CERROR("Can't unpack obd_quotactl\n");
2094                         rc = -EPROTO;
2095                 }
2096         } else if (!rc) {
2097                 CERROR("Can't unpack obd_quotactl\n");
2098                 rc = -EPROTO;
2099         }
2100         ptlrpc_req_finished(req);
2101
2102         return rc;
2103 }
2104
2105 static int mdc_ioc_swap_layouts(struct obd_export *exp,
2106                                 struct md_op_data *op_data)
2107 {
2108         LIST_HEAD(cancels);
2109         struct ptlrpc_request   *req;
2110         int                      rc, count;
2111         struct mdc_swap_layouts *msl, *payload;
2112
2113         msl = op_data->op_data;
2114
2115         /* When the MDT will get the MDS_SWAP_LAYOUTS RPC the
2116          * first thing it will do is to cancel the 2 layout
2117          * locks hold by this client.
2118          * So the client must cancel its layout locks on the 2 fids
2119          * with the request RPC to avoid extra RPC round trips
2120          */
2121         count = mdc_resource_get_unused(exp, &op_data->op_fid1, &cancels,
2122                                         LCK_CR, MDS_INODELOCK_LAYOUT |
2123                                         MDS_INODELOCK_XATTR);
2124         count += mdc_resource_get_unused(exp, &op_data->op_fid2, &cancels,
2125                                          LCK_CR, MDS_INODELOCK_LAYOUT |
2126                                          MDS_INODELOCK_XATTR);
2127
2128         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
2129                                    &RQF_MDS_SWAP_LAYOUTS);
2130         if (!req) {
2131                 ldlm_lock_list_put(&cancels, l_bl_ast, count);
2132                 return -ENOMEM;
2133         }
2134
2135         rc = mdc_prep_elc_req(exp, req, MDS_SWAP_LAYOUTS, &cancels, count);
2136         if (rc) {
2137                 ptlrpc_request_free(req);
2138                 return rc;
2139         }
2140
2141         mdc_swap_layouts_pack(req, op_data);
2142
2143         payload = req_capsule_client_get(&req->rq_pill, &RMF_SWAP_LAYOUTS);
2144         LASSERT(payload);
2145
2146         *payload = *msl;
2147
2148         ptlrpc_request_set_replen(req);
2149
2150         rc = ptlrpc_queue_wait(req);
2151
2152         ptlrpc_req_finished(req);
2153         return rc;
2154 }
2155
2156 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2157                          void *karg, void __user *uarg)
2158 {
2159         struct obd_device *obd = exp->exp_obd;
2160         struct obd_ioctl_data *data = karg;
2161         struct obd_import *imp = obd->u.cli.cl_import;
2162         int rc;
2163
2164         if (!try_module_get(THIS_MODULE)) {
2165                 CERROR("%s: cannot get module '%s'\n", obd->obd_name,
2166                        module_name(THIS_MODULE));
2167                 return -EINVAL;
2168         }
2169         switch (cmd) {
2170         case OBD_IOC_CHANGELOG_SEND:
2171                 rc = mdc_ioc_changelog_send(obd, karg);
2172                 goto out;
2173         case OBD_IOC_CHANGELOG_CLEAR: {
2174                 struct ioc_changelog *icc = karg;
2175                 struct changelog_setinfo cs = {
2176                         .cs_recno = icc->icc_recno,
2177                         .cs_id = icc->icc_id
2178                 };
2179
2180                 rc = obd_set_info_async(NULL, exp, strlen(KEY_CHANGELOG_CLEAR),
2181                                         KEY_CHANGELOG_CLEAR, sizeof(cs), &cs,
2182                                         NULL);
2183                 goto out;
2184         }
2185         case OBD_IOC_FID2PATH:
2186                 rc = mdc_ioc_fid2path(exp, karg);
2187                 goto out;
2188         case LL_IOC_HSM_CT_START:
2189                 rc = mdc_ioc_hsm_ct_start(exp, karg);
2190                 /* ignore if it was already registered on this MDS. */
2191                 if (rc == -EEXIST)
2192                         rc = 0;
2193                 goto out;
2194         case LL_IOC_HSM_PROGRESS:
2195                 rc = mdc_ioc_hsm_progress(exp, karg);
2196                 goto out;
2197         case LL_IOC_HSM_STATE_GET:
