lustre/mdc/mdc_request.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2016, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  */
  32
  33 #define DEBUG_SUBSYSTEM S_MDC
  34
  35 #include <linux/init.h>
  36 #include <linux/kthread.h>
  37 #include <linux/module.h>
  38 #include <linux/pagemap.h>
  39 #include <linux/user_namespace.h>
  40 #include <linux/utsname.h>
  41 #ifdef HAVE_UIDGID_HEADER
  42 # include <linux/uidgid.h>
  43 #endif
  44
  45 #include <lustre/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_ioctl.h>
  53 #include <lustre_kernelcomm.h>
  54 #include <lustre_lmv.h>
  55 #include <lustre_log.h>
  56 #include <uapi/linux/lustre_param.h>
  57 #include <lustre_swab.h>
  58 #include <obd_class.h>
  59
  60 #include "mdc_internal.h"
  61
  62 #define REQUEST_MINOR 244
  63
  64 static int mdc_cleanup(struct obd_device *obd);
  65
  66 static inline int mdc_queue_wait(struct ptlrpc_request *req)
  67 {
  68         struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
  69         int rc;
  70
  71         /* obd_get_request_slot() ensures that this client has no more
  72          * than cl_max_rpcs_in_flight RPCs simultaneously inf light
  73          * against an MDT. */
  74         rc = obd_get_request_slot(cli);
  75         if (rc != 0)
  76                 return rc;
  77
  78         rc = ptlrpc_queue_wait(req);
  79         obd_put_request_slot(cli);
  80
  81         return rc;
  82 }
  83
  84 /*
  85  * Send MDS_GET_ROOT RPC to fetch root FID.
  86  *
  87  * If \a fileset is not NULL it should contain a subdirectory off
  88  * the ROOT/ directory to be mounted on the client. Return the FID
  89  * of the subdirectory to the client to mount onto its mountpoint.
  90  *
  91  * \param[in]   imp     MDC import
  92  * \param[in]   fileset fileset name, which could be NULL
  93  * \param[out]  rootfid root FID of this mountpoint
  94  * \param[out]  pc      root capa will be unpacked and saved in this pointer
  95  *
  96  * \retval      0 on success, negative errno on failure
  97  */
  98 static int mdc_get_root(struct obd_export *exp, const char *fileset,
  99                          struct lu_fid *rootfid)
 100 {
 101         struct ptlrpc_request   *req;
 102         struct mdt_body         *body;
 103         int                      rc;
 104
 105         ENTRY;
 106
 107         if (fileset && !(exp_connect_flags(exp) & OBD_CONNECT_SUBTREE))
 108                 RETURN(-ENOTSUPP);
 109
 110         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 111                                 &RQF_MDS_GET_ROOT);
 112         if (req == NULL)
 113                 RETURN(-ENOMEM);
 114
 115         if (fileset != NULL)
 116                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 117                                      strlen(fileset) + 1);
 118         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_ROOT);
 119         if (rc) {
 120                 ptlrpc_request_free(req);
 121                 RETURN(rc);
 122         }
 123         mdc_pack_body(req, NULL, 0, 0, -1, 0);
 124         if (fileset != NULL) {
 125                 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 126
 127                 memcpy(name, fileset, strlen(fileset));
 128         }
 129         lustre_msg_add_flags(req->rq_reqmsg, LUSTRE_IMP_FULL);
 130         req->rq_send_state = LUSTRE_IMP_FULL;
 131
 132         ptlrpc_request_set_replen(req);
 133
 134         rc = ptlrpc_queue_wait(req);
 135         if (rc)
 136                 GOTO(out, rc);
 137
 138         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 139         if (body == NULL)
 140                 GOTO(out, rc = -EPROTO);
 141
 142         *rootfid = body->mbo_fid1;
 143         CDEBUG(D_NET, "root fid="DFID", last_committed=%llu\n",
 144                PFID(rootfid), lustre_msg_get_last_committed(req->rq_repmsg));
 145         EXIT;
 146 out:
 147         ptlrpc_req_finished(req);
 148
 149         return rc;
 150 }
 151
 152 /*
 153  * This function now is known to always saying that it will receive 4 buffers
 154  * from server. Even for cases when acl_size and md_size is zero, RPC header
 155  * will contain 4 fields and RPC itself will contain zero size fields. This is
 156  * because mdt_getattr*() _always_ returns 4 fields, but if acl is not needed
 157  * and thus zero, it shrinks it, making zero size. The same story about
 158  * md_size. And this is course of problem when client waits for smaller number
 159  * of fields. This issue will be fixed later when client gets aware of RPC
 160  * layouts.  --umka
 161  */
 162 static int mdc_getattr_common(struct obd_export *exp,
 163                               struct ptlrpc_request *req)
 164 {
 165         struct req_capsule *pill = &req->rq_pill;
 166         struct mdt_body    *body;
 167         void               *eadata;
 168         int                 rc;
 169         ENTRY;
 170
 171         /* Request message already built. */
 172         rc = ptlrpc_queue_wait(req);
 173         if (rc != 0)
 174                 RETURN(rc);
 175
 176         /* sanity check for the reply */
 177         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 178         if (body == NULL)
 179                 RETURN(-EPROTO);
 180
 181         CDEBUG(D_NET, "mode: %o\n", body->mbo_mode);
 182
 183         mdc_update_max_ea_from_body(exp, body);
 184         if (body->mbo_eadatasize != 0) {
 185                 eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
 186                                                       body->mbo_eadatasize);
 187                 if (eadata == NULL)
 188                         RETURN(-EPROTO);
 189         }
 190
 191         RETURN(0);
 192 }
 193
 194 static int mdc_getattr(struct obd_export *exp, struct md_op_data *op_data,
 195                        struct ptlrpc_request **request)
 196 {
 197         struct ptlrpc_request *req;
 198         int                    rc;
 199         ENTRY;
 200
 201         /* Single MDS without an LMV case */
 202         if (op_data->op_flags & MF_GET_MDT_IDX) {
 203                 op_data->op_mds = 0;
 204                 RETURN(0);
 205         }
 206         *request = NULL;
 207         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
 208         if (req == NULL)
 209                 RETURN(-ENOMEM);
 210
 211         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
 212         if (rc) {
 213                 ptlrpc_request_free(req);
 214                 RETURN(rc);
 215         }
 216
 217         mdc_pack_body(req, &op_data->op_fid1, op_data->op_valid,
 218                       op_data->op_mode, -1, 0);
 219
 220         req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
 221                              req->rq_import->imp_connect_data.ocd_max_easize);
 222         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 223                              op_data->op_mode);
 224         ptlrpc_request_set_replen(req);
 225
 226         rc = mdc_getattr_common(exp, req);
 227         if (rc)
 228                 ptlrpc_req_finished(req);
 229         else
 230                 *request = req;
 231         RETURN(rc);
 232 }
 233
 234 static int mdc_getattr_name(struct obd_export *exp, struct md_op_data *op_data,
 235                             struct ptlrpc_request **request)
 236 {
 237         struct ptlrpc_request *req;
 238         int                    rc;
 239         ENTRY;
 240
 241         *request = NULL;
 242         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 243                                    &RQF_MDS_GETATTR_NAME);
 244         if (req == NULL)
 245                 RETURN(-ENOMEM);
 246
 247         req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 248                              op_data->op_namelen + 1);
 249
 250         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR_NAME);
 251         if (rc) {
 252                 ptlrpc_request_free(req);
 253                 RETURN(rc);
 254         }
 255
 256         mdc_pack_body(req, &op_data->op_fid1, op_data->op_valid,
 257                       op_data->op_mode, op_data->op_suppgids[0], 0);
 258
 259         if (op_data->op_name) {
 260                 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 261                 LASSERT(strnlen(op_data->op_name, op_data->op_namelen) ==
 262                                 op_data->op_namelen);
 263                 memcpy(name, op_data->op_name, op_data->op_namelen);
 264         }
 265
 266         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 267                              op_data->op_mode);
 268         req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
 269                              req->rq_import->imp_connect_data.ocd_max_easize);
 270         ptlrpc_request_set_replen(req);
 271
 272         rc = mdc_getattr_common(exp, req);
 273         if (rc)
 274                 ptlrpc_req_finished(req);
 275         else
 276                 *request = req;
 277         RETURN(rc);
 278 }
 279
 280 static int mdc_xattr_common(struct obd_export *exp,const struct req_format *fmt,
 281                             const struct lu_fid *fid, int opcode, u64 valid,
 282                             const char *xattr_name, const char *input,
 283                             int input_size, int output_size, int flags,
 284                             __u32 suppgid, struct ptlrpc_request **request)
 285 {
 286         struct ptlrpc_request *req;
 287         int   xattr_namelen = 0;
 288         char *tmp;
 289         int   rc;
 290         ENTRY;
 291
 292         *request = NULL;
 293         req = ptlrpc_request_alloc(class_exp2cliimp(exp), fmt);
 294         if (req == NULL)
 295                 RETURN(-ENOMEM);
 296
 297         if (xattr_name) {
 298                 xattr_namelen = strlen(xattr_name) + 1;
 299                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 300                                      xattr_namelen);
 301         }
 302         if (input_size) {
 303                 LASSERT(input);
 304                 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
 305                                      input_size);
 306         }
 307
 308         /* Flush local XATTR locks to get rid of a possible cancel RPC */
 309         if (opcode == MDS_REINT && fid_is_sane(fid) &&
 310             exp->exp_connect_data.ocd_ibits_known & MDS_INODELOCK_XATTR) {
 311                 struct list_head cancels = LIST_HEAD_INIT(cancels);
 312                 int count;
 313
 314                 /* Without that packing would fail */
 315                 if (input_size == 0)
 316                         req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
 317                                              RCL_CLIENT, 0);
 318
 319                 count = mdc_resource_get_unused(exp, fid,
 320                                                 &cancels, LCK_EX,
 321                                                 MDS_INODELOCK_XATTR);
 322
 323                 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
 324                 if (rc) {
 325                         ptlrpc_request_free(req);
 326                         RETURN(rc);
 327                 }
 328         } else {
 329                 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode);
 330                 if (rc) {
 331                         ptlrpc_request_free(req);
 332                         RETURN(rc);
 333                 }
 334         }
 335
 336         if (opcode == MDS_REINT) {
 337                 struct mdt_rec_setxattr *rec;
 338
 339                 CLASSERT(sizeof(struct mdt_rec_setxattr) ==
 340                          sizeof(struct mdt_rec_reint));
 341                 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
 342                 rec->sx_opcode = REINT_SETXATTR;
 343                 rec->sx_fsuid  = from_kuid(&init_user_ns, current_fsuid());
 344                 rec->sx_fsgid  = from_kgid(&init_user_ns, current_fsgid());
 345                 rec->sx_cap    = cfs_curproc_cap_pack();
 346                 rec->sx_suppgid1 = suppgid;
 347                 rec->sx_suppgid2 = -1;
 348                 rec->sx_fid    = *fid;
 349                 rec->sx_valid  = valid | OBD_MD_FLCTIME;
 350                 rec->sx_time   = ktime_get_real_seconds();
 351                 rec->sx_size   = output_size;
 352                 rec->sx_flags  = flags;
 353         } else {
 354                 mdc_pack_body(req, fid, valid, output_size, suppgid, flags);
 355         }
 356
 357         if (xattr_name) {
 358                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 359                 memcpy(tmp, xattr_name, xattr_namelen);
 360         }
 361         if (input_size) {
 362                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
 363                 memcpy(tmp, input, input_size);
 364         }
 365
 366         if (req_capsule_has_field(&req->rq_pill, &RMF_EADATA, RCL_SERVER))
 367                 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
 368                                      RCL_SERVER, output_size);
 369         ptlrpc_request_set_replen(req);
 370
 371         /* make rpc */
 372         if (opcode == MDS_REINT)
 373                 mdc_get_mod_rpc_slot(req, NULL);
 374
 375         rc = ptlrpc_queue_wait(req);
 376
 377         if (opcode == MDS_REINT)
 378                 mdc_put_mod_rpc_slot(req, NULL);
 379
 380         if (rc)
