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, 2016, 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/init.h>
  36 #include <linux/kthread.h>
  37 #include <linux/module.h>
  38 #include <linux/pagemap.h>
  39 #include <linux/user_namespace.h>
  40 #include <linux/utsname.h>
  41 #ifdef HAVE_UIDGID_HEADER
  42 # include <linux/uidgid.h>
  43 #endif
  44
  45 #include <lustre_errno.h>
  46
  47 #include <cl_object.h>
  48 #include <llog_swab.h>
  49 #include <lprocfs_status.h>
  50 #include <lustre_acl.h>
  51 #include <lustre_fid.h>
  52 #include <uapi/linux/lustre/lustre_ioctl.h>
  53 #include <lustre_kernelcomm.h>
  54 #include <lustre_lmv.h>
  55 #include <lustre_log.h>
  56 #include <uapi/linux/lustre/lustre_param.h>
  57 #include <lustre_swab.h>
  58 #include <obd_class.h>
  59 #include <lustre_osc.h>
  60
  61 #include "mdc_internal.h"
  62
  63 #define REQUEST_MINOR 244
  64
  65 static int mdc_cleanup(struct obd_device *obd);
  66
  67 static inline int mdc_queue_wait(struct ptlrpc_request *req)
  68 {
  69         struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
  70         int rc;
  71
  72         /* obd_get_request_slot() ensures that this client has no more
  73          * than cl_max_rpcs_in_flight RPCs simultaneously inf light
  74          * against an MDT. */
  75         rc = obd_get_request_slot(cli);
  76         if (rc != 0)
  77                 return rc;
  78
  79         rc = ptlrpc_queue_wait(req);
  80         obd_put_request_slot(cli);
  81
  82         return rc;
  83 }
  84
  85 /*
  86  * Send MDS_GET_ROOT RPC to fetch root FID.
  87  *
  88  * If \a fileset is not NULL it should contain a subdirectory off
  89  * the ROOT/ directory to be mounted on the client. Return the FID
  90  * of the subdirectory to the client to mount onto its mountpoint.
  91  *
  92  * \param[in]   imp     MDC import
  93  * \param[in]   fileset fileset name, which could be NULL
  94  * \param[out]  rootfid root FID of this mountpoint
  95  * \param[out]  pc      root capa will be unpacked and saved in this pointer
  96  *
  97  * \retval      0 on success, negative errno on failure
  98  */
  99 static int mdc_get_root(struct obd_export *exp, const char *fileset,
 100                          struct lu_fid *rootfid)
 101 {
 102         struct ptlrpc_request   *req;
 103         struct mdt_body         *body;
 104         int                      rc;
 105
 106         ENTRY;
 107
 108         if (fileset && !(exp_connect_flags(exp) & OBD_CONNECT_SUBTREE))
 109                 RETURN(-ENOTSUPP);
 110
 111         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 112                                 &RQF_MDS_GET_ROOT);
 113         if (req == NULL)
 114                 RETURN(-ENOMEM);
 115
 116         if (fileset != NULL)
 117                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 118                                      strlen(fileset) + 1);
 119         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_ROOT);
 120         if (rc) {
 121                 ptlrpc_request_free(req);
 122                 RETURN(rc);
 123         }
 124         mdc_pack_body(req, NULL, 0, 0, -1, 0);
 125         if (fileset != NULL) {
 126                 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 127
 128                 memcpy(name, fileset, strlen(fileset));
 129         }
 130         lustre_msg_add_flags(req->rq_reqmsg, LUSTRE_IMP_FULL);
 131         req->rq_send_state = LUSTRE_IMP_FULL;
 132
 133         ptlrpc_request_set_replen(req);
 134
 135         rc = ptlrpc_queue_wait(req);
 136         if (rc)
 137                 GOTO(out, rc);
 138
 139         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 140         if (body == NULL)
 141                 GOTO(out, rc = -EPROTO);
 142
 143         *rootfid = body->mbo_fid1;
 144         CDEBUG(D_NET, "root fid="DFID", last_committed=%llu\n",
 145                PFID(rootfid), lustre_msg_get_last_committed(req->rq_repmsg));
 146         EXIT;
 147 out:
 148         ptlrpc_req_finished(req);
 149
 150         return rc;
 151 }
 152
 153 /*
 154  * This function now is known to always saying that it will receive 4 buffers
 155  * from server. Even for cases when acl_size and md_size is zero, RPC header
 156  * will contain 4 fields and RPC itself will contain zero size fields. This is
 157  * because mdt_getattr*() _always_ returns 4 fields, but if acl is not needed
 158  * and thus zero, it shrinks it, making zero size. The same story about
 159  * md_size. And this is course of problem when client waits for smaller number
 160  * of fields. This issue will be fixed later when client gets aware of RPC
 161  * layouts.  --umka
 162  */
 163 static int mdc_getattr_common(struct obd_export *exp,
 164                               struct ptlrpc_request *req)
 165 {
 166         struct req_capsule *pill = &req->rq_pill;
 167         struct mdt_body    *body;
 168         void               *eadata;
 169         int                 rc;
 170         ENTRY;
 171
 172         /* Request message already built. */
 173         rc = ptlrpc_queue_wait(req);
 174         if (rc != 0)
 175                 RETURN(rc);
 176
 177         /* sanity check for the reply */
 178         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 179         if (body == NULL)
 180                 RETURN(-EPROTO);
 181
 182         CDEBUG(D_NET, "mode: %o\n", body->mbo_mode);
 183
 184         mdc_update_max_ea_from_body(exp, body);
 185         if (body->mbo_eadatasize != 0) {
 186                 eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
 187                                                       body->mbo_eadatasize);
 188                 if (eadata == NULL)
 189                         RETURN(-EPROTO);
 190         }
 191
 192         RETURN(0);
 193 }
 194
 195 static int mdc_getattr(struct obd_export *exp, struct md_op_data *op_data,
 196                        struct ptlrpc_request **request)
 197 {
 198         struct ptlrpc_request *req;
 199         int                    rc;
 200         ENTRY;
 201
 202         /* Single MDS without an LMV case */
 203         if (op_data->op_flags & MF_GET_MDT_IDX) {
 204                 op_data->op_mds = 0;
 205                 RETURN(0);
 206         }
 207         *request = NULL;
 208         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
 209         if (req == NULL)
 210                 RETURN(-ENOMEM);
 211
 212         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
 213         if (rc) {
 214                 ptlrpc_request_free(req);
 215                 RETURN(rc);
 216         }
 217
 218         mdc_pack_body(req, &op_data->op_fid1, op_data->op_valid,
 219                       op_data->op_mode, -1, 0);
 220
 221         req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
 222                              req->rq_import->imp_connect_data.ocd_max_easize);
 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_getattr_name(struct obd_export *exp, struct md_op_data *op_data,
 236                             struct ptlrpc_request **request)
 237 {
 238         struct ptlrpc_request *req;
 239         int                    rc;
 240         ENTRY;
 241
 242         *request = NULL;
 243         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 244                                    &RQF_MDS_GETATTR_NAME);
 245         if (req == NULL)
 246                 RETURN(-ENOMEM);
 247
 248         req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 249                              op_data->op_namelen + 1);
 250
 251         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR_NAME);
 252         if (rc) {
 253                 ptlrpc_request_free(req);
 254                 RETURN(rc);
 255         }
 256
 257         mdc_pack_body(req, &op_data->op_fid1, op_data->op_valid,
 258                       op_data->op_mode, op_data->op_suppgids[0], 0);
 259
 260         if (op_data->op_name) {
 261                 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 262                 LASSERT(strnlen(op_data->op_name, op_data->op_namelen) ==
 263                                 op_data->op_namelen);
 264                 memcpy(name, op_data->op_name, op_data->op_namelen);
 265         }
 266
 267         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 268                              op_data->op_mode);
 269         req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
 270                              req->rq_import->imp_connect_data.ocd_max_easize);
 271         ptlrpc_request_set_replen(req);
 272
 273         rc = mdc_getattr_common(exp, req);
 274         if (rc)
 275                 ptlrpc_req_finished(req);
 276         else
 277                 *request = req;
 278         RETURN(rc);
 279 }
 280
 281 static int mdc_xattr_common(struct obd_export *exp,const struct req_format *fmt,
 282                             const struct lu_fid *fid, int opcode, u64 valid,
 283                             const char *xattr_name, const char *input,
 284                             int input_size, int output_size, int flags,
 285                             __u32 suppgid, struct ptlrpc_request **request)
 286 {
 287         struct ptlrpc_request *req;
 288         int   xattr_namelen = 0;
 289         char *tmp;
 290         int   rc;
 291         ENTRY;
 292
 293         *request = NULL;
 294         req = ptlrpc_request_alloc(class_exp2cliimp(exp), fmt);
 295         if (req == NULL)
 296                 RETURN(-ENOMEM);
 297
 298         if (xattr_name) {
 299                 xattr_namelen = strlen(xattr_name) + 1;
 300                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 301                                      xattr_namelen);
 302         }
 303         if (input_size) {
 304                 LASSERT(input);
 305                 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
 306                                      input_size);
 307         }
 308
 309         /* Flush local XATTR locks to get rid of a possible cancel RPC */
 310         if (opcode == MDS_REINT && fid_is_sane(fid) &&
 311             exp->exp_connect_data.ocd_ibits_known & MDS_INODELOCK_XATTR) {
 312                 struct list_head cancels = LIST_HEAD_INIT(cancels);
 313                 int count;
 314
 315                 /* Without that packing would fail */
 316                 if (input_size == 0)
 317                         req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
 318                                              RCL_CLIENT, 0);
 319
 320                 count = mdc_resource_get_unused(exp, fid,
 321                                                 &cancels, LCK_EX,
 322                                                 MDS_INODELOCK_XATTR);
 323
 324                 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
 325                 if (rc) {
 326                         ptlrpc_request_free(req);
 327                         RETURN(rc);
 328                 }
 329         } else {
 330                 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode);
 331                 if (rc) {
 332                         ptlrpc_request_free(req);
 333                         RETURN(rc);
 334                 }
 335         }
 336
 337         if (opcode == MDS_REINT) {
 338                 struct mdt_rec_setxattr *rec;
 339
 340                 CLASSERT(sizeof(struct mdt_rec_setxattr) ==
 341                          sizeof(struct mdt_rec_reint));
 342                 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
 343                 rec->sx_opcode = REINT_SETXATTR;
 344                 rec->sx_fsuid  = from_kuid(&init_user_ns, current_fsuid());
 345                 rec->sx_fsgid  = from_kgid(&init_user_ns, current_fsgid());
 346                 rec->sx_cap    = cfs_curproc_cap_pack();
 347                 rec->sx_suppgid1 = suppgid;
 348                 rec->sx_suppgid2 = -1;
 349                 rec->sx_fid    = *fid;
 350                 rec->sx_valid  = valid | OBD_MD_FLCTIME;
 351                 rec->sx_time   = ktime_get_real_seconds();
 352                 rec->sx_size   = output_size;
 353                 rec->sx_flags  = flags;
 354         } else {
 355                 mdc_pack_body(req, fid, valid, output_size, suppgid, flags);
 356         }
 357
 358         if (xattr_name) {
 359                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 360                 memcpy(tmp, xattr_name, xattr_namelen);
 361         }
 362         if (input_size) {
 363                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
 364                 memcpy(tmp, input, input_size);
 365         }
 366
 367         if (req_capsule_has_field(&req->rq_pill, &RMF_EADATA, RCL_SERVER))
 368                 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
 369                                      RCL_SERVER, output_size);
 370         ptlrpc_request_set_replen(req);
 371
 372         /* make rpc */
 373         if (opcode == MDS_REINT)
 374                 mdc_get_mod_rpc_slot(req, NULL);
 375
 376         rc = ptlrpc_queue_wait(req);
 377
 378         if (opcode == MDS_REINT)
 379                 mdc_put_mod_rpc_slot(req, NULL);
 380
 381         if (rc)
 382                 ptlrpc_req_finished(req);
 383         else
 384                 *request = req;
 385         RETURN(rc);
 386 }
 387