2198                 rc = mdc_ioc_hsm_state_get(exp, karg);
2199                 goto out;
2200         case LL_IOC_HSM_STATE_SET:
2201                 rc = mdc_ioc_hsm_state_set(exp, karg);
2202                 goto out;
2203         case LL_IOC_HSM_ACTION:
2204                 rc = mdc_ioc_hsm_current_action(exp, karg);
2205                 goto out;
2206         case LL_IOC_HSM_REQUEST:
2207                 rc = mdc_ioc_hsm_request(exp, karg);
2208                 goto out;
2209         case OBD_IOC_CLIENT_RECOVER:
2210                 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1, 0);
2211                 if (rc < 0)
2212                         goto out;
2213                 rc = 0;
2214                 goto out;
2215         case IOC_OSC_SET_ACTIVE:
2216                 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
2217                 goto out;
2218         case OBD_IOC_POLL_QUOTACHECK:
2219                 rc = mdc_quota_poll_check(exp, (struct if_quotacheck *)karg);
2220                 goto out;
2221         case OBD_IOC_PING_TARGET:
2222                 rc = ptlrpc_obd_ping(obd);
2223                 goto out;
2224         /*
2225          * Normally IOC_OBD_STATFS, OBD_IOC_QUOTACTL iocontrol are handled by
2226          * LMV instead of MDC. But when the cluster is upgraded from 1.8,
2227          * there'd be no LMV layer thus we might be called here. Eventually
2228          * this code should be removed.
2229          * bz20731, LU-592.
2230          */
2231         case IOC_OBD_STATFS: {
2232                 struct obd_statfs stat_buf = {0};
2233
2234                 if (*((__u32 *)data->ioc_inlbuf2) != 0) {
2235                         rc = -ENODEV;
2236                         goto out;
2237                 }
2238
2239                 /* copy UUID */
2240                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd),
2241                                  min_t(size_t, data->ioc_plen2,
2242                                        sizeof(struct obd_uuid)))) {
2243                         rc = -EFAULT;
2244                         goto out;
2245                 }
2246
2247                 rc = mdc_statfs(NULL, obd->obd_self_export, &stat_buf,
2248                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
2249                                 0);
2250                 if (rc != 0)
2251                         goto out;
2252
2253                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
2254                                  min_t(size_t, data->ioc_plen1,
2255                                        sizeof(stat_buf)))) {
2256                         rc = -EFAULT;
2257                         goto out;
2258                 }
2259
2260                 rc = 0;
2261                 goto out;
2262         }
2263         case OBD_IOC_QUOTACTL: {
2264                 struct if_quotactl *qctl = karg;
2265                 struct obd_quotactl *oqctl;
2266
2267                 oqctl = kzalloc(sizeof(*oqctl), GFP_NOFS);
2268                 if (!oqctl) {
2269                         rc = -ENOMEM;
2270                         goto out;
2271                 }
2272
2273                 QCTL_COPY(oqctl, qctl);
2274                 rc = obd_quotactl(exp, oqctl);
2275                 if (rc == 0) {
2276                         QCTL_COPY(qctl, oqctl);
2277                         qctl->qc_valid = QC_MDTIDX;
2278                         qctl->obd_uuid = obd->u.cli.cl_target_uuid;
2279                 }
2280
2281                 kfree(oqctl);
2282                 goto out;
2283         }
2284         case LL_IOC_GET_CONNECT_FLAGS:
2285                 if (copy_to_user(uarg, exp_connect_flags_ptr(exp),
2286                                  sizeof(*exp_connect_flags_ptr(exp)))) {
2287                         rc = -EFAULT;
2288                         goto out;
2289                 }
2290
2291                 rc = 0;
2292                 goto out;
2293         case LL_IOC_LOV_SWAP_LAYOUTS:
2294                 rc = mdc_ioc_swap_layouts(exp, karg);
2295                 goto out;
2296         default:
2297                 CERROR("unrecognised ioctl: cmd = %#x\n", cmd);
