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, 2017, 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         size_t                 u32_count = 0;
 765         int                    rc;
 766         int                    saved_rc = 0;
 767         ENTRY;
 768
 769         CDEBUG(D_INODE, "%s: "DFID" file closed with intent: %x\n",
 770                exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
 771                op_data->op_bias);
 772
 773         if (op_data->op_bias & MDS_CLOSE_INTENT) {
 774                 req_fmt = &RQF_MDS_CLOSE_INTENT;
 775                 if (op_data->op_bias & MDS_HSM_RELEASE) {
 776                         /* allocate a FID for volatile file */
 777                         rc = mdc_fid_alloc(NULL, exp, &op_data->op_fid2,
 778                                            op_data);
 779                         if (rc < 0) {
 780                                 CERROR("%s: "DFID" allocating FID: rc = %d\n",
 781                                        obd->obd_name, PFID(&op_data->op_fid1),
 782                                        rc);
 783                                 /* save the errcode and proceed to close */
 784                                 saved_rc = rc;
 785                         }
 786                 }
 787                 if (op_data->op_bias & MDS_CLOSE_RESYNC_DONE) {
 788                         size_t count = op_data->op_data_size / sizeof(__u32);
 789
 790                         if (count > INLINE_RESYNC_ARRAY_SIZE)
 791                                 u32_count = count;
 792                 }
 793         } else {
 794                 req_fmt = &RQF_MDS_CLOSE;
 795         }
 796
 797         *request = NULL;
 798         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_CLOSE))
 799                 req = NULL;
 800         else
 801                 req = ptlrpc_request_alloc(class_exp2cliimp(exp), req_fmt);
 802
 803         /* Ensure that this close's handle is fixed up during replay. */
 804         if (likely(mod != NULL)) {
 805                 LASSERTF(mod->mod_open_req != NULL &&
 806                          mod->mod_open_req->rq_type != LI_POISON,
 807                          "POISONED open %p!\n", mod->mod_open_req);
 808
 809                 mod->mod_close_req = req;
 810
 811                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
 812                 /* We no longer want to preserve this open for replay even
 813                  * though the open was committed. b=3632, b=3633 */
 814                 spin_lock(&mod->mod_open_req->rq_lock);
 815                 mod->mod_open_req->rq_replay = 0;
 816                 spin_unlock(&mod->mod_open_req->rq_lock);
 817         } else {
 818                 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
 819         }
 820         if (req == NULL) {
 821                 /**
 822                  * TODO: repeat close after errors
 823                  */
 824                 CWARN("%s: close of FID "DFID" failed, file reference will be "
 825                       "dropped when this client unmounts or is evicted\n",
 826                       obd->obd_name, PFID(&op_data->op_fid1));
 827                 GOTO(out, rc = -ENOMEM);
 828         }
 829
 830         if (u32_count > 0)
 831                 req_capsule_set_size(&req->rq_pill, &RMF_U32, RCL_CLIENT,
 832                                      u32_count * sizeof(__u32));
 833
 834         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
 835         if (rc) {
 836                 ptlrpc_request_free(req);
 837                 req = NULL;
 838                 GOTO(out, rc);
 839         }
 840
 841         /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
 842          * portal whose threads are not taking any DLM locks and are therefore
 843          * always progressing */
 844         req->rq_request_portal = MDS_READPAGE_PORTAL;
 845         ptlrpc_at_set_req_timeout(req);
 846
 847
 848         mdc_close_pack(req, op_data);
 849
 850         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 851                              obd->u.cli.cl_default_mds_easize);
 852
 853         ptlrpc_request_set_replen(req);
 854
 855         mdc_get_mod_rpc_slot(req, NULL);
 856         rc = ptlrpc_queue_wait(req);
 857         mdc_put_mod_rpc_slot(req, NULL);
 858
 859         if (req->rq_repmsg == NULL) {
 860                 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
 861                        req->rq_status);
 862                 if (rc == 0)
 863                         rc = req->rq_status ?: -EIO;
 864         } else if (rc == 0 || rc == -EAGAIN) {
 865                 struct mdt_body *body;
 866
 867                 rc = lustre_msg_get_status(req->rq_repmsg);
 868                 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
 869                         DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
 870                                   "= %d", rc);
 871                         if (rc > 0)
 872                                 rc = -rc;
 873                 }
 874                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 875                 if (body == NULL)
 876                         rc = -EPROTO;
 877         } else if (rc == -ESTALE) {
 878                 /**
 879                  * it can be allowed error after 3633 if open was committed and
 880                  * server failed before close was sent. Let's check if mod
 881                  * exists and return no error in that case
 882                  */
 883                 if (mod) {
 884                         DEBUG_REQ(D_HA, req, "Reset ESTALE = %d", rc);
 885                         LASSERT(mod->mod_open_req != NULL);
 886                         if (mod->mod_open_req->rq_committed)
 887                                 rc = 0;
 888                 }
 889         }
 890
 891 out:
 892         if (mod) {
 893                 if (rc != 0)
 894                         mod->mod_close_req = NULL;
 895                 /* Since now, mod is accessed through open_req only,
 896                  * thus close req does not keep a reference on mod anymore. */
 897                 obd_mod_put(mod);
 898         }
 899         *request = req;
 900
 901         RETURN(rc < 0 ? rc : saved_rc);
 902 }
 903
 904 static int mdc_getpage(struct obd_export *exp, const struct lu_fid *fid,
 905                        u64 offset, struct page **pages, int npages,
 906                        struct ptlrpc_request **request)
 907 {
 908         struct ptlrpc_request   *req;
 909         struct ptlrpc_bulk_desc *desc;
 910         int                      i;
 911         wait_queue_head_t        waitq;
 912         int                      resends = 0;
 913         struct l_wait_info       lwi;
 914         int                      rc;
 915         ENTRY;
 916
 917         *request = NULL;
 918         init_waitqueue_head(&waitq);
 919
 920 restart_bulk:
 921         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
 922         if (req == NULL)
 923                 RETURN(-ENOMEM);
 924
 925         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
 926         if (rc) {
 927                 ptlrpc_request_free(req);
 928                 RETURN(rc);
 929         }
 930
 931         req->rq_request_portal = MDS_READPAGE_PORTAL;
 932         ptlrpc_at_set_req_timeout(req);
 933
 934         desc = ptlrpc_prep_bulk_imp(req, npages, 1,
 935                                     PTLRPC_BULK_PUT_SINK | PTLRPC_BULK_BUF_KIOV,
 936                                     MDS_BULK_PORTAL,
 937                                     &ptlrpc_bulk_kiov_pin_ops);
 938         if (desc == NULL) {
 939                 ptlrpc_req_finished(req);
 940                 RETURN(-ENOMEM);
 941         }
 942
 943         /* NB req now owns desc and will free it when it gets freed */
 944         for (i = 0; i < npages; i++)
 945                 desc->bd_frag_ops->add_kiov_frag(desc, pages[i], 0,
 946                                                  PAGE_SIZE);
 947
 948         mdc_readdir_pack(req, offset, PAGE_SIZE * npages, fid);
 949
 950         ptlrpc_request_set_replen(req);
 951         rc = ptlrpc_queue_wait(req);
 952         if (rc) {
 953                 ptlrpc_req_finished(req);
 954                 if (rc != -ETIMEDOUT)
 955                         RETURN(rc);
 956
 957                 resends++;
 958                 if (!client_should_resend(resends, &exp->exp_obd->u.cli)) {
 959                         CERROR("%s: too many resend retries: rc = %d\n",
 960                                exp->exp_obd->obd_name, -EIO);
 961                         RETURN(-EIO);
 962                 }
 963                 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(resends), NULL, NULL,
 964                                        NULL);
 965                 l_wait_event(waitq, 0, &lwi);
 966
 967                 goto restart_bulk;
 968         }
 969
 970         rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
 971                                           req->rq_bulk->bd_nob_transferred);
 972         if (rc < 0) {
 973                 ptlrpc_req_finished(req);
 974                 RETURN(rc);
 975         }
 976
 977         if (req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK) {
 978                 CERROR("%s: unexpected bytes transferred: %d (%ld expected)\n",
 979                        exp->exp_obd->obd_name, req->rq_bulk->bd_nob_transferred,
 980                        PAGE_SIZE * npages);
 981                 ptlrpc_req_finished(req);
 982                 RETURN(-EPROTO);
 983         }
 984
 985         *request = req;
 986         RETURN(0);
 987 }
 988
 989 static void mdc_release_page(struct page *page, int remove)
 990 {
 991         if (remove) {
 992                 lock_page(page);
 993                 if (likely(page->mapping != NULL))
 994                         truncate_complete_page(page->mapping, page);
 995                 unlock_page(page);
 996         }
 997         put_page(page);
 998 }
 999
1000 static struct page *mdc_page_locate(struct address_space *mapping, __u64 *hash,
1001                                     __u64 *start, __u64 *end, int hash64)
1002 {
1003         /*
1004          * Complement of hash is used as an index so that
1005          * radix_tree_gang_lookup() can be used to find a page with starting
1006          * hash _smaller_ than one we are looking for.