 388 static int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
 389                         u64 valid, const char *xattr_name,
 390                         const char *input, int input_size, int output_size,
 391                         int flags, __u32 suppgid,
 392                         struct ptlrpc_request **request)
 393 {
 394         return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
 395                                 fid, MDS_REINT, valid, xattr_name,
 396                                 input, input_size, output_size, flags,
 397                                 suppgid, request);
 398 }
 399
 400 static int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
 401                         u64 valid, const char *xattr_name,
 402                         const char *input, int input_size, int output_size,
 403                         int flags, struct ptlrpc_request **request)
 404 {
 405         return mdc_xattr_common(exp, &RQF_MDS_GETXATTR,
 406                                 fid, MDS_GETXATTR, valid, xattr_name,
 407                                 input, input_size, output_size, flags,
 408                                 -1, request);
 409 }
 410
 411 #ifdef CONFIG_FS_POSIX_ACL
 412 static int mdc_unpack_acl(struct ptlrpc_request *req, struct lustre_md *md)
 413 {
 414         struct req_capsule     *pill = &req->rq_pill;
 415         struct mdt_body        *body = md->body;
 416         struct posix_acl       *acl;
 417         void                   *buf;
 418         int                     rc;
 419         ENTRY;
 420
 421         if (!body->mbo_aclsize)
 422                 RETURN(0);
 423
 424         buf = req_capsule_server_sized_get(pill, &RMF_ACL, body->mbo_aclsize);
 425
 426         if (!buf)
 427                 RETURN(-EPROTO);
 428
 429         acl = posix_acl_from_xattr(&init_user_ns, buf, body->mbo_aclsize);
 430         if (acl == NULL)
 431                 RETURN(0);
 432         if (IS_ERR(acl)) {
 433                 rc = PTR_ERR(acl);
 434                 CERROR("convert xattr to acl: %d\n", rc);
 435                 RETURN(rc);
 436         }
 437
 438         rc = posix_acl_valid(&init_user_ns, acl);
 439         if (rc) {
 440                 CERROR("validate acl: %d\n", rc);
 441                 posix_acl_release(acl);
 442                 RETURN(rc);
 443         }
 444
 445         md->posix_acl = acl;
 446         RETURN(0);
 447 }
 448 #else
 449 #define mdc_unpack_acl(req, md) 0
 450 #endif
 451
 452 int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
 453                       struct obd_export *dt_exp, struct obd_export *md_exp,
 454                       struct lustre_md *md)
 455 {
 456         struct req_capsule *pill = &req->rq_pill;
 457         int rc;
 458         ENTRY;
 459
 460         LASSERT(md);
 461         memset(md, 0, sizeof(*md));
 462
 463         md->body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 464         LASSERT(md->body != NULL);
 465
 466         if (md->body->mbo_valid & OBD_MD_FLEASIZE) {
 467                 if (!S_ISREG(md->body->mbo_mode)) {
 468                         CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, should be a "
 469                                "regular file, but is not\n");
 470                         GOTO(out, rc = -EPROTO);
 471                 }
 472
 473                 if (md->body->mbo_eadatasize == 0) {
 474                         CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, "
 475                                "but eadatasize 0\n");
 476                         GOTO(out, rc = -EPROTO);
 477                 }
 478
 479                 md->layout.lb_len = md->body->mbo_eadatasize;
 480                 md->layout.lb_buf = req_capsule_server_sized_get(pill,
 481                                                         &RMF_MDT_MD,
 482                                                         md->layout.lb_len);
 483                 if (md->layout.lb_buf == NULL)
 484                         GOTO(out, rc = -EPROTO);
 485         } else if (md->body->mbo_valid & OBD_MD_FLDIREA) {
 486                 const union lmv_mds_md *lmv;
 487                 size_t lmv_size;
 488
 489                 if (!S_ISDIR(md->body->mbo_mode)) {
 490                         CDEBUG(D_INFO, "OBD_MD_FLDIREA set, should be a "
 491                                "directory, but is not\n");
 492                         GOTO(out, rc = -EPROTO);
 493                 }
 494
 495                 lmv_size = md->body->mbo_eadatasize;
 496                 if (lmv_size == 0) {
 497                         CDEBUG(D_INFO, "OBD_MD_FLDIREA is set, "
 498                                "but eadatasize 0\n");
 499                         RETURN(-EPROTO);
 500                 }
 501
 502                 if (md->body->mbo_valid & OBD_MD_MEA) {
 503                         lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
 504                                                            lmv_size);
 505                         if (lmv == NULL)
 506                                 GOTO(out, rc = -EPROTO);
 507
 508                         rc = md_unpackmd(md_exp, &md->lmv, lmv, lmv_size);
 509                         if (rc < 0)
 510                                 GOTO(out, rc);
 511
 512                         if (rc < (typeof(rc))sizeof(*md->lmv)) {
 513                                 CDEBUG(D_INFO, "size too small:  "
 514                                        "rc < sizeof(*md->lmv) (%d < %d)\n",
 515                                         rc, (int)sizeof(*md->lmv));
 516                                 GOTO(out, rc = -EPROTO);
 517                         }
 518                 }
 519         }
 520         rc = 0;
 521
 522         if (md->body->mbo_valid & OBD_MD_FLACL) {
 523                 /* for ACL, it's possible that FLACL is set but aclsize is zero.
 524                  * only when aclsize != 0 there's an actual segment for ACL
 525                  * in reply buffer.
 526                  */
 527                 if (md->body->mbo_aclsize) {
 528                         rc = mdc_unpack_acl(req, md);
 529                         if (rc)
 530                                 GOTO(out, rc);
 531 #ifdef CONFIG_FS_POSIX_ACL
 532                 } else {
 533                         md->posix_acl = NULL;
 534 #endif
 535                 }
 536         }
 537
 538         EXIT;
 539 out:
 540         if (rc) {
 541 #ifdef CONFIG_FS_POSIX_ACL
 542                 posix_acl_release(md->posix_acl);
 543 #endif
 544         }
 545         return rc;
 546 }
 547
 548 int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
 549 {
 550         ENTRY;
 551         RETURN(0);
 552 }
 553
 554 void mdc_replay_open(struct ptlrpc_request *req)
 555 {
 556         struct md_open_data *mod = req->rq_cb_data;
 557         struct ptlrpc_request *close_req;
 558         struct obd_client_handle *och;
 559         struct lustre_handle old;
 560         struct mdt_body *body;
 561         ENTRY;
 562
 563         if (mod == NULL) {
 564                 DEBUG_REQ(D_ERROR, req,
 565                           "Can't properly replay without open data.");
 566                 EXIT;
 567                 return;
 568         }
 569
 570         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 571         LASSERT(body != NULL);
 572
 573         spin_lock(&req->rq_lock);
 574         och = mod->mod_och;
 575         if (och && och->och_fh.cookie)
 576                 req->rq_early_free_repbuf = 1;
 577         else
 578                 req->rq_early_free_repbuf = 0;
 579         spin_unlock(&req->rq_lock);
 580
 581         if (req->rq_early_free_repbuf) {
 582                 struct lustre_handle *file_fh;
 583
 584                 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
 585
 586                 file_fh = &och->och_fh;
 587                 CDEBUG(D_HA, "updating handle from %#llx to %#llx\n",
 588                        file_fh->cookie, body->mbo_handle.cookie);
 589                 old = *file_fh;
 590                 *file_fh = body->mbo_handle;
 591         }
 592
 593         close_req = mod->mod_close_req;
 594         if (close_req) {
 595                 __u32 opc = lustre_msg_get_opc(close_req->rq_reqmsg);
 596                 struct mdt_ioepoch *epoch;
 597
 598                 LASSERT(opc == MDS_CLOSE);
 599                 epoch = req_capsule_client_get(&close_req->rq_pill,
 600                                                &RMF_MDT_EPOCH);
 601                 LASSERT(epoch);
 602
 603                 if (req->rq_early_free_repbuf)
 604                         LASSERT(!memcmp(&old, &epoch->mio_handle, sizeof(old)));
 605
 606                 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
 607                 epoch->mio_handle = body->mbo_handle;
 608         }
 609         EXIT;
 610 }
 611
 612 void mdc_commit_open(struct ptlrpc_request *req)
 613 {
 614         struct md_open_data *mod = req->rq_cb_data;
 615         if (mod == NULL)
 616                 return;
 617
 618         /**
 619          * No need to touch md_open_data::mod_och, it holds a reference on
 620          * \var mod and will zero references to each other, \var mod will be
 621          * freed after that when md_open_data::mod_och will put the reference.
 622          */
 623
 624         /**
 625          * Do not let open request to disappear as it still may be needed
 626          * for close rpc to happen (it may happen on evict only, otherwise
 627          * ptlrpc_request::rq_replay does not let mdc_commit_open() to be
 628          * called), just mark this rpc as committed to distinguish these 2
 629          * cases, see mdc_close() for details. The open request reference will
 630          * be put along with freeing \var mod.
 631          */
 632         ptlrpc_request_addref(req);
 633         spin_lock(&req->rq_lock);
 634         req->rq_committed = 1;
 635         spin_unlock(&req->rq_lock);
 636         req->rq_cb_data = NULL;
 637         obd_mod_put(mod);
 638 }
 639
 640 int mdc_set_open_replay_data(struct obd_export *exp,
 641                              struct obd_client_handle *och,
 642                              struct lookup_intent *it)
 643 {
 644         struct md_open_data     *mod;
 645         struct mdt_rec_create   *rec;
 646         struct mdt_body         *body;
 647         struct ptlrpc_request   *open_req = it->it_request;
 648         struct obd_import       *imp = open_req->rq_import;
 649         ENTRY;
 650
 651         if (!open_req->rq_replay)
 652                 RETURN(0);
 653
 654         rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
 655         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
 656         LASSERT(rec != NULL);
 657         /* Incoming message in my byte order (it's been swabbed). */
 658         /* Outgoing messages always in my byte order. */
 659         LASSERT(body != NULL);
 660
 661         /* Only if the import is replayable, we set replay_open data */
 662         if (och && imp->imp_replayable) {
 663                 mod = obd_mod_alloc();
 664                 if (mod == NULL) {
 665                         DEBUG_REQ(D_ERROR, open_req,
 666                                   "Can't allocate md_open_data");
 667                         RETURN(0);
 668                 }
 669
 670                 /**
 671                  * Take a reference on \var mod, to be freed on mdc_close().
 672                  * It protects \var mod from being freed on eviction (commit
 673                  * callback is called despite rq_replay flag).
 674                  * Another reference for \var och.
 675                  */
 676                 obd_mod_get(mod);
 677                 obd_mod_get(mod);
 678
 679                 spin_lock(&open_req->rq_lock);
 680                 och->och_mod = mod;
 681                 mod->mod_och = och;
 682                 mod->mod_is_create = it_disposition(it, DISP_OPEN_CREATE) ||
 683                                      it_disposition(it, DISP_OPEN_STRIPE);
 684                 mod->mod_open_req = open_req;
 685                 open_req->rq_cb_data = mod;
 686                 open_req->rq_commit_cb = mdc_commit_open;
 687                 open_req->rq_early_free_repbuf = 1;
 688                 spin_unlock(&open_req->rq_lock);
 689         }
 690
 691         rec->cr_fid2 = body->mbo_fid1;
 692         rec->cr_ioepoch = body->mbo_ioepoch;
 693         rec->cr_old_handle.cookie = body->mbo_handle.cookie;
 694         open_req->rq_replay_cb = mdc_replay_open;
 695         if (!fid_is_sane(&body->mbo_fid1)) {
 696                 DEBUG_REQ(D_ERROR, open_req, "Saving replay request with "
 697                           "insane fid");
 698                 LBUG();
 699         }
 700
 701         DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
 702         RETURN(0);
 703 }
 704
 705 static void mdc_free_open(struct md_open_data *mod)
 706 {
 707         int committed = 0;
 708
 709         if (mod->mod_is_create == 0 &&
 710             imp_connect_disp_stripe(mod->mod_open_req->rq_import))
 711                 committed = 1;
 712
 713         /**
 714          * No reason to asssert here if the open request has
 715          * rq_replay == 1. It means that mdc_close failed, and
 716          * close request wasn`t sent. It is not fatal to client.