2298                 rc = -ENOTTY;
2299                 goto out;
2300         }
2301 out:
2302         module_put(THIS_MODULE);
2303
2304         return rc;
2305 }
2306
2307 static int mdc_get_info_rpc(struct obd_export *exp,
2308                             u32 keylen, void *key,
2309                             int vallen, void *val)
2310 {
2311         struct obd_import      *imp = class_exp2cliimp(exp);
2312         struct ptlrpc_request  *req;
2313         char               *tmp;
2314         int                  rc = -EINVAL;
2315
2316         req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
2317         if (!req)
2318                 return -ENOMEM;
2319
2320         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
2321                              RCL_CLIENT, keylen);
2322         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
2323                              RCL_CLIENT, sizeof(__u32));
2324
2325         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
2326         if (rc) {
2327                 ptlrpc_request_free(req);
2328                 return rc;
2329         }
2330
2331         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
2332         memcpy(tmp, key, keylen);
2333         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
2334         memcpy(tmp, &vallen, sizeof(__u32));
2335
2336         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
2337                              RCL_SERVER, vallen);
2338         ptlrpc_request_set_replen(req);
2339
2340         rc = ptlrpc_queue_wait(req);
2341         /* -EREMOTE means the get_info result is partial, and it needs to
2342          * continue on another MDT, see fid2path part in lmv_iocontrol
2343          */
2344         if (rc == 0 || rc == -EREMOTE) {
2345                 tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
2346                 memcpy(val, tmp, vallen);
2347                 if (ptlrpc_rep_need_swab(req)) {
2348                         if (KEY_IS(KEY_FID2PATH))
2349                                 lustre_swab_fid2path(val);
2350                 }
2351         }
2352         ptlrpc_req_finished(req);
2353
2354         return rc;
2355 }
2356
2357 static void lustre_swab_hai(struct hsm_action_item *h)
2358 {
2359         __swab32s(&h->hai_len);
2360         __swab32s(&h->hai_action);
2361         lustre_swab_lu_fid(&h->hai_fid);
2362         lustre_swab_lu_fid(&h->hai_dfid);
2363         __swab64s(&h->hai_cookie);
2364         __swab64s(&h->hai_extent.offset);
2365         __swab64s(&h->hai_extent.length);
2366         __swab64s(&h->hai_gid);
2367 }
2368
2369 static void lustre_swab_hal(struct hsm_action_list *h)
2370 {
2371         struct hsm_action_item  *hai;
2372         u32 i;
2373
2374         __swab32s(&h->hal_version);
2375         __swab32s(&h->hal_count);
2376         __swab32s(&h->hal_archive_id);
2377         __swab64s(&h->hal_flags);
2378         hai = hai_first(h);
2379         for (i = 0; i < h->hal_count; i++, hai = hai_next(hai))
2380                 lustre_swab_hai(hai);
2381 }
2382
2383 static void lustre_swab_kuch(struct kuc_hdr *l)
2384 {
2385         __swab16s(&l->kuc_magic);
2386         /* __u8 l->kuc_transport */
2387         __swab16s(&l->kuc_msgtype);
2388         __swab16s(&l->kuc_msglen);
2389 }
2390
2391 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2392                                 struct lustre_kernelcomm *lk)
2393 {
2394         struct obd_import  *imp = class_exp2cliimp(exp);
2395         __u32               archive = lk->lk_data;
2396         int                 rc = 0;
2397
2398         if (lk->lk_group != KUC_GRP_HSM) {
2399                 CERROR("Bad copytool group %d\n", lk->lk_group);
2400                 return -EINVAL;
2401         }
2402
2403         CDEBUG(D_HSM, "CT start r%d w%d u%d g%d f%#x\n", lk->lk_rfd, lk->lk_wfd,
2404                lk->lk_uid, lk->lk_group, lk->lk_flags);
2405
2406         if (lk->lk_flags & LK_FLG_STOP) {
2407                 /* Unregister with the coordinator */
2408                 rc = mdc_ioc_hsm_ct_unregister(imp);
2409         } else {
2410                 rc = mdc_ioc_hsm_ct_register(imp, archive);
2411         }