1007          */
1008         unsigned long offset = hash_x_index(*hash, hash64);
1009         struct page *page;
1010         int found;
1011
1012         spin_lock_irq(&mapping->tree_lock);
1013         found = radix_tree_gang_lookup(&mapping->page_tree,
1014                                        (void **)&page, offset, 1);
1015         if (found > 0 && !radix_tree_exceptional_entry(page)) {
1016                 struct lu_dirpage *dp;
1017
1018                 get_page(page);
1019                 spin_unlock_irq(&mapping->tree_lock);
1020                 /*
1021                  * In contrast to find_lock_page() we are sure that directory
1022                  * page cannot be truncated (while DLM lock is held) and,
1023                  * hence, can avoid restart.
1024                  *
1025                  * In fact, page cannot be locked here at all, because
1026                  * mdc_read_page_remote does synchronous io.
1027                  */
1028                 wait_on_page_locked(page);
1029                 if (PageUptodate(page)) {
1030                         dp = kmap(page);
1031                         if (BITS_PER_LONG == 32 && hash64) {
1032                                 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1033                                 *end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
1034                                 *hash  = *hash >> 32;
1035                         } else {
1036                                 *start = le64_to_cpu(dp->ldp_hash_start);
1037                                 *end   = le64_to_cpu(dp->ldp_hash_end);
1038                         }
1039                         if (unlikely(*start == 1 && *hash == 0))
1040                                 *hash = *start;
1041                         else
1042                                 LASSERTF(*start <= *hash, "start = %#llx"
1043                                          ",end = %#llx,hash = %#llx\n",
1044                                          *start, *end, *hash);
1045                         CDEBUG(D_VFSTRACE, "offset %lx [%#llx %#llx],"
1046                               " hash %#llx\n", offset, *start, *end, *hash);
1047                         if (*hash > *end) {
1048                                 kunmap(page);
1049                                 mdc_release_page(page, 0);
1050                                 page = NULL;
1051                         } else if (*end != *start && *hash == *end) {
1052                                 /*
1053                                  * upon hash collision, remove this page,
1054                                  * otherwise put page reference, and
1055                                  * mdc_read_page_remote() will issue RPC to
1056                                  * fetch the page we want.
1057                                  */
1058                                 kunmap(page);
1059                                 mdc_release_page(page,
1060                                     le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
1061                                 page = NULL;
1062                         }
1063                 } else {
1064                         put_page(page);
1065                         page = ERR_PTR(-EIO);
1066                 }
1067         } else {
1068                 spin_unlock_irq(&mapping->tree_lock);
1069                 page = NULL;
1070         }
1071         return page;
1072 }
1073
1074 /*
1075  * Adjust a set of pages, each page containing an array of lu_dirpages,
1076  * so that each page can be used as a single logical lu_dirpage.
1077  *
1078  * A lu_dirpage is laid out as follows, where s = ldp_hash_start,
1079  * e = ldp_hash_end, f = ldp_flags, p = padding, and each "ent" is a
1080  * struct lu_dirent.  It has size up to LU_PAGE_SIZE. The ldp_hash_end
1081  * value is used as a cookie to request the next lu_dirpage in a
1082  * directory listing that spans multiple pages (two in this example):
1083  *   ________
1084  *  |        |
1085  * .|--------v-------   -----.
1086  * |s|e|f|p|ent|ent| ... |ent|
1087  * '--|--------------   -----'   Each PAGE contains a single
1088  *    '------.                   lu_dirpage.
1089  * .---------v-------   -----.
1090  * |s|e|f|p|ent| 0 | ... | 0 |
1091  * '-----------------   -----'
1092  *
1093  * However, on hosts where the native VM page size (PAGE_SIZE) is
1094  * larger than LU_PAGE_SIZE, a single host page may contain multiple
1095  * lu_dirpages. After reading the lu_dirpages from the MDS, the
1096  * ldp_hash_end of the first lu_dirpage refers to the one immediately
1097  * after it in the same PAGE (arrows simplified for brevity, but
1098  * in general e0==s1, e1==s2, etc.):
1099  *
1100  * .--------------------   -----.
1101  * |s0|e0|f0|p|ent|ent| ... |ent|
1102  * |---v----------------   -----|
1103  * |s1|e1|f1|p|ent|ent| ... |ent|
1104  * |---v----------------   -----|  Here, each PAGE contains
1105  *             ...                 multiple lu_dirpages.
1106  * |---v----------------   -----|
1107  * |s'|e'|f'|p|ent|ent| ... |ent|
1108  * '---|----------------   -----'
1109  *     v
1110  * .----------------------------.
1111  * |        next PAGE           |
1112  *
1113  * This structure is transformed into a single logical lu_dirpage as follows:
1114  *
1115  * - Replace e0 with e' so the request for the next lu_dirpage gets the page
1116  *   labeled 'next PAGE'.
1117  *
1118  * - Copy the LDF_COLLIDE flag from f' to f0 to correctly reflect whether
1119  *   a hash collision with the next page exists.
1120  *
1121  * - Adjust the lde_reclen of the ending entry of each lu_dirpage to span
1122  *   to the first entry of the next lu_dirpage.
1123  */
1124 #if PAGE_SIZE > LU_PAGE_SIZE
1125 static void mdc_adjust_dirpages(struct page **pages, int cfs_pgs, int lu_pgs)
1126 {
1127         int i;
1128
1129         for (i = 0; i < cfs_pgs; i++) {
1130                 struct lu_dirpage       *dp = kmap(pages[i]);
1131                 struct lu_dirpage       *first = dp;
1132                 struct lu_dirent        *end_dirent = NULL;
1133                 struct lu_dirent        *ent;
1134                 __u64           hash_end = le64_to_cpu(dp->ldp_hash_end);
1135                 __u32           flags = le32_to_cpu(dp->ldp_flags);
1136
1137                 while (--lu_pgs > 0) {
1138                         ent = lu_dirent_start(dp);
1139                         for (end_dirent = ent; ent != NULL;
1140                              end_dirent = ent, ent = lu_dirent_next(ent));
1141
1142                         /* Advance dp to next lu_dirpage. */
1143                         dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE);
1144
1145                         /* Check if we've reached the end of the PAGE. */
1146                         if (!((unsigned long)dp & ~PAGE_MASK))
1147                                 break;
1148
1149                         /* Save the hash and flags of this lu_dirpage. */
1150                         hash_end = le64_to_cpu(dp->ldp_hash_end);
1151                         flags = le32_to_cpu(dp->ldp_flags);
1152
1153                         /* Check if lu_dirpage contains no entries. */
1154                         if (end_dirent == NULL)
1155                                 break;
1156
1157                         /* Enlarge the end entry lde_reclen from 0 to
1158                          * first entry of next lu_dirpage. */
1159                         LASSERT(le16_to_cpu(end_dirent->lde_reclen) == 0);
1160                         end_dirent->lde_reclen =
1161                                 cpu_to_le16((char *)(dp->ldp_entries) -
1162                                             (char *)end_dirent);
1163                 }
1164
1165                 first->ldp_hash_end = hash_end;
1166                 first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE);
1167                 first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE);
1168
1169                 kunmap(pages[i]);
1170         }
1171         LASSERTF(lu_pgs == 0, "left = %d\n", lu_pgs);
1172 }
1173 #else
1174 #define mdc_adjust_dirpages(pages, cfs_pgs, lu_pgs) do {} while (0)
1175 #endif  /* PAGE_SIZE > LU_PAGE_SIZE */
1176
1177 /* parameters for readdir page */
1178 struct readpage_param {
1179         struct md_op_data       *rp_mod;
1180         __u64                   rp_off;
1181         int                     rp_hash64;
1182         struct obd_export       *rp_exp;
1183         struct md_callback      *rp_cb;
1184 };
1185
1186 #ifndef HAVE_DELETE_FROM_PAGE_CACHE
1187 static inline void delete_from_page_cache(struct page *page)
1188 {
1189         remove_from_page_cache(page);
1190         put_page(page);
1191 }
1192 #endif
1193
1194 /**
1195  * Read pages from server.
1196  *
1197  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
1198  * a header lu_dirpage which describes the start/end hash, and whether this
1199  * page is empty (contains no dir entry) or hash collide with next page.
1200  * After client receives reply, several pages will be integrated into dir page
1201  * in PAGE_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the
1202  * lu_dirpage for this integrated page will be adjusted.