 717          * The worst thing is eviction if the client gets open lock
 718          **/
 719
 720         DEBUG_REQ(D_RPCTRACE, mod->mod_open_req, "free open request rq_replay"
 721                   "= %d\n", mod->mod_open_req->rq_replay);
 722
 723         ptlrpc_request_committed(mod->mod_open_req, committed);
 724         if (mod->mod_close_req)
 725                 ptlrpc_request_committed(mod->mod_close_req, committed);
 726 }
 727
 728 int mdc_clear_open_replay_data(struct obd_export *exp,
 729                                struct obd_client_handle *och)
 730 {
 731         struct md_open_data *mod = och->och_mod;
 732         ENTRY;
 733
 734         /**
 735          * It is possible to not have \var mod in a case of eviction between
 736          * lookup and ll_file_open().
 737          **/
 738         if (mod == NULL)
 739                 RETURN(0);
 740
 741         LASSERT(mod != LP_POISON);
 742         LASSERT(mod->mod_open_req != NULL);
 743
 744         spin_lock(&mod->mod_open_req->rq_lock);
 745         if (mod->mod_och)
 746                 mod->mod_och->och_fh.cookie = 0;
 747         mod->mod_open_req->rq_early_free_repbuf = 0;
 748         spin_unlock(&mod->mod_open_req->rq_lock);
 749         mdc_free_open(mod);
 750
 751         mod->mod_och = NULL;
 752         och->och_mod = NULL;
 753         obd_mod_put(mod);
 754
 755         RETURN(0);
 756 }
 757
 758 static int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
 759                      struct md_open_data *mod, struct ptlrpc_request **request)
 760 {
 761         struct obd_device     *obd = class_exp2obd(exp);
 762         struct ptlrpc_request *req;
 763         struct req_format     *req_fmt;
 764         int                    rc;
 765         int                    saved_rc = 0;
 766         ENTRY;
 767
 768         if (op_data->op_bias & MDS_HSM_RELEASE) {
 769                 req_fmt = &RQF_MDS_INTENT_CLOSE;
 770
 771                 /* allocate a FID for volatile file */
 772                 rc = mdc_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
 773                 if (rc < 0) {
 774                         CERROR("%s: "DFID" failed to allocate FID: %d\n",
 775                                obd->obd_name, PFID(&op_data->op_fid1), rc);
 776                         /* save the errcode and proceed to close */
 777                         saved_rc = rc;
 778                 }
 779         } else if (op_data->op_bias & (MDS_CLOSE_LAYOUT_SWAP |
 780                                        MDS_CLOSE_LAYOUT_MERGE)) {
 781                 req_fmt = &RQF_MDS_INTENT_CLOSE;
 782         } else {
 783                 req_fmt = &RQF_MDS_CLOSE;
 784         }
 785
 786         *request = NULL;
 787         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_CLOSE))
 788                 req = NULL;
 789         else
 790                 req = ptlrpc_request_alloc(class_exp2cliimp(exp), req_fmt);
 791
 792         /* Ensure that this close's handle is fixed up during replay. */
 793         if (likely(mod != NULL)) {
 794                 LASSERTF(mod->mod_open_req != NULL &&
 795                          mod->mod_open_req->rq_type != LI_POISON,
 796                          "POISONED open %p!\n", mod->mod_open_req);
 797
 798                 mod->mod_close_req = req;
 799
 800                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
 801                 /* We no longer want to preserve this open for replay even
 802                  * though the open was committed. b=3632, b=3633 */
 803                 spin_lock(&mod->mod_open_req->rq_lock);
 804                 mod->mod_open_req->rq_replay = 0;
 805                 spin_unlock(&mod->mod_open_req->rq_lock);
 806         } else {
 807                 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
 808         }
 809         if (req == NULL) {
 810                 /**
 811                  * TODO: repeat close after errors
 812                  */
 813                 CWARN("%s: close of FID "DFID" failed, file reference will be "
 814                       "dropped when this client unmounts or is evicted\n",
 815                       obd->obd_name, PFID(&op_data->op_fid1));
 816                 GOTO(out, rc = -ENOMEM);
 817         }
 818
 819         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
 820         if (rc) {
 821                 ptlrpc_request_free(req);
 822                 req = NULL;
 823                 GOTO(out, rc);
 824         }
 825
 826         /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
 827          * portal whose threads are not taking any DLM locks and are therefore
 828          * always progressing */
 829         req->rq_request_portal = MDS_READPAGE_PORTAL;
 830         ptlrpc_at_set_req_timeout(req);
 831
 832
 833         mdc_close_pack(req, op_data);
 834
 835         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 836                              obd->u.cli.cl_default_mds_easize);
 837
 838         ptlrpc_request_set_replen(req);
 839
 840         mdc_get_mod_rpc_slot(req, NULL);
 841         rc = ptlrpc_queue_wait(req);
 842         mdc_put_mod_rpc_slot(req, NULL);
 843
 844         if (req->rq_repmsg == NULL) {
 845                 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
 846                        req->rq_status);
 847                 if (rc == 0)
 848                         rc = req->rq_status ?: -EIO;
 849         } else if (rc == 0 || rc == -EAGAIN) {
 850                 struct mdt_body *body;
 851
 852                 rc = lustre_msg_get_status(req->rq_repmsg);
 853                 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
 854                         DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
 855                                   "= %d", rc);
 856                         if (rc > 0)
 857                                 rc = -rc;
 858                 }
 859                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 860                 if (body == NULL)
 861                         rc = -EPROTO;
 862         } else if (rc == -ESTALE) {
 863                 /**
 864                  * it can be allowed error after 3633 if open was committed and
 865                  * server failed before close was sent. Let's check if mod
 866                  * exists and return no error in that case
 867                  */
 868                 if (mod) {
 869                         DEBUG_REQ(D_HA, req, "Reset ESTALE = %d", rc);
 870                         LASSERT(mod->mod_open_req != NULL);
 871                         if (mod->mod_open_req->rq_committed)
 872                                 rc = 0;
 873                 }
 874         }
 875
 876 out:
 877         if (mod) {
 878                 if (rc != 0)
 879                         mod->mod_close_req = NULL;
 880                 /* Since now, mod is accessed through open_req only,
 881                  * thus close req does not keep a reference on mod anymore. */
 882                 obd_mod_put(mod);
 883         }
 884         *request = req;
 885
 886         RETURN(rc < 0 ? rc : saved_rc);
 887 }
 888
 889 static int mdc_getpage(struct obd_export *exp, const struct lu_fid *fid,
 890                        u64 offset, struct page **pages, int npages,
 891                        struct ptlrpc_request **request)
 892 {
 893         struct ptlrpc_request   *req;
 894         struct ptlrpc_bulk_desc *desc;
 895         int                      i;
 896         wait_queue_head_t        waitq;
 897         int                      resends = 0;
 898         struct l_wait_info       lwi;
 899         int                      rc;
 900         ENTRY;
 901
 902         *request = NULL;
 903         init_waitqueue_head(&waitq);
 904
 905 restart_bulk:
 906         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
 907         if (req == NULL)
 908                 RETURN(-ENOMEM);
 909
 910         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
 911         if (rc) {
 912                 ptlrpc_request_free(req);
 913                 RETURN(rc);
 914         }
 915
 916         req->rq_request_portal = MDS_READPAGE_PORTAL;
 917         ptlrpc_at_set_req_timeout(req);
 918
 919         desc = ptlrpc_prep_bulk_imp(req, npages, 1,
 920                                     PTLRPC_BULK_PUT_SINK | PTLRPC_BULK_BUF_KIOV,
 921                                     MDS_BULK_PORTAL,
 922                                     &ptlrpc_bulk_kiov_pin_ops);
 923         if (desc == NULL) {
 924                 ptlrpc_req_finished(req);
 925                 RETURN(-ENOMEM);
 926         }
 927
 928         /* NB req now owns desc and will free it when it gets freed */
 929         for (i = 0; i < npages; i++)
 930                 desc->bd_frag_ops->add_kiov_frag(desc, pages[i], 0,
 931                                                  PAGE_SIZE);
 932
 933         mdc_readdir_pack(req, offset, PAGE_SIZE * npages, fid);
 934
 935         ptlrpc_request_set_replen(req);
 936         rc = ptlrpc_queue_wait(req);
 937         if (rc) {
 938                 ptlrpc_req_finished(req);
 939                 if (rc != -ETIMEDOUT)
 940                         RETURN(rc);
 941
 942                 resends++;
 943                 if (!client_should_resend(resends, &exp->exp_obd->u.cli)) {
 944                         CERROR("%s: too many resend retries: rc = %d\n",
 945                                exp->exp_obd->obd_name, -EIO);
 946                         RETURN(-EIO);
 947                 }
 948                 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(resends), NULL, NULL,
 949                                        NULL);
 950                 l_wait_event(waitq, 0, &lwi);
 951
 952                 goto restart_bulk;
 953         }
 954
 955         rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
 956                                           req->rq_bulk->bd_nob_transferred);
 957         if (rc < 0) {
 958                 ptlrpc_req_finished(req);
 959                 RETURN(rc);
 960         }
 961
 962         if (req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK) {
 963                 CERROR("%s: unexpected bytes transferred: %d (%ld expected)\n",
 964                        exp->exp_obd->obd_name, req->rq_bulk->bd_nob_transferred,
 965                        PAGE_SIZE * npages);
 966                 ptlrpc_req_finished(req);
 967                 RETURN(-EPROTO);
 968         }
 969
 970         *request = req;
 971         RETURN(0);
 972 }
 973
 974 static void mdc_release_page(struct page *page, int remove)
 975 {
 976         if (remove) {
 977                 lock_page(page);
 978                 if (likely(page->mapping != NULL))
 979                         truncate_complete_page(page->mapping, page);
 980                 unlock_page(page);
 981         }
 982         put_page(page);
 983 }
 984
 985 static struct page *mdc_page_locate(struct address_space *mapping, __u64 *hash,
 986                                     __u64 *start, __u64 *end, int hash64)
 987 {
 988         /*
 989          * Complement of hash is used as an index so that
 990          * radix_tree_gang_lookup() can be used to find a page with starting
 991          * hash _smaller_ than one we are looking for.
 992          */
 993         unsigned long offset = hash_x_index(*hash, hash64);
 994         struct page *page;
 995         int found;
 996
 997         spin_lock_irq(&mapping->tree_lock);
 998         found = radix_tree_gang_lookup(&mapping->page_tree,
 999                                        (void **)&page, offset, 1);
1000         if (found > 0 && !radix_tree_exceptional_entry(page)) {
1001                 struct lu_dirpage *dp;
1002
1003                 get_page(page);
1004                 spin_unlock_irq(&mapping->tree_lock);
1005                 /*
1006                  * In contrast to find_lock_page() we are sure that directory
1007                  * page cannot be truncated (while DLM lock is held) and,
1008                  * hence, can avoid restart.
1009                  *
1010                  * In fact, page cannot be locked here at all, because
1011                  * mdc_read_page_remote does synchronous io.
1012                  */
1013                 wait_on_page_locked(page);
1014                 if (PageUptodate(page)) {
1015                         dp = kmap(page);
1016                         if (BITS_PER_LONG == 32 && hash64) {
1017                                 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1018                                 *end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
1019                                 *hash  = *hash >> 32;
1020                         } else {
1021                                 *start = le64_to_cpu(dp->ldp_hash_start);
1022                                 *end   = le64_to_cpu(dp->ldp_hash_end);
1023                         }
1024                         if (unlikely(*start == 1 && *hash == 0))
1025                                 *hash = *start;
1026                         else
1027                                 LASSERTF(*start <= *hash, "start = %#llx"
1028                                          ",end = %#llx,hash = %#llx\n",
1029                                          *start, *end, *hash);
1030                         CDEBUG(D_VFSTRACE, "offset %lx [%#llx %#llx],"
1031                               " hash %#llx\n", offset, *start, *end, *hash);
1032                         if (*hash > *end) {
1033                                 kunmap(page);
1034                                 mdc_release_page(page, 0);
1035                                 page = NULL;
1036                         } else if (*end != *start && *hash == *end) {
1037                                 /*
1038                                  * upon hash collision, remove this page,
1039                                  * otherwise put page reference, and
1040                                  * mdc_read_page_remote() will issue RPC to
1041                                  * fetch the page we want.
1042                                  */
1043                                 kunmap(page);
1044                                 mdc_release_page(page,
1045                                     le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
1046                                 page = NULL;
1047                         }
1048                 } else {
1049                         put_page(page);
1050                         page = ERR_PTR(-EIO);
1051                 }
1052         } else {
1053                 spin_unlock_irq(&mapping->tree_lock);
1054                 page = NULL;
1055         }
1056         return page;
1057 }
1058
1059 /*
1060  * Adjust a set of pages, each page containing an array of lu_dirpages,
1061  * so that each page can be used as a single logical lu_dirpage.
