lustre/mdc/mdc_request.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2016, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  */
  32
  33 #define DEBUG_SUBSYSTEM S_MDC
  34
  35 #include <linux/init.h>
  36 #include <linux/kthread.h>
  37 #include <linux/module.h>
  38 #include <linux/pagemap.h>
  39 #include <linux/user_namespace.h>
  40 #include <linux/utsname.h>
  41 #ifdef HAVE_UIDGID_HEADER
  42 # include <linux/uidgid.h>
  43 #endif
  44
  45 #include <lustre_errno.h>
  46
  47 #include <cl_object.h>
  48 #include <llog_swab.h>
  49 #include <lprocfs_status.h>
  50 #include <lustre_acl.h>
  51 #include <lustre_fid.h>
  52 #include <uapi/linux/lustre/lustre_ioctl.h>
  53 #include <lustre_kernelcomm.h>
  54 #include <lustre_lmv.h>
  55 #include <lustre_log.h>
  56 #include <uapi/linux/lustre/lustre_param.h>
  57 #include <lustre_swab.h>
  58 #include <obd_class.h>
  59 #include <lustre_osc.h>
  60
  61 #include "mdc_internal.h"
  62
  63 #define REQUEST_MINOR 244
  64
  65 static int mdc_cleanup(struct obd_device *obd);
  66
  67 static inline int mdc_queue_wait(struct ptlrpc_request *req)
  68 {
  69         struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
  70         int rc;
  71
  72         /* obd_get_request_slot() ensures that this client has no more
  73          * than cl_max_rpcs_in_flight RPCs simultaneously inf light
  74          * against an MDT. */
  75         rc = obd_get_request_slot(cli);
  76         if (rc != 0)
  77                 return rc;
  78
  79         rc = ptlrpc_queue_wait(req);
  80         obd_put_request_slot(cli);
  81
  82         return rc;
  83 }
  84
  85 /*
  86  * Send MDS_GET_ROOT RPC to fetch root FID.
  87  *
  88  * If \a fileset is not NULL it should contain a subdirectory off
  89  * the ROOT/ directory to be mounted on the client. Return the FID
  90  * of the subdirectory to the client to mount onto its mountpoint.
  91  *
  92  * \param[in]   imp     MDC import
  93  * \param[in]   fileset fileset name, which could be NULL
  94  * \param[out]  rootfid root FID of this mountpoint
  95  * \param[out]  pc      root capa will be unpacked and saved in this pointer
  96  *
  97  * \retval      0 on success, negative errno on failure
  98  */
  99 static int mdc_get_root(struct obd_export *exp, const char *fileset,
 100                          struct lu_fid *rootfid)
 101 {
 102         struct ptlrpc_request   *req;
 103         struct mdt_body         *body;
 104         int                      rc;
 105
 106         ENTRY;
 107
 108         if (fileset && !(exp_connect_flags(exp) & OBD_CONNECT_SUBTREE))
 109                 RETURN(-ENOTSUPP);
 110
 111         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 112                                 &RQF_MDS_GET_ROOT);
 113         if (req == NULL)
 114                 RETURN(-ENOMEM);
 115
 116         if (fileset != NULL)
 117                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 118                                      strlen(fileset) + 1);
 119         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_ROOT);
 120         if (rc) {
 121                 ptlrpc_request_free(req);
 122                 RETURN(rc);
 123         }
 124         mdc_pack_body(req, NULL, 0, 0, -1, 0);
 125         if (fileset != NULL) {
 126                 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 127
 128                 memcpy(name, fileset, strlen(fileset));
 129         }
 130         lustre_msg_add_flags(req->rq_reqmsg, LUSTRE_IMP_FULL);
 131         req->rq_send_state = LUSTRE_IMP_FULL;
 132
 133         ptlrpc_request_set_replen(req);
 134
 135         rc = ptlrpc_queue_wait(req);
 136         if (rc)
 137                 GOTO(out, rc);
 138
 139         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 140         if (body == NULL)
 141                 GOTO(out, rc = -EPROTO);
 142
 143         *rootfid = body->mbo_fid1;
 144         CDEBUG(D_NET, "root fid="DFID", last_committed=%llu\n",
 145                PFID(rootfid), lustre_msg_get_last_committed(req->rq_repmsg));
 146         EXIT;
 147 out:
 148         ptlrpc_req_finished(req);
 149
 150         return rc;
 151 }
 152
 153 /*
 154  * This function now is known to always saying that it will receive 4 buffers
 155  * from server. Even for cases when acl_size and md_size is zero, RPC header
 156  * will contain 4 fields and RPC itself will contain zero size fields. This is
 157  * because mdt_getattr*() _always_ returns 4 fields, but if acl is not needed
 158  * and thus zero, it shrinks it, making zero size. The same story about
 159  * md_size. And this is course of problem when client waits for smaller number
 160  * of fields. This issue will be fixed later when client gets aware of RPC
 161  * layouts.  --umka
 162  */
 163 static int mdc_getattr_common(struct obd_export *exp,
 164                               struct ptlrpc_request *req)
 165 {