 381                 ptlrpc_req_finished(req);
 382         else
 383                 *request = req;
 384         RETURN(rc);
 385 }
 386
 387 static int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
 388                         u64 valid, const char *xattr_name,
 389                         const char *input, int input_size, int output_size,
 390                         int flags, __u32 suppgid,
 391                         struct ptlrpc_request **request)
 392 {
 393         return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
 394                                 fid, MDS_REINT, valid, xattr_name,
 395                                 input, input_size, output_size, flags,
 396                                 suppgid, request);
 397 }
 398
 399 static int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
 400                         u64 valid, const char *xattr_name,
 401                         const char *input, int input_size, int output_size,
 402                         int flags, struct ptlrpc_request **request)
 403 {
 404         return mdc_xattr_common(exp, &RQF_MDS_GETXATTR,
 405                                 fid, MDS_GETXATTR, valid, xattr_name,
 406                                 input, input_size, output_size, flags,
 407                                 -1, request);
 408 }
 409
 410 #ifdef CONFIG_FS_POSIX_ACL
 411 static int mdc_unpack_acl(struct ptlrpc_request *req, struct lustre_md *md)
 412 {
 413         struct req_capsule     *pill = &req->rq_pill;
 414         struct mdt_body        *body = md->body;
 415         struct posix_acl       *acl;
 416         void                   *buf;
 417         int                     rc;
 418         ENTRY;
 419
 420         if (!body->mbo_aclsize)
 421                 RETURN(0);
 422
 423         buf = req_capsule_server_sized_get(pill, &RMF_ACL, body->mbo_aclsize);
 424
 425         if (!buf)
 426                 RETURN(-EPROTO);
 427
 428         acl = posix_acl_from_xattr(&init_user_ns, buf, body->mbo_aclsize);
 429         if (acl == NULL)
 430                 RETURN(0);
 431         if (IS_ERR(acl)) {
 432                 rc = PTR_ERR(acl);
 433                 CERROR("convert xattr to acl: %d\n", rc);
 434                 RETURN(rc);
 435         }
 436
 437         rc = posix_acl_valid(&init_user_ns, acl);
 438         if (rc) {
 439                 CERROR("validate acl: %d\n", rc);
 440                 posix_acl_release(acl);
 441                 RETURN(rc);
 442         }
 443
 444         md->posix_acl = acl;
 445         RETURN(0);
 446 }
 447 #else
 448 #define mdc_unpack_acl(req, md) 0
 449 #endif
 450
 451 int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
 452                       struct obd_export *dt_exp, struct obd_export *md_exp,
 453                       struct lustre_md *md)
 454 {
 455         struct req_capsule *pill = &req->rq_pill;
 456         int rc;
 457         ENTRY;
 458
 459         LASSERT(md);
 460         memset(md, 0, sizeof(*md));
 461
 462         md->body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 463         LASSERT(md->body != NULL);
 464
 465         if (md->body->mbo_valid & OBD_MD_FLEASIZE) {
 466                 if (!S_ISREG(md->body->mbo_mode)) {
 467                         CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, should be a "
 468                                "regular file, but is not\n");
 469                         GOTO(out, rc = -EPROTO);
 470                 }
 471
 472                 if (md->body->mbo_eadatasize == 0) {
 473                         CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, "
 474                                "but eadatasize 0\n");
 475                         GOTO(out, rc = -EPROTO);
 476                 }
 477
 478                 md->layout.lb_len = md->body->mbo_eadatasize;
 479                 md->layout.lb_buf = req_capsule_server_sized_get(pill,
 480                                                         &RMF_MDT_MD,
 481                                                         md->layout.lb_len);
 482                 if (md->layout.lb_buf == NULL)
 483                         GOTO(out, rc = -EPROTO);
 484         } else if (md->body->mbo_valid & OBD_MD_FLDIREA) {
 485                 const union lmv_mds_md *lmv;
 486                 size_t lmv_size;
 487
 488                 if (!S_ISDIR(md->body->mbo_mode)) {
 489                         CDEBUG(D_INFO, "OBD_MD_FLDIREA set, should be a "
 490                                "directory, but is not\n");
 491                         GOTO(out, rc = -EPROTO);
 492                 }
 493
 494                 lmv_size = md->body->mbo_eadatasize;
 495                 if (lmv_size == 0) {
 496                         CDEBUG(D_INFO, "OBD_MD_FLDIREA is set, "
 497                                "but eadatasize 0\n");
 498                         RETURN(-EPROTO);
 499                 }
 500
 501                 if (md->body->mbo_valid & OBD_MD_MEA) {
 502                         lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
 503                                                            lmv_size);
 504                         if (lmv == NULL)
 505                                 GOTO(out, rc = -EPROTO);
 506
 507                         rc = md_unpackmd(md_exp, &md->lmv, lmv, lmv_size);
 508                         if (rc < 0)
 509                                 GOTO(out, rc);
 510
 511                         if (rc < (typeof(rc))sizeof(*md->lmv)) {
 512                                 CDEBUG(D_INFO, "size too small:  "
 513                                        "rc < sizeof(*md->lmv) (%d < %d)\n",
 514                                         rc, (int)sizeof(*md->lmv));
 515                                 GOTO(out, rc = -EPROTO);
 516                         }
 517                 }
 518         }
 519         rc = 0;
 520
 521         if (md->body->mbo_valid & OBD_MD_FLACL) {
 522                 /* for ACL, it's possible that FLACL is set but aclsize is zero.
 523                  * only when aclsize != 0 there's an actual segment for ACL
 524                  * in reply buffer.
 525                  */
 526                 if (md->body->mbo_aclsize) {
 527                         rc = mdc_unpack_acl(req, md);
 528                         if (rc)
 529                                 GOTO(out, rc);
 530 #ifdef CONFIG_FS_POSIX_ACL
 531                 } else {
 532                         md->posix_acl = NULL;
 533 #endif
 534                 }
 535         }
 536
 537         EXIT;
 538 out:
 539         if (rc) {
 540 #ifdef CONFIG_FS_POSIX_ACL
 541                 posix_acl_release(md->posix_acl);
 542 #endif
 543         }
 544         return rc;
 545 }
 546
 547 int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
 548 {
 549         ENTRY;
 550         RETURN(0);
 551 }
 552
 553 void mdc_replay_open(struct ptlrpc_request *req)
 554 {
 555         struct md_open_data *mod = req->rq_cb_data;
 556         struct ptlrpc_request *close_req;
 557         struct obd_client_handle *och;
 558         struct lustre_handle old;
 559         struct mdt_body *body;
 560         ENTRY;
 561
 562         if (mod == NULL) {
 563                 DEBUG_REQ(D_ERROR, req,
 564                           "Can't properly replay without open data.");
 565                 EXIT;
 566                 return;
 567         }
 568
 569         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 570         LASSERT(body != NULL);
 571
 572         spin_lock(&req->rq_lock);
 573         och = mod->mod_och;
 574         if (och && och->och_fh.cookie)
 575                 req->rq_early_free_repbuf = 1;
 576         else
 577                 req->rq_early_free_repbuf = 0;
 578         spin_unlock(&req->rq_lock);
 579
 580         if (req->rq_early_free_repbuf) {
 581                 struct lustre_handle *file_fh;
 582
 583                 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
 584
 585                 file_fh = &och->och_fh;
 586                 CDEBUG(D_HA, "updating handle from %#llx to %#llx\n",
 587                        file_fh->cookie, body->mbo_handle.cookie);
 588                 old = *file_fh;
 589                 *file_fh = body->mbo_handle;
 590         }
 591
 592         close_req = mod->mod_close_req;
 593         if (close_req) {
 594                 __u32 opc = lustre_msg_get_opc(close_req->rq_reqmsg);
 595                 struct mdt_ioepoch *epoch;
 596
 597                 LASSERT(opc == MDS_CLOSE);
 598                 epoch = req_capsule_client_get(&close_req->rq_pill,
 599                                                &RMF_MDT_EPOCH);
 600                 LASSERT(epoch);
 601
 602                 if (req->rq_early_free_repbuf)
 603                         LASSERT(!memcmp(&old, &epoch->mio_handle, sizeof(old)));
 604
 605                 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
 606                 epoch->mio_handle = body->mbo_handle;
 607         }
 608         EXIT;
 609 }
 610
 611 void mdc_commit_open(struct ptlrpc_request *req)
 612 {
 613         struct md_open_data *mod = req->rq_cb_data;
 614         if (mod == NULL)
 615                 return;
 616
 617         /**
 618          * No need to touch md_open_data::mod_och, it holds a reference on
 619          * \var mod and will zero references to each other, \var mod will be
 620          * freed after that when md_open_data::mod_och will put the reference.
 621          */
 622
 623         /**
 624          * Do not let open request to disappear as it still may be needed
 625          * for close rpc to happen (it may happen on evict only, otherwise
 626          * ptlrpc_request::rq_replay does not let mdc_commit_open() to be
 627          * called), just mark this rpc as committed to distinguish these 2
 628          * cases, see mdc_close() for details. The open request reference will
 629          * be put along with freeing \var mod.
 630          */
 631         ptlrpc_request_addref(req);
 632         spin_lock(&req->rq_lock);
 633         req->rq_committed = 1;
 634         spin_unlock(&req->rq_lock);
 635         req->rq_cb_data = NULL;
 636         obd_mod_put(mod);
 637 }
 638
 639 int mdc_set_open_replay_data(struct obd_export *exp,
 640                              struct obd_client_handle *och,
 641                              struct lookup_intent *it)
 642 {
 643         struct md_open_data     *mod;
 644         struct mdt_rec_create   *rec;
 645         struct mdt_body         *body;
 646         struct ptlrpc_request   *open_req = it->it_request;
 647         struct obd_import       *imp = open_req->rq_import;
 648         ENTRY;
 649
 650         if (!open_req->rq_replay)
 651                 RETURN(0);
 652
 653         rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
 654         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
 655         LASSERT(rec != NULL);
 656         /* Incoming message in my byte order (it's been swabbed). */
 657         /* Outgoing messages always in my byte order. */
 658         LASSERT(body != NULL);
 659
 660         /* Only if the import is replayable, we set replay_open data */
 661         if (och && imp->imp_replayable) {
 662                 mod = obd_mod_alloc();
 663                 if (mod == NULL) {
 664                         DEBUG_REQ(D_ERROR, open_req,
 665                                   "Can't allocate md_open_data");
 666                         RETURN(0);
 667                 }
 668
 669                 /**
 670                  * Take a reference on \var mod, to be freed on mdc_close().
 671                  * It protects \var mod from being freed on eviction (commit
 672                  * callback is called despite rq_replay flag).
 673                  * Another reference for \var och.
 674                  */
 675                 obd_mod_get(mod);
 676                 obd_mod_get(mod);
 677
 678                 spin_lock(&open_req->rq_lock);
 679                 och->och_mod = mod;
 680                 mod->mod_och = och;
 681                 mod->mod_is_create = it_disposition(it, DISP_OPEN_CREATE) ||
 682                                      it_disposition(it, DISP_OPEN_STRIPE);
 683                 mod->mod_open_req = open_req;
 684                 open_req->rq_cb_data = mod;
 685                 open_req->rq_commit_cb = mdc_commit_open;
 686                 open_req->rq_early_free_repbuf = 1;
 687                 spin_unlock(&open_req->rq_lock);
 688         }
 689
 690         rec->cr_fid2 = body->mbo_fid1;
 691         rec->cr_ioepoch = body->mbo_ioepoch;
 692         rec->cr_old_handle.cookie = body->mbo_handle.cookie;
 693         open_req->rq_replay_cb = mdc_replay_open;
 694         if (!fid_is_sane(&body->mbo_fid1)) {
 695                 DEBUG_REQ(D_ERROR, open_req, "Saving replay request with "
 696                           "insane fid");
 697                 LBUG();
 698         }
 699
 700         DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
 701         RETURN(0);
 702 }
 703
 704 static void mdc_free_open(struct md_open_data *mod)
 705 {
 706         int committed = 0;
 707
 708         if (mod->mod_is_create == 0 &&
 709             imp_connect_disp_stripe(mod->mod_open_req->rq_import))
 710                 committed = 1;
 711
 712         /**
 713          * No reason to asssert here if the open request has
 714          * rq_replay == 1. It means that mdc_close failed, and
 715          * close request wasn`t sent. It is not fatal to client.