2412
2413         return rc;
2414 }
2415
2416 /**
2417  * Send a message to any listening copytools
2418  * @param val KUC message (kuc_hdr + hsm_action_list)
2419  * @param len total length of message
2420  */
2421 static int mdc_hsm_copytool_send(size_t len, void *val)
2422 {
2423         struct kuc_hdr          *lh = (struct kuc_hdr *)val;
2424         struct hsm_action_list  *hal = (struct hsm_action_list *)(lh + 1);
2425
2426         if (len < sizeof(*lh) + sizeof(*hal)) {
2427                 CERROR("Short HSM message %zu < %zu\n", len,
2428                        sizeof(*lh) + sizeof(*hal));
2429                 return -EPROTO;
2430         }
2431         if (lh->kuc_magic == __swab16(KUC_MAGIC)) {
2432                 lustre_swab_kuch(lh);
2433                 lustre_swab_hal(hal);
2434         } else if (lh->kuc_magic != KUC_MAGIC) {
2435                 CERROR("Bad magic %x!=%x\n", lh->kuc_magic, KUC_MAGIC);
2436                 return -EPROTO;
2437         }
2438
2439         CDEBUG(D_HSM,
2440                "Received message mg=%x t=%d m=%d l=%d actions=%d on %s\n",
2441                lh->kuc_magic, lh->kuc_transport, lh->kuc_msgtype,
2442                lh->kuc_msglen, hal->hal_count, hal->hal_fsname);
2443
2444         /* Broadcast to HSM listeners */
2445         return libcfs_kkuc_group_put(KUC_GRP_HSM, lh);
2446 }
2447
2448 /**
2449  * callback function passed to kuc for re-registering each HSM copytool
2450  * running on MDC, after MDT shutdown/recovery.
2451  * @param data copytool registration data
2452  * @param cb_arg callback argument (obd_import)
2453  */
2454 static int mdc_hsm_ct_reregister(void *data, void *cb_arg)
2455 {
2456         struct kkuc_ct_data     *kcd = data;
2457         struct obd_import       *imp = (struct obd_import *)cb_arg;
2458         int                      rc;
2459
2460         if (!kcd || kcd->kcd_magic != KKUC_CT_DATA_MAGIC)
2461                 return -EPROTO;
2462
2463         if (!obd_uuid_equals(&kcd->kcd_uuid, &imp->imp_obd->obd_uuid))
2464                 return 0;
2465
2466         CDEBUG(D_HA, "%s: recover copytool registration to MDT (archive=%#x)\n",
2467                imp->imp_obd->obd_name, kcd->kcd_archive);
2468         rc = mdc_ioc_hsm_ct_register(imp, kcd->kcd_archive);
2469
2470         /* ignore error if the copytool is already registered */
2471         return (rc == -EEXIST) ? 0 : rc;
2472 }
2473
2474 static int mdc_set_info_async(const struct lu_env *env,
2475                               struct obd_export *exp,
2476                               u32 keylen, void *key,
2477                               u32 vallen, void *val,
2478                               struct ptlrpc_request_set *set)
2479 {
2480         struct obd_import       *imp = class_exp2cliimp(exp);
2481         int                      rc;
2482
2483         if (KEY_IS(KEY_READ_ONLY)) {
2484                 if (vallen != sizeof(int))
2485                         return -EINVAL;
2486
2487                 spin_lock(&imp->imp_lock);
2488                 if (*((int *)val)) {
2489                         imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
2490                         imp->imp_connect_data.ocd_connect_flags |=
2491                                                         OBD_CONNECT_RDONLY;
2492                 } else {
2493                         imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
2494                         imp->imp_connect_data.ocd_connect_flags &=
2495                                                         ~OBD_CONNECT_RDONLY;
2496                 }
2497                 spin_unlock(&imp->imp_lock);
2498
2499                 return do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2500                                          keylen, key, vallen, val, set);
2501         }
2502         if (KEY_IS(KEY_SPTLRPC_CONF)) {
2503                 sptlrpc_conf_client_adapt(exp->exp_obd);
2504                 return 0;
2505         }
2506         if (KEY_IS(KEY_FLUSH_CTX)) {
2507                 sptlrpc_import_flush_my_ctx(imp);
2508                 return 0;
2509         }