 166         struct req_capsule *pill = &req->rq_pill;
 167         struct mdt_body    *body;
 168         void               *eadata;
 169         int                 rc;
 170         ENTRY;
 171
 172         /* Request message already built. */
 173         rc = ptlrpc_queue_wait(req);
 174         if (rc != 0)
 175                 RETURN(rc);
 176
 177         /* sanity check for the reply */
 178         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 179         if (body == NULL)
 180                 RETURN(-EPROTO);
 181
 182         CDEBUG(D_NET, "mode: %o\n", body->mbo_mode);
 183
 184         mdc_update_max_ea_from_body(exp, body);
 185         if (body->mbo_eadatasize != 0) {
 186                 eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
 187                                                       body->mbo_eadatasize);
 188                 if (eadata == NULL)
 189                         RETURN(-EPROTO);
 190         }
 191
 192         RETURN(0);
 193 }
 194
 195 static int mdc_getattr(struct obd_export *exp, struct md_op_data *op_data,
 196                        struct ptlrpc_request **request)
 197 {
 198         struct ptlrpc_request *req;
 199         int                    rc;
 200         ENTRY;
 201
 202         /* Single MDS without an LMV case */
 203         if (op_data->op_flags & MF_GET_MDT_IDX) {
 204                 op_data->op_mds = 0;
 205                 RETURN(0);
 206         }
 207         *request = NULL;
 208         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
 209         if (req == NULL)
 210                 RETURN(-ENOMEM);
 211
 212         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
 213         if (rc) {
 214                 ptlrpc_request_free(req);
 215                 RETURN(rc);
 216         }
 217
 218         mdc_pack_body(req, &op_data->op_fid1, op_data->op_valid,
 219                       op_data->op_mode, -1, 0);
 220
 221         req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
 222                              req->rq_import->imp_connect_data.ocd_max_easize);
 223         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 224                              op_data->op_mode);
 225         ptlrpc_request_set_replen(req);
 226
 227         rc = mdc_getattr_common(exp, req);
 228         if (rc)
 229                 ptlrpc_req_finished(req);
 230         else
 231                 *request = req;
 232         RETURN(rc);
 233 }
 234
 235 static int mdc_getattr_name(struct obd_export *exp, struct md_op_data *op_data,
 236                             struct ptlrpc_request **request)
 237 {
 238         struct ptlrpc_request *req;
 239         int                    rc;
 240         ENTRY;
 241
 242         *request = NULL;
 243         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 244                                    &RQF_MDS_GETATTR_NAME);
 245         if (req == NULL)
 246                 RETURN(-ENOMEM);
 247
 248         req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 249                              op_data->op_namelen + 1);
 250
 251         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR_NAME);
 252         if (rc) {
 253                 ptlrpc_request_free(req);
 254                 RETURN(rc);
 255         }
 256
 257         mdc_pack_body(req, &op_data->op_fid1, op_data->op_valid,
 258                       op_data->op_mode, op_data->op_suppgids[0], 0);
 259
 260         if (op_data->op_name) {
 261                 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 262                 LASSERT(strnlen(op_data->op_name, op_data->op_namelen) ==
 263                                 op_data->op_namelen);
 264                 memcpy(name, op_data->op_name, op_data->op_namelen);
 265         }
 266
 267         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 268                              op_data->op_mode);
 269         req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
 270                              req->rq_import->imp_connect_data.ocd_max_easize);
 271         ptlrpc_request_set_replen(req);
 272
 273         rc = mdc_getattr_common(exp, req);
 274         if (rc)
 275                 ptlrpc_req_finished(req);
 276         else
 277                 *request = req;
 278         RETURN(rc);
 279 }
 280
 281 static int mdc_xattr_common(struct obd_export *exp,const struct req_format *fmt,
 282                             const struct lu_fid *fid, int opcode, u64 valid,
 283                             const char *xattr_name, const char *input,
 284                             int input_size, int output_size, int flags,
 285                             __u32 suppgid, struct ptlrpc_request **request)
 286 {
 287         struct ptlrpc_request *req;
 288         int   xattr_namelen = 0;
 289         char *tmp;
 290         int   rc;
 291         ENTRY;
 292
 293         *request = NULL;
 294         req = ptlrpc_request_alloc(class_exp2cliimp(exp), fmt);