1062  *
1063  * A lu_dirpage is laid out as follows, where s = ldp_hash_start,
1064  * e = ldp_hash_end, f = ldp_flags, p = padding, and each "ent" is a
1065  * struct lu_dirent.  It has size up to LU_PAGE_SIZE. The ldp_hash_end
1066  * value is used as a cookie to request the next lu_dirpage in a
1067  * directory listing that spans multiple pages (two in this example):
1068  *   ________
1069  *  |        |
1070  * .|--------v-------   -----.
1071  * |s|e|f|p|ent|ent| ... |ent|
1072  * '--|--------------   -----'   Each PAGE contains a single
1073  *    '------.                   lu_dirpage.
1074  * .---------v-------   -----.
1075  * |s|e|f|p|ent| 0 | ... | 0 |
1076  * '-----------------   -----'
1077  *
1078  * However, on hosts where the native VM page size (PAGE_SIZE) is
1079  * larger than LU_PAGE_SIZE, a single host page may contain multiple
1080  * lu_dirpages. After reading the lu_dirpages from the MDS, the
1081  * ldp_hash_end of the first lu_dirpage refers to the one immediately
1082  * after it in the same PAGE (arrows simplified for brevity, but
1083  * in general e0==s1, e1==s2, etc.):
1084  *
1085  * .--------------------   -----.
1086  * |s0|e0|f0|p|ent|ent| ... |ent|
1087  * |---v----------------   -----|
1088  * |s1|e1|f1|p|ent|ent| ... |ent|
1089  * |---v----------------   -----|  Here, each PAGE contains
1090  *             ...                 multiple lu_dirpages.
1091  * |---v----------------   -----|
1092  * |s'|e'|f'|p|ent|ent| ... |ent|
1093  * '---|----------------   -----'
1094  *     v
1095  * .----------------------------.
1096  * |        next PAGE           |
1097  *
1098  * This structure is transformed into a single logical lu_dirpage as follows:
1099  *
1100  * - Replace e0 with e' so the request for the next lu_dirpage gets the page
1101  *   labeled 'next PAGE'.
1102  *
1103  * - Copy the LDF_COLLIDE flag from f' to f0 to correctly reflect whether
1104  *   a hash collision with the next page exists.
1105  *
1106  * - Adjust the lde_reclen of the ending entry of each lu_dirpage to span
1107  *   to the first entry of the next lu_dirpage.
1108  */
1109 #if PAGE_SIZE > LU_PAGE_SIZE
1110 static void mdc_adjust_dirpages(struct page **pages, int cfs_pgs, int lu_pgs)
1111 {
1112         int i;
1113
1114         for (i = 0; i < cfs_pgs; i++) {
1115                 struct lu_dirpage       *dp = kmap(pages[i]);
1116                 struct lu_dirpage       *first = dp;
1117                 struct lu_dirent        *end_dirent = NULL;
1118                 struct lu_dirent        *ent;
1119                 __u64           hash_end = le64_to_cpu(dp->ldp_hash_end);
1120                 __u32           flags = le32_to_cpu(dp->ldp_flags);
1121
1122                 while (--lu_pgs > 0) {
1123                         ent = lu_dirent_start(dp);
1124                         for (end_dirent = ent; ent != NULL;
1125                              end_dirent = ent, ent = lu_dirent_next(ent));
1126
1127                         /* Advance dp to next lu_dirpage. */
1128                         dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE);
1129
1130                         /* Check if we've reached the end of the PAGE. */
1131                         if (!((unsigned long)dp & ~PAGE_MASK))
1132                                 break;
1133
1134                         /* Save the hash and flags of this lu_dirpage. */
1135                         hash_end = le64_to_cpu(dp->ldp_hash_end);
1136                         flags = le32_to_cpu(dp->ldp_flags);
1137
1138                         /* Check if lu_dirpage contains no entries. */
1139                         if (end_dirent == NULL)
1140                                 break;
1141
1142                         /* Enlarge the end entry lde_reclen from 0 to
1143                          * first entry of next lu_dirpage. */
1144                         LASSERT(le16_to_cpu(end_dirent->lde_reclen) == 0);
1145                         end_dirent->lde_reclen =
1146                                 cpu_to_le16((char *)(dp->ldp_entries) -
1147                                             (char *)end_dirent);
1148                 }
1149
1150                 first->ldp_hash_end = hash_end;
1151                 first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE);
1152                 first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE);
1153
1154                 kunmap(pages[i]);
1155         }
1156         LASSERTF(lu_pgs == 0, "left = %d\n", lu_pgs);
1157 }
1158 #else
1159 #define mdc_adjust_dirpages(pages, cfs_pgs, lu_pgs) do {} while (0)
1160 #endif  /* PAGE_SIZE > LU_PAGE_SIZE */
1161
1162 /* parameters for readdir page */
1163 struct readpage_param {
1164         struct md_op_data       *rp_mod;
1165         __u64                   rp_off;
1166         int                     rp_hash64;
1167         struct obd_export       *rp_exp;
1168         struct md_callback      *rp_cb;
1169 };
1170
1171 #ifndef HAVE_DELETE_FROM_PAGE_CACHE
1172 static inline void delete_from_page_cache(struct page *page)
1173 {
1174         remove_from_page_cache(page);
1175         put_page(page);
1176 }
1177 #endif
1178
1179 /**
1180  * Read pages from server.
1181  *
1182  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
1183  * a header lu_dirpage which describes the start/end hash, and whether this
1184  * page is empty (contains no dir entry) or hash collide with next page.
1185  * After client receives reply, several pages will be integrated into dir page
1186  * in PAGE_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the
1187  * lu_dirpage for this integrated page will be adjusted.
1188  **/
1189 static int mdc_read_page_remote(void *data, struct page *page0)
1190 {
1191         struct readpage_param *rp = data;
1192         struct page **page_pool;
1193         struct page *page;
1194         struct lu_dirpage *dp;
1195         struct md_op_data *op_data = rp->rp_mod;
1196         struct ptlrpc_request *req;
1197         int max_pages;
1198         struct inode *inode;
1199         struct lu_fid *fid;
1200         int rd_pgs = 0; /* number of pages actually read */
1201         int npages;
1202         int i;
1203         int rc;
1204         ENTRY;
1205
1206         max_pages = rp->rp_exp->exp_obd->u.cli.cl_max_pages_per_rpc;
1207         inode = op_data->op_data;
1208         fid = &op_data->op_fid1;
1209         LASSERT(inode != NULL);
1210
1211         OBD_ALLOC(page_pool, sizeof(page_pool[0]) * max_pages);
1212         if (page_pool != NULL) {
1213                 page_pool[0] = page0;
1214         } else {
1215                 page_pool = &page0;
1216                 max_pages = 1;
1217         }
1218
1219         for (npages = 1; npages < max_pages; npages++) {
1220                 page = page_cache_alloc_cold(inode->i_mapping);
1221                 if (page == NULL)
1222                         break;
1223                 page_pool[npages] = page;
1224         }
1225
1226         rc = mdc_getpage(rp->rp_exp, fid, rp->rp_off, page_pool, npages, &req);
1227         if (rc < 0) {
1228                 /* page0 is special, which was added into page cache early */
1229                 delete_from_page_cache(page0);
1230         } else {
1231                 int lu_pgs;
1232
1233                 rd_pgs = (req->rq_bulk->bd_nob_transferred + PAGE_SIZE - 1) >>
1234                         PAGE_SHIFT;
1235                 lu_pgs = req->rq_bulk->bd_nob_transferred >> LU_PAGE_SHIFT;
1236                 LASSERT(!(req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK));
1237
1238                 CDEBUG(D_INODE, "read %d(%d) pages\n", rd_pgs, lu_pgs);
1239
1240                 mdc_adjust_dirpages(page_pool, rd_pgs, lu_pgs);
1241
1242                 SetPageUptodate(page0);
1243         }
1244         unlock_page(page0);
1245
1246         ptlrpc_req_finished(req);
1247         CDEBUG(D_CACHE, "read %d/%d pages\n", rd_pgs, npages);
1248         for (i = 1; i < npages; i++) {
1249                 unsigned long   offset;
1250                 __u64           hash;
1251                 int ret;
1252
1253                 page = page_pool[i];
1254
1255                 if (rc < 0 || i >= rd_pgs) {
1256                         put_page(page);
1257                         continue;
1258                 }
1259
1260                 SetPageUptodate(page);
1261
1262                 dp = kmap(page);
1263                 hash = le64_to_cpu(dp->ldp_hash_start);
1264                 kunmap(page);
1265
1266                 offset = hash_x_index(hash, rp->rp_hash64);
1267
1268                 prefetchw(&page->flags);
1269                 ret = add_to_page_cache_lru(page, inode->i_mapping, offset,
1270                                             GFP_KERNEL);
1271                 if (ret == 0)
1272                         unlock_page(page);
1273                 else
1274                         CDEBUG(D_VFSTRACE, "page %lu add to page cache failed:"
1275                                " rc = %d\n", offset, ret);
1276                 put_page(page);
1277         }
1278
1279         if (page_pool != &page0)
1280                 OBD_FREE(page_pool, sizeof(page_pool[0]) * max_pages);
1281
1282         RETURN(rc);
1283 }
1284
1285 /**
1286  * Read dir page from cache first, if it can not find it, read it from
1287  * server and add into the cache.
1288  *
1289  * \param[in] exp       MDC export
1290  * \param[in] op_data   client MD stack parameters, transfering parameters
1291  *                      between different layers on client MD stack.
1292  * \param[in] cb_op     callback required for ldlm lock enqueue during
1293  *                      read page
1294  * \param[in] hash_offset the hash offset of the page to be read
1295  * \param[in] ppage     the page to be read
1296  *
1297  * retval               = 0 get the page successfully
1298  *                      errno(<0) get the page failed
1299  */
1300 static int mdc_read_page(struct obd_export *exp, struct md_op_data *op_data,
1301                          struct md_callback *cb_op, __u64 hash_offset,
1302                          struct page **ppage)
1303 {
1304         struct lookup_intent    it = { .it_op = IT_READDIR };
1305         struct page             *page;
1306         struct inode            *dir = op_data->op_data;
1307         struct address_space    *mapping;
1308         struct lu_dirpage       *dp;
1309         __u64                   start = 0;
1310         __u64                   end = 0;
1311         struct lustre_handle    lockh;
1312         struct ptlrpc_request   *enq_req = NULL;
1313         struct readpage_param   rp_param;
1314         int rc;
1315
1316         ENTRY;
1317
1318         *ppage = NULL;
1319
1320         LASSERT(dir != NULL);
1321         mapping = dir->i_mapping;
1322
1323         rc = mdc_intent_lock(exp, op_data, &it, &enq_req,
1324                              cb_op->md_blocking_ast, 0);
1325         if (enq_req != NULL)
1326                 ptlrpc_req_finished(enq_req);
1327
1328         if (rc < 0) {
1329                 CERROR("%s: "DFID" lock enqueue fails: rc = %d\n",
1330                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1), rc);
1331                 RETURN(rc);
1332         }
1333
1334         rc = 0;
1335         lockh.cookie = it.it_lock_handle;
1336         mdc_set_lock_data(exp, &lockh, dir, NULL);
1337
1338         rp_param.rp_off = hash_offset;
1339         rp_param.rp_hash64 = op_data->op_cli_flags & CLI_HASH64;
1340         page = mdc_page_locate(mapping, &rp_param.rp_off, &start, &end,
1341                                rp_param.rp_hash64);
1342         if (IS_ERR(page)) {
1343                 CERROR("%s: dir page locate: "DFID" at %llu: rc %ld\n",
1344                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1345                        rp_param.rp_off, PTR_ERR(page));
1346                 GOTO(out_unlock, rc = PTR_ERR(page));
1347         } else if (page != NULL) {
1348                 /*
1349                  * XXX nikita: not entirely correct handling of a corner case:
1350                  * suppose hash chain of entries with hash value HASH crosses
1351                  * border between pages P0 and P1. First both P0 and P1 are
1352                  * cached, seekdir() is called for some entry from the P0 part
1353                  * of the chain. Later P0 goes out of cache. telldir(HASH)
1354                  * happens and finds P1, as it starts with matching hash
1355                  * value. Remaining entries from P0 part of the chain are
1356                  * skipped. (Is that really a bug?)