2510         if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2511                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2512                                        keylen, key, vallen, val, set);
2513                 return rc;
2514         }
2515         if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) {
2516                 rc = mdc_hsm_copytool_send(vallen, val);
2517                 return rc;
2518         }
2519         if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2520                 u32 *default_easize = val;
2521
2522                 exp->exp_obd->u.cli.cl_default_mds_easize = *default_easize;
2523                 return 0;
2524         }
2525
2526         CERROR("Unknown key %s\n", (char *)key);
2527         return -EINVAL;
2528 }
2529
2530 static int mdc_get_info(const struct lu_env *env, struct obd_export *exp,
2531                         __u32 keylen, void *key, __u32 *vallen, void *val,
2532                         struct lov_stripe_md *lsm)
2533 {
2534         int rc = -EINVAL;
2535
2536         if (KEY_IS(KEY_MAX_EASIZE)) {
2537                 u32 mdsize, *max_easize;
2538
2539                 if (*vallen != sizeof(int))
2540                         return -EINVAL;
2541                 mdsize = *(u32 *)val;
2542                 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
2543                         exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
2544                 max_easize = val;
2545                 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
2546                 return 0;
2547         } else if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2548                 u32 *default_easize;
2549
2550                 if (*vallen != sizeof(int))
2551                         return -EINVAL;
2552                 default_easize = val;
2553                 *default_easize = exp->exp_obd->u.cli.cl_default_mds_easize;
2554                 return 0;
2555         } else if (KEY_IS(KEY_CONN_DATA)) {
2556                 struct obd_import *imp = class_exp2cliimp(exp);
2557                 struct obd_connect_data *data = val;
2558
2559                 if (*vallen != sizeof(*data))
2560                         return -EINVAL;
2561
2562                 *data = imp->imp_connect_data;
2563                 return 0;
2564         } else if (KEY_IS(KEY_TGT_COUNT)) {
2565                 *((u32 *)val) = 1;
2566                 return 0;
2567         }
2568
2569         rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
2570
2571         return rc;
2572 }
2573
2574 static int mdc_sync(struct obd_export *exp, const struct lu_fid *fid,
2575                     struct ptlrpc_request **request)
2576 {
2577         struct ptlrpc_request *req;
2578         int                 rc;
2579
2580         *request = NULL;
2581         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
2582         if (!req)
2583                 return -ENOMEM;
2584
2585         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
2586         if (rc) {
2587                 ptlrpc_request_free(req);
2588                 return rc;
2589         }
2590
2591         mdc_pack_body(req, fid, 0, 0, -1, 0);
2592
2593         ptlrpc_request_set_replen(req);
2594
2595         rc = ptlrpc_queue_wait(req);
2596         if (rc)
2597                 ptlrpc_req_finished(req);
2598         else
2599                 *request = req;
2600         return rc;
2601 }
2602
2603 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
2604                             enum obd_import_event event)
2605 {
2606         int rc = 0;
2607
2608         LASSERT(imp->imp_obd == obd);
2609
2610         switch (event) {
2611         case IMP_EVENT_DISCON: {
2612 #if 0
2613                 /* XXX Pass event up to OBDs stack. used only for FLD now */
2614                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DISCON, NULL);
2615 #endif
2616                 break;
2617         }
2618         case IMP_EVENT_INACTIVE: {
2619                 struct client_obd *cli = &obd->u.cli;
2620                 /*
2621                  * Flush current sequence to make client obtain new one
2622                  * from server in case of disconnect/reconnect.