 295         if (req == NULL)
 296                 RETURN(-ENOMEM);
 297
 298         if (xattr_name) {
 299                 xattr_namelen = strlen(xattr_name) + 1;
 300                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 301                                      xattr_namelen);
 302         }
 303         if (input_size) {
 304                 LASSERT(input);
 305                 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
 306                                      input_size);
 307         }
 308
 309         /* Flush local XATTR locks to get rid of a possible cancel RPC */
 310         if (opcode == MDS_REINT && fid_is_sane(fid) &&
 311             exp->exp_connect_data.ocd_ibits_known & MDS_INODELOCK_XATTR) {
 312                 struct list_head cancels = LIST_HEAD_INIT(cancels);
 313                 int count;
 314
 315                 /* Without that packing would fail */
 316                 if (input_size == 0)
 317                         req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
 318                                              RCL_CLIENT, 0);
 319
 320                 count = mdc_resource_get_unused(exp, fid,
 321                                                 &cancels, LCK_EX,
 322                                                 MDS_INODELOCK_XATTR);
 323
 324                 rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
 325                 if (rc) {
 326                         ptlrpc_request_free(req);
 327                         RETURN(rc);
 328                 }
 329         } else {
 330                 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode);
 331                 if (rc) {
 332                         ptlrpc_request_free(req);
 333                         RETURN(rc);
 334                 }
 335         }
 336
 337         if (opcode == MDS_REINT) {
 338                 struct mdt_rec_setxattr *rec;
 339
 340                 CLASSERT(sizeof(struct mdt_rec_setxattr) ==
 341                          sizeof(struct mdt_rec_reint));
 342                 rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
 343                 rec->sx_opcode = REINT_SETXATTR;
 344                 rec->sx_fsuid  = from_kuid(&init_user_ns, current_fsuid());
 345                 rec->sx_fsgid  = from_kgid(&init_user_ns, current_fsgid());
 346                 rec->sx_cap    = cfs_curproc_cap_pack();
 347                 rec->sx_suppgid1 = suppgid;
 348                 rec->sx_suppgid2 = -1;
 349                 rec->sx_fid    = *fid;
 350                 rec->sx_valid  = valid | OBD_MD_FLCTIME;
 351                 rec->sx_time   = ktime_get_real_seconds();
 352                 rec->sx_size   = output_size;
 353                 rec->sx_flags  = flags;
 354         } else {
 355                 mdc_pack_body(req, fid, valid, output_size, suppgid, flags);
 356         }
 357
 358         if (xattr_name) {
 359                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 360                 memcpy(tmp, xattr_name, xattr_namelen);
 361         }
 362         if (input_size) {
 363                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_EADATA);
 364                 memcpy(tmp, input, input_size);
 365         }
 366
 367         if (req_capsule_has_field(&req->rq_pill, &RMF_EADATA, RCL_SERVER))
 368                 req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
 369                                      RCL_SERVER, output_size);
 370         ptlrpc_request_set_replen(req);
 371
 372         /* make rpc */
 373         if (opcode == MDS_REINT)
 374                 mdc_get_mod_rpc_slot(req, NULL);
 375
 376         rc = ptlrpc_queue_wait(req);
 377
 378         if (opcode == MDS_REINT)
 379                 mdc_put_mod_rpc_slot(req, NULL);
 380
 381         if (rc)
 382                 ptlrpc_req_finished(req);
 383         else
 384                 *request = req;
 385         RETURN(rc);
 386 }
 387
 388 static int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
 389                         u64 valid, const char *xattr_name,
 390                         const char *input, int input_size, int output_size,
 391                         int flags, __u32 suppgid,
 392                         struct ptlrpc_request **request)
 393 {
 394         return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
 395                                 fid, MDS_REINT, valid, xattr_name,
 396                                 input, input_size, output_size, flags,
 397                                 suppgid, request);
 398 }
 399
 400 static int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
 401                         u64 valid, const char *xattr_name,
 402                         const char *input, int input_size, int output_size,
 403                         int flags, struct ptlrpc_request **request)
 404 {