 716          * The worst thing is eviction if the client gets open lock
 717          **/
 718
 719         DEBUG_REQ(D_RPCTRACE, mod->mod_open_req, "free open request rq_replay"
 720                   "= %d\n", mod->mod_open_req->rq_replay);
 721
 722         ptlrpc_request_committed(mod->mod_open_req, committed);
 723         if (mod->mod_close_req)
 724                 ptlrpc_request_committed(mod->mod_close_req, committed);
 725 }
 726
 727 int mdc_clear_open_replay_data(struct obd_export *exp,
 728                                struct obd_client_handle *och)
 729 {
 730         struct md_open_data *mod = och->och_mod;
 731         ENTRY;
 732
 733         /**
 734          * It is possible to not have \var mod in a case of eviction between
 735          * lookup and ll_file_open().
 736          **/
 737         if (mod == NULL)
 738                 RETURN(0);
 739
 740         LASSERT(mod != LP_POISON);
 741         LASSERT(mod->mod_open_req != NULL);
 742
 743         spin_lock(&mod->mod_open_req->rq_lock);
 744         if (mod->mod_och)
 745                 mod->mod_och->och_fh.cookie = 0;
 746         mod->mod_open_req->rq_early_free_repbuf = 0;
 747         spin_unlock(&mod->mod_open_req->rq_lock);
 748         mdc_free_open(mod);
 749
 750         mod->mod_och = NULL;
 751         och->och_mod = NULL;
 752         obd_mod_put(mod);
 753
 754         RETURN(0);
 755 }
 756
 757 static int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
 758                      struct md_open_data *mod, struct ptlrpc_request **request)
 759 {
 760         struct obd_device     *obd = class_exp2obd(exp);
 761         struct ptlrpc_request *req;
 762         struct req_format     *req_fmt;
 763         int                    rc;
 764         int                    saved_rc = 0;
 765         ENTRY;
 766
 767         if (op_data->op_bias & MDS_HSM_RELEASE) {
 768                 req_fmt = &RQF_MDS_INTENT_CLOSE;
 769
 770                 /* allocate a FID for volatile file */
 771                 rc = mdc_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
 772                 if (rc < 0) {
 773                         CERROR("%s: "DFID" failed to allocate FID: %d\n",
 774                                obd->obd_name, PFID(&op_data->op_fid1), rc);
 775                         /* save the errcode and proceed to close */
 776                         saved_rc = rc;
 777                 }
 778         } else if (op_data->op_bias & MDS_CLOSE_LAYOUT_SWAP) {
 779                 req_fmt = &RQF_MDS_INTENT_CLOSE;
 780         } else {
 781                 req_fmt = &RQF_MDS_CLOSE;
 782         }
 783
 784         *request = NULL;
 785         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_CLOSE))
 786                 req = NULL;
 787         else
 788                 req = ptlrpc_request_alloc(class_exp2cliimp(exp), req_fmt);
 789
 790         /* Ensure that this close's handle is fixed up during replay. */
 791         if (likely(mod != NULL)) {
 792                 LASSERTF(mod->mod_open_req != NULL &&
 793                          mod->mod_open_req->rq_type != LI_POISON,
 794                          "POISONED open %p!\n", mod->mod_open_req);
 795
 796                 mod->mod_close_req = req;
 797
 798                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
 799                 /* We no longer want to preserve this open for replay even
 800                  * though the open was committed. b=3632, b=3633 */
 801                 spin_lock(&mod->mod_open_req->rq_lock);
 802                 mod->mod_open_req->rq_replay = 0;
 803                 spin_unlock(&mod->mod_open_req->rq_lock);
 804         } else {
 805                 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
 806         }
 807         if (req == NULL) {
 808                 /**
 809                  * TODO: repeat close after errors
 810                  */
 811                 CWARN("%s: close of FID "DFID" failed, file reference will be "
 812                       "dropped when this client unmounts or is evicted\n",
 813                       obd->obd_name, PFID(&op_data->op_fid1));
 814                 GOTO(out, rc = -ENOMEM);
 815         }
 816
 817         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
 818         if (rc) {
 819                 ptlrpc_request_free(req);
 820                 req = NULL;
 821                 GOTO(out, rc);
 822         }
 823
 824         /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
 825          * portal whose threads are not taking any DLM locks and are therefore
 826          * always progressing */
 827         req->rq_request_portal = MDS_READPAGE_PORTAL;
 828         ptlrpc_at_set_req_timeout(req);
 829
 830
 831         mdc_close_pack(req, op_data);
 832
 833         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 834                              obd->u.cli.cl_default_mds_easize);
 835
 836         ptlrpc_request_set_replen(req);
 837
 838         mdc_get_mod_rpc_slot(req, NULL);
 839         rc = ptlrpc_queue_wait(req);
 840         mdc_put_mod_rpc_slot(req, NULL);
 841
 842         if (req->rq_repmsg == NULL) {
 843                 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
 844                        req->rq_status);
 845                 if (rc == 0)
 846                         rc = req->rq_status ?: -EIO;
 847         } else if (rc == 0 || rc == -EAGAIN) {
 848                 struct mdt_body *body;
 849
 850                 rc = lustre_msg_get_status(req->rq_repmsg);
 851                 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
 852                         DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
 853                                   "= %d", rc);
 854                         if (rc > 0)
 855                                 rc = -rc;
 856                 }
 857                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 858                 if (body == NULL)
 859                         rc = -EPROTO;
 860         } else if (rc == -ESTALE) {
 861                 /**
 862                  * it can be allowed error after 3633 if open was committed and
 863                  * server failed before close was sent. Let's check if mod
 864                  * exists and return no error in that case
 865                  */
 866                 if (mod) {
 867                         DEBUG_REQ(D_HA, req, "Reset ESTALE = %d", rc);
 868                         LASSERT(mod->mod_open_req != NULL);
 869                         if (mod->mod_open_req->rq_committed)
 870                                 rc = 0;
 871                 }
 872         }
 873
 874 out:
 875         if (mod) {
 876                 if (rc != 0)
 877                         mod->mod_close_req = NULL;
 878                 /* Since now, mod is accessed through open_req only,
 879                  * thus close req does not keep a reference on mod anymore. */
 880                 obd_mod_put(mod);
 881         }
 882         *request = req;
 883
 884         RETURN(rc < 0 ? rc : saved_rc);
 885 }
 886
 887 static int mdc_getpage(struct obd_export *exp, const struct lu_fid *fid,
 888                        u64 offset, struct page **pages, int npages,
 889                        struct ptlrpc_request **request)
 890 {
 891         struct ptlrpc_request   *req;
 892         struct ptlrpc_bulk_desc *desc;
 893         int                      i;
 894         wait_queue_head_t        waitq;
 895         int                      resends = 0;
 896         struct l_wait_info       lwi;
 897         int                      rc;
 898         ENTRY;
 899
 900         *request = NULL;
 901         init_waitqueue_head(&waitq);
 902
 903 restart_bulk:
 904         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
 905         if (req == NULL)
 906                 RETURN(-ENOMEM);
 907
 908         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
 909         if (rc) {
 910                 ptlrpc_request_free(req);
 911                 RETURN(rc);
 912         }
 913
 914         req->rq_request_portal = MDS_READPAGE_PORTAL;
 915         ptlrpc_at_set_req_timeout(req);
 916
 917         desc = ptlrpc_prep_bulk_imp(req, npages, 1,
 918                                     PTLRPC_BULK_PUT_SINK | PTLRPC_BULK_BUF_KIOV,
 919                                     MDS_BULK_PORTAL,
 920                                     &ptlrpc_bulk_kiov_pin_ops);
 921         if (desc == NULL) {
 922                 ptlrpc_req_finished(req);
 923                 RETURN(-ENOMEM);
 924         }
 925
 926         /* NB req now owns desc and will free it when it gets freed */
 927         for (i = 0; i < npages; i++)
 928                 desc->bd_frag_ops->add_kiov_frag(desc, pages[i], 0,
 929                                                  PAGE_SIZE);
 930
 931         mdc_readdir_pack(req, offset, PAGE_SIZE * npages, fid);
 932
 933         ptlrpc_request_set_replen(req);
 934         rc = ptlrpc_queue_wait(req);
 935         if (rc) {
 936                 ptlrpc_req_finished(req);
 937                 if (rc != -ETIMEDOUT)
 938                         RETURN(rc);
 939
 940                 resends++;
 941                 if (!client_should_resend(resends, &exp->exp_obd->u.cli)) {
 942                         CERROR("%s: too many resend retries: rc = %d\n",
 943                                exp->exp_obd->obd_name, -EIO);
 944                         RETURN(-EIO);
 945                 }
 946                 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(resends), NULL, NULL,
 947                                        NULL);
 948                 l_wait_event(waitq, 0, &lwi);
 949
 950                 goto restart_bulk;
 951         }
 952
 953         rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
 954                                           req->rq_bulk->bd_nob_transferred);
 955         if (rc < 0) {
 956                 ptlrpc_req_finished(req);
 957                 RETURN(rc);
 958         }
 959
 960         if (req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK) {
 961                 CERROR("%s: unexpected bytes transferred: %d (%ld expected)\n",
 962                        exp->exp_obd->obd_name, req->rq_bulk->bd_nob_transferred,
 963                        PAGE_SIZE * npages);
 964                 ptlrpc_req_finished(req);
 965                 RETURN(-EPROTO);
 966         }
 967
 968         *request = req;
 969         RETURN(0);
 970 }
 971
 972 static void mdc_release_page(struct page *page, int remove)
 973 {
 974         if (remove) {
 975                 lock_page(page);
 976                 if (likely(page->mapping != NULL))
 977                         truncate_complete_page(page->mapping, page);
 978                 unlock_page(page);
 979         }
 980         put_page(page);
 981 }
 982
 983 static struct page *mdc_page_locate(struct address_space *mapping, __u64 *hash,
 984                                     __u64 *start, __u64 *end, int hash64)
 985 {
 986         /*
 987          * Complement of hash is used as an index so that
 988          * radix_tree_gang_lookup() can be used to find a page with starting
 989          * hash _smaller_ than one we are looking for.
 990          */
 991         unsigned long offset = hash_x_index(*hash, hash64);
 992         struct page *page;
 993         int found;
 994
 995         spin_lock_irq(&mapping->tree_lock);
 996         found = radix_tree_gang_lookup(&mapping->page_tree,
 997                                        (void **)&page, offset, 1);
 998         if (found > 0 && !radix_tree_exceptional_entry(page)) {
 999                 struct lu_dirpage *dp;
1000
1001                 get_page(page);
1002                 spin_unlock_irq(&mapping->tree_lock);
1003                 /*
1004                  * In contrast to find_lock_page() we are sure that directory
1005                  * page cannot be truncated (while DLM lock is held) and,
1006                  * hence, can avoid restart.
1007                  *
1008                  * In fact, page cannot be locked here at all, because
1009                  * mdc_read_page_remote does synchronous io.
1010                  */
1011                 wait_on_page_locked(page);
1012                 if (PageUptodate(page)) {
1013                         dp = kmap(page);
1014                         if (BITS_PER_LONG == 32 && hash64) {
1015                                 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1016                                 *end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
1017                                 *hash  = *hash >> 32;
1018                         } else {
1019                                 *start = le64_to_cpu(dp->ldp_hash_start);
1020                                 *end   = le64_to_cpu(dp->ldp_hash_end);
1021                         }
1022                         if (unlikely(*start == 1 && *hash == 0))
1023                                 *hash = *start;
1024                         else
1025                                 LASSERTF(*start <= *hash, "start = %#llx"
1026                                          ",end = %#llx,hash = %#llx\n",
1027                                          *start, *end, *hash);
1028                         CDEBUG(D_VFSTRACE, "offset %lx [%#llx %#llx],"
1029                               " hash %#llx\n", offset, *start, *end, *hash);
1030                         if (*hash > *end) {
1031                                 kunmap(page);
1032                                 mdc_release_page(page, 0);
1033                                 page = NULL;
1034                         } else if (*end != *start && *hash == *end) {
1035                                 /*
1036                                  * upon hash collision, remove this page,
1037                                  * otherwise put page reference, and
1038                                  * mdc_read_page_remote() will issue RPC to
1039                                  * fetch the page we want.