1203  **/
1204 static int mdc_read_page_remote(void *data, struct page *page0)
1205 {
1206         struct readpage_param *rp = data;
1207         struct page **page_pool;
1208         struct page *page;
1209         struct lu_dirpage *dp;
1210         struct md_op_data *op_data = rp->rp_mod;
1211         struct ptlrpc_request *req;
1212         int max_pages;
1213         struct inode *inode;
1214         struct lu_fid *fid;
1215         int rd_pgs = 0; /* number of pages actually read */
1216         int npages;
1217         int i;
1218         int rc;
1219         ENTRY;
1220
1221         max_pages = rp->rp_exp->exp_obd->u.cli.cl_max_pages_per_rpc;
1222         inode = op_data->op_data;
1223         fid = &op_data->op_fid1;
1224         LASSERT(inode != NULL);
1225
1226         OBD_ALLOC(page_pool, sizeof(page_pool[0]) * max_pages);
1227         if (page_pool != NULL) {
1228                 page_pool[0] = page0;
1229         } else {
1230                 page_pool = &page0;
1231                 max_pages = 1;
1232         }
1233
1234         for (npages = 1; npages < max_pages; npages++) {
1235                 page = page_cache_alloc_cold(inode->i_mapping);
1236                 if (page == NULL)
1237                         break;
1238                 page_pool[npages] = page;
1239         }
1240
1241         rc = mdc_getpage(rp->rp_exp, fid, rp->rp_off, page_pool, npages, &req);
1242         if (rc < 0) {
1243                 /* page0 is special, which was added into page cache early */
1244                 delete_from_page_cache(page0);
1245         } else {
1246                 int lu_pgs;
1247
1248                 rd_pgs = (req->rq_bulk->bd_nob_transferred + PAGE_SIZE - 1) >>
1249                         PAGE_SHIFT;
1250                 lu_pgs = req->rq_bulk->bd_nob_transferred >> LU_PAGE_SHIFT;
1251                 LASSERT(!(req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK));
1252
1253                 CDEBUG(D_INODE, "read %d(%d) pages\n", rd_pgs, lu_pgs);
1254
1255                 mdc_adjust_dirpages(page_pool, rd_pgs, lu_pgs);
1256
1257                 SetPageUptodate(page0);
1258         }
1259         unlock_page(page0);
1260
1261         ptlrpc_req_finished(req);
1262         CDEBUG(D_CACHE, "read %d/%d pages\n", rd_pgs, npages);
1263         for (i = 1; i < npages; i++) {
1264                 unsigned long   offset;
1265                 __u64           hash;
1266                 int ret;
1267
1268                 page = page_pool[i];
1269
1270                 if (rc < 0 || i >= rd_pgs) {
1271                         put_page(page);
1272                         continue;
1273                 }
1274
1275                 SetPageUptodate(page);
1276
1277                 dp = kmap(page);
1278                 hash = le64_to_cpu(dp->ldp_hash_start);
1279                 kunmap(page);
1280
1281                 offset = hash_x_index(hash, rp->rp_hash64);
1282
1283                 prefetchw(&page->flags);
1284                 ret = add_to_page_cache_lru(page, inode->i_mapping, offset,
1285                                             GFP_KERNEL);
1286                 if (ret == 0)
1287                         unlock_page(page);
1288                 else
1289                         CDEBUG(D_VFSTRACE, "page %lu add to page cache failed:"
1290                                " rc = %d\n", offset, ret);
1291                 put_page(page);
1292         }
1293
1294         if (page_pool != &page0)
1295                 OBD_FREE(page_pool, sizeof(page_pool[0]) * max_pages);
1296
1297         RETURN(rc);
1298 }
1299
1300 /**
1301  * Read dir page from cache first, if it can not find it, read it from
1302  * server and add into the cache.
1303  *
1304  * \param[in] exp       MDC export
1305  * \param[in] op_data   client MD stack parameters, transfering parameters
1306  *                      between different layers on client MD stack.
1307  * \param[in] cb_op     callback required for ldlm lock enqueue during
1308  *                      read page
1309  * \param[in] hash_offset the hash offset of the page to be read
1310  * \param[in] ppage     the page to be read
1311  *
1312  * retval               = 0 get the page successfully
1313  *                      errno(<0) get the page failed
1314  */
1315 static int mdc_read_page(struct obd_export *exp, struct md_op_data *op_data,
1316                          struct md_callback *cb_op, __u64 hash_offset,
1317                          struct page **ppage)
1318 {
1319         struct lookup_intent    it = { .it_op = IT_READDIR };
1320         struct page             *page;
1321         struct inode            *dir = op_data->op_data;
1322         struct address_space    *mapping;
1323         struct lu_dirpage       *dp;
1324         __u64                   start = 0;
1325         __u64                   end = 0;
1326         struct lustre_handle    lockh;
1327         struct ptlrpc_request   *enq_req = NULL;
1328         struct readpage_param   rp_param;
1329         int rc;
1330
1331         ENTRY;
1332
1333         *ppage = NULL;
1334
1335         LASSERT(dir != NULL);
1336         mapping = dir->i_mapping;
1337
1338         rc = mdc_intent_lock(exp, op_data, &it, &enq_req,
1339                              cb_op->md_blocking_ast, 0);
1340         if (enq_req != NULL)
1341                 ptlrpc_req_finished(enq_req);
1342
1343         if (rc < 0) {
1344                 CERROR("%s: "DFID" lock enqueue fails: rc = %d\n",
1345                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1), rc);
1346                 RETURN(rc);
1347         }
1348
1349         rc = 0;
1350         lockh.cookie = it.it_lock_handle;
1351         mdc_set_lock_data(exp, &lockh, dir, NULL);
1352
1353         rp_param.rp_off = hash_offset;
1354         rp_param.rp_hash64 = op_data->op_cli_flags & CLI_HASH64;
1355         page = mdc_page_locate(mapping, &rp_param.rp_off, &start, &end,
1356                                rp_param.rp_hash64);
1357         if (IS_ERR(page)) {
1358                 CERROR("%s: dir page locate: "DFID" at %llu: rc %ld\n",
1359                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1360                        rp_param.rp_off, PTR_ERR(page));
1361                 GOTO(out_unlock, rc = PTR_ERR(page));
1362         } else if (page != NULL) {
1363                 /*
1364                  * XXX nikita: not entirely correct handling of a corner case:
1365                  * suppose hash chain of entries with hash value HASH crosses
1366                  * border between pages P0 and P1. First both P0 and P1 are
1367                  * cached, seekdir() is called for some entry from the P0 part
1368                  * of the chain. Later P0 goes out of cache. telldir(HASH)
1369                  * happens and finds P1, as it starts with matching hash
1370                  * value. Remaining entries from P0 part of the chain are
1371                  * skipped. (Is that really a bug?)
1372                  *
1373                  * Possible solutions: 0. don't cache P1 is such case, handle
1374                  * it as an "overflow" page. 1. invalidate all pages at
1375                  * once. 2. use HASH|1 as an index for P1.
1376                  */
1377                 GOTO(hash_collision, page);
1378         }
1379
1380         rp_param.rp_exp = exp;
1381         rp_param.rp_mod = op_data;
1382         page = read_cache_page(mapping,
1383                                hash_x_index(rp_param.rp_off,
1384                                             rp_param.rp_hash64),
1385                                mdc_read_page_remote, &rp_param);
1386         if (IS_ERR(page)) {
1387                 CDEBUG(D_INFO, "%s: read cache page: "DFID" at %llu: %ld\n",
1388                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1389                        rp_param.rp_off, PTR_ERR(page));
1390                 GOTO(out_unlock, rc = PTR_ERR(page));
1391         }
1392
1393         wait_on_page_locked(page);
1394         (void)kmap(page);
1395         if (!PageUptodate(page)) {
1396                 CERROR("%s: page not updated: "DFID" at %llu: rc %d\n",
1397                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1398                        rp_param.rp_off, -5);
1399                 goto fail;
1400         }
1401         if (!PageChecked(page))
1402                 SetPageChecked(page);
1403         if (PageError(page)) {
1404                 CERROR("%s: page error: "DFID" at %llu: rc %d\n",
1405                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1406                        rp_param.rp_off, -5);
1407                 goto fail;
1408         }
1409
1410 hash_collision:
1411         dp = page_address(page);
1412         if (BITS_PER_LONG == 32 && rp_param.rp_hash64) {
1413                 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1414                 end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
1415                 rp_param.rp_off = hash_offset >> 32;
1416         } else {
1417                 start = le64_to_cpu(dp->ldp_hash_start);
1418                 end   = le64_to_cpu(dp->ldp_hash_end);
1419                 rp_param.rp_off = hash_offset;
1420         }
1421         if (end == start) {
1422                 LASSERT(start == rp_param.rp_off);
1423                 CWARN("Page-wide hash collision: %#lx\n", (unsigned long)end);
1424 #if BITS_PER_LONG == 32
1425                 CWARN("Real page-wide hash collision at [%llu %llu] with "
1426                       "hash %llu\n", le64_to_cpu(dp->ldp_hash_start),
1427                       le64_to_cpu(dp->ldp_hash_end), hash_offset);
1428 #endif
1429
1430                 /*
1431                  * Fetch whole overflow chain...
1432                  *
1433                  * XXX not yet.