2623                  */
2624                 if (cli->cl_seq)
2625                         seq_client_flush(cli->cl_seq);
2626
2627                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
2628                 break;
2629         }
2630         case IMP_EVENT_INVALIDATE: {
2631                 struct ldlm_namespace *ns = obd->obd_namespace;
2632
2633                 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2634
2635                 break;
2636         }
2637         case IMP_EVENT_ACTIVE:
2638                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
2639                 /* redo the kuc registration after reconnecting */
2640                 if (rc == 0)
2641                         /* re-register HSM agents */
2642                         rc = libcfs_kkuc_group_foreach(KUC_GRP_HSM,
2643                                                        mdc_hsm_ct_reregister,
2644                                                        (void *)imp);
2645                 break;
2646         case IMP_EVENT_OCD:
2647                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
2648                 break;
2649         case IMP_EVENT_DEACTIVATE:
2650         case IMP_EVENT_ACTIVATE:
2651                 break;
2652         default:
2653                 CERROR("Unknown import event %x\n", event);
2654                 LBUG();
2655         }
2656         return rc;
2657 }
2658
2659 int mdc_fid_alloc(const struct lu_env *env, struct obd_export *exp,
2660                   struct lu_fid *fid, struct md_op_data *op_data)
2661 {
2662         struct client_obd *cli = &exp->exp_obd->u.cli;
2663         struct lu_client_seq *seq = cli->cl_seq;
2664
2665         return seq_client_alloc_fid(env, seq, fid);
2666 }
2667
2668 static struct obd_uuid *mdc_get_uuid(struct obd_export *exp)
2669 {
2670         struct client_obd *cli = &exp->exp_obd->u.cli;
2671
2672         return &cli->cl_target_uuid;
2673 }
2674
2675 /**
2676  * Determine whether the lock can be canceled before replaying it during
2677  * recovery, non zero value will be return if the lock can be canceled,
2678  * or zero returned for not
2679  */
2680 static int mdc_cancel_weight(struct ldlm_lock *lock)
2681 {
2682         if (lock->l_resource->lr_type != LDLM_IBITS)
2683                 return 0;
2684
2685         /* FIXME: if we ever get into a situation where there are too many
2686          * opened files with open locks on a single node, then we really
2687          * should replay these open locks to reget it
2688          */
2689         if (lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_OPEN)
2690                 return 0;
2691
2692         return 1;
2693 }
2694
2695 static int mdc_resource_inode_free(struct ldlm_resource *res)
2696 {
2697         if (res->lr_lvb_inode)
2698                 res->lr_lvb_inode = NULL;
2699
2700         return 0;
2701 }
2702
2703 static struct ldlm_valblock_ops inode_lvbo = {
2704         .lvbo_free = mdc_resource_inode_free,
2705 };
2706
2707 static int mdc_llog_init(struct obd_device *obd)
2708 {
2709         struct obd_llog_group   *olg = &obd->obd_olg;
2710         struct llog_ctxt        *ctxt;
2711         int                      rc;
2712
2713         rc = llog_setup(NULL, obd, olg, LLOG_CHANGELOG_REPL_CTXT, obd,
2714                         &llog_client_ops);
2715         if (rc)
2716                 return rc;
2717
2718         ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
2719         llog_initiator_connect(ctxt);
2720         llog_ctxt_put(ctxt);
2721
2722         return 0;
2723 }
2724
2725 static void mdc_llog_finish(struct obd_device *obd)
2726 {
2727         struct llog_ctxt *ctxt;
2728
2729         ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
2730         if (ctxt)
2731                 llog_cleanup(NULL, ctxt);
2732 }
2733
2734 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