1040                                  */
1041                                 kunmap(page);
1042                                 mdc_release_page(page,
1043                                     le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
1044                                 page = NULL;
1045                         }
1046                 } else {
1047                         put_page(page);
1048                         page = ERR_PTR(-EIO);
1049                 }
1050         } else {
1051                 spin_unlock_irq(&mapping->tree_lock);
1052                 page = NULL;
1053         }
1054         return page;
1055 }
1056
1057 /*
1058  * Adjust a set of pages, each page containing an array of lu_dirpages,
1059  * so that each page can be used as a single logical lu_dirpage.
1060  *
1061  * A lu_dirpage is laid out as follows, where s = ldp_hash_start,
1062  * e = ldp_hash_end, f = ldp_flags, p = padding, and each "ent" is a
1063  * struct lu_dirent.  It has size up to LU_PAGE_SIZE. The ldp_hash_end
1064  * value is used as a cookie to request the next lu_dirpage in a
1065  * directory listing that spans multiple pages (two in this example):
1066  *   ________
1067  *  |        |
1068  * .|--------v-------   -----.
1069  * |s|e|f|p|ent|ent| ... |ent|
1070  * '--|--------------   -----'   Each PAGE contains a single
1071  *    '------.                   lu_dirpage.
1072  * .---------v-------   -----.
1073  * |s|e|f|p|ent| 0 | ... | 0 |
1074  * '-----------------   -----'
1075  *
1076  * However, on hosts where the native VM page size (PAGE_SIZE) is
1077  * larger than LU_PAGE_SIZE, a single host page may contain multiple
1078  * lu_dirpages. After reading the lu_dirpages from the MDS, the
1079  * ldp_hash_end of the first lu_dirpage refers to the one immediately
1080  * after it in the same PAGE (arrows simplified for brevity, but
1081  * in general e0==s1, e1==s2, etc.):
1082  *
1083  * .--------------------   -----.
1084  * |s0|e0|f0|p|ent|ent| ... |ent|
1085  * |---v----------------   -----|
1086  * |s1|e1|f1|p|ent|ent| ... |ent|
1087  * |---v----------------   -----|  Here, each PAGE contains
1088  *             ...                 multiple lu_dirpages.
1089  * |---v----------------   -----|
1090  * |s'|e'|f'|p|ent|ent| ... |ent|
1091  * '---|----------------   -----'
1092  *     v
1093  * .----------------------------.
1094  * |        next PAGE           |
1095  *
1096  * This structure is transformed into a single logical lu_dirpage as follows:
1097  *
1098  * - Replace e0 with e' so the request for the next lu_dirpage gets the page
1099  *   labeled 'next PAGE'.
1100  *
1101  * - Copy the LDF_COLLIDE flag from f' to f0 to correctly reflect whether
1102  *   a hash collision with the next page exists.
1103  *
1104  * - Adjust the lde_reclen of the ending entry of each lu_dirpage to span
1105  *   to the first entry of the next lu_dirpage.
1106  */
1107 #if PAGE_SIZE > LU_PAGE_SIZE
1108 static void mdc_adjust_dirpages(struct page **pages, int cfs_pgs, int lu_pgs)
1109 {
1110         int i;
1111
1112         for (i = 0; i < cfs_pgs; i++) {
1113                 struct lu_dirpage       *dp = kmap(pages[i]);
1114                 struct lu_dirpage       *first = dp;
1115                 struct lu_dirent        *end_dirent = NULL;
1116                 struct lu_dirent        *ent;
1117                 __u64           hash_end = le64_to_cpu(dp->ldp_hash_end);
1118                 __u32           flags = le32_to_cpu(dp->ldp_flags);
1119
1120                 while (--lu_pgs > 0) {
1121                         ent = lu_dirent_start(dp);
1122                         for (end_dirent = ent; ent != NULL;
1123                              end_dirent = ent, ent = lu_dirent_next(ent));
1124
1125                         /* Advance dp to next lu_dirpage. */
1126                         dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE);
1127
1128                         /* Check if we've reached the end of the PAGE. */
1129                         if (!((unsigned long)dp & ~PAGE_MASK))
1130                                 break;
1131
1132                         /* Save the hash and flags of this lu_dirpage. */
1133                         hash_end = le64_to_cpu(dp->ldp_hash_end);
1134                         flags = le32_to_cpu(dp->ldp_flags);
1135
1136                         /* Check if lu_dirpage contains no entries. */
1137                         if (end_dirent == NULL)
1138                                 break;
1139
1140                         /* Enlarge the end entry lde_reclen from 0 to
1141                          * first entry of next lu_dirpage. */
1142                         LASSERT(le16_to_cpu(end_dirent->lde_reclen) == 0);
1143                         end_dirent->lde_reclen =
1144                                 cpu_to_le16((char *)(dp->ldp_entries) -
1145                                             (char *)end_dirent);
1146                 }
1147
1148                 first->ldp_hash_end = hash_end;
1149                 first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE);
1150                 first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE);
1151
1152                 kunmap(pages[i]);
1153         }
1154         LASSERTF(lu_pgs == 0, "left = %d\n", lu_pgs);
1155 }
1156 #else
1157 #define mdc_adjust_dirpages(pages, cfs_pgs, lu_pgs) do {} while (0)
1158 #endif  /* PAGE_SIZE > LU_PAGE_SIZE */
1159
1160 /* parameters for readdir page */
1161 struct readpage_param {
1162         struct md_op_data       *rp_mod;
1163         __u64                   rp_off;
1164         int                     rp_hash64;
1165         struct obd_export       *rp_exp;
1166         struct md_callback      *rp_cb;
1167 };
1168
1169 #ifndef HAVE_DELETE_FROM_PAGE_CACHE
1170 static inline void delete_from_page_cache(struct page *page)
1171 {
1172         remove_from_page_cache(page);
1173         put_page(page);
1174 }
1175 #endif
1176
1177 /**
1178  * Read pages from server.
1179  *
1180  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
1181  * a header lu_dirpage which describes the start/end hash, and whether this
1182  * page is empty (contains no dir entry) or hash collide with next page.
1183  * After client receives reply, several pages will be integrated into dir page
1184  * in PAGE_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the
1185  * lu_dirpage for this integrated page will be adjusted.
1186  **/
1187 static int mdc_read_page_remote(void *data, struct page *page0)
1188 {
1189         struct readpage_param *rp = data;
1190         struct page **page_pool;
1191         struct page *page;
1192         struct lu_dirpage *dp;
1193         struct md_op_data *op_data = rp->rp_mod;
1194         struct ptlrpc_request *req;
1195         int max_pages;
1196         struct inode *inode;
1197         struct lu_fid *fid;
1198         int rd_pgs = 0; /* number of pages actually read */
1199         int npages;
1200         int i;
1201         int rc;
1202         ENTRY;
1203
1204         max_pages = rp->rp_exp->exp_obd->u.cli.cl_max_pages_per_rpc;
1205         inode = op_data->op_data;
1206         fid = &op_data->op_fid1;
1207         LASSERT(inode != NULL);
1208
1209         OBD_ALLOC(page_pool, sizeof(page_pool[0]) * max_pages);
1210         if (page_pool != NULL) {
1211                 page_pool[0] = page0;
1212         } else {
1213                 page_pool = &page0;
1214                 max_pages = 1;
1215         }
1216
1217         for (npages = 1; npages < max_pages; npages++) {
1218                 page = page_cache_alloc_cold(inode->i_mapping);
1219                 if (page == NULL)
1220                         break;
1221                 page_pool[npages] = page;
1222         }
1223
1224         rc = mdc_getpage(rp->rp_exp, fid, rp->rp_off, page_pool, npages, &req);
1225         if (rc < 0) {
1226                 /* page0 is special, which was added into page cache early */
1227                 delete_from_page_cache(page0);
1228         } else {
1229                 int lu_pgs;
1230
1231                 rd_pgs = (req->rq_bulk->bd_nob_transferred + PAGE_SIZE - 1) >>
1232                         PAGE_SHIFT;
1233                 lu_pgs = req->rq_bulk->bd_nob_transferred >> LU_PAGE_SHIFT;
1234                 LASSERT(!(req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK));
1235
1236                 CDEBUG(D_INODE, "read %d(%d) pages\n", rd_pgs, lu_pgs);
1237
1238                 mdc_adjust_dirpages(page_pool, rd_pgs, lu_pgs);
1239
1240                 SetPageUptodate(page0);
1241         }
1242         unlock_page(page0);
1243
1244         ptlrpc_req_finished(req);
1245         CDEBUG(D_CACHE, "read %d/%d pages\n", rd_pgs, npages);
1246         for (i = 1; i < npages; i++) {
1247                 unsigned long   offset;
1248                 __u64           hash;
1249                 int ret;
1250
1251                 page = page_pool[i];
1252
1253                 if (rc < 0 || i >= rd_pgs) {
1254                         put_page(page);
1255                         continue;
1256                 }
1257
1258                 SetPageUptodate(page);
1259
1260                 dp = kmap(page);
1261                 hash = le64_to_cpu(dp->ldp_hash_start);
1262                 kunmap(page);
1263
1264                 offset = hash_x_index(hash, rp->rp_hash64);
1265
1266                 prefetchw(&page->flags);
1267                 ret = add_to_page_cache_lru(page, inode->i_mapping, offset,
1268                                             GFP_KERNEL);
1269                 if (ret == 0)
1270                         unlock_page(page);
1271                 else
1272                         CDEBUG(D_VFSTRACE, "page %lu add to page cache failed:"
1273                                " rc = %d\n", offset, ret);
1274                 put_page(page);
1275         }
1276
1277         if (page_pool != &page0)
1278                 OBD_FREE(page_pool, sizeof(page_pool[0]) * max_pages);
1279
1280         RETURN(rc);
1281 }
1282
1283 /**
1284  * Read dir page from cache first, if it can not find it, read it from
1285  * server and add into the cache.
1286  *
1287  * \param[in] exp       MDC export
1288  * \param[in] op_data   client MD stack parameters, transfering parameters
1289  *                      between different layers on client MD stack.
1290  * \param[in] cb_op     callback required for ldlm lock enqueue during
1291  *                      read page
1292  * \param[in] hash_offset the hash offset of the page to be read
1293  * \param[in] ppage     the page to be read
1294  *
1295  * retval               = 0 get the page successfully
1296  *                      errno(<0) get the page failed
1297  */
1298 static int mdc_read_page(struct obd_export *exp, struct md_op_data *op_data,
1299                          struct md_callback *cb_op, __u64 hash_offset,
1300                          struct page **ppage)
1301 {
1302         struct lookup_intent    it = { .it_op = IT_READDIR };
1303         struct page             *page;
1304         struct inode            *dir = op_data->op_data;
1305         struct address_space    *mapping;
1306         struct lu_dirpage       *dp;
1307         __u64                   start = 0;
1308         __u64                   end = 0;
1309         struct lustre_handle    lockh;
1310         struct ptlrpc_request   *enq_req = NULL;
1311         struct readpage_param   rp_param;
1312         int rc;
1313
1314         ENTRY;
1315
1316         *ppage = NULL;
1317
1318         LASSERT(dir != NULL);
1319         mapping = dir->i_mapping;
1320
1321         rc = mdc_intent_lock(exp, op_data, &it, &enq_req,
1322                              cb_op->md_blocking_ast, 0);
1323         if (enq_req != NULL)
1324                 ptlrpc_req_finished(enq_req);
1325
1326         if (rc < 0) {
1327                 CERROR("%s: "DFID" lock enqueue fails: rc = %d\n",
1328                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1), rc);
1329                 RETURN(rc);
1330         }
1331
1332         rc = 0;
1333         lockh.cookie = it.it_lock_handle;
1334         mdc_set_lock_data(exp, &lockh, dir, NULL);
1335
1336         rp_param.rp_off = hash_offset;
1337         rp_param.rp_hash64 = op_data->op_cli_flags & CLI_HASH64;
1338         page = mdc_page_locate(mapping, &rp_param.rp_off, &start, &end,
1339                                rp_param.rp_hash64);
1340         if (IS_ERR(page)) {
1341                 CERROR("%s: dir page locate: "DFID" at %llu: rc %ld\n",
1342                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1343                        rp_param.rp_off, PTR_ERR(page));
1344                 GOTO(out_unlock, rc = PTR_ERR(page));
1345         } else if (page != NULL) {
1346                 /*
1347                  * XXX nikita: not entirely correct handling of a corner case:
1348                  * suppose hash chain of entries with hash value HASH crosses
1349                  * border between pages P0 and P1. First both P0 and P1 are
1350                  * cached, seekdir() is called for some entry from the P0 part
1351                  * of the chain. Later P0 goes out of cache. telldir(HASH)
1352                  * happens and finds P1, as it starts with matching hash
1353                  * value. Remaining entries from P0 part of the chain are
1354                  * skipped. (Is that really a bug?)