1434                  */
1435                 goto fail;
1436         }
1437         *ppage = page;
1438 out_unlock:
1439         ldlm_lock_decref(&lockh, it.it_lock_mode);
1440         return rc;
1441 fail:
1442         kunmap(page);
1443         mdc_release_page(page, 1);
1444         rc = -EIO;
1445         goto out_unlock;
1446 }
1447
1448 static int mdc_statfs(const struct lu_env *env,
1449                       struct obd_export *exp, struct obd_statfs *osfs,
1450                       time64_t max_age, __u32 flags)
1451 {
1452         struct obd_device     *obd = class_exp2obd(exp);
1453         struct ptlrpc_request *req;
1454         struct obd_statfs     *msfs;
1455         struct obd_import     *imp = NULL;
1456         int                    rc;
1457         ENTRY;
1458
1459         /*
1460          * Since the request might also come from lprocfs, so we need
1461          * sync this with client_disconnect_export Bug15684
1462          */
1463         down_read(&obd->u.cli.cl_sem);
1464         if (obd->u.cli.cl_import)
1465                 imp = class_import_get(obd->u.cli.cl_import);
1466         up_read(&obd->u.cli.cl_sem);
1467         if (!imp)
1468                 RETURN(-ENODEV);
1469
1470         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_STATFS,
1471                                         LUSTRE_MDS_VERSION, MDS_STATFS);
1472         if (req == NULL)
1473                 GOTO(output, rc = -ENOMEM);
1474
1475         ptlrpc_request_set_replen(req);
1476
1477         if (flags & OBD_STATFS_NODELAY) {
1478                 /* procfs requests not want stay in wait for avoid deadlock */
1479                 req->rq_no_resend = 1;
1480                 req->rq_no_delay = 1;
1481         }
1482
1483         rc = ptlrpc_queue_wait(req);
1484         if (rc) {
1485                 /* check connection error first */
1486                 if (imp->imp_connect_error)
1487                         rc = imp->imp_connect_error;
1488                 GOTO(out, rc);
1489         }
1490
1491         msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1492         if (msfs == NULL)
1493                 GOTO(out, rc = -EPROTO);
1494
1495         *osfs = *msfs;
1496         EXIT;
1497 out:
1498         ptlrpc_req_finished(req);
1499 output:
1500         class_import_put(imp);
1501         return rc;
1502 }
1503
1504 static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
1505 {
1506         __u32 keylen, vallen;
1507         void *key;
1508         int rc;
1509
1510         if (gf->gf_pathlen > PATH_MAX)
1511                 RETURN(-ENAMETOOLONG);
1512         if (gf->gf_pathlen < 2)
1513                 RETURN(-EOVERFLOW);
1514
1515         /* Key is KEY_FID2PATH + getinfo_fid2path description */
1516         keylen = cfs_size_round(sizeof(KEY_FID2PATH) + sizeof(*gf) +
1517                                 sizeof(struct lu_fid));
1518         OBD_ALLOC(key, keylen);
1519         if (key == NULL)
1520                 RETURN(-ENOMEM);
1521         memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
1522         memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf));
1523         memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf),
1524                gf->gf_u.gf_root_fid, sizeof(struct lu_fid));
1525         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
1526                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
1527
1528         if (!fid_is_sane(&gf->gf_fid))
1529                 GOTO(out, rc = -EINVAL);
1530
1531         /* Val is struct getinfo_fid2path result plus path */
1532         vallen = sizeof(*gf) + gf->gf_pathlen;
1533
1534         rc = obd_get_info(NULL, exp, keylen, key, &vallen, gf);
1535         if (rc != 0 && rc != -EREMOTE)
1536                 GOTO(out, rc);
1537
1538         if (vallen <= sizeof(*gf))
1539                 GOTO(out, rc = -EPROTO);
1540         if (vallen > sizeof(*gf) + gf->gf_pathlen)
1541                 GOTO(out, rc = -EOVERFLOW);
1542
1543         CDEBUG(D_IOCTL, "path got "DFID" from %llu #%d: %s\n",
1544                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno,
1545                gf->gf_pathlen < 512 ? gf->gf_u.gf_path :
1546                /* only log the last 512 characters of the path */
1547                gf->gf_u.gf_path + gf->gf_pathlen - 512);
1548
1549 out:
1550         OBD_FREE(key, keylen);
1551         return rc;
1552 }
1553
1554 static int mdc_ioc_hsm_progress(struct obd_export *exp,
1555                                 struct hsm_progress_kernel *hpk)
1556 {
1557         struct obd_import               *imp = class_exp2cliimp(exp);
1558         struct hsm_progress_kernel      *req_hpk;
1559         struct ptlrpc_request           *req;
1560         int                              rc;
1561         ENTRY;
1562
1563         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_PROGRESS,
1564                                         LUSTRE_MDS_VERSION, MDS_HSM_PROGRESS);
1565         if (req == NULL)
1566                 GOTO(out, rc = -ENOMEM);
1567
1568         mdc_pack_body(req, NULL, 0, 0, -1, 0);
1569
1570         /* Copy hsm_progress struct */
1571         req_hpk = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_PROGRESS);
1572         if (req_hpk == NULL)
1573                 GOTO(out, rc = -EPROTO);
1574
1575         *req_hpk = *hpk;
1576         req_hpk->hpk_errval = lustre_errno_hton(hpk->hpk_errval);
1577
1578         ptlrpc_request_set_replen(req);
1579
1580         mdc_get_mod_rpc_slot(req, NULL);
1581         rc = ptlrpc_queue_wait(req);
1582         mdc_put_mod_rpc_slot(req, NULL);
1583
1584         GOTO(out, rc);
1585 out:
1586         ptlrpc_req_finished(req);
1587         return rc;
1588 }
1589
1590 static int mdc_ioc_hsm_ct_register(struct obd_import *imp, __u32 archives)
1591 {
1592         __u32                   *archive_mask;
1593         struct ptlrpc_request   *req;
1594         int                      rc;
1595         ENTRY;
1596
1597         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_REGISTER,
1598                                         LUSTRE_MDS_VERSION,
1599                                         MDS_HSM_CT_REGISTER);
1600         if (req == NULL)
1601                 GOTO(out, rc = -ENOMEM);
1602
1603         mdc_pack_body(req, NULL, 0, 0, -1, 0);
1604
1605         /* Copy hsm_progress struct */
1606         archive_mask = req_capsule_client_get(&req->rq_pill,
1607                                               &RMF_MDS_HSM_ARCHIVE);
1608         if (archive_mask == NULL)
1609                 GOTO(out, rc = -EPROTO);
1610
1611         *archive_mask = archives;
1612
1613         ptlrpc_request_set_replen(req);
1614
1615         rc = mdc_queue_wait(req);
1616         GOTO(out, rc);
1617 out:
1618         ptlrpc_req_finished(req);
1619         return rc;
1620 }
1621
1622 static int mdc_ioc_hsm_current_action(struct obd_export *exp,
1623                                       struct md_op_data *op_data)
1624 {
1625         struct hsm_current_action       *hca = op_data->op_data;
1626         struct hsm_current_action       *req_hca;
1627         struct ptlrpc_request           *req;
1628         int                              rc;
1629         ENTRY;
1630
1631         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1632                                    &RQF_MDS_HSM_ACTION);
1633         if (req == NULL)
1634                 RETURN(-ENOMEM);
1635
1636         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_ACTION);
1637         if (rc) {
1638                 ptlrpc_request_free(req);
1639                 RETURN(rc);
1640         }
1641
1642         mdc_pack_body(req, &op_data->op_fid1, 0, 0,
1643                       op_data->op_suppgids[0], 0);
1644
1645         ptlrpc_request_set_replen(req);
1646
1647         rc = mdc_queue_wait(req);
1648         if (rc)
1649                 GOTO(out, rc);
1650
1651         req_hca = req_capsule_server_get(&req->rq_pill,
1652                                          &RMF_MDS_HSM_CURRENT_ACTION);
1653         if (req_hca == NULL)
1654                 GOTO(out, rc = -EPROTO);
1655
1656         *hca = *req_hca;
1657
1658         EXIT;
1659 out:
1660         ptlrpc_req_finished(req);
1661         return rc;
1662 }
1663
1664 static int mdc_ioc_hsm_ct_unregister(struct obd_import *imp)
1665 {
1666         struct ptlrpc_request   *req;
1667         int                      rc;
1668         ENTRY;
1669
1670         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_UNREGISTER,
1671                                         LUSTRE_MDS_VERSION,
1672                                         MDS_HSM_CT_UNREGISTER);
1673         if (req == NULL)
1674                 GOTO(out, rc = -ENOMEM);
1675
1676         mdc_pack_body(req, NULL, 0, 0, -1, 0);
1677
1678         ptlrpc_request_set_replen(req);
1679
1680         rc = mdc_queue_wait(req);
1681         GOTO(out, rc);
1682 out:
1683         ptlrpc_req_finished(req);
1684         return rc;
1685 }
1686
1687 static int mdc_ioc_hsm_state_get(struct obd_export *exp,
1688                                  struct md_op_data *op_data)
1689 {
1690         struct hsm_user_state   *hus = op_data->op_data;
1691         struct hsm_user_state   *req_hus;
1692         struct ptlrpc_request   *req;
1693         int                      rc;
1694         ENTRY;
1695
1696         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1697                                    &RQF_MDS_HSM_STATE_GET);
1698         if (req == NULL)
1699                 RETURN(-ENOMEM);
1700
1701         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_GET);
1702         if (rc != 0) {
1703                 ptlrpc_request_free(req);
1704                 RETURN(rc);
1705         }
1706
1707         mdc_pack_body(req, &op_data->op_fid1, 0, 0,
1708                       op_data->op_suppgids[0], 0);
1709
1710         ptlrpc_request_set_replen(req);
1711
1712         rc = mdc_queue_wait(req);
1713         if (rc)
1714                 GOTO(out, rc);
1715
1716         req_hus = req_capsule_server_get(&req->rq_pill, &RMF_HSM_USER_STATE);