1357                  *
1358                  * Possible solutions: 0. don't cache P1 is such case, handle
1359                  * it as an "overflow" page. 1. invalidate all pages at
1360                  * once. 2. use HASH|1 as an index for P1.
1361                  */
1362                 GOTO(hash_collision, page);
1363         }
1364
1365         rp_param.rp_exp = exp;
1366         rp_param.rp_mod = op_data;
1367         page = read_cache_page(mapping,
1368                                hash_x_index(rp_param.rp_off,
1369                                             rp_param.rp_hash64),
1370                                mdc_read_page_remote, &rp_param);
1371         if (IS_ERR(page)) {
1372                 CDEBUG(D_INFO, "%s: read cache page: "DFID" at %llu: %ld\n",
1373                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1374                        rp_param.rp_off, PTR_ERR(page));
1375                 GOTO(out_unlock, rc = PTR_ERR(page));
1376         }
1377
1378         wait_on_page_locked(page);
1379         (void)kmap(page);
1380         if (!PageUptodate(page)) {
1381                 CERROR("%s: page not updated: "DFID" at %llu: rc %d\n",
1382                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1383                        rp_param.rp_off, -5);
1384                 goto fail;
1385         }
1386         if (!PageChecked(page))
1387                 SetPageChecked(page);
1388         if (PageError(page)) {
1389                 CERROR("%s: page error: "DFID" at %llu: rc %d\n",
1390                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1391                        rp_param.rp_off, -5);
1392                 goto fail;
1393         }
1394
1395 hash_collision:
1396         dp = page_address(page);
1397         if (BITS_PER_LONG == 32 && rp_param.rp_hash64) {
1398                 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1399                 end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
1400                 rp_param.rp_off = hash_offset >> 32;
1401         } else {
1402                 start = le64_to_cpu(dp->ldp_hash_start);
1403                 end   = le64_to_cpu(dp->ldp_hash_end);
1404                 rp_param.rp_off = hash_offset;
1405         }
1406         if (end == start) {
1407                 LASSERT(start == rp_param.rp_off);
1408                 CWARN("Page-wide hash collision: %#lx\n", (unsigned long)end);
1409 #if BITS_PER_LONG == 32
1410                 CWARN("Real page-wide hash collision at [%llu %llu] with "
1411                       "hash %llu\n", le64_to_cpu(dp->ldp_hash_start),
1412                       le64_to_cpu(dp->ldp_hash_end), hash_offset);
1413 #endif
1414
1415                 /*
1416                  * Fetch whole overflow chain...
1417                  *
1418                  * XXX not yet.
1419                  */
1420                 goto fail;
1421         }
1422         *ppage = page;
1423 out_unlock:
1424         ldlm_lock_decref(&lockh, it.it_lock_mode);
1425         return rc;
1426 fail:
1427         kunmap(page);
1428         mdc_release_page(page, 1);
1429         rc = -EIO;
1430         goto out_unlock;
1431 }
1432
1433
1434 static int mdc_statfs(const struct lu_env *env,
1435                       struct obd_export *exp, struct obd_statfs *osfs,
1436                       __u64 max_age, __u32 flags)
1437 {
1438         struct obd_device     *obd = class_exp2obd(exp);
1439         struct ptlrpc_request *req;
1440         struct obd_statfs     *msfs;
1441         struct obd_import     *imp = NULL;
1442         int                    rc;
1443         ENTRY;
1444
1445         /*
1446          * Since the request might also come from lprocfs, so we need
1447          * sync this with client_disconnect_export Bug15684
1448          */
1449         down_read(&obd->u.cli.cl_sem);
1450         if (obd->u.cli.cl_import)
1451                 imp = class_import_get(obd->u.cli.cl_import);
1452         up_read(&obd->u.cli.cl_sem);
1453         if (!imp)
1454                 RETURN(-ENODEV);
1455
1456         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_STATFS,
1457                                         LUSTRE_MDS_VERSION, MDS_STATFS);
1458         if (req == NULL)
1459                 GOTO(output, rc = -ENOMEM);
1460
1461         ptlrpc_request_set_replen(req);
1462
1463         if (flags & OBD_STATFS_NODELAY) {
1464                 /* procfs requests not want stay in wait for avoid deadlock */
1465                 req->rq_no_resend = 1;
1466                 req->rq_no_delay = 1;
1467         }
1468
1469         rc = ptlrpc_queue_wait(req);
1470         if (rc) {
1471                 /* check connection error first */
1472                 if (imp->imp_connect_error)
1473                         rc = imp->imp_connect_error;
1474                 GOTO(out, rc);
1475         }
1476
1477         msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1478         if (msfs == NULL)
1479                 GOTO(out, rc = -EPROTO);
1480
1481         *osfs = *msfs;
1482         EXIT;
1483 out:
1484         ptlrpc_req_finished(req);
1485 output:
1486         class_import_put(imp);
1487         return rc;
1488 }
1489
1490 static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
1491 {
1492         __u32 keylen, vallen;
1493         void *key;
1494         int rc;
1495
1496         if (gf->gf_pathlen > PATH_MAX)
1497                 RETURN(-ENAMETOOLONG);
1498         if (gf->gf_pathlen < 2)
1499                 RETURN(-EOVERFLOW);
1500
1501         /* Key is KEY_FID2PATH + getinfo_fid2path description */
1502         keylen = cfs_size_round(sizeof(KEY_FID2PATH) + sizeof(*gf) +
1503                                 sizeof(struct lu_fid));
1504         OBD_ALLOC(key, keylen);
1505         if (key == NULL)
1506                 RETURN(-ENOMEM);
1507         memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
1508         memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf));
1509         memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf),
1510                gf->gf_u.gf_root_fid, sizeof(struct lu_fid));
1511         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
1512                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
1513
1514         if (!fid_is_sane(&gf->gf_fid))
1515                 GOTO(out, rc = -EINVAL);
1516
1517         /* Val is struct getinfo_fid2path result plus path */
1518         vallen = sizeof(*gf) + gf->gf_pathlen;
1519
1520         rc = obd_get_info(NULL, exp, keylen, key, &vallen, gf);
1521         if (rc != 0 && rc != -EREMOTE)
1522                 GOTO(out, rc);
1523
1524         if (vallen <= sizeof(*gf))
1525                 GOTO(out, rc = -EPROTO);
1526         if (vallen > sizeof(*gf) + gf->gf_pathlen)
1527                 GOTO(out, rc = -EOVERFLOW);
1528
1529         CDEBUG(D_IOCTL, "path got "DFID" from %llu #%d: %s\n",
1530                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno,
1531                gf->gf_pathlen < 512 ? gf->gf_u.gf_path :
1532                /* only log the last 512 characters of the path */
1533                gf->gf_u.gf_path + gf->gf_pathlen - 512);
1534
1535 out:
1536         OBD_FREE(key, keylen);
1537         return rc;
1538 }
1539
1540 static int mdc_ioc_hsm_progress(struct obd_export *exp,
1541                                 struct hsm_progress_kernel *hpk)
1542 {
1543         struct obd_import               *imp = class_exp2cliimp(exp);
1544         struct hsm_progress_kernel      *req_hpk;
1545         struct ptlrpc_request           *req;
1546         int                              rc;
1547         ENTRY;
1548
1549         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_PROGRESS,
1550                                         LUSTRE_MDS_VERSION, MDS_HSM_PROGRESS);
1551         if (req == NULL)
1552                 GOTO(out, rc = -ENOMEM);
1553
1554         mdc_pack_body(req, NULL, 0, 0, -1, 0);
1555
1556         /* Copy hsm_progress struct */
1557         req_hpk = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_PROGRESS);
1558         if (req_hpk == NULL)
1559                 GOTO(out, rc = -EPROTO);
1560
1561         *req_hpk = *hpk;
1562         req_hpk->hpk_errval = lustre_errno_hton(hpk->hpk_errval);
1563
1564         ptlrpc_request_set_replen(req);
1565
1566         mdc_get_mod_rpc_slot(req, NULL);
1567         rc = ptlrpc_queue_wait(req);
1568         mdc_put_mod_rpc_slot(req, NULL);
1569
1570         GOTO(out, rc);
1571 out:
1572         ptlrpc_req_finished(req);
1573         return rc;
1574 }
1575
1576 static int mdc_ioc_hsm_ct_register(struct obd_import *imp, __u32 archives)
1577 {
1578         __u32                   *archive_mask;
1579         struct ptlrpc_request   *req;
1580         int                      rc;
1581         ENTRY;
1582
1583         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_REGISTER,
1584                                         LUSTRE_MDS_VERSION,
1585                                         MDS_HSM_CT_REGISTER);
1586         if (req == NULL)
1587                 GOTO(out, rc = -ENOMEM);
1588
1589         mdc_pack_body(req, NULL, 0, 0, -1, 0);
1590
1591         /* Copy hsm_progress struct */
1592         archive_mask = req_capsule_client_get(&req->rq_pill,
1593                                               &RMF_MDS_HSM_ARCHIVE);
1594         if (archive_mask == NULL)
1595                 GOTO(out, rc = -EPROTO);
1596
1597         *archive_mask = archives;
1598
1599         ptlrpc_request_set_replen(req);
1600
1601         rc = mdc_queue_wait(req);
1602         GOTO(out, rc);
1603 out:
1604         ptlrpc_req_finished(req);
1605         return rc;
1606 }
1607
1608 static int mdc_ioc_hsm_current_action(struct obd_export *exp,
1609                                       struct md_op_data *op_data)
1610 {
1611         struct hsm_current_action       *hca = op_data->op_data;
1612         struct hsm_current_action       *req_hca;
1613         struct ptlrpc_request           *req;
1614         int                              rc;
1615         ENTRY;
1616
1617         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1618                                    &RQF_MDS_HSM_ACTION);
1619         if (req == NULL)
1620                 RETURN(-ENOMEM);
1621
1622         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_ACTION);
1623         if (rc) {
1624                 ptlrpc_request_free(req);
1625                 RETURN(rc);
1626         }
1627
1628         mdc_pack_body(req, &op_data->op_fid1, 0, 0,
1629                       op_data->op_suppgids[0], 0);
1630
1631         ptlrpc_request_set_replen(req);
1632
1633         rc = mdc_queue_wait(req);
1634         if (rc)
1635                 GOTO(out, rc);
1636
1637         req_hca = req_capsule_server_get(&req->rq_pill,
1638                                          &RMF_MDS_HSM_CURRENT_ACTION);
1639         if (req_hca == NULL)
1640                 GOTO(out, rc = -EPROTO);
1641
1642         *hca = *req_hca;
1643
1644         EXIT;
1645 out:
1646         ptlrpc_req_finished(req);
1647         return rc;
1648 }
1649
1650 static int mdc_ioc_hsm_ct_unregister(struct obd_import *imp)
1651 {
1652         struct ptlrpc_request   *req;
1653         int                      rc;
1654         ENTRY;
1655
1656         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_UNREGISTER,
1657                                         LUSTRE_MDS_VERSION,
1658                                         MDS_HSM_CT_UNREGISTER);
1659         if (req == NULL)
1660                 GOTO(out, rc = -ENOMEM);