1355                  *
1356                  * Possible solutions: 0. don't cache P1 is such case, handle
1357                  * it as an "overflow" page. 1. invalidate all pages at
1358                  * once. 2. use HASH|1 as an index for P1.
1359                  */
1360                 GOTO(hash_collision, page);
1361         }
1362
1363         rp_param.rp_exp = exp;
1364         rp_param.rp_mod = op_data;
1365         page = read_cache_page(mapping,
1366                                hash_x_index(rp_param.rp_off,
1367                                             rp_param.rp_hash64),
1368                                mdc_read_page_remote, &rp_param);
1369         if (IS_ERR(page)) {
1370                 CDEBUG(D_INFO, "%s: read cache page: "DFID" at %llu: %ld\n",
1371                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1372                        rp_param.rp_off, PTR_ERR(page));
1373                 GOTO(out_unlock, rc = PTR_ERR(page));
1374         }
1375
1376         wait_on_page_locked(page);
1377         (void)kmap(page);
1378         if (!PageUptodate(page)) {
1379                 CERROR("%s: page not updated: "DFID" at %llu: rc %d\n",
1380                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1381                        rp_param.rp_off, -5);
1382                 goto fail;
1383         }
1384         if (!PageChecked(page))
1385                 SetPageChecked(page);
1386         if (PageError(page)) {
1387                 CERROR("%s: page error: "DFID" at %llu: rc %d\n",
1388                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1389                        rp_param.rp_off, -5);
1390                 goto fail;
1391         }
1392
1393 hash_collision:
1394         dp = page_address(page);
1395         if (BITS_PER_LONG == 32 && rp_param.rp_hash64) {
1396                 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1397                 end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
1398                 rp_param.rp_off = hash_offset >> 32;
1399         } else {
1400                 start = le64_to_cpu(dp->ldp_hash_start);
1401                 end   = le64_to_cpu(dp->ldp_hash_end);
1402                 rp_param.rp_off = hash_offset;
1403         }
1404         if (end == start) {
1405                 LASSERT(start == rp_param.rp_off);
1406                 CWARN("Page-wide hash collision: %#lx\n", (unsigned long)end);
1407 #if BITS_PER_LONG == 32
1408                 CWARN("Real page-wide hash collision at [%llu %llu] with "
1409                       "hash %llu\n", le64_to_cpu(dp->ldp_hash_start),
1410                       le64_to_cpu(dp->ldp_hash_end), hash_offset);
1411 #endif
1412
1413                 /*
1414                  * Fetch whole overflow chain...
1415                  *
1416                  * XXX not yet.
1417                  */
1418                 goto fail;
1419         }
1420         *ppage = page;
1421 out_unlock:
1422         ldlm_lock_decref(&lockh, it.it_lock_mode);
1423         return rc;
1424 fail:
1425         kunmap(page);
1426         mdc_release_page(page, 1);
1427         rc = -EIO;
1428         goto out_unlock;
1429 }
1430
1431
1432 static int mdc_statfs(const struct lu_env *env,
1433                       struct obd_export *exp, struct obd_statfs *osfs,
1434                       __u64 max_age, __u32 flags)
1435 {
1436         struct obd_device     *obd = class_exp2obd(exp);
1437         struct ptlrpc_request *req;
1438         struct obd_statfs     *msfs;
1439         struct obd_import     *imp = NULL;
1440         int                    rc;
1441         ENTRY;
1442
1443         /*
1444          * Since the request might also come from lprocfs, so we need
1445          * sync this with client_disconnect_export Bug15684
1446          */
1447         down_read(&obd->u.cli.cl_sem);
1448         if (obd->u.cli.cl_import)
1449                 imp = class_import_get(obd->u.cli.cl_import);
1450         up_read(&obd->u.cli.cl_sem);
1451         if (!imp)
1452                 RETURN(-ENODEV);
1453
1454         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_STATFS,
1455                                         LUSTRE_MDS_VERSION, MDS_STATFS);
1456         if (req == NULL)
1457                 GOTO(output, rc = -ENOMEM);
1458
1459         ptlrpc_request_set_replen(req);
1460
1461         if (flags & OBD_STATFS_NODELAY) {
1462                 /* procfs requests not want stay in wait for avoid deadlock */
1463                 req->rq_no_resend = 1;
1464                 req->rq_no_delay = 1;
1465         }
1466
1467         rc = ptlrpc_queue_wait(req);
1468         if (rc) {
1469                 /* check connection error first */
1470                 if (imp->imp_connect_error)
1471                         rc = imp->imp_connect_error;
1472                 GOTO(out, rc);
1473         }
1474
1475         msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1476         if (msfs == NULL)
1477                 GOTO(out, rc = -EPROTO);
1478
1479         *osfs = *msfs;
1480         EXIT;
1481 out:
1482         ptlrpc_req_finished(req);
1483 output:
1484         class_import_put(imp);
1485         return rc;
1486 }
1487
1488 static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
1489 {
1490         __u32 keylen, vallen;
1491         void *key;
1492         int rc;
1493
1494         if (gf->gf_pathlen > PATH_MAX)
1495                 RETURN(-ENAMETOOLONG);
1496         if (gf->gf_pathlen < 2)
1497                 RETURN(-EOVERFLOW);
1498
1499         /* Key is KEY_FID2PATH + getinfo_fid2path description */
1500         keylen = cfs_size_round(sizeof(KEY_FID2PATH) + sizeof(*gf) +
1501                                 sizeof(struct lu_fid));
1502         OBD_ALLOC(key, keylen);
1503         if (key == NULL)
1504                 RETURN(-ENOMEM);
1505         memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
1506         memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf));
1507         memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf),
1508                gf->gf_u.gf_root_fid, sizeof(struct lu_fid));
1509         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
1510                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
1511
1512         if (!fid_is_sane(&gf->gf_fid))
1513                 GOTO(out, rc = -EINVAL);
1514
1515         /* Val is struct getinfo_fid2path result plus path */
1516         vallen = sizeof(*gf) + gf->gf_pathlen;
1517
1518         rc = obd_get_info(NULL, exp, keylen, key, &vallen, gf);
1519         if (rc != 0 && rc != -EREMOTE)
1520                 GOTO(out, rc);
1521
1522         if (vallen <= sizeof(*gf))
1523                 GOTO(out, rc = -EPROTO);
1524         if (vallen > sizeof(*gf) + gf->gf_pathlen)
1525                 GOTO(out, rc = -EOVERFLOW);
1526
1527         CDEBUG(D_IOCTL, "path got "DFID" from %llu #%d: %s\n",
1528                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno,
1529                gf->gf_pathlen < 512 ? gf->gf_u.gf_path :
1530                /* only log the last 512 characters of the path */
1531                gf->gf_u.gf_path + gf->gf_pathlen - 512);
1532
1533 out:
1534         OBD_FREE(key, keylen);
1535         return rc;
1536 }
1537
1538 static int mdc_ioc_hsm_progress(struct obd_export *exp,
1539                                 struct hsm_progress_kernel *hpk)
1540 {
1541         struct obd_import               *imp = class_exp2cliimp(exp);
1542         struct hsm_progress_kernel      *req_hpk;
1543         struct ptlrpc_request           *req;
1544         int                              rc;
1545         ENTRY;
1546
1547         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_PROGRESS,
1548                                         LUSTRE_MDS_VERSION, MDS_HSM_PROGRESS);
1549         if (req == NULL)
1550                 GOTO(out, rc = -ENOMEM);
1551
1552         mdc_pack_body(req, NULL, 0, 0, -1, 0);
1553
1554         /* Copy hsm_progress struct */
1555         req_hpk = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_PROGRESS);
1556         if (req_hpk == NULL)
1557                 GOTO(out, rc = -EPROTO);
1558
1559         *req_hpk = *hpk;
1560         req_hpk->hpk_errval = lustre_errno_hton(hpk->hpk_errval);
1561
1562         ptlrpc_request_set_replen(req);
1563
1564         mdc_get_mod_rpc_slot(req, NULL);
1565         rc = ptlrpc_queue_wait(req);
1566         mdc_put_mod_rpc_slot(req, NULL);
1567
1568         GOTO(out, rc);
1569 out:
1570         ptlrpc_req_finished(req);
1571         return rc;
1572 }
1573
1574 static int mdc_ioc_hsm_ct_register(struct obd_import *imp, __u32 archives)
1575 {
1576         __u32                   *archive_mask;
1577         struct ptlrpc_request   *req;
1578         int                      rc;
1579         ENTRY;
1580
1581         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_REGISTER,
1582                                         LUSTRE_MDS_VERSION,
1583                                         MDS_HSM_CT_REGISTER);
1584         if (req == NULL)
1585                 GOTO(out, rc = -ENOMEM);
1586
1587         mdc_pack_body(req, NULL, 0, 0, -1, 0);
1588
1589         /* Copy hsm_progress struct */
1590         archive_mask = req_capsule_client_get(&req->rq_pill,
1591                                               &RMF_MDS_HSM_ARCHIVE);
1592         if (archive_mask == NULL)
1593                 GOTO(out, rc = -EPROTO);
1594
1595         *archive_mask = archives;
1596
1597         ptlrpc_request_set_replen(req);
1598
1599         rc = mdc_queue_wait(req);
1600         GOTO(out, rc);
1601 out:
1602         ptlrpc_req_finished(req);
1603         return rc;
1604 }
1605
1606 static int mdc_ioc_hsm_current_action(struct obd_export *exp,
1607                                       struct md_op_data *op_data)
1608 {
1609         struct hsm_current_action       *hca = op_data->op_data;
1610         struct hsm_current_action       *req_hca;
1611         struct ptlrpc_request           *req;
1612         int                              rc;
1613         ENTRY;
1614
1615         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1616                                    &RQF_MDS_HSM_ACTION);
1617         if (req == NULL)
1618                 RETURN(-ENOMEM);
1619
1620         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_ACTION);
1621         if (rc) {
1622                 ptlrpc_request_free(req);
1623                 RETURN(rc);
1624         }
1625
1626         mdc_pack_body(req, &op_data->op_fid1, 0, 0,
1627                       op_data->op_suppgids[0], 0);
1628
1629         ptlrpc_request_set_replen(req);
1630
1631         rc = mdc_queue_wait(req);
1632         if (rc)
1633                 GOTO(out, rc);
1634
1635         req_hca = req_capsule_server_get(&req->rq_pill,
1636                                          &RMF_MDS_HSM_CURRENT_ACTION);
1637         if (req_hca == NULL)
1638                 GOTO(out, rc = -EPROTO);
1639
1640         *hca = *req_hca;
1641
1642         EXIT;
1643 out:
1644         ptlrpc_req_finished(req);
1645         return rc;
1646 }
1647
1648 static int mdc_ioc_hsm_ct_unregister(struct obd_import *imp)
1649 {
1650         struct ptlrpc_request   *req;
1651         int                      rc;
1652         ENTRY;
1653
1654         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_UNREGISTER,
1655                                         LUSTRE_MDS_VERSION,
1656                                         MDS_HSM_CT_UNREGISTER);
1657         if (req == NULL)
1658                 GOTO(out, rc = -ENOMEM);