 405         return mdc_xattr_common(exp, &RQF_MDS_GETXATTR,
 406                                 fid, MDS_GETXATTR, valid, xattr_name,
 407                                 input, input_size, output_size, flags,
 408                                 -1, request);
 409 }
 410
 411 #ifdef CONFIG_FS_POSIX_ACL
 412 static int mdc_unpack_acl(struct ptlrpc_request *req, struct lustre_md *md)
 413 {
 414         struct req_capsule     *pill = &req->rq_pill;
 415         struct mdt_body        *body = md->body;
 416         struct posix_acl       *acl;
 417         void                   *buf;
 418         int                     rc;
 419         ENTRY;
 420
 421         if (!body->mbo_aclsize)
 422                 RETURN(0);
 423
 424         buf = req_capsule_server_sized_get(pill, &RMF_ACL, body->mbo_aclsize);
 425
 426         if (!buf)
 427                 RETURN(-EPROTO);
 428
 429         acl = posix_acl_from_xattr(&init_user_ns, buf, body->mbo_aclsize);
 430         if (acl == NULL)
 431                 RETURN(0);
 432         if (IS_ERR(acl)) {
 433                 rc = PTR_ERR(acl);
 434                 CERROR("convert xattr to acl: %d\n", rc);
 435                 RETURN(rc);
 436         }
 437
 438         rc = posix_acl_valid(&init_user_ns, acl);
 439         if (rc) {
 440                 CERROR("validate acl: %d\n", rc);
 441                 posix_acl_release(acl);
 442                 RETURN(rc);
 443         }
 444
 445         md->posix_acl = acl;
 446         RETURN(0);
 447 }
 448 #else
 449 #define mdc_unpack_acl(req, md) 0
 450 #endif
 451
 452 int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
 453                       struct obd_export *dt_exp, struct obd_export *md_exp,
 454                       struct lustre_md *md)
 455 {
 456         struct req_capsule *pill = &req->rq_pill;
 457         int rc;
 458         ENTRY;
 459
 460         LASSERT(md);
 461         memset(md, 0, sizeof(*md));
 462
 463         md->body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 464         LASSERT(md->body != NULL);
 465
 466         if (md->body->mbo_valid & OBD_MD_FLEASIZE) {
 467                 if (!S_ISREG(md->body->mbo_mode)) {
 468                         CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, should be a "
 469                                "regular file, but is not\n");
 470                         GOTO(out, rc = -EPROTO);
 471                 }
 472
 473                 if (md->body->mbo_eadatasize == 0) {
 474                         CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, "
 475                                "but eadatasize 0\n");
 476                         GOTO(out, rc = -EPROTO);
 477                 }
 478
 479                 md->layout.lb_len = md->body->mbo_eadatasize;
 480                 md->layout.lb_buf = req_capsule_server_sized_get(pill,
 481                                                         &RMF_MDT_MD,
 482                                                         md->layout.lb_len);
 483                 if (md->layout.lb_buf == NULL)
 484                         GOTO(out, rc = -EPROTO);
 485         } else if (md->body->mbo_valid & OBD_MD_FLDIREA) {
 486                 const union lmv_mds_md *lmv;
 487                 size_t lmv_size;
 488
 489                 if (!S_ISDIR(md->body->mbo_mode)) {
 490                         CDEBUG(D_INFO, "OBD_MD_FLDIREA set, should be a "
 491                                "directory, but is not\n");
 492                         GOTO(out, rc = -EPROTO);
 493                 }
 494
 495                 lmv_size = md->body->mbo_eadatasize;
 496                 if (lmv_size == 0) {
 497                         CDEBUG(D_INFO, "OBD_MD_FLDIREA is set, "
 498                                "but eadatasize 0\n");
 499                         RETURN(-EPROTO);
 500                 }
 501
 502                 if (md->body->mbo_valid & OBD_MD_MEA) {
 503                         lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
 504                                                            lmv_size);
 505                         if (lmv == NULL)
 506                                 GOTO(out, rc = -EPROTO);
 507
 508                         rc = md_unpackmd(md_exp, &md->lmv, lmv, lmv_size);
 509                         if (rc < 0)
 510                                 GOTO(out, rc);
 511
 512                         if (rc < (typeof(rc))sizeof(*md->lmv)) {
 513                                 CDEBUG(D_INFO, "size too small:  "
 514                                        "rc < sizeof(*md->lmv) (%d < %d)\n",
 515                                         rc, (int)sizeof(*md->lmv));
 516                                 GOTO(out, rc = -EPROTO);
 517                         }
 518                 }
 519         }
 520         rc = 0;
 521
 522         if (md->body->mbo_valid & OBD_MD_FLACL) {
 523                 /* for ACL, it's possible that FLACL is set but aclsize is zero.