1661
1662         mdc_pack_body(req, NULL, 0, 0, -1, 0);
1663
1664         ptlrpc_request_set_replen(req);
1665
1666         rc = mdc_queue_wait(req);
1667         GOTO(out, rc);
1668 out:
1669         ptlrpc_req_finished(req);
1670         return rc;
1671 }
1672
1673 static int mdc_ioc_hsm_state_get(struct obd_export *exp,
1674                                  struct md_op_data *op_data)
1675 {
1676         struct hsm_user_state   *hus = op_data->op_data;
1677         struct hsm_user_state   *req_hus;
1678         struct ptlrpc_request   *req;
1679         int                      rc;
1680         ENTRY;
1681
1682         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1683                                    &RQF_MDS_HSM_STATE_GET);
1684         if (req == NULL)
1685                 RETURN(-ENOMEM);
1686
1687         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_GET);
1688         if (rc != 0) {
1689                 ptlrpc_request_free(req);
1690                 RETURN(rc);
1691         }
1692
1693         mdc_pack_body(req, &op_data->op_fid1, 0, 0,
1694                       op_data->op_suppgids[0], 0);
1695
1696         ptlrpc_request_set_replen(req);
1697
1698         rc = mdc_queue_wait(req);
1699         if (rc)
1700                 GOTO(out, rc);
1701
1702         req_hus = req_capsule_server_get(&req->rq_pill, &RMF_HSM_USER_STATE);
1703         if (req_hus == NULL)
1704                 GOTO(out, rc = -EPROTO);
1705
1706         *hus = *req_hus;
1707
1708         EXIT;
1709 out:
1710         ptlrpc_req_finished(req);
1711         return rc;
1712 }
1713
1714 static int mdc_ioc_hsm_state_set(struct obd_export *exp,
1715                                  struct md_op_data *op_data)
1716 {
1717         struct hsm_state_set    *hss = op_data->op_data;
1718         struct hsm_state_set    *req_hss;
1719         struct ptlrpc_request   *req;
1720         int                      rc;
1721         ENTRY;
1722
1723         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1724                                    &RQF_MDS_HSM_STATE_SET);
1725         if (req == NULL)
1726                 RETURN(-ENOMEM);
1727
1728         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_SET);
1729         if (rc) {
1730                 ptlrpc_request_free(req);
1731                 RETURN(rc);
1732         }
1733
1734         mdc_pack_body(req, &op_data->op_fid1, 0, 0,
1735                       op_data->op_suppgids[0], 0);
1736
1737         /* Copy states */
1738         req_hss = req_capsule_client_get(&req->rq_pill, &RMF_HSM_STATE_SET);
1739         if (req_hss == NULL)
1740                 GOTO(out, rc = -EPROTO);
1741         *req_hss = *hss;
1742
1743         ptlrpc_request_set_replen(req);
1744
1745         mdc_get_mod_rpc_slot(req, NULL);
1746         rc = ptlrpc_queue_wait(req);
1747         mdc_put_mod_rpc_slot(req, NULL);
1748
1749         GOTO(out, rc);
1750 out:
1751         ptlrpc_req_finished(req);
1752         return rc;
1753 }
1754
1755 static int mdc_ioc_hsm_request(struct obd_export *exp,
1756                                struct hsm_user_request *hur)
1757 {
1758         struct obd_import       *imp = class_exp2cliimp(exp);
1759         struct ptlrpc_request   *req;
1760         struct hsm_request      *req_hr;
1761         struct hsm_user_item    *req_hui;
1762         char                    *req_opaque;
1763         int                      rc;
1764         ENTRY;
1765
1766         req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_REQUEST);
1767         if (req == NULL)
1768                 GOTO(out, rc = -ENOMEM);
1769
1770         req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM, RCL_CLIENT,
1771                              hur->hur_request.hr_itemcount
1772                              * sizeof(struct hsm_user_item));
1773         req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA, RCL_CLIENT,
1774                              hur->hur_request.hr_data_len);
1775
1776         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_REQUEST);
1777         if (rc) {
1778                 ptlrpc_request_free(req);
1779                 RETURN(rc);
1780         }
1781
1782         mdc_pack_body(req, NULL, 0, 0, -1, 0);
1783
1784         /* Copy hsm_request struct */
1785         req_hr = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_REQUEST);
1786         if (req_hr == NULL)
1787                 GOTO(out, rc = -EPROTO);
1788         *req_hr = hur->hur_request;
1789
1790         /* Copy hsm_user_item structs */
1791         req_hui = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM);
1792         if (req_hui == NULL)
1793                 GOTO(out, rc = -EPROTO);
1794         memcpy(req_hui, hur->hur_user_item,
1795                hur->hur_request.hr_itemcount * sizeof(struct hsm_user_item));
1796
1797         /* Copy opaque field */
1798         req_opaque = req_capsule_client_get(&req->rq_pill, &RMF_GENERIC_DATA);
1799         if (req_opaque == NULL)
1800                 GOTO(out, rc = -EPROTO);
1801         memcpy(req_opaque, hur_data(hur), hur->hur_request.hr_data_len);
1802
1803         ptlrpc_request_set_replen(req);
1804
1805         mdc_get_mod_rpc_slot(req, NULL);
1806         rc = ptlrpc_queue_wait(req);
1807         mdc_put_mod_rpc_slot(req, NULL);
1808
1809         GOTO(out, rc);
1810
1811 out:
1812         ptlrpc_req_finished(req);
1813         return rc;
1814 }
1815
1816 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
1817                                 struct lustre_kernelcomm *lk);
1818
1819 static int mdc_quotactl(struct obd_device *unused, struct obd_export *exp,
1820                         struct obd_quotactl *oqctl)
1821 {
1822         struct ptlrpc_request   *req;
1823         struct obd_quotactl     *oqc;
1824         int                      rc;
1825         ENTRY;
1826
1827         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
1828                                         &RQF_MDS_QUOTACTL, LUSTRE_MDS_VERSION,
1829                                         MDS_QUOTACTL);
1830         if (req == NULL)
1831                 RETURN(-ENOMEM);
1832
1833         oqc = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1834         *oqc = *oqctl;
1835
1836         ptlrpc_request_set_replen(req);
1837         ptlrpc_at_set_req_timeout(req);
1838         req->rq_no_resend = 1;
1839
1840         rc = ptlrpc_queue_wait(req);
1841         if (rc)
1842                 CERROR("ptlrpc_queue_wait failed, rc: %d\n", rc);
1843
1844         if (req->rq_repmsg &&
1845             (oqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL))) {
1846                 *oqctl = *oqc;
1847         } else if (!rc) {
1848                 CERROR ("Can't unpack obd_quotactl\n");
1849                 rc = -EPROTO;
1850         }
1851         ptlrpc_req_finished(req);
1852
1853         RETURN(rc);
1854 }
1855
1856 static int mdc_ioc_swap_layouts(struct obd_export *exp,
1857                                 struct md_op_data *op_data)
1858 {
1859         struct list_head cancels = LIST_HEAD_INIT(cancels);
1860         struct ptlrpc_request   *req;
1861         int                      rc, count;
1862         struct mdc_swap_layouts *msl, *payload;
1863         ENTRY;
1864
1865         msl = op_data->op_data;
1866
1867         /* When the MDT will get the MDS_SWAP_LAYOUTS RPC the
1868          * first thing it will do is to cancel the 2 layout
1869          * locks held by this client.
1870          * So the client must cancel its layout locks on the 2 fids
1871          * with the request RPC to avoid extra RPC round trips.
1872          */
1873         count = mdc_resource_get_unused(exp, &op_data->op_fid1, &cancels,
1874                                         LCK_EX, MDS_INODELOCK_LAYOUT |
1875                                         MDS_INODELOCK_XATTR);
1876         count += mdc_resource_get_unused(exp, &op_data->op_fid2, &cancels,
1877                                          LCK_EX, MDS_INODELOCK_LAYOUT |
1878                                          MDS_INODELOCK_XATTR);
1879
1880         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1881                                    &RQF_MDS_SWAP_LAYOUTS);
1882         if (req == NULL) {
1883                 ldlm_lock_list_put(&cancels, l_bl_ast, count);
1884                 RETURN(-ENOMEM);
1885         }
1886
1887         rc = mdc_prep_elc_req(exp, req, MDS_SWAP_LAYOUTS, &cancels, count);
1888         if (rc) {
1889                 ptlrpc_request_free(req);
1890                 RETURN(rc);
1891         }
1892
1893         mdc_swap_layouts_pack(req, op_data);
1894
1895         payload = req_capsule_client_get(&req->rq_pill, &RMF_SWAP_LAYOUTS);
1896         LASSERT(payload);
1897
1898         *payload = *msl;
1899
1900         ptlrpc_request_set_replen(req);
1901
1902         rc = ptlrpc_queue_wait(req);
1903         if (rc)
1904                 GOTO(out, rc);
1905         EXIT;
1906
1907 out:
1908         ptlrpc_req_finished(req);
1909         return rc;
1910 }
1911
1912 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1913                          void *karg, void __user *uarg)
1914 {
1915         struct obd_device *obd = exp->exp_obd;
1916         struct obd_ioctl_data *data = karg;
1917         struct obd_import *imp = obd->u.cli.cl_import;
1918         int rc;
1919         ENTRY;
1920
1921         if (!try_module_get(THIS_MODULE)) {
1922                 CERROR("%s: cannot get module '%s'\n", obd->obd_name,
1923                        module_name(THIS_MODULE));
1924                 return -EINVAL;
1925         }
1926         switch (cmd) {
1927         case OBD_IOC_FID2PATH:
1928                 rc = mdc_ioc_fid2path(exp, karg);
1929                 GOTO(out, rc);
1930         case LL_IOC_HSM_CT_START:
1931                 rc = mdc_ioc_hsm_ct_start(exp, karg);
1932                 /* ignore if it was already registered on this MDS. */
1933                 if (rc == -EEXIST)
1934                         rc = 0;
1935                 GOTO(out, rc);
1936         case LL_IOC_HSM_PROGRESS:
1937                 rc = mdc_ioc_hsm_progress(exp, karg);
1938                 GOTO(out, rc);
1939         case LL_IOC_HSM_STATE_GET:
1940                 rc = mdc_ioc_hsm_state_get(exp, karg);
1941                 GOTO(out, rc);
1942         case LL_IOC_HSM_STATE_SET:
1943                 rc = mdc_ioc_hsm_state_set(exp, karg);
1944                 GOTO(out, rc);
1945         case LL_IOC_HSM_ACTION:
1946                 rc = mdc_ioc_hsm_current_action(exp, karg);
1947                 GOTO(out, rc);
1948         case LL_IOC_HSM_REQUEST:
1949                 rc = mdc_ioc_hsm_request(exp, karg);
1950                 GOTO(out, rc);
1951         case OBD_IOC_CLIENT_RECOVER:
1952                 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1, 0);
1953                 if (rc < 0)
1954                         GOTO(out, rc);
1955                 GOTO(out, rc = 0);
1956         case IOC_OSC_SET_ACTIVE:
1957                 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
1958                 GOTO(out, rc);
1959         case OBD_IOC_PING_TARGET:
1960                 rc = ptlrpc_obd_ping(obd);
1961                 GOTO(out, rc);
1962         /*
1963          * Normally IOC_OBD_STATFS, OBD_IOC_QUOTACTL iocontrol are handled by
1964          * LMV instead of MDC. But when the cluster is upgraded from 1.8,
1965          * there'd be no LMV layer thus we might be called here. Eventually
1966          * this code should be removed.
1967          * bz20731, LU-592.