1659
1660         mdc_pack_body(req, NULL, 0, 0, -1, 0);
1661
1662         ptlrpc_request_set_replen(req);
1663
1664         rc = mdc_queue_wait(req);
1665         GOTO(out, rc);
1666 out:
1667         ptlrpc_req_finished(req);
1668         return rc;
1669 }
1670
1671 static int mdc_ioc_hsm_state_get(struct obd_export *exp,
1672                                  struct md_op_data *op_data)
1673 {
1674         struct hsm_user_state   *hus = op_data->op_data;
1675         struct hsm_user_state   *req_hus;
1676         struct ptlrpc_request   *req;
1677         int                      rc;
1678         ENTRY;
1679
1680         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1681                                    &RQF_MDS_HSM_STATE_GET);
1682         if (req == NULL)
1683                 RETURN(-ENOMEM);
1684
1685         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_GET);
1686         if (rc != 0) {
1687                 ptlrpc_request_free(req);
1688                 RETURN(rc);
1689         }
1690
1691         mdc_pack_body(req, &op_data->op_fid1, 0, 0,
1692                       op_data->op_suppgids[0], 0);
1693
1694         ptlrpc_request_set_replen(req);
1695
1696         rc = mdc_queue_wait(req);
1697         if (rc)
1698                 GOTO(out, rc);
1699
1700         req_hus = req_capsule_server_get(&req->rq_pill, &RMF_HSM_USER_STATE);
1701         if (req_hus == NULL)
1702                 GOTO(out, rc = -EPROTO);
1703
1704         *hus = *req_hus;
1705
1706         EXIT;
1707 out:
1708         ptlrpc_req_finished(req);
1709         return rc;
1710 }
1711
1712 static int mdc_ioc_hsm_state_set(struct obd_export *exp,
1713                                  struct md_op_data *op_data)
1714 {
1715         struct hsm_state_set    *hss = op_data->op_data;
1716         struct hsm_state_set    *req_hss;
1717         struct ptlrpc_request   *req;
1718         int                      rc;
1719         ENTRY;
1720
1721         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1722                                    &RQF_MDS_HSM_STATE_SET);
1723         if (req == NULL)
1724                 RETURN(-ENOMEM);
1725
1726         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_SET);
1727         if (rc) {
1728                 ptlrpc_request_free(req);
1729                 RETURN(rc);
1730         }
1731
1732         mdc_pack_body(req, &op_data->op_fid1, 0, 0,
1733                       op_data->op_suppgids[0], 0);
1734
1735         /* Copy states */
1736         req_hss = req_capsule_client_get(&req->rq_pill, &RMF_HSM_STATE_SET);
1737         if (req_hss == NULL)
1738                 GOTO(out, rc = -EPROTO);
1739         *req_hss = *hss;
1740
1741         ptlrpc_request_set_replen(req);
1742
1743         mdc_get_mod_rpc_slot(req, NULL);
1744         rc = ptlrpc_queue_wait(req);
1745         mdc_put_mod_rpc_slot(req, NULL);
1746
1747         GOTO(out, rc);
1748 out:
1749         ptlrpc_req_finished(req);
1750         return rc;
1751 }
1752
1753 static int mdc_ioc_hsm_request(struct obd_export *exp,
1754                                struct hsm_user_request *hur)
1755 {
1756         struct obd_import       *imp = class_exp2cliimp(exp);
1757         struct ptlrpc_request   *req;
1758         struct hsm_request      *req_hr;
1759         struct hsm_user_item    *req_hui;
1760         char                    *req_opaque;
1761         int                      rc;
1762         ENTRY;
1763
1764         req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_REQUEST);
1765         if (req == NULL)
1766                 GOTO(out, rc = -ENOMEM);
1767
1768         req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM, RCL_CLIENT,
1769                              hur->hur_request.hr_itemcount
1770                              * sizeof(struct hsm_user_item));
1771         req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA, RCL_CLIENT,
1772                              hur->hur_request.hr_data_len);
1773
1774         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_REQUEST);
1775         if (rc) {
1776                 ptlrpc_request_free(req);
1777                 RETURN(rc);
1778         }
1779
1780         mdc_pack_body(req, NULL, 0, 0, -1, 0);
1781
1782         /* Copy hsm_request struct */
1783         req_hr = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_REQUEST);
1784         if (req_hr == NULL)
1785                 GOTO(out, rc = -EPROTO);
1786         *req_hr = hur->hur_request;
1787
1788         /* Copy hsm_user_item structs */
1789         req_hui = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM);
1790         if (req_hui == NULL)
1791                 GOTO(out, rc = -EPROTO);
1792         memcpy(req_hui, hur->hur_user_item,
1793                hur->hur_request.hr_itemcount * sizeof(struct hsm_user_item));
1794
1795         /* Copy opaque field */
1796         req_opaque = req_capsule_client_get(&req->rq_pill, &RMF_GENERIC_DATA);
1797         if (req_opaque == NULL)
1798                 GOTO(out, rc = -EPROTO);
1799         memcpy(req_opaque, hur_data(hur), hur->hur_request.hr_data_len);
1800
1801         ptlrpc_request_set_replen(req);
1802
1803         mdc_get_mod_rpc_slot(req, NULL);
1804         rc = ptlrpc_queue_wait(req);
1805         mdc_put_mod_rpc_slot(req, NULL);
1806
1807         GOTO(out, rc);
1808
1809 out:
1810         ptlrpc_req_finished(req);
1811         return rc;
1812 }
1813
1814 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
1815                                 struct lustre_kernelcomm *lk);
1816
1817 static int mdc_quotactl(struct obd_device *unused, struct obd_export *exp,
1818                         struct obd_quotactl *oqctl)
1819 {
1820         struct ptlrpc_request   *req;
1821         struct obd_quotactl     *oqc;
1822         int                      rc;
1823         ENTRY;
1824
1825         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
1826                                         &RQF_MDS_QUOTACTL, LUSTRE_MDS_VERSION,
1827                                         MDS_QUOTACTL);
1828         if (req == NULL)
1829                 RETURN(-ENOMEM);
1830
1831         oqc = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1832         *oqc = *oqctl;
1833
1834         ptlrpc_request_set_replen(req);
1835         ptlrpc_at_set_req_timeout(req);
1836         req->rq_no_resend = 1;
1837
1838         rc = ptlrpc_queue_wait(req);
1839         if (rc)
1840                 CERROR("ptlrpc_queue_wait failed, rc: %d\n", rc);
1841
1842         if (req->rq_repmsg &&
1843             (oqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL))) {
1844                 *oqctl = *oqc;
1845         } else if (!rc) {
1846                 CERROR ("Can't unpack obd_quotactl\n");
1847                 rc = -EPROTO;
1848         }
1849         ptlrpc_req_finished(req);
1850
1851         RETURN(rc);
1852 }
1853
1854 static int mdc_ioc_swap_layouts(struct obd_export *exp,
1855                                 struct md_op_data *op_data)
1856 {
1857         struct list_head cancels = LIST_HEAD_INIT(cancels);
1858         struct ptlrpc_request   *req;
1859         int                      rc, count;
1860         struct mdc_swap_layouts *msl, *payload;
1861         ENTRY;
1862
1863         msl = op_data->op_data;
1864
1865         /* When the MDT will get the MDS_SWAP_LAYOUTS RPC the
1866          * first thing it will do is to cancel the 2 layout
1867          * locks held by this client.
1868          * So the client must cancel its layout locks on the 2 fids
1869          * with the request RPC to avoid extra RPC round trips.
1870          */
1871         count = mdc_resource_get_unused(exp, &op_data->op_fid1, &cancels,
1872                                         LCK_EX, MDS_INODELOCK_LAYOUT |
1873                                         MDS_INODELOCK_XATTR);
1874         count += mdc_resource_get_unused(exp, &op_data->op_fid2, &cancels,
1875                                          LCK_EX, MDS_INODELOCK_LAYOUT |
1876                                          MDS_INODELOCK_XATTR);
1877
1878         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1879                                    &RQF_MDS_SWAP_LAYOUTS);
1880         if (req == NULL) {
1881                 ldlm_lock_list_put(&cancels, l_bl_ast, count);
1882                 RETURN(-ENOMEM);
1883         }
1884
1885         rc = mdc_prep_elc_req(exp, req, MDS_SWAP_LAYOUTS, &cancels, count);
1886         if (rc) {
1887                 ptlrpc_request_free(req);
1888                 RETURN(rc);
1889         }
1890
1891         mdc_swap_layouts_pack(req, op_data);
1892
1893         payload = req_capsule_client_get(&req->rq_pill, &RMF_SWAP_LAYOUTS);
1894         LASSERT(payload);
1895
1896         *payload = *msl;
1897
1898         ptlrpc_request_set_replen(req);
1899
1900         rc = ptlrpc_queue_wait(req);
1901         if (rc)
1902                 GOTO(out, rc);
1903         EXIT;
1904
1905 out:
1906         ptlrpc_req_finished(req);
1907         return rc;
1908 }
1909
1910 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1911                          void *karg, void __user *uarg)
1912 {
1913         struct obd_device *obd = exp->exp_obd;
1914         struct obd_ioctl_data *data = karg;
1915         struct obd_import *imp = obd->u.cli.cl_import;
1916         int rc;
1917         ENTRY;
1918
1919         if (!try_module_get(THIS_MODULE)) {
1920                 CERROR("%s: cannot get module '%s'\n", obd->obd_name,
1921                        module_name(THIS_MODULE));
1922                 return -EINVAL;
1923         }
1924         switch (cmd) {
1925         case OBD_IOC_FID2PATH:
1926                 rc = mdc_ioc_fid2path(exp, karg);
1927                 GOTO(out, rc);
1928         case LL_IOC_HSM_CT_START:
1929                 rc = mdc_ioc_hsm_ct_start(exp, karg);
1930                 /* ignore if it was already registered on this MDS. */
1931                 if (rc == -EEXIST)
1932                         rc = 0;
1933                 GOTO(out, rc);
1934         case LL_IOC_HSM_PROGRESS:
1935                 rc = mdc_ioc_hsm_progress(exp, karg);
1936                 GOTO(out, rc);
1937         case LL_IOC_HSM_STATE_GET:
1938                 rc = mdc_ioc_hsm_state_get(exp, karg);
1939                 GOTO(out, rc);
1940         case LL_IOC_HSM_STATE_SET:
1941                 rc = mdc_ioc_hsm_state_set(exp, karg);
1942                 GOTO(out, rc);
1943         case LL_IOC_HSM_ACTION:
1944                 rc = mdc_ioc_hsm_current_action(exp, karg);
1945                 GOTO(out, rc);
1946         case LL_IOC_HSM_REQUEST:
1947                 rc = mdc_ioc_hsm_request(exp, karg);
1948                 GOTO(out, rc);
1949         case OBD_IOC_CLIENT_RECOVER:
1950                 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1, 0);
1951                 if (rc < 0)
1952                         GOTO(out, rc);
1953                 GOTO(out, rc = 0);
1954         case IOC_OSC_SET_ACTIVE:
1955                 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
1956                 GOTO(out, rc);
1957         case OBD_IOC_PING_TARGET:
1958                 rc = ptlrpc_obd_ping(obd);
1959                 GOTO(out, rc);
1960         /*
1961          * Normally IOC_OBD_STATFS, OBD_IOC_QUOTACTL iocontrol are handled by
1962          * LMV instead of MDC. But when the cluster is upgraded from 1.8,
1963          * there'd be no LMV layer thus we might be called here. Eventually
1964          * this code should be removed.
1965          * bz20731, LU-592.