2735 {
2736         struct client_obd *cli = &obd->u.cli;
2737         struct lprocfs_static_vars lvars = { NULL };
2738         int rc;
2739
2740         cli->cl_rpc_lock = kzalloc(sizeof(*cli->cl_rpc_lock), GFP_NOFS);
2741         if (!cli->cl_rpc_lock)
2742                 return -ENOMEM;
2743         mdc_init_rpc_lock(cli->cl_rpc_lock);
2744
2745         rc = ptlrpcd_addref();
2746         if (rc < 0)
2747                 goto err_rpc_lock;
2748
2749         cli->cl_close_lock = kzalloc(sizeof(*cli->cl_close_lock), GFP_NOFS);
2750         if (!cli->cl_close_lock) {
2751                 rc = -ENOMEM;
2752                 goto err_ptlrpcd_decref;
2753         }
2754         mdc_init_rpc_lock(cli->cl_close_lock);
2755
2756         rc = client_obd_setup(obd, cfg);
2757         if (rc)
2758                 goto err_close_lock;
2759         lprocfs_mdc_init_vars(&lvars);
2760         lprocfs_obd_setup(obd, lvars.obd_vars, lvars.sysfs_vars);
2761         sptlrpc_lprocfs_cliobd_attach(obd);
2762         ptlrpc_lprocfs_register_obd(obd);
2763
2764         ns_register_cancel(obd->obd_namespace, mdc_cancel_weight);
2765
2766         obd->obd_namespace->ns_lvbo = &inode_lvbo;
2767
2768         rc = mdc_llog_init(obd);
2769         if (rc) {
2770                 mdc_cleanup(obd);
2771                 CERROR("failed to setup llogging subsystems\n");
2772         }
2773
2774         return rc;
2775
2776 err_close_lock:
2777         kfree(cli->cl_close_lock);
2778 err_ptlrpcd_decref:
2779         ptlrpcd_decref();
2780 err_rpc_lock:
2781         kfree(cli->cl_rpc_lock);
2782         return rc;
2783 }
2784
2785 /* Initialize the default and maximum LOV EA and cookie sizes.  This allows
2786  * us to make MDS RPCs with large enough reply buffers to hold a default
2787  * sized EA and cookie without having to calculate this (via a call into the
2788  * LOV + OSCs) each time we make an RPC.  The maximum size is also tracked
2789  * but not used to avoid wastefully vmalloc()'ing large reply buffers when
2790  * a large number of stripes is possible.  If a larger reply buffer is
2791  * required it will be reallocated in the ptlrpc layer due to overflow.
2792  */
2793 static int mdc_init_ea_size(struct obd_export *exp, u32 easize, u32 def_easize,
2794                             u32 cookiesize, u32 def_cookiesize)
2795 {
2796         struct obd_device *obd = exp->exp_obd;
2797         struct client_obd *cli = &obd->u.cli;
2798
2799         if (cli->cl_max_mds_easize < easize)
2800                 cli->cl_max_mds_easize = easize;
2801
2802         if (cli->cl_default_mds_easize < def_easize)
2803                 cli->cl_default_mds_easize = def_easize;
2804
2805         if (cli->cl_max_mds_cookiesize < cookiesize)
2806                 cli->cl_max_mds_cookiesize = cookiesize;
2807
2808         if (cli->cl_default_mds_cookiesize < def_cookiesize)
2809                 cli->cl_default_mds_cookiesize = def_cookiesize;
2810
2811         return 0;
2812 }
2813
2814 static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2815 {
2816         switch (stage) {
2817         case OBD_CLEANUP_EARLY:
2818                 break;
2819         case OBD_CLEANUP_EXPORTS:
2820                 /* Failsafe, ok if racy */
2821                 if (obd->obd_type->typ_refcnt <= 1)
2822                         libcfs_kkuc_group_rem(0, KUC_GRP_HSM);
2823
2824                 obd_cleanup_client_import(obd);
2825                 ptlrpc_lprocfs_unregister_obd(obd);
2826                 lprocfs_obd_cleanup(obd);
2827
2828                 mdc_llog_finish(obd);
2829                 break;
2830         }
2831         return 0;
2832 }
2833
2834 static int mdc_cleanup(struct obd_device *obd)
2835 {
2836         struct client_obd *cli = &obd->u.cli;
2837
2838         kfree(cli->cl_rpc_lock);
2839         kfree(cli->cl_close_lock);
2840
2841         ptlrpcd_decref();
2842