1717         if (req_hus == NULL)
1718                 GOTO(out, rc = -EPROTO);
1719
1720         *hus = *req_hus;
1721
1722         EXIT;
1723 out:
1724         ptlrpc_req_finished(req);
1725         return rc;
1726 }
1727
1728 static int mdc_ioc_hsm_state_set(struct obd_export *exp,
1729                                  struct md_op_data *op_data)
1730 {
1731         struct hsm_state_set    *hss = op_data->op_data;
1732         struct hsm_state_set    *req_hss;
1733         struct ptlrpc_request   *req;
1734         int                      rc;
1735         ENTRY;
1736
1737         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1738                                    &RQF_MDS_HSM_STATE_SET);
1739         if (req == NULL)
1740                 RETURN(-ENOMEM);
1741
1742         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_SET);
1743         if (rc) {
1744                 ptlrpc_request_free(req);
1745                 RETURN(rc);
1746         }
1747
1748         mdc_pack_body(req, &op_data->op_fid1, 0, 0,
1749                       op_data->op_suppgids[0], 0);
1750
1751         /* Copy states */
1752         req_hss = req_capsule_client_get(&req->rq_pill, &RMF_HSM_STATE_SET);
1753         if (req_hss == NULL)
1754                 GOTO(out, rc = -EPROTO);
1755         *req_hss = *hss;
1756
1757         ptlrpc_request_set_replen(req);
1758
1759         mdc_get_mod_rpc_slot(req, NULL);
1760         rc = ptlrpc_queue_wait(req);
1761         mdc_put_mod_rpc_slot(req, NULL);
1762
1763         GOTO(out, rc);
1764 out:
1765         ptlrpc_req_finished(req);
1766         return rc;
1767 }
1768
1769 static int mdc_ioc_hsm_request(struct obd_export *exp,
1770                                struct hsm_user_request *hur)
1771 {
1772         struct obd_import       *imp = class_exp2cliimp(exp);
1773         struct ptlrpc_request   *req;
1774         struct hsm_request      *req_hr;
1775         struct hsm_user_item    *req_hui;
1776         char                    *req_opaque;
1777         int                      rc;
1778         ENTRY;
1779
1780         req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_REQUEST);
1781         if (req == NULL)
1782                 GOTO(out, rc = -ENOMEM);
1783
1784         req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM, RCL_CLIENT,
1785                              hur->hur_request.hr_itemcount
1786                              * sizeof(struct hsm_user_item));
1787         req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA, RCL_CLIENT,
1788                              hur->hur_request.hr_data_len);
1789
1790         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_REQUEST);
1791         if (rc) {
1792                 ptlrpc_request_free(req);
1793                 RETURN(rc);
1794         }
1795
1796         mdc_pack_body(req, NULL, 0, 0, -1, 0);
1797
1798         /* Copy hsm_request struct */
1799         req_hr = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_REQUEST);
1800         if (req_hr == NULL)
1801                 GOTO(out, rc = -EPROTO);
1802         *req_hr = hur->hur_request;
1803
1804         /* Copy hsm_user_item structs */
1805         req_hui = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM);
1806         if (req_hui == NULL)
1807                 GOTO(out, rc = -EPROTO);
1808         memcpy(req_hui, hur->hur_user_item,
1809                hur->hur_request.hr_itemcount * sizeof(struct hsm_user_item));
1810
1811         /* Copy opaque field */
1812         req_opaque = req_capsule_client_get(&req->rq_pill, &RMF_GENERIC_DATA);
1813         if (req_opaque == NULL)
1814                 GOTO(out, rc = -EPROTO);
1815         memcpy(req_opaque, hur_data(hur), hur->hur_request.hr_data_len);
1816
1817         ptlrpc_request_set_replen(req);
1818
1819         mdc_get_mod_rpc_slot(req, NULL);
1820         rc = ptlrpc_queue_wait(req);
1821         mdc_put_mod_rpc_slot(req, NULL);
1822
1823         GOTO(out, rc);
1824
1825 out:
1826         ptlrpc_req_finished(req);
1827         return rc;
1828 }
1829
1830 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
1831                                 struct lustre_kernelcomm *lk);
1832
1833 static int mdc_quotactl(struct obd_device *unused, struct obd_export *exp,
1834                         struct obd_quotactl *oqctl)
1835 {
1836         struct ptlrpc_request   *req;
1837         struct obd_quotactl     *oqc;
1838         int                      rc;
1839         ENTRY;
1840
1841         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
1842                                         &RQF_MDS_QUOTACTL, LUSTRE_MDS_VERSION,
1843                                         MDS_QUOTACTL);
1844         if (req == NULL)
1845                 RETURN(-ENOMEM);
1846
1847         oqc = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1848         *oqc = *oqctl;
1849
1850         ptlrpc_request_set_replen(req);
1851         ptlrpc_at_set_req_timeout(req);
1852         req->rq_no_resend = 1;
1853
1854         rc = ptlrpc_queue_wait(req);
1855         if (rc)
1856                 CERROR("ptlrpc_queue_wait failed, rc: %d\n", rc);
1857
1858         if (req->rq_repmsg &&
1859             (oqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL))) {
1860                 *oqctl = *oqc;
1861         } else if (!rc) {
1862                 CERROR ("Can't unpack obd_quotactl\n");
1863                 rc = -EPROTO;
1864         }
1865         ptlrpc_req_finished(req);
1866
1867         RETURN(rc);
1868 }
1869
1870 static int mdc_ioc_swap_layouts(struct obd_export *exp,
1871                                 struct md_op_data *op_data)
1872 {
1873         struct list_head cancels = LIST_HEAD_INIT(cancels);
1874         struct ptlrpc_request   *req;
1875         int                      rc, count;
1876         struct mdc_swap_layouts *msl, *payload;
1877         ENTRY;
1878
1879         msl = op_data->op_data;
1880
1881         /* When the MDT will get the MDS_SWAP_LAYOUTS RPC the
1882          * first thing it will do is to cancel the 2 layout
1883          * locks held by this client.
1884          * So the client must cancel its layout locks on the 2 fids
1885          * with the request RPC to avoid extra RPC round trips.
1886          */
1887         count = mdc_resource_get_unused(exp, &op_data->op_fid1, &cancels,
1888                                         LCK_EX, MDS_INODELOCK_LAYOUT |
1889                                         MDS_INODELOCK_XATTR);
1890         count += mdc_resource_get_unused(exp, &op_data->op_fid2, &cancels,
1891                                          LCK_EX, MDS_INODELOCK_LAYOUT |
1892                                          MDS_INODELOCK_XATTR);
1893
1894         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1895                                    &RQF_MDS_SWAP_LAYOUTS);
1896         if (req == NULL) {
1897                 ldlm_lock_list_put(&cancels, l_bl_ast, count);
1898                 RETURN(-ENOMEM);
1899         }
1900
1901         rc = mdc_prep_elc_req(exp, req, MDS_SWAP_LAYOUTS, &cancels, count);
1902         if (rc) {
1903                 ptlrpc_request_free(req);
1904                 RETURN(rc);
1905         }
1906
1907         mdc_swap_layouts_pack(req, op_data);
1908
1909         payload = req_capsule_client_get(&req->rq_pill, &RMF_SWAP_LAYOUTS);
1910         LASSERT(payload);
1911
1912         *payload = *msl;
1913
1914         ptlrpc_request_set_replen(req);
1915
1916         rc = ptlrpc_queue_wait(req);
1917         if (rc)
1918                 GOTO(out, rc);
1919         EXIT;
1920
1921 out:
1922         ptlrpc_req_finished(req);
1923         return rc;
1924 }
1925
1926 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1927                          void *karg, void __user *uarg)
1928 {
1929         struct obd_device *obd = exp->exp_obd;
1930         struct obd_ioctl_data *data = karg;
1931         struct obd_import *imp = obd->u.cli.cl_import;
1932         int rc;
1933         ENTRY;
1934
1935         if (!try_module_get(THIS_MODULE)) {
1936                 CERROR("%s: cannot get module '%s'\n", obd->obd_name,
1937                        module_name(THIS_MODULE));
1938                 return -EINVAL;
1939         }
1940         switch (cmd) {
1941         case OBD_IOC_FID2PATH:
1942                 rc = mdc_ioc_fid2path(exp, karg);
1943                 GOTO(out, rc);
1944         case LL_IOC_HSM_CT_START:
1945                 rc = mdc_ioc_hsm_ct_start(exp, karg);
1946                 /* ignore if it was already registered on this MDS. */
1947                 if (rc == -EEXIST)
1948                         rc = 0;
1949                 GOTO(out, rc);
1950         case LL_IOC_HSM_PROGRESS:
1951                 rc = mdc_ioc_hsm_progress(exp, karg);
1952                 GOTO(out, rc);
1953         case LL_IOC_HSM_STATE_GET:
1954                 rc = mdc_ioc_hsm_state_get(exp, karg);
1955                 GOTO(out, rc);
1956         case LL_IOC_HSM_STATE_SET:
1957                 rc = mdc_ioc_hsm_state_set(exp, karg);
1958                 GOTO(out, rc);
1959         case LL_IOC_HSM_ACTION:
1960                 rc = mdc_ioc_hsm_current_action(exp, karg);
1961                 GOTO(out, rc);
1962         case LL_IOC_HSM_REQUEST:
1963                 rc = mdc_ioc_hsm_request(exp, karg);
1964                 GOTO(out, rc);
1965         case OBD_IOC_CLIENT_RECOVER:
1966                 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1, 0);
1967                 if (rc < 0)
1968                         GOTO(out, rc);
1969                 GOTO(out, rc = 0);
1970         case IOC_OSC_SET_ACTIVE:
1971                 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
1972                 GOTO(out, rc);
1973         case OBD_IOC_PING_TARGET:
1974                 rc = ptlrpc_obd_ping(obd);
1975                 GOTO(out, rc);
1976         /*
1977          * Normally IOC_OBD_STATFS, OBD_IOC_QUOTACTL iocontrol are handled by
1978          * LMV instead of MDC. But when the cluster is upgraded from 1.8,
1979          * there'd be no LMV layer thus we might be called here. Eventually
1980          * this code should be removed.