1966          */
1967         case IOC_OBD_STATFS: {
1968                 struct obd_statfs stat_buf = {0};
1969
1970                 if (*((__u32 *) data->ioc_inlbuf2) != 0)
1971                         GOTO(out, rc = -ENODEV);
1972
1973                 /* copy UUID */
1974                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd),
1975                                  min((int)data->ioc_plen2,
1976                                      (int)sizeof(struct obd_uuid))))
1977                         GOTO(out, rc = -EFAULT);
1978
1979                 rc = mdc_statfs(NULL, obd->obd_self_export, &stat_buf,
1980                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1981                                 0);
1982                 if (rc != 0)
1983                         GOTO(out, rc);
1984
1985                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1986                                      min((int) data->ioc_plen1,
1987                                          (int) sizeof(stat_buf))))
1988                         GOTO(out, rc = -EFAULT);
1989
1990                 GOTO(out, rc = 0);
1991         }
1992         case OBD_IOC_QUOTACTL: {
1993                 struct if_quotactl *qctl = karg;
1994                 struct obd_quotactl *oqctl;
1995
1996                 OBD_ALLOC_PTR(oqctl);
1997                 if (oqctl == NULL)
1998                         GOTO(out, rc = -ENOMEM);
1999
2000                 QCTL_COPY(oqctl, qctl);
2001                 rc = obd_quotactl(exp, oqctl);
2002                 if (rc == 0) {
2003                         QCTL_COPY(qctl, oqctl);
2004                         qctl->qc_valid = QC_MDTIDX;
2005                         qctl->obd_uuid = obd->u.cli.cl_target_uuid;
2006                 }
2007
2008                 OBD_FREE_PTR(oqctl);
2009                 GOTO(out, rc);
2010         }
2011         case LL_IOC_GET_CONNECT_FLAGS:
2012                 if (copy_to_user(uarg, exp_connect_flags_ptr(exp),
2013                                  sizeof(*exp_connect_flags_ptr(exp))))
2014                         GOTO(out, rc = -EFAULT);
2015
2016                 GOTO(out, rc = 0);
2017         case LL_IOC_LOV_SWAP_LAYOUTS:
2018                 rc = mdc_ioc_swap_layouts(exp, karg);
2019                 GOTO(out, rc);
2020         default:
2021                 CERROR("unrecognised ioctl: cmd = %#x\n", cmd);
2022                 GOTO(out, rc = -ENOTTY);
2023         }
2024 out:
2025         module_put(THIS_MODULE);
2026
2027         return rc;
2028 }
2029
2030 static int mdc_get_info_rpc(struct obd_export *exp,
2031                             u32 keylen, void *key,
2032                             u32 vallen, void *val)
2033 {
2034         struct obd_import      *imp = class_exp2cliimp(exp);
2035         struct ptlrpc_request  *req;
2036         char                   *tmp;
2037         int                     rc = -EINVAL;
2038         ENTRY;
2039
2040         req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
2041         if (req == NULL)
2042                 RETURN(-ENOMEM);
2043
2044         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
2045                              RCL_CLIENT, keylen);
2046         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
2047                              RCL_CLIENT, sizeof(vallen));
2048
2049         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
2050         if (rc) {
2051                 ptlrpc_request_free(req);
2052                 RETURN(rc);
2053         }
2054
2055         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
2056         memcpy(tmp, key, keylen);
2057         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
2058         memcpy(tmp, &vallen, sizeof(vallen));
2059
2060         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
2061                              RCL_SERVER, vallen);
2062         ptlrpc_request_set_replen(req);
2063
2064         rc = ptlrpc_queue_wait(req);
2065         /* -EREMOTE means the get_info result is partial, and it needs to
2066          * continue on another MDT, see fid2path part in lmv_iocontrol */
2067         if (rc == 0 || rc == -EREMOTE) {
2068                 tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
2069                 memcpy(val, tmp, vallen);
2070                 if (ptlrpc_rep_need_swab(req)) {
2071                         if (KEY_IS(KEY_FID2PATH))
2072                                 lustre_swab_fid2path(val);
2073                 }
2074         }
2075         ptlrpc_req_finished(req);
2076
2077         RETURN(rc);
2078 }
2079
2080 static void lustre_swab_hai(struct hsm_action_item *h)
2081 {
2082         __swab32s(&h->hai_len);
2083         __swab32s(&h->hai_action);
2084         lustre_swab_lu_fid(&h->hai_fid);
2085         lustre_swab_lu_fid(&h->hai_dfid);
2086         __swab64s(&h->hai_cookie);
2087         __swab64s(&h->hai_extent.offset);
2088         __swab64s(&h->hai_extent.length);
2089         __swab64s(&h->hai_gid);
2090 }
2091
2092 static void lustre_swab_hal(struct hsm_action_list *h)
2093 {
2094         struct hsm_action_item  *hai;
2095         __u32                    i;
2096
2097         __swab32s(&h->hal_version);
2098         __swab32s(&h->hal_count);
2099         __swab32s(&h->hal_archive_id);
2100         __swab64s(&h->hal_flags);
2101         hai = hai_first(h);
2102         for (i = 0; i < h->hal_count; i++, hai = hai_next(hai))
2103                 lustre_swab_hai(hai);
2104 }
2105
2106 static void lustre_swab_kuch(struct kuc_hdr *l)
2107 {
2108         __swab16s(&l->kuc_magic);
2109         /* __u8 l->kuc_transport */
2110         __swab16s(&l->kuc_msgtype);
2111         __swab16s(&l->kuc_msglen);
2112 }
2113
2114 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2115                                 struct lustre_kernelcomm *lk)
2116 {
2117         struct obd_import  *imp = class_exp2cliimp(exp);
2118         __u32               archive = lk->lk_data;
2119         int                 rc = 0;
2120
2121         if (lk->lk_group != KUC_GRP_HSM) {
2122                 CERROR("Bad copytool group %d\n", lk->lk_group);
2123                 return -EINVAL;
2124         }
2125
2126         CDEBUG(D_HSM, "CT start r%d w%d u%d g%d f%#x\n", lk->lk_rfd, lk->lk_wfd,
2127                lk->lk_uid, lk->lk_group, lk->lk_flags);
2128
2129         if (lk->lk_flags & LK_FLG_STOP) {
2130                 /* Unregister with the coordinator */
2131                 rc = mdc_ioc_hsm_ct_unregister(imp);
2132         } else {
2133                 rc = mdc_ioc_hsm_ct_register(imp, archive);
2134         }
2135
2136         return rc;
2137 }
2138
2139 /**
2140  * Send a message to any listening copytools
2141  * @param val KUC message (kuc_hdr + hsm_action_list)
2142  * @param len total length of message
2143  */
2144 static int mdc_hsm_copytool_send(size_t len, void *val)
2145 {
2146         struct kuc_hdr          *lh = (struct kuc_hdr *)val;
2147         struct hsm_action_list  *hal = (struct hsm_action_list *)(lh + 1);
2148         int                      rc;
2149         ENTRY;
2150
2151         if (len < sizeof(*lh) + sizeof(*hal)) {
2152                 CERROR("Short HSM message %zu < %zu\n", len,
2153                        sizeof(*lh) + sizeof(*hal));
2154                 RETURN(-EPROTO);
2155         }
2156         if (lh->kuc_magic == __swab16(KUC_MAGIC)) {
2157                 lustre_swab_kuch(lh);
2158                 lustre_swab_hal(hal);
2159         } else if (lh->kuc_magic != KUC_MAGIC) {
2160                 CERROR("Bad magic %x!=%x\n", lh->kuc_magic, KUC_MAGIC);
2161                 RETURN(-EPROTO);
2162         }
2163
2164         CDEBUG(D_HSM, " Received message mg=%x t=%d m=%d l=%d actions=%d "
2165                "on %s\n",
2166                lh->kuc_magic, lh->kuc_transport, lh->kuc_msgtype,
2167                lh->kuc_msglen, hal->hal_count, hal->hal_fsname);
2168
2169         /* Broadcast to HSM listeners */
2170         rc = libcfs_kkuc_group_put(KUC_GRP_HSM, lh);
2171
2172         RETURN(rc);
2173 }
2174
2175 /**
2176  * callback function passed to kuc for re-registering each HSM copytool
2177  * running on MDC, after MDT shutdown/recovery.
2178  * @param data copytool registration data
2179  * @param cb_arg callback argument (obd_import)
2180  */
2181 static int mdc_hsm_ct_reregister(void *data, void *cb_arg)
2182 {
2183         struct kkuc_ct_data     *kcd = data;
2184         struct obd_import       *imp = (struct obd_import *)cb_arg;
2185         int                      rc;
2186
2187         if (kcd == NULL || kcd->kcd_magic != KKUC_CT_DATA_MAGIC)
2188                 return -EPROTO;
2189
2190         if (!obd_uuid_equals(&kcd->kcd_uuid, &imp->imp_obd->obd_uuid))
2191                 return 0;
2192
2193         CDEBUG(D_HA, "%s: recover copytool registration to MDT (archive=%#x)\n",
2194                imp->imp_obd->obd_name, kcd->kcd_archive);
2195         rc = mdc_ioc_hsm_ct_register(imp, kcd->kcd_archive);
2196
2197         /* ignore error if the copytool is already registered */
2198         return (rc == -EEXIST) ? 0 : rc;
2199 }
2200
2201 /**
2202  * Re-establish all kuc contexts with MDT
2203  * after MDT shutdown/recovery.
2204  */
2205 static int mdc_kuc_reregister(struct obd_import *imp)
2206 {
2207         /* re-register HSM agents */
2208         return libcfs_kkuc_group_foreach(KUC_GRP_HSM, mdc_hsm_ct_reregister,
2209                                          (void *)imp);
2210 }
2211
2212 static int mdc_set_info_async(const struct lu_env *env,
2213                               struct obd_export *exp,
2214                               u32 keylen, void *key,
2215                               u32 vallen, void *val,
2216                               struct ptlrpc_request_set *set)
2217 {
2218         struct obd_import       *imp = class_exp2cliimp(exp);
2219         int                      rc;
2220         ENTRY;
2221
2222         if (KEY_IS(KEY_READ_ONLY)) {
2223                 if (vallen != sizeof(int))
2224                         RETURN(-EINVAL);
2225
2226                 spin_lock(&imp->imp_lock);
2227                 if (*((int *)val)) {
2228                         imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
2229                         imp->imp_connect_data.ocd_connect_flags |=
2230                                                         OBD_CONNECT_RDONLY;
2231                 } else {
2232                         imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
2233                         imp->imp_connect_data.ocd_connect_flags &=
2234                                                         ~OBD_CONNECT_RDONLY;
2235                 }
2236                 spin_unlock(&imp->imp_lock);
2237
2238                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2239                                        keylen, key, vallen, val, set);
2240                 RETURN(rc);
2241         }
2242         if (KEY_IS(KEY_SPTLRPC_CONF)) {
2243                 sptlrpc_conf_client_adapt(exp->exp_obd);
2244                 RETURN(0);
2245         }
2246         if (KEY_IS(KEY_FLUSH_CTX)) {
2247                 sptlrpc_import_flush_my_ctx(imp);
2248                 RETURN(0);
2249         }
2250         if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2251                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2252                                        keylen, key, vallen, val, set);
2253                 RETURN(rc);
2254         }
2255         if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) {
2256                 rc = mdc_hsm_copytool_send(vallen, val);
2257                 RETURN(rc);
2258         }
2259
2260         if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2261                 __u32 *default_easize = val;
2262
2263                 exp->exp_obd->u.cli.cl_default_mds_easize = *default_easize;
2264                 RETURN(0);
2265         }
2266
2267         CERROR("Unknown key %s\n", (char *)key);
2268         RETURN(-EINVAL);
2269 }
2270
2271 static int mdc_get_info(const struct lu_env *env, struct obd_export *exp,
2272                         __u32 keylen, void *key, __u32 *vallen, void *val)
2273 {
2274         int rc = -EINVAL;
2275
2276         if (KEY_IS(KEY_MAX_EASIZE)) {
2277                 __u32 mdsize, *max_easize;
2278
2279                 if (*vallen != sizeof(int))
2280                         RETURN(-EINVAL);
2281                 mdsize = *(__u32 *)val;
2282                 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
2283                         exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
2284                 max_easize = val;
2285                 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
2286                 RETURN(0);
2287         } else if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2288                 __u32 *default_easize;
2289
2290                 if (*vallen != sizeof(int))
2291                         RETURN(-EINVAL);
2292                 default_easize = val;
2293                 *default_easize = exp->exp_obd->u.cli.cl_default_mds_easize;
2294                 RETURN(0);
2295         } else if (KEY_IS(KEY_CONN_DATA)) {
2296                 struct obd_import *imp = class_exp2cliimp(exp);
2297                 struct obd_connect_data *data = val;
2298
2299                 if (*vallen != sizeof(*data))
2300                         RETURN(-EINVAL);
2301
2302                 *data = imp->imp_connect_data;
2303                 RETURN(0);
2304         } else if (KEY_IS(KEY_TGT_COUNT)) {
2305                 *((__u32 *)val) = 1;
2306                 RETURN(0);
2307         }
2308
2309         rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
2310
2311         RETURN(rc);
2312 }
2313
2314 static int mdc_fsync(struct obd_export *exp, const struct lu_fid *fid,
2315                      struct ptlrpc_request **request)
2316 {
2317         struct ptlrpc_request *req;
2318         int                    rc;
2319         ENTRY;
2320
2321         *request = NULL;
2322         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
2323         if (req == NULL)
2324                 RETURN(-ENOMEM);
2325
2326         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
2327         if (rc) {
2328                 ptlrpc_request_free(req);
2329                 RETURN(rc);
2330         }
2331
2332         mdc_pack_body(req, fid, 0, 0, -1, 0);
2333
2334         ptlrpc_request_set_replen(req);
2335
2336         rc = ptlrpc_queue_wait(req);
2337         if (rc)
2338                 ptlrpc_req_finished(req);
2339         else
2340                 *request = req;
2341         RETURN(rc);
2342 }
2343
2344 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
2345                             enum obd_import_event event)
2346 {
2347         int rc = 0;
2348
2349         LASSERT(imp->imp_obd == obd);
2350
2351         switch (event) {
2352
2353         case IMP_EVENT_INACTIVE: {
2354                 struct client_obd *cli = &obd->u.cli;
2355                 /*
2356                  * Flush current sequence to make client obtain new one
2357                  * from server in case of disconnect/reconnect.