1968          */
1969         case IOC_OBD_STATFS: {
1970                 struct obd_statfs stat_buf = {0};
1971
1972                 if (*((__u32 *) data->ioc_inlbuf2) != 0)
1973                         GOTO(out, rc = -ENODEV);
1974
1975                 /* copy UUID */
1976                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd),
1977                                  min((int)data->ioc_plen2,
1978                                      (int)sizeof(struct obd_uuid))))
1979                         GOTO(out, rc = -EFAULT);
1980
1981                 rc = mdc_statfs(NULL, obd->obd_self_export, &stat_buf,
1982                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1983                                 0);
1984                 if (rc != 0)
1985                         GOTO(out, rc);
1986
1987                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1988                                      min((int) data->ioc_plen1,
1989                                          (int) sizeof(stat_buf))))
1990                         GOTO(out, rc = -EFAULT);
1991
1992                 GOTO(out, rc = 0);
1993         }
1994         case OBD_IOC_QUOTACTL: {
1995                 struct if_quotactl *qctl = karg;
1996                 struct obd_quotactl *oqctl;
1997
1998                 OBD_ALLOC_PTR(oqctl);
1999                 if (oqctl == NULL)
2000                         GOTO(out, rc = -ENOMEM);
2001
2002                 QCTL_COPY(oqctl, qctl);
2003                 rc = obd_quotactl(exp, oqctl);
2004                 if (rc == 0) {
2005                         QCTL_COPY(qctl, oqctl);
2006                         qctl->qc_valid = QC_MDTIDX;
2007                         qctl->obd_uuid = obd->u.cli.cl_target_uuid;
2008                 }
2009
2010                 OBD_FREE_PTR(oqctl);
2011                 GOTO(out, rc);
2012         }
2013         case LL_IOC_GET_CONNECT_FLAGS:
2014                 if (copy_to_user(uarg, exp_connect_flags_ptr(exp),
2015                                  sizeof(*exp_connect_flags_ptr(exp))))
2016                         GOTO(out, rc = -EFAULT);
2017
2018                 GOTO(out, rc = 0);
2019         case LL_IOC_LOV_SWAP_LAYOUTS:
2020                 rc = mdc_ioc_swap_layouts(exp, karg);
2021                 GOTO(out, rc);
2022         default:
2023                 CERROR("unrecognised ioctl: cmd = %#x\n", cmd);
2024                 GOTO(out, rc = -ENOTTY);
2025         }
2026 out:
2027         module_put(THIS_MODULE);
2028
2029         return rc;
2030 }
2031
2032 static int mdc_get_info_rpc(struct obd_export *exp,
2033                             u32 keylen, void *key,
2034                             u32 vallen, void *val)
2035 {
2036         struct obd_import      *imp = class_exp2cliimp(exp);
2037         struct ptlrpc_request  *req;
2038         char                   *tmp;
2039         int                     rc = -EINVAL;
2040         ENTRY;
2041
2042         req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
2043         if (req == NULL)
2044                 RETURN(-ENOMEM);
2045
2046         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
2047                              RCL_CLIENT, keylen);
2048         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
2049                              RCL_CLIENT, sizeof(vallen));
2050
2051         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
2052         if (rc) {
2053                 ptlrpc_request_free(req);
2054                 RETURN(rc);
2055         }
2056
2057         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
2058         memcpy(tmp, key, keylen);
2059         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
2060         memcpy(tmp, &vallen, sizeof(vallen));
2061
2062         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
2063                              RCL_SERVER, vallen);
2064         ptlrpc_request_set_replen(req);
2065
2066         rc = ptlrpc_queue_wait(req);
2067         /* -EREMOTE means the get_info result is partial, and it needs to
2068          * continue on another MDT, see fid2path part in lmv_iocontrol */
2069         if (rc == 0 || rc == -EREMOTE) {
2070                 tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
2071                 memcpy(val, tmp, vallen);
2072                 if (ptlrpc_rep_need_swab(req)) {
2073                         if (KEY_IS(KEY_FID2PATH))
2074                                 lustre_swab_fid2path(val);
2075                 }
2076         }
2077         ptlrpc_req_finished(req);
2078
2079         RETURN(rc);
2080 }
2081
2082 static void lustre_swab_hai(struct hsm_action_item *h)
2083 {
2084         __swab32s(&h->hai_len);
2085         __swab32s(&h->hai_action);
2086         lustre_swab_lu_fid(&h->hai_fid);
2087         lustre_swab_lu_fid(&h->hai_dfid);
2088         __swab64s(&h->hai_cookie);
2089         __swab64s(&h->hai_extent.offset);
2090         __swab64s(&h->hai_extent.length);
2091         __swab64s(&h->hai_gid);
2092 }
2093
2094 static void lustre_swab_hal(struct hsm_action_list *h)
2095 {
2096         struct hsm_action_item  *hai;
2097         __u32                    i;
2098
2099         __swab32s(&h->hal_version);
2100         __swab32s(&h->hal_count);
2101         __swab32s(&h->hal_archive_id);
2102         __swab64s(&h->hal_flags);
2103         hai = hai_first(h);
2104         for (i = 0; i < h->hal_count; i++, hai = hai_next(hai))
2105                 lustre_swab_hai(hai);
2106 }
2107
2108 static void lustre_swab_kuch(struct kuc_hdr *l)
2109 {
2110         __swab16s(&l->kuc_magic);
2111         /* __u8 l->kuc_transport */
2112         __swab16s(&l->kuc_msgtype);
2113         __swab16s(&l->kuc_msglen);
2114 }
2115
2116 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2117                                 struct lustre_kernelcomm *lk)
2118 {
2119         struct obd_import  *imp = class_exp2cliimp(exp);
2120         __u32               archive = lk->lk_data;
2121         int                 rc = 0;
2122
2123         if (lk->lk_group != KUC_GRP_HSM) {
2124                 CERROR("Bad copytool group %d\n", lk->lk_group);
2125                 return -EINVAL;
2126         }
2127
2128         CDEBUG(D_HSM, "CT start r%d w%d u%d g%d f%#x\n", lk->lk_rfd, lk->lk_wfd,
2129                lk->lk_uid, lk->lk_group, lk->lk_flags);
2130
2131         if (lk->lk_flags & LK_FLG_STOP) {
2132                 /* Unregister with the coordinator */
2133                 rc = mdc_ioc_hsm_ct_unregister(imp);
2134         } else {
2135                 rc = mdc_ioc_hsm_ct_register(imp, archive);
2136         }
2137
2138         return rc;
2139 }
2140
2141 /**
2142  * Send a message to any listening copytools
2143  * @param val KUC message (kuc_hdr + hsm_action_list)
2144  * @param len total length of message
2145  */
2146 static int mdc_hsm_copytool_send(size_t len, void *val)
2147 {
2148         struct kuc_hdr          *lh = (struct kuc_hdr *)val;
2149         struct hsm_action_list  *hal = (struct hsm_action_list *)(lh + 1);
2150         int                      rc;
2151         ENTRY;
2152
2153         if (len < sizeof(*lh) + sizeof(*hal)) {
2154                 CERROR("Short HSM message %zu < %zu\n", len,
2155                        sizeof(*lh) + sizeof(*hal));
2156                 RETURN(-EPROTO);
2157         }
2158         if (lh->kuc_magic == __swab16(KUC_MAGIC)) {
2159                 lustre_swab_kuch(lh);
2160                 lustre_swab_hal(hal);
2161         } else if (lh->kuc_magic != KUC_MAGIC) {
2162                 CERROR("Bad magic %x!=%x\n", lh->kuc_magic, KUC_MAGIC);
2163                 RETURN(-EPROTO);
2164         }
2165
2166         CDEBUG(D_HSM, " Received message mg=%x t=%d m=%d l=%d actions=%d "
2167                "on %s\n",
2168                lh->kuc_magic, lh->kuc_transport, lh->kuc_msgtype,
2169                lh->kuc_msglen, hal->hal_count, hal->hal_fsname);
2170
2171         /* Broadcast to HSM listeners */
2172         rc = libcfs_kkuc_group_put(KUC_GRP_HSM, lh);
2173
2174         RETURN(rc);
2175 }
2176
2177 /**
2178  * callback function passed to kuc for re-registering each HSM copytool
2179  * running on MDC, after MDT shutdown/recovery.
2180  * @param data copytool registration data
2181  * @param cb_arg callback argument (obd_import)
2182  */
2183 static int mdc_hsm_ct_reregister(void *data, void *cb_arg)
2184 {
2185         struct kkuc_ct_data     *kcd = data;
2186         struct obd_import       *imp = (struct obd_import *)cb_arg;
2187         int                      rc;
2188
2189         if (kcd == NULL || kcd->kcd_magic != KKUC_CT_DATA_MAGIC)
2190                 return -EPROTO;
2191
2192         if (!obd_uuid_equals(&kcd->kcd_uuid, &imp->imp_obd->obd_uuid))
2193                 return 0;
2194
2195         CDEBUG(D_HA, "%s: recover copytool registration to MDT (archive=%#x)\n",
2196                imp->imp_obd->obd_name, kcd->kcd_archive);
2197         rc = mdc_ioc_hsm_ct_register(imp, kcd->kcd_archive);
2198
2199         /* ignore error if the copytool is already registered */
2200         return (rc == -EEXIST) ? 0 : rc;
2201 }
2202
2203 /**
2204  * Re-establish all kuc contexts with MDT
2205  * after MDT shutdown/recovery.
2206  */
2207 static int mdc_kuc_reregister(struct obd_import *imp)
2208 {
2209         /* re-register HSM agents */
2210         return libcfs_kkuc_group_foreach(KUC_GRP_HSM, mdc_hsm_ct_reregister,
2211                                          (void *)imp);
2212 }
2213
2214 static int mdc_set_info_async(const struct lu_env *env,
2215                               struct obd_export *exp,
2216                               u32 keylen, void *key,
2217                               u32 vallen, void *val,
2218                               struct ptlrpc_request_set *set)
2219 {
2220         struct obd_import       *imp = class_exp2cliimp(exp);
2221         int                      rc;
2222         ENTRY;
2223
2224         if (KEY_IS(KEY_READ_ONLY)) {
2225                 if (vallen != sizeof(int))
2226                         RETURN(-EINVAL);
2227
2228                 spin_lock(&imp->imp_lock);
2229                 if (*((int *)val)) {
2230                         imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
2231                         imp->imp_connect_data.ocd_connect_flags |=
2232                                                         OBD_CONNECT_RDONLY;
2233                 } else {
2234                         imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
2235                         imp->imp_connect_data.ocd_connect_flags &=
2236                                                         ~OBD_CONNECT_RDONLY;
2237                 }
2238                 spin_unlock(&imp->imp_lock);
2239
2240                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2241                                        keylen, key, vallen, val, set);
2242                 RETURN(rc);
2243         }
2244         if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2245                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2246                                        keylen, key, vallen, val, set);
2247                 RETURN(rc);
2248         }
2249         if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) {
2250                 rc = mdc_hsm_copytool_send(vallen, val);
2251                 RETURN(rc);
2252         }
2253
2254         if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2255                 __u32 *default_easize = val;
2256
2257                 exp->exp_obd->u.cli.cl_default_mds_easize = *default_easize;
2258                 RETURN(0);
2259         }
2260
2261         rc = osc_set_info_async(env, exp, keylen, key, vallen, val, set);
2262         RETURN(rc);
2263 }
2264
2265 static int mdc_get_info(const struct lu_env *env, struct obd_export *exp,
2266                         __u32 keylen, void *key, __u32 *vallen, void *val)
2267 {
2268         int rc = -EINVAL;
2269
2270         if (KEY_IS(KEY_MAX_EASIZE)) {
2271                 __u32 mdsize, *max_easize;
2272
2273                 if (*vallen != sizeof(int))
2274                         RETURN(-EINVAL);
2275                 mdsize = *(__u32 *)val;
2276                 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
2277                         exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
2278                 max_easize = val;
2279                 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
2280                 RETURN(0);
2281         } else if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2282                 __u32 *default_easize;
2283
2284                 if (*vallen != sizeof(int))
2285                         RETURN(-EINVAL);
2286                 default_easize = val;
2287                 *default_easize = exp->exp_obd->u.cli.cl_default_mds_easize;
2288                 RETURN(0);
2289         } else if (KEY_IS(KEY_CONN_DATA)) {
2290                 struct obd_import *imp = class_exp2cliimp(exp);
2291                 struct obd_connect_data *data = val;
2292
2293                 if (*vallen != sizeof(*data))
2294                         RETURN(-EINVAL);
2295
2296                 *data = imp->imp_connect_data;
2297                 RETURN(0);
2298         } else if (KEY_IS(KEY_TGT_COUNT)) {
2299                 *((__u32 *)val) = 1;
2300                 RETURN(0);
2301         }
2302
2303         rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
2304
2305         RETURN(rc);
2306 }
2307
2308 static int mdc_fsync(struct obd_export *exp, const struct lu_fid *fid,
2309                      struct ptlrpc_request **request)
2310 {
2311         struct ptlrpc_request *req;
2312         int                    rc;
2313         ENTRY;
2314
2315         *request = NULL;
2316         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
2317         if (req == NULL)
2318                 RETURN(-ENOMEM);
2319
2320         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
2321         if (rc) {
2322                 ptlrpc_request_free(req);
2323                 RETURN(rc);
2324         }
2325
2326         mdc_pack_body(req, fid, 0, 0, -1, 0);
2327
2328         ptlrpc_request_set_replen(req);
2329
2330         rc = ptlrpc_queue_wait(req);
2331         if (rc)
2332                 ptlrpc_req_finished(req);
2333         else
2334                 *request = req;
2335         RETURN(rc);
2336 }
2337
2338 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
2339                             enum obd_import_event event)
2340 {
2341         struct client_obd *cli = &obd->u.cli;
2342         int rc = 0;
2343
2344         LASSERT(imp->imp_obd == obd);
2345
2346         switch (event) {
2347         case IMP_EVENT_DISCON:
2348                 spin_lock(&cli->cl_loi_list_lock);
2349                 cli->cl_avail_grant = 0;
2350                 cli->cl_lost_grant = 0;
2351                 spin_unlock(&cli->cl_loi_list_lock);
2352                 break;
2353         case IMP_EVENT_INACTIVE:
2354                 /*
2355                  * Flush current sequence to make client obtain new one
2356                  * from server in case of disconnect/reconnect.