1981          * bz20731, LU-592.
1982          */
1983         case IOC_OBD_STATFS: {
1984                 struct obd_statfs stat_buf = {0};
1985
1986                 if (*((__u32 *) data->ioc_inlbuf2) != 0)
1987                         GOTO(out, rc = -ENODEV);
1988
1989                 /* copy UUID */
1990                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd),
1991                                  min((int)data->ioc_plen2,
1992                                      (int)sizeof(struct obd_uuid))))
1993                         GOTO(out, rc = -EFAULT);
1994
1995                 rc = mdc_statfs(NULL, obd->obd_self_export, &stat_buf,
1996                                 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
1997                                 0);
1998                 if (rc != 0)
1999                         GOTO(out, rc);
2000
2001                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
2002                                      min((int) data->ioc_plen1,
2003                                          (int) sizeof(stat_buf))))
2004                         GOTO(out, rc = -EFAULT);
2005
2006                 GOTO(out, rc = 0);
2007         }
2008         case OBD_IOC_QUOTACTL: {
2009                 struct if_quotactl *qctl = karg;
2010                 struct obd_quotactl *oqctl;
2011
2012                 OBD_ALLOC_PTR(oqctl);
2013                 if (oqctl == NULL)
2014                         GOTO(out, rc = -ENOMEM);
2015
2016                 QCTL_COPY(oqctl, qctl);
2017                 rc = obd_quotactl(exp, oqctl);
2018                 if (rc == 0) {
2019                         QCTL_COPY(qctl, oqctl);
2020                         qctl->qc_valid = QC_MDTIDX;
2021                         qctl->obd_uuid = obd->u.cli.cl_target_uuid;
2022                 }
2023
2024                 OBD_FREE_PTR(oqctl);
2025                 GOTO(out, rc);
2026         }
2027         case LL_IOC_GET_CONNECT_FLAGS:
2028                 if (copy_to_user(uarg, exp_connect_flags_ptr(exp),
2029                                  sizeof(*exp_connect_flags_ptr(exp))))
2030                         GOTO(out, rc = -EFAULT);
2031
2032                 GOTO(out, rc = 0);
2033         case LL_IOC_LOV_SWAP_LAYOUTS:
2034                 rc = mdc_ioc_swap_layouts(exp, karg);
2035                 GOTO(out, rc);
2036         default:
2037                 CERROR("unrecognised ioctl: cmd = %#x\n", cmd);
2038                 GOTO(out, rc = -ENOTTY);
2039         }
2040 out:
2041         module_put(THIS_MODULE);
2042
2043         return rc;
2044 }
2045
2046 static int mdc_get_info_rpc(struct obd_export *exp,
2047                             u32 keylen, void *key,
2048                             u32 vallen, void *val)
2049 {
2050         struct obd_import      *imp = class_exp2cliimp(exp);
2051         struct ptlrpc_request  *req;
2052         char                   *tmp;
2053         int                     rc = -EINVAL;
2054         ENTRY;
2055
2056         req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
2057         if (req == NULL)
2058                 RETURN(-ENOMEM);
2059
2060         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
2061                              RCL_CLIENT, keylen);
2062         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
2063                              RCL_CLIENT, sizeof(vallen));
2064
2065         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
2066         if (rc) {
2067                 ptlrpc_request_free(req);
2068                 RETURN(rc);
2069         }
2070
2071         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
2072         memcpy(tmp, key, keylen);
2073         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
2074         memcpy(tmp, &vallen, sizeof(vallen));
2075
2076         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
2077                              RCL_SERVER, vallen);
2078         ptlrpc_request_set_replen(req);
2079
2080         rc = ptlrpc_queue_wait(req);
2081         /* -EREMOTE means the get_info result is partial, and it needs to
2082          * continue on another MDT, see fid2path part in lmv_iocontrol */
2083         if (rc == 0 || rc == -EREMOTE) {
2084                 tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
2085                 memcpy(val, tmp, vallen);
2086                 if (ptlrpc_rep_need_swab(req)) {
2087                         if (KEY_IS(KEY_FID2PATH))
2088                                 lustre_swab_fid2path(val);
2089                 }
2090         }
2091         ptlrpc_req_finished(req);
2092
2093         RETURN(rc);
2094 }
2095
2096 static void lustre_swab_hai(struct hsm_action_item *h)
2097 {
2098         __swab32s(&h->hai_len);
2099         __swab32s(&h->hai_action);
2100         lustre_swab_lu_fid(&h->hai_fid);
2101         lustre_swab_lu_fid(&h->hai_dfid);
2102         __swab64s(&h->hai_cookie);
2103         __swab64s(&h->hai_extent.offset);
2104         __swab64s(&h->hai_extent.length);
2105         __swab64s(&h->hai_gid);
2106 }
2107
2108 static void lustre_swab_hal(struct hsm_action_list *h)
2109 {
2110         struct hsm_action_item  *hai;
2111         __u32                    i;
2112
2113         __swab32s(&h->hal_version);
2114         __swab32s(&h->hal_count);
2115         __swab32s(&h->hal_archive_id);
2116         __swab64s(&h->hal_flags);
2117         hai = hai_first(h);
2118         for (i = 0; i < h->hal_count; i++, hai = hai_next(hai))
2119                 lustre_swab_hai(hai);
2120 }
2121
2122 static void lustre_swab_kuch(struct kuc_hdr *l)
2123 {
2124         __swab16s(&l->kuc_magic);
2125         /* __u8 l->kuc_transport */
2126         __swab16s(&l->kuc_msgtype);
2127         __swab16s(&l->kuc_msglen);
2128 }
2129
2130 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2131                                 struct lustre_kernelcomm *lk)
2132 {
2133         struct obd_import  *imp = class_exp2cliimp(exp);
2134         __u32               archive = lk->lk_data;
2135         int                 rc = 0;
2136
2137         if (lk->lk_group != KUC_GRP_HSM) {
2138                 CERROR("Bad copytool group %d\n", lk->lk_group);
2139                 return -EINVAL;
2140         }
2141
2142         CDEBUG(D_HSM, "CT start r%d w%d u%d g%d f%#x\n", lk->lk_rfd, lk->lk_wfd,
2143                lk->lk_uid, lk->lk_group, lk->lk_flags);
2144
2145         if (lk->lk_flags & LK_FLG_STOP) {
2146                 /* Unregister with the coordinator */
2147                 rc = mdc_ioc_hsm_ct_unregister(imp);
2148         } else {
2149                 rc = mdc_ioc_hsm_ct_register(imp, archive);
2150         }
2151
2152         return rc;
2153 }
2154
2155 /**
2156  * Send a message to any listening copytools
2157  * @param val KUC message (kuc_hdr + hsm_action_list)
2158  * @param len total length of message
2159  */
2160 static int mdc_hsm_copytool_send(const struct obd_uuid *uuid,
2161                                  size_t len, void *val)
2162 {
2163         struct kuc_hdr          *lh = (struct kuc_hdr *)val;
2164         struct hsm_action_list  *hal = (struct hsm_action_list *)(lh + 1);
2165         int                      rc;
2166         ENTRY;
2167
2168         if (len < sizeof(*lh) + sizeof(*hal)) {
2169                 CERROR("Short HSM message %zu < %zu\n", len,
2170                        sizeof(*lh) + sizeof(*hal));
2171                 RETURN(-EPROTO);
2172         }
2173         if (lh->kuc_magic == __swab16(KUC_MAGIC)) {
2174                 lustre_swab_kuch(lh);
2175                 lustre_swab_hal(hal);
2176         } else if (lh->kuc_magic != KUC_MAGIC) {
2177                 CERROR("Bad magic %x!=%x\n", lh->kuc_magic, KUC_MAGIC);
2178                 RETURN(-EPROTO);
2179         }
2180
2181         CDEBUG(D_HSM, " Received message mg=%x t=%d m=%d l=%d actions=%d "
2182                "on %s\n",
2183                lh->kuc_magic, lh->kuc_transport, lh->kuc_msgtype,
2184                lh->kuc_msglen, hal->hal_count, hal->hal_fsname);
2185
2186         /* Broadcast to HSM listeners */
2187         rc = libcfs_kkuc_group_put(uuid, KUC_GRP_HSM, lh);
2188
2189         RETURN(rc);
2190 }
2191
2192 /**
2193  * callback function passed to kuc for re-registering each HSM copytool
2194  * running on MDC, after MDT shutdown/recovery.