2843         return client_obd_cleanup(obd);
2844 }
2845
2846 static int mdc_process_config(struct obd_device *obd, u32 len, void *buf)
2847 {
2848         struct lustre_cfg *lcfg = buf;
2849         struct lprocfs_static_vars lvars = { NULL };
2850         int rc = 0;
2851
2852         lprocfs_mdc_init_vars(&lvars);
2853         switch (lcfg->lcfg_command) {
2854         default:
2855                 rc = class_process_proc_param(PARAM_MDC, lvars.obd_vars,
2856                                               lcfg, obd);
2857                 if (rc > 0)
2858                         rc = 0;
2859                 break;
2860         }
2861         return rc;
2862 }
2863
2864 static struct obd_ops mdc_obd_ops = {
2865         .owner          = THIS_MODULE,
2866         .setup          = mdc_setup,
2867         .precleanup     = mdc_precleanup,
2868         .cleanup        = mdc_cleanup,
2869         .add_conn       = client_import_add_conn,
2870         .del_conn       = client_import_del_conn,
2871         .connect        = client_connect_import,
2872         .disconnect     = client_disconnect_export,
2873         .iocontrol      = mdc_iocontrol,
2874         .set_info_async = mdc_set_info_async,
2875         .statfs         = mdc_statfs,
2876         .fid_init       = client_fid_init,
2877         .fid_fini       = client_fid_fini,
2878         .fid_alloc      = mdc_fid_alloc,
2879         .import_event   = mdc_import_event,
2880         .get_info       = mdc_get_info,
2881         .process_config = mdc_process_config,
2882         .get_uuid       = mdc_get_uuid,
2883         .quotactl       = mdc_quotactl,
2884         .quotacheck     = mdc_quotacheck
2885 };
2886
2887 static struct md_ops mdc_md_ops = {
2888         .getstatus              = mdc_getstatus,
2889         .null_inode             = mdc_null_inode,
2890         .close                  = mdc_close,
2891         .create                 = mdc_create,
2892         .done_writing           = mdc_done_writing,
2893         .enqueue                = mdc_enqueue,
2894         .getattr                = mdc_getattr,
2895         .getattr_name           = mdc_getattr_name,
2896         .intent_lock            = mdc_intent_lock,
2897         .link                   = mdc_link,
2898         .rename                 = mdc_rename,
2899         .setattr                = mdc_setattr,
2900         .setxattr               = mdc_setxattr,
2901         .getxattr               = mdc_getxattr,
2902         .sync                   = mdc_sync,
2903         .read_page              = mdc_read_page,
2904         .unlink                 = mdc_unlink,
2905         .cancel_unused          = mdc_cancel_unused,
2906         .init_ea_size           = mdc_init_ea_size,
2907         .set_lock_data          = mdc_set_lock_data,
2908         .lock_match             = mdc_lock_match,
2909         .get_lustre_md          = mdc_get_lustre_md,
2910         .free_lustre_md         = mdc_free_lustre_md,
2911         .set_open_replay_data   = mdc_set_open_replay_data,
2912         .clear_open_replay_data = mdc_clear_open_replay_data,
2913         .intent_getattr_async   = mdc_intent_getattr_async,
2914         .revalidate_lock        = mdc_revalidate_lock
2915 };
2916
2917 static int __init mdc_init(void)
2918 {
2919         struct lprocfs_static_vars lvars = { NULL };
2920
2921         lprocfs_mdc_init_vars(&lvars);
2922
2923         return class_register_type(&mdc_obd_ops, &mdc_md_ops,
2924                                  LUSTRE_MDC_NAME, NULL);
2925 }
2926
2927 static void /*__exit*/ mdc_exit(void)
2928 {
2929         class_unregister_type(LUSTRE_MDC_NAME);
2930 }
2931
2932 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
2933 MODULE_DESCRIPTION("Lustre Metadata Client");
2934 MODULE_VERSION(LUSTRE_VERSION_STRING);
2935 MODULE_LICENSE("GPL");
2936
2937 module_init(mdc_init);
2938 module_exit(mdc_exit);