2357                  */
2358                 down_read(&cli->cl_seq_rwsem);
2359                 if (cli->cl_seq)
2360                         seq_client_flush(cli->cl_seq);
2361                 up_read(&cli->cl_seq_rwsem);
2362
2363                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE);
2364                 break;
2365         case IMP_EVENT_INVALIDATE: {
2366                 struct ldlm_namespace *ns = obd->obd_namespace;
2367                 struct lu_env *env;
2368                 __u16 refcheck;
2369
2370                 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2371
2372                 env = cl_env_get(&refcheck);
2373                 if (!IS_ERR(env)) {
2374                         /* Reset grants. All pages go to failing rpcs due to
2375                          * the invalid import.
2376                          */
2377                         osc_io_unplug(env, cli, NULL);
2378
2379                         cfs_hash_for_each_nolock(ns->ns_rs_hash,
2380                                                  osc_ldlm_resource_invalidate,
2381                                                  env, 0);
2382                         cl_env_put(env, &refcheck);
2383                         ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2384                 } else {
2385                         rc = PTR_ERR(env);
2386                 }
2387                 break;
2388         }
2389         case IMP_EVENT_ACTIVE:
2390                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE);
2391                 /* redo the kuc registration after reconnecting */
2392                 if (rc == 0)
2393                         rc = mdc_kuc_reregister(imp);
2394                 break;
2395         case IMP_EVENT_OCD: {
2396                 struct obd_connect_data *ocd = &imp->imp_connect_data;
2397
2398                 if (OCD_HAS_FLAG(ocd, GRANT))
2399                         osc_init_grant(cli, ocd);
2400
2401                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD);
2402                 break;
2403         }
2404         case IMP_EVENT_DEACTIVATE:
2405         case IMP_EVENT_ACTIVATE:
2406                 break;
2407         default:
2408                 CERROR("Unknown import event %x\n", event);
2409                 LBUG();
2410         }
2411         RETURN(rc);
2412 }
2413
2414 int mdc_fid_alloc(const struct lu_env *env, struct obd_export *exp,
2415                   struct lu_fid *fid, struct md_op_data *op_data)
2416 {
2417         struct client_obd *cli = &exp->exp_obd->u.cli;
2418         int rc = -EIO;
2419
2420         ENTRY;
2421
2422         down_read(&cli->cl_seq_rwsem);
2423         if (cli->cl_seq)
2424                 rc = seq_client_alloc_fid(env, cli->cl_seq, fid);
2425         up_read(&cli->cl_seq_rwsem);
2426
2427         RETURN(rc);
2428 }
2429
2430 static struct obd_uuid *mdc_get_uuid(struct obd_export *exp)
2431 {
2432         struct client_obd *cli = &exp->exp_obd->u.cli;
2433         return &cli->cl_target_uuid;
2434 }
2435
2436 /**
2437  * Determine whether the lock can be canceled before replaying it during
2438  * recovery, non zero value will be return if the lock can be canceled,
2439  * or zero returned for not
2440  */
2441 static int mdc_cancel_weight(struct ldlm_lock *lock)
2442 {
2443         if (lock->l_resource->lr_type != LDLM_IBITS)
2444                 RETURN(0);
2445
2446         /* FIXME: if we ever get into a situation where there are too many
2447          * opened files with open locks on a single node, then we really
2448          * should replay these open locks to reget it */
2449         if (lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_OPEN)
2450                 RETURN(0);
2451
2452         RETURN(1);
2453 }
2454
2455 static int mdc_resource_inode_free(struct ldlm_resource *res)
2456 {
2457         if (res->lr_lvb_inode)
2458                 res->lr_lvb_inode = NULL;
2459
2460         return 0;
2461 }
2462
2463 static struct ldlm_valblock_ops inode_lvbo = {
2464         .lvbo_free = mdc_resource_inode_free
2465 };
2466
2467 static int mdc_llog_init(struct obd_device *obd)
2468 {
2469         struct obd_llog_group   *olg = &obd->obd_olg;
2470         struct llog_ctxt        *ctxt;
2471         int                      rc;
2472
2473         ENTRY;
2474
2475         rc = llog_setup(NULL, obd, olg, LLOG_CHANGELOG_REPL_CTXT, obd,
2476                         &llog_client_ops);
2477         if (rc < 0)
2478                 RETURN(rc);
2479
2480         ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
2481         llog_initiator_connect(ctxt);
2482         llog_ctxt_put(ctxt);
2483
2484         RETURN(0);
2485 }
2486
2487 static void mdc_llog_finish(struct obd_device *obd)
2488 {
2489         struct llog_ctxt *ctxt;
2490
2491         ENTRY;
2492
2493         ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
2494         if (ctxt != NULL)
2495                 llog_cleanup(NULL, ctxt);
2496
2497         EXIT;
2498 }
2499
2500 int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
2501 {
2502         int rc;
2503
2504         ENTRY;
2505
2506         rc = osc_setup_common(obd, cfg);
2507         if (rc < 0)
2508                 RETURN(rc);
2509
2510 #ifdef CONFIG_PROC_FS
2511         obd->obd_vars = lprocfs_mdc_obd_vars;
2512         lprocfs_obd_setup(obd, false);
2513         lprocfs_alloc_md_stats(obd, 0);
2514 #endif
2515
2516         sptlrpc_lprocfs_cliobd_attach(obd);
2517         ptlrpc_lprocfs_register_obd(obd);
2518
2519         ns_register_cancel(obd->obd_namespace, mdc_cancel_weight);
2520
2521         obd->obd_namespace->ns_lvbo = &inode_lvbo;
2522
2523         rc = mdc_llog_init(obd);
2524         if (rc) {
2525                 CERROR("%s: failed to setup llogging subsystems: rc = %d\n",
2526                        obd->obd_name, rc);
2527                 GOTO(err_llog_cleanup, rc);
2528         }
2529
2530         rc = mdc_changelog_cdev_init(obd);
2531         if (rc) {
2532                 CERROR("%s: failed to setup changelog char device: rc = %d\n",
2533                        obd->obd_name, rc);
2534                 GOTO(err_changelog_cleanup, rc);
2535         }
2536
2537         RETURN(rc);
2538
2539 err_changelog_cleanup:
2540         mdc_llog_finish(obd);
2541 err_llog_cleanup:
2542         ptlrpc_lprocfs_unregister_obd(obd);
2543         lprocfs_obd_cleanup(obd);
2544         lprocfs_free_md_stats(obd);
2545
2546         osc_cleanup_common(obd);
2547         return rc;
2548 }
2549
2550 /* Initialize the default and maximum LOV EA sizes.  This allows
2551  * us to make MDS RPCs with large enough reply buffers to hold a default
2552  * sized EA without having to calculate this (via a call into the
2553  * LOV + OSCs) each time we make an RPC.  The maximum size is also tracked
2554  * but not used to avoid wastefully vmalloc()'ing large reply buffers when
2555  * a large number of stripes is possible.  If a larger reply buffer is
2556  * required it will be reallocated in the ptlrpc layer due to overflow.
2557  */
2558 static int mdc_init_ea_size(struct obd_export *exp, __u32 easize,
2559                             __u32 def_easize)
2560 {
2561         struct obd_device *obd = exp->exp_obd;
2562         struct client_obd *cli = &obd->u.cli;
2563         ENTRY;
2564
2565         if (cli->cl_max_mds_easize < easize)
2566                 cli->cl_max_mds_easize = easize;
2567
2568         if (cli->cl_default_mds_easize < def_easize)
2569                 cli->cl_default_mds_easize = def_easize;
2570
2571         RETURN(0);
2572 }
2573
2574 static int mdc_precleanup(struct obd_device *obd)
2575 {
2576         ENTRY;
2577
2578         osc_precleanup_common(obd);
2579
2580         /* Failsafe, ok if racy */
2581         if (obd->obd_type->typ_refcnt <= 1)
2582                 libcfs_kkuc_group_rem(0, KUC_GRP_HSM);
2583
2584         mdc_changelog_cdev_finish(obd);
2585
2586         obd_cleanup_client_import(obd);
2587         ptlrpc_lprocfs_unregister_obd(obd);
2588         lprocfs_obd_cleanup(obd);
2589         lprocfs_free_md_stats(obd);
2590         mdc_llog_finish(obd);
2591         RETURN(0);
2592 }
2593
2594 static int mdc_cleanup(struct obd_device *obd)
2595 {
2596         return osc_cleanup_common(obd);
2597 }
2598
2599 int mdc_process_config(struct obd_device *obd, size_t len, void *buf)
2600 {
2601         struct lustre_cfg *lcfg = buf;
2602         int rc;
2603
2604         rc = class_process_proc_param(PARAM_MDC, obd->obd_vars, lcfg, obd);
2605         return (rc > 0 ? 0: rc);
2606 }
2607
2608 static struct obd_ops mdc_obd_ops = {
2609         .o_owner            = THIS_MODULE,
2610         .o_setup            = mdc_setup,
2611         .o_precleanup       = mdc_precleanup,
2612         .o_cleanup          = mdc_cleanup,
2613         .o_add_conn         = client_import_add_conn,
2614         .o_del_conn         = client_import_del_conn,
2615         .o_connect          = client_connect_import,
2616         .o_reconnect        = osc_reconnect,
2617         .o_disconnect       = osc_disconnect,
2618         .o_iocontrol        = mdc_iocontrol,
2619         .o_set_info_async   = mdc_set_info_async,
2620         .o_statfs           = mdc_statfs,
2621         .o_fid_init         = client_fid_init,
2622         .o_fid_fini         = client_fid_fini,
2623         .o_fid_alloc        = mdc_fid_alloc,
2624         .o_import_event     = mdc_import_event,
2625         .o_get_info         = mdc_get_info,
2626         .o_process_config   = mdc_process_config,
2627         .o_get_uuid         = mdc_get_uuid,
2628         .o_quotactl         = mdc_quotactl,
2629 };
2630
2631 static struct md_ops mdc_md_ops = {
2632         .m_get_root         = mdc_get_root,
2633         .m_null_inode       = mdc_null_inode,
2634         .m_close            = mdc_close,
2635         .m_create           = mdc_create,
2636         .m_enqueue          = mdc_enqueue,
2637         .m_getattr          = mdc_getattr,
2638         .m_getattr_name     = mdc_getattr_name,
2639         .m_intent_lock      = mdc_intent_lock,
2640         .m_link             = mdc_link,
2641         .m_rename           = mdc_rename,
2642         .m_setattr          = mdc_setattr,
2643         .m_setxattr         = mdc_setxattr,
2644         .m_getxattr         = mdc_getxattr,
2645         .m_fsync                = mdc_fsync,
2646         .m_read_page            = mdc_read_page,
2647         .m_unlink           = mdc_unlink,
2648         .m_cancel_unused    = mdc_cancel_unused,
2649         .m_init_ea_size     = mdc_init_ea_size,
2650         .m_set_lock_data    = mdc_set_lock_data,
2651         .m_lock_match       = mdc_lock_match,
2652         .m_get_lustre_md    = mdc_get_lustre_md,
2653         .m_free_lustre_md   = mdc_free_lustre_md,
2654         .m_set_open_replay_data = mdc_set_open_replay_data,
2655         .m_clear_open_replay_data = mdc_clear_open_replay_data,
2656         .m_intent_getattr_async = mdc_intent_getattr_async,
2657         .m_revalidate_lock      = mdc_revalidate_lock
2658 };
2659
2660 static int __init mdc_init(void)
2661 {
2662         return class_register_type(&mdc_obd_ops, &mdc_md_ops, true, NULL,
2663                                    LUSTRE_MDC_NAME, &mdc_device_type);
2664 }
2665
2666 static void __exit mdc_exit(void)
2667 {
2668         class_unregister_type(LUSTRE_MDC_NAME);
2669 }
2670
2671 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
2672 MODULE_DESCRIPTION("Lustre Metadata Client");
2673 MODULE_VERSION(LUSTRE_VERSION_STRING);
2674 MODULE_LICENSE("GPL");
2675
2676 module_init(mdc_init);
2677 module_exit(mdc_exit);