2195  * @param data copytool registration data
2196  * @param cb_arg callback argument (obd_import)
2197  */
2198 static int mdc_hsm_ct_reregister(void *data, void *cb_arg)
2199 {
2200         struct kkuc_ct_data     *kcd = data;
2201         struct obd_import       *imp = (struct obd_import *)cb_arg;
2202         int                      rc;
2203
2204         if (kcd == NULL || kcd->kcd_magic != KKUC_CT_DATA_MAGIC)
2205                 return -EPROTO;
2206
2207         CDEBUG(D_HA, "%s: recover copytool registration to MDT (archive=%#x)\n",
2208                imp->imp_obd->obd_name, kcd->kcd_archive);
2209         rc = mdc_ioc_hsm_ct_register(imp, kcd->kcd_archive);
2210
2211         /* ignore error if the copytool is already registered */
2212         return (rc == -EEXIST) ? 0 : rc;
2213 }
2214
2215 /**
2216  * Re-establish all kuc contexts with MDT
2217  * after MDT shutdown/recovery.
2218  */
2219 static int mdc_kuc_reregister(struct obd_import *imp)
2220 {
2221         /* re-register HSM agents */
2222         return libcfs_kkuc_group_foreach(&imp->imp_obd->obd_uuid, KUC_GRP_HSM,
2223                                          mdc_hsm_ct_reregister, imp);
2224 }
2225
2226 static int mdc_set_info_async(const struct lu_env *env,
2227                               struct obd_export *exp,
2228                               u32 keylen, void *key,
2229                               u32 vallen, void *val,
2230                               struct ptlrpc_request_set *set)
2231 {
2232         struct obd_import       *imp = class_exp2cliimp(exp);
2233         int                      rc;
2234         ENTRY;
2235
2236         if (KEY_IS(KEY_READ_ONLY)) {
2237                 if (vallen != sizeof(int))
2238                         RETURN(-EINVAL);
2239
2240                 spin_lock(&imp->imp_lock);
2241                 if (*((int *)val)) {
2242                         imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
2243                         imp->imp_connect_data.ocd_connect_flags |=
2244                                                         OBD_CONNECT_RDONLY;
2245                 } else {
2246                         imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
2247                         imp->imp_connect_data.ocd_connect_flags &=
2248                                                         ~OBD_CONNECT_RDONLY;
2249                 }
2250                 spin_unlock(&imp->imp_lock);
2251
2252                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2253                                        keylen, key, vallen, val, set);
2254                 RETURN(rc);
2255         }
2256         if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2257                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2258                                        keylen, key, vallen, val, set);
2259                 RETURN(rc);
2260         }
2261         if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) {
2262                 rc = mdc_hsm_copytool_send(&imp->imp_obd->obd_uuid, vallen,
2263                                            val);
2264                 RETURN(rc);
2265         }
2266
2267         if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2268                 __u32 *default_easize = val;
2269
2270                 exp->exp_obd->u.cli.cl_default_mds_easize = *default_easize;
2271                 RETURN(0);
2272         }
2273
2274         rc = osc_set_info_async(env, exp, keylen, key, vallen, val, set);
2275         RETURN(rc);
2276 }
2277
2278 static int mdc_get_info(const struct lu_env *env, struct obd_export *exp,
2279                         __u32 keylen, void *key, __u32 *vallen, void *val)
2280 {
2281         int rc = -EINVAL;
2282
2283         if (KEY_IS(KEY_MAX_EASIZE)) {
2284                 __u32 mdsize, *max_easize;
2285
2286                 if (*vallen != sizeof(int))
2287                         RETURN(-EINVAL);
2288                 mdsize = *(__u32 *)val;
2289                 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
2290                         exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
2291                 max_easize = val;
2292                 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
2293                 RETURN(0);
2294         } else if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2295                 __u32 *default_easize;
2296
2297                 if (*vallen != sizeof(int))
2298                         RETURN(-EINVAL);
2299                 default_easize = val;
2300                 *default_easize = exp->exp_obd->u.cli.cl_default_mds_easize;
2301                 RETURN(0);
2302         } else if (KEY_IS(KEY_CONN_DATA)) {
2303                 struct obd_import *imp = class_exp2cliimp(exp);
2304                 struct obd_connect_data *data = val;
2305
2306                 if (*vallen != sizeof(*data))
2307                         RETURN(-EINVAL);
2308
2309                 *data = imp->imp_connect_data;
2310                 RETURN(0);
2311         } else if (KEY_IS(KEY_TGT_COUNT)) {
2312                 *((__u32 *)val) = 1;
2313                 RETURN(0);
2314         }
2315
2316         rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
2317
2318         RETURN(rc);
2319 }
2320
2321 static int mdc_fsync(struct obd_export *exp, const struct lu_fid *fid,
2322                      struct ptlrpc_request **request)
2323 {
2324         struct ptlrpc_request *req;
2325         int                    rc;
2326         ENTRY;
2327
2328         *request = NULL;
2329         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
2330         if (req == NULL)
2331                 RETURN(-ENOMEM);
2332
2333         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
2334         if (rc) {
2335                 ptlrpc_request_free(req);
2336                 RETURN(rc);
2337         }
2338
2339         mdc_pack_body(req, fid, 0, 0, -1, 0);
2340
2341         ptlrpc_request_set_replen(req);
2342
2343         rc = ptlrpc_queue_wait(req);
2344         if (rc)
2345                 ptlrpc_req_finished(req);
2346         else
2347                 *request = req;
2348         RETURN(rc);
2349 }
2350
2351 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
2352                             enum obd_import_event event)
2353 {
2354         struct client_obd *cli = &obd->u.cli;
2355         int rc = 0;
2356
2357         LASSERT(imp->imp_obd == obd);
2358
2359         switch (event) {
2360         case IMP_EVENT_DISCON:
2361                 spin_lock(&cli->cl_loi_list_lock);
2362                 cli->cl_avail_grant = 0;
2363                 cli->cl_lost_grant = 0;
2364                 spin_unlock(&cli->cl_loi_list_lock);
2365                 break;
2366         case IMP_EVENT_INACTIVE:
2367                 /*
2368                  * Flush current sequence to make client obtain new one
2369                  * from server in case of disconnect/reconnect.
2370                  */
2371                 down_read(&cli->cl_seq_rwsem);
2372                 if (cli->cl_seq)
2373                         seq_client_flush(cli->cl_seq);
2374                 up_read(&cli->cl_seq_rwsem);
2375
2376                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE);
2377                 break;
2378         case IMP_EVENT_INVALIDATE: {
2379                 struct ldlm_namespace *ns = obd->obd_namespace;
2380                 struct lu_env *env;
2381                 __u16 refcheck;
2382
2383                 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2384
2385                 env = cl_env_get(&refcheck);
2386                 if (!IS_ERR(env)) {
2387                         /* Reset grants. All pages go to failing rpcs due to
2388                          * the invalid import.