2358                  */
2359                 down_read(&cli->cl_seq_rwsem);
2360                 if (cli->cl_seq)
2361                         seq_client_flush(cli->cl_seq);
2362                 up_read(&cli->cl_seq_rwsem);
2363
2364                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE);
2365                 break;
2366         }
2367         case IMP_EVENT_INVALIDATE: {
2368                 struct ldlm_namespace *ns = obd->obd_namespace;
2369
2370                 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2371
2372                 break;
2373         }
2374         case IMP_EVENT_ACTIVE:
2375                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE);
2376                 /* redo the kuc registration after reconnecting */
2377                 if (rc == 0)
2378                         rc = mdc_kuc_reregister(imp);
2379                 break;
2380         case IMP_EVENT_OCD:
2381                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD);
2382                 break;
2383         case IMP_EVENT_DISCON:
2384         case IMP_EVENT_DEACTIVATE:
2385         case IMP_EVENT_ACTIVATE:
2386                 break;
2387         default:
2388                 CERROR("Unknown import event %x\n", event);
2389                 LBUG();
2390         }
2391         RETURN(rc);
2392 }
2393
2394 int mdc_fid_alloc(const struct lu_env *env, struct obd_export *exp,
2395                   struct lu_fid *fid, struct md_op_data *op_data)
2396 {
2397         struct client_obd *cli = &exp->exp_obd->u.cli;
2398         int rc = -EIO;
2399
2400         ENTRY;
2401
2402         down_read(&cli->cl_seq_rwsem);
2403         if (cli->cl_seq)
2404                 rc = seq_client_alloc_fid(env, cli->cl_seq, fid);
2405         up_read(&cli->cl_seq_rwsem);
2406
2407         RETURN(rc);
2408 }
2409
2410 static struct obd_uuid *mdc_get_uuid(struct obd_export *exp)
2411 {
2412         struct client_obd *cli = &exp->exp_obd->u.cli;
2413         return &cli->cl_target_uuid;
2414 }
2415
2416 /**
2417  * Determine whether the lock can be canceled before replaying it during
2418  * recovery, non zero value will be return if the lock can be canceled,
2419  * or zero returned for not
2420  */
2421 static int mdc_cancel_weight(struct ldlm_lock *lock)
2422 {
2423         if (lock->l_resource->lr_type != LDLM_IBITS)
2424                 RETURN(0);
2425
2426         /* FIXME: if we ever get into a situation where there are too many
2427          * opened files with open locks on a single node, then we really
2428          * should replay these open locks to reget it */
2429         if (lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_OPEN)
2430                 RETURN(0);
2431
2432         RETURN(1);
2433 }
2434
2435 static int mdc_resource_inode_free(struct ldlm_resource *res)
2436 {
2437         if (res->lr_lvb_inode)
2438                 res->lr_lvb_inode = NULL;
2439
2440         return 0;
2441 }
2442
2443 static struct ldlm_valblock_ops inode_lvbo = {
2444         .lvbo_free = mdc_resource_inode_free
2445 };
2446
2447 static int mdc_llog_init(struct obd_device *obd)
2448 {
2449         struct obd_llog_group   *olg = &obd->obd_olg;
2450         struct llog_ctxt        *ctxt;
2451         int                      rc;
2452
2453         ENTRY;
2454
2455         rc = llog_setup(NULL, obd, olg, LLOG_CHANGELOG_REPL_CTXT, obd,
2456                         &llog_client_ops);
2457         if (rc < 0)
2458                 RETURN(rc);
2459
2460         ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
2461         llog_initiator_connect(ctxt);
2462         llog_ctxt_put(ctxt);
2463
2464         RETURN(0);
2465 }
2466
2467 static void mdc_llog_finish(struct obd_device *obd)
2468 {
2469         struct llog_ctxt *ctxt;
2470
2471         ENTRY;
2472
2473         ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
2474         if (ctxt != NULL)
2475                 llog_cleanup(NULL, ctxt);
2476
2477         EXIT;
2478 }
2479
2480 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
2481 {
2482         int                             rc;
2483         ENTRY;
2484
2485         rc = ptlrpcd_addref();
2486         if (rc < 0)
2487                 RETURN(rc);
2488
2489         rc = client_obd_setup(obd, cfg);
2490         if (rc)
2491                 GOTO(err_ptlrpcd_decref, rc);
2492 #ifdef CONFIG_PROC_FS
2493         obd->obd_vars = lprocfs_mdc_obd_vars;
2494         lprocfs_obd_setup(obd, false);
2495         lprocfs_alloc_md_stats(obd, 0);
2496 #endif
2497         sptlrpc_lprocfs_cliobd_attach(obd);
2498         ptlrpc_lprocfs_register_obd(obd);
2499
2500         ns_register_cancel(obd->obd_namespace, mdc_cancel_weight);
2501
2502         obd->obd_namespace->ns_lvbo = &inode_lvbo;
2503
2504         rc = mdc_llog_init(obd);
2505         if (rc) {
2506                 CERROR("%s: failed to setup llogging subsystems: rc = %d\n",
2507                        obd->obd_name, rc);
2508                 GOTO(err_mdc_cleanup, rc);
2509         }
2510
2511         rc = mdc_changelog_cdev_init(obd);
2512         if (rc) {
2513                 CERROR("%s: failed to setup changelog char device: rc = %d\n",
2514                        obd->obd_name, rc);
2515                 GOTO(err_mdc_cleanup, rc);
2516         }
2517
2518         EXIT;
2519 err_mdc_cleanup:
2520         if (rc)
2521                 client_obd_cleanup(obd);
2522
2523 err_ptlrpcd_decref:
2524         if (rc)
2525                 ptlrpcd_decref();
2526
2527         return rc;
2528 }
2529
2530 /* Initialize the default and maximum LOV EA sizes.  This allows
2531  * us to make MDS RPCs with large enough reply buffers to hold a default
2532  * sized EA without having to calculate this (via a call into the
2533  * LOV + OSCs) each time we make an RPC.  The maximum size is also tracked
2534  * but not used to avoid wastefully vmalloc()'ing large reply buffers when
2535  * a large number of stripes is possible.  If a larger reply buffer is
2536  * required it will be reallocated in the ptlrpc layer due to overflow.
2537  */
2538 static int mdc_init_ea_size(struct obd_export *exp, __u32 easize,
2539                             __u32 def_easize)
2540 {
2541         struct obd_device *obd = exp->exp_obd;
2542         struct client_obd *cli = &obd->u.cli;
2543         ENTRY;
2544
2545         if (cli->cl_max_mds_easize < easize)
2546                 cli->cl_max_mds_easize = easize;
2547
2548         if (cli->cl_default_mds_easize < def_easize)
2549                 cli->cl_default_mds_easize = def_easize;
2550
2551         RETURN(0);
2552 }
2553
2554 static int mdc_precleanup(struct obd_device *obd)
2555 {
2556         ENTRY;
2557
2558         /* Failsafe, ok if racy */
2559         if (obd->obd_type->typ_refcnt <= 1)
2560                 libcfs_kkuc_group_rem(0, KUC_GRP_HSM);
2561
2562         mdc_changelog_cdev_finish(obd);
2563
2564         obd_cleanup_client_import(obd);
2565         ptlrpc_lprocfs_unregister_obd(obd);
2566         lprocfs_obd_cleanup(obd);
2567         lprocfs_free_md_stats(obd);
2568         mdc_llog_finish(obd);
2569         RETURN(0);
2570 }
2571
2572 static int mdc_cleanup(struct obd_device *obd)
2573 {
2574         ptlrpcd_decref();
2575
2576         return client_obd_cleanup(obd);
2577 }
2578
2579 static int mdc_process_config(struct obd_device *obd, size_t len, void *buf)
2580 {
2581         struct lustre_cfg *lcfg = buf;
2582         int rc = class_process_proc_param(PARAM_MDC, obd->obd_vars, lcfg, obd);
2583         return (rc > 0 ? 0: rc);
2584 }
2585
2586 static struct obd_ops mdc_obd_ops = {
2587         .o_owner            = THIS_MODULE,
2588         .o_setup            = mdc_setup,
2589         .o_precleanup       = mdc_precleanup,
2590         .o_cleanup          = mdc_cleanup,
2591         .o_add_conn         = client_import_add_conn,
2592         .o_del_conn         = client_import_del_conn,
2593         .o_connect          = client_connect_import,
2594         .o_disconnect       = client_disconnect_export,
2595         .o_iocontrol        = mdc_iocontrol,
2596         .o_set_info_async   = mdc_set_info_async,
2597         .o_statfs           = mdc_statfs,
2598         .o_fid_init         = client_fid_init,
2599         .o_fid_fini         = client_fid_fini,
2600         .o_fid_alloc        = mdc_fid_alloc,
2601         .o_import_event     = mdc_import_event,
2602         .o_get_info         = mdc_get_info,
2603         .o_process_config   = mdc_process_config,
2604         .o_get_uuid         = mdc_get_uuid,
2605         .o_quotactl         = mdc_quotactl,
2606 };
2607
2608 static struct md_ops mdc_md_ops = {
2609         .m_get_root         = mdc_get_root,
2610         .m_null_inode       = mdc_null_inode,
2611         .m_close            = mdc_close,
2612         .m_create           = mdc_create,
2613         .m_enqueue          = mdc_enqueue,
2614         .m_getattr          = mdc_getattr,
2615         .m_getattr_name     = mdc_getattr_name,
2616         .m_intent_lock      = mdc_intent_lock,
2617         .m_link             = mdc_link,
2618         .m_rename           = mdc_rename,
2619         .m_setattr          = mdc_setattr,
2620         .m_setxattr         = mdc_setxattr,
2621         .m_getxattr         = mdc_getxattr,
2622         .m_fsync                = mdc_fsync,
2623         .m_read_page            = mdc_read_page,
2624         .m_unlink           = mdc_unlink,
2625         .m_cancel_unused    = mdc_cancel_unused,
2626         .m_init_ea_size     = mdc_init_ea_size,
2627         .m_set_lock_data    = mdc_set_lock_data,
2628         .m_lock_match       = mdc_lock_match,
2629         .m_get_lustre_md    = mdc_get_lustre_md,
2630         .m_free_lustre_md   = mdc_free_lustre_md,
2631         .m_set_open_replay_data = mdc_set_open_replay_data,
2632         .m_clear_open_replay_data = mdc_clear_open_replay_data,
2633         .m_intent_getattr_async = mdc_intent_getattr_async,
2634         .m_revalidate_lock      = mdc_revalidate_lock
2635 };
2636
2637 static int __init mdc_init(void)
2638 {
2639         return class_register_type(&mdc_obd_ops, &mdc_md_ops, true, NULL,
2640                                    LUSTRE_MDC_NAME, NULL);
2641 }
2642
2643 static void __exit mdc_exit(void)
2644 {
2645         class_unregister_type(LUSTRE_MDC_NAME);
2646 }
2647
2648 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
2649 MODULE_DESCRIPTION("Lustre Metadata Client");
2650 MODULE_VERSION(LUSTRE_VERSION_STRING);
2651 MODULE_LICENSE("GPL");
2652
2653 module_init(mdc_init);
2654 module_exit(mdc_exit);