2389                          */
2390                         osc_io_unplug(env, cli, NULL);
2391
2392                         cfs_hash_for_each_nolock(ns->ns_rs_hash,
2393                                                  osc_ldlm_resource_invalidate,
2394                                                  env, 0);
2395                         cl_env_put(env, &refcheck);
2396                         ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2397                 } else {
2398                         rc = PTR_ERR(env);
2399                 }
2400                 break;
2401         }
2402         case IMP_EVENT_ACTIVE:
2403                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE);
2404                 /* redo the kuc registration after reconnecting */
2405                 if (rc == 0)
2406                         rc = mdc_kuc_reregister(imp);
2407                 break;
2408         case IMP_EVENT_OCD: {
2409                 struct obd_connect_data *ocd = &imp->imp_connect_data;
2410
2411                 if (OCD_HAS_FLAG(ocd, GRANT))
2412                         osc_init_grant(cli, ocd);
2413
2414                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD);
2415                 break;
2416         }
2417         case IMP_EVENT_DEACTIVATE:
2418         case IMP_EVENT_ACTIVATE:
2419                 break;
2420         default:
2421                 CERROR("Unknown import event %x\n", event);
2422                 LBUG();
2423         }
2424         RETURN(rc);
2425 }
2426
2427 int mdc_fid_alloc(const struct lu_env *env, struct obd_export *exp,
2428                   struct lu_fid *fid, struct md_op_data *op_data)
2429 {
2430         struct client_obd *cli = &exp->exp_obd->u.cli;
2431         int rc = -EIO;
2432
2433         ENTRY;
2434
2435         down_read(&cli->cl_seq_rwsem);
2436         if (cli->cl_seq)
2437                 rc = seq_client_alloc_fid(env, cli->cl_seq, fid);
2438         up_read(&cli->cl_seq_rwsem);
2439
2440         RETURN(rc);
2441 }
2442
2443 static struct obd_uuid *mdc_get_uuid(struct obd_export *exp)
2444 {
2445         struct client_obd *cli = &exp->exp_obd->u.cli;
2446         return &cli->cl_target_uuid;
2447 }
2448
2449 /**
2450  * Determine whether the lock can be canceled before replaying it during
2451  * recovery, non zero value will be return if the lock can be canceled,
2452  * or zero returned for not
2453  */
2454 static int mdc_cancel_weight(struct ldlm_lock *lock)
2455 {
2456         if (lock->l_resource->lr_type != LDLM_IBITS)
2457                 RETURN(0);
2458
2459         /* FIXME: if we ever get into a situation where there are too many
2460          * opened files with open locks on a single node, then we really
2461          * should replay these open locks to reget it */
2462         if (lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_OPEN)
2463                 RETURN(0);
2464
2465         RETURN(1);
2466 }
2467
2468 static int mdc_resource_inode_free(struct ldlm_resource *res)
2469 {
2470         if (res->lr_lvb_inode)
2471                 res->lr_lvb_inode = NULL;
2472
2473         return 0;
2474 }
2475
2476 static struct ldlm_valblock_ops inode_lvbo = {
2477         .lvbo_free = mdc_resource_inode_free
2478 };
2479
2480 static int mdc_llog_init(struct obd_device *obd)
2481 {
2482         struct obd_llog_group   *olg = &obd->obd_olg;
2483         struct llog_ctxt        *ctxt;
2484         int                      rc;
2485
2486         ENTRY;
2487
2488         rc = llog_setup(NULL, obd, olg, LLOG_CHANGELOG_REPL_CTXT, obd,
2489                         &llog_client_ops);
2490         if (rc < 0)
2491                 RETURN(rc);
2492
2493         ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
2494         llog_initiator_connect(ctxt);
2495         llog_ctxt_put(ctxt);
2496
2497         RETURN(0);
2498 }
2499
2500 static void mdc_llog_finish(struct obd_device *obd)
2501 {
2502         struct llog_ctxt *ctxt;
2503
2504         ENTRY;
2505
2506         ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
2507         if (ctxt != NULL)
2508                 llog_cleanup(NULL, ctxt);
2509
2510         EXIT;
2511 }
2512
2513 int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
2514 {
2515         int rc;
2516
2517         ENTRY;
2518
2519         rc = osc_setup_common(obd, cfg);
2520         if (rc < 0)
2521                 RETURN(rc);
2522
2523 #ifdef CONFIG_PROC_FS
2524         obd->obd_vars = lprocfs_mdc_obd_vars;
2525         lprocfs_obd_setup(obd, false);
2526         lprocfs_alloc_md_stats(obd, 0);
2527 #endif
2528
2529         sptlrpc_lprocfs_cliobd_attach(obd);
2530         ptlrpc_lprocfs_register_obd(obd);
2531
2532         ns_register_cancel(obd->obd_namespace, mdc_cancel_weight);
2533
2534         obd->obd_namespace->ns_lvbo = &inode_lvbo;
2535
2536         rc = mdc_llog_init(obd);
2537         if (rc) {
2538                 CERROR("%s: failed to setup llogging subsystems: rc = %d\n",
2539                        obd->obd_name, rc);
2540                 GOTO(err_llog_cleanup, rc);
2541         }
2542
2543         rc = mdc_changelog_cdev_init(obd);
2544         if (rc) {
2545                 CERROR("%s: failed to setup changelog char device: rc = %d\n",
2546                        obd->obd_name, rc);
2547                 GOTO(err_changelog_cleanup, rc);
2548         }
2549
2550         RETURN(rc);
2551
2552 err_changelog_cleanup:
2553         mdc_llog_finish(obd);
2554 err_llog_cleanup:
2555         ptlrpc_lprocfs_unregister_obd(obd);
2556         lprocfs_free_md_stats(obd);
2557
2558         osc_cleanup_common(obd);
2559         return rc;
2560 }
2561
2562 /* Initialize the default and maximum LOV EA sizes.  This allows
2563  * us to make MDS RPCs with large enough reply buffers to hold a default
2564  * sized EA without having to calculate this (via a call into the
2565  * LOV + OSCs) each time we make an RPC.  The maximum size is also tracked
2566  * but not used to avoid wastefully vmalloc()'ing large reply buffers when
2567  * a large number of stripes is possible.  If a larger reply buffer is
2568  * required it will be reallocated in the ptlrpc layer due to overflow.
2569  */
2570 static int mdc_init_ea_size(struct obd_export *exp, __u32 easize,
2571                             __u32 def_easize)
2572 {
2573         struct obd_device *obd = exp->exp_obd;
2574         struct client_obd *cli = &obd->u.cli;
2575         ENTRY;
2576
2577         if (cli->cl_max_mds_easize < easize)
2578                 cli->cl_max_mds_easize = easize;
2579
2580         if (cli->cl_default_mds_easize < def_easize)
2581                 cli->cl_default_mds_easize = def_easize;
2582
2583         RETURN(0);
2584 }
2585
2586 static int mdc_precleanup(struct obd_device *obd)
2587 {
2588         ENTRY;
2589
2590         osc_precleanup_common(obd);
2591         mdc_changelog_cdev_finish(obd);
2592
2593         obd_cleanup_client_import(obd);
2594         ptlrpc_lprocfs_unregister_obd(obd);
2595         lprocfs_free_md_stats(obd);
2596         mdc_llog_finish(obd);
2597         RETURN(0);
2598 }
2599
2600 static int mdc_cleanup(struct obd_device *obd)
2601 {
2602         return osc_cleanup_common(obd);
2603 }
2604
2605 int mdc_process_config(struct obd_device *obd, size_t len, void *buf)
2606 {
2607         struct lustre_cfg *lcfg = buf;
2608         int rc;
2609
2610         rc = class_process_proc_param(PARAM_MDC, obd->obd_vars, lcfg, obd);
2611         return (rc > 0 ? 0: rc);
2612 }
2613
2614 static struct obd_ops mdc_obd_ops = {
2615         .o_owner            = THIS_MODULE,
2616         .o_setup            = mdc_setup,
2617         .o_precleanup       = mdc_precleanup,
2618         .o_cleanup          = mdc_cleanup,
2619         .o_add_conn         = client_import_add_conn,
2620         .o_del_conn         = client_import_del_conn,
2621         .o_connect          = client_connect_import,
2622         .o_reconnect        = osc_reconnect,
2623         .o_disconnect       = osc_disconnect,
2624         .o_iocontrol        = mdc_iocontrol,
2625         .o_set_info_async   = mdc_set_info_async,
2626         .o_statfs           = mdc_statfs,
2627         .o_fid_init         = client_fid_init,
2628         .o_fid_fini         = client_fid_fini,
2629         .o_fid_alloc        = mdc_fid_alloc,
2630         .o_import_event     = mdc_import_event,
2631         .o_get_info         = mdc_get_info,
2632         .o_process_config   = mdc_process_config,
2633         .o_get_uuid         = mdc_get_uuid,
2634         .o_quotactl         = mdc_quotactl,
2635 };
2636
2637 static struct md_ops mdc_md_ops = {
2638         .m_get_root         = mdc_get_root,
2639         .m_null_inode       = mdc_null_inode,
2640         .m_close            = mdc_close,
2641         .m_create           = mdc_create,
2642         .m_enqueue          = mdc_enqueue,
2643         .m_getattr          = mdc_getattr,
2644         .m_getattr_name     = mdc_getattr_name,
2645         .m_intent_lock      = mdc_intent_lock,
2646         .m_link             = mdc_link,
2647         .m_rename           = mdc_rename,
2648         .m_setattr          = mdc_setattr,
2649         .m_setxattr         = mdc_setxattr,
2650         .m_getxattr         = mdc_getxattr,
2651         .m_fsync                = mdc_fsync,
2652         .m_file_resync          = mdc_file_resync,
2653         .m_read_page            = mdc_read_page,
2654         .m_unlink           = mdc_unlink,
2655         .m_cancel_unused    = mdc_cancel_unused,
2656         .m_init_ea_size     = mdc_init_ea_size,
2657         .m_set_lock_data    = mdc_set_lock_data,
2658         .m_lock_match       = mdc_lock_match,
2659         .m_get_lustre_md    = mdc_get_lustre_md,
2660         .m_free_lustre_md   = mdc_free_lustre_md,
2661         .m_set_open_replay_data = mdc_set_open_replay_data,
2662         .m_clear_open_replay_data = mdc_clear_open_replay_data,
2663         .m_intent_getattr_async = mdc_intent_getattr_async,
2664         .m_revalidate_lock      = mdc_revalidate_lock
2665 };
2666
2667 static int __init mdc_init(void)
2668 {
2669         return class_register_type(&mdc_obd_ops, &mdc_md_ops, true, NULL,
2670                                    LUSTRE_MDC_NAME, &mdc_device_type);
2671 }
2672
2673 static void __exit mdc_exit(void)
2674 {
2675         class_unregister_type(LUSTRE_MDC_NAME);
2676 }
2677
2678 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
2679 MODULE_DESCRIPTION("Lustre Metadata Client");
2680 MODULE_VERSION(LUSTRE_VERSION_STRING);
2681 MODULE_LICENSE("GPL");
2682
2683 module_init(mdc_init);
2684 module_exit(mdc_exit);