 524                  * only when aclsize != 0 there's an actual segment for ACL
 525                  * in reply buffer.
 526                  */
 527                 if (md->body->mbo_aclsize) {
 528                         rc = mdc_unpack_acl(req, md);
 529                         if (rc)
 530                                 GOTO(out, rc);
 531 #ifdef CONFIG_FS_POSIX_ACL
 532                 } else {
 533                         md->posix_acl = NULL;
 534 #endif
 535                 }
 536         }
 537
 538         EXIT;
 539 out:
 540         if (rc) {
 541 #ifdef CONFIG_FS_POSIX_ACL
 542                 posix_acl_release(md->posix_acl);
 543 #endif
 544         }
 545         return rc;
 546 }
 547
 548 int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
 549 {
 550         ENTRY;
 551         RETURN(0);
 552 }
 553
 554 void mdc_replay_open(struct ptlrpc_request *req)
 555 {
 556         struct md_open_data *mod = req->rq_cb_data;
 557         struct ptlrpc_request *close_req;
 558         struct obd_client_handle *och;
 559         struct lustre_handle old;
 560         struct mdt_body *body;
 561         ENTRY;
 562
 563         if (mod == NULL) {
 564                 DEBUG_REQ(D_ERROR, req,
 565                           "Can't properly replay without open data.");
 566                 EXIT;
 567                 return;
 568         }
 569
 570         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 571         LASSERT(body != NULL);
 572
 573         spin_lock(&req->rq_lock);
 574         och = mod->mod_och;
 575         if (och && och->och_fh.cookie)
 576                 req->rq_early_free_repbuf = 1;
 577         else
 578                 req->rq_early_free_repbuf = 0;
 579         spin_unlock(&req->rq_lock);
 580
 581         if (req->rq_early_free_repbuf) {
 582                 struct lustre_handle *file_fh;
 583
 584                 LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
 585
 586                 file_fh = &och->och_fh;
 587                 CDEBUG(D_HA, "updating handle from %#llx to %#llx\n",
 588                        file_fh->cookie, body->mbo_handle.cookie);
 589                 old = *file_fh;
 590                 *file_fh = body->mbo_handle;
 591         }
 592
 593         close_req = mod->mod_close_req;
 594         if (close_req) {
 595                 __u32 opc = lustre_msg_get_opc(close_req->rq_reqmsg);
 596                 struct mdt_ioepoch *epoch;
 597
 598                 LASSERT(opc == MDS_CLOSE);
 599                 epoch = req_capsule_client_get(&close_req->rq_pill,
 600                                                &RMF_MDT_EPOCH);
 601                 LASSERT(epoch);
 602
 603                 if (req->rq_early_free_repbuf)
 604                         LASSERT(!memcmp(&old, &epoch->mio_handle, sizeof(old)));
 605
 606                 DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
 607                 epoch->mio_handle = body->mbo_handle;
 608         }
 609         EXIT;
 610 }
 611
 612 void mdc_commit_open(struct ptlrpc_request *req)
 613 {
 614         struct md_open_data *mod = req->rq_cb_data;
 615         if (mod == NULL)
 616                 return;
 617
 618         /**
 619          * No need to touch md_open_data::mod_och, it holds a reference on
 620          * \var mod and will zero references to each other, \var mod will be
 621          * freed after that when md_open_data::mod_och will put the reference.
 622          */
 623
 624         /**
 625          * Do not let open request to disappear as it still may be needed
 626          * for close rpc to happen (it may happen on evict only, otherwise
 627          * ptlrpc_request::rq_replay does not let mdc_commit_open() to be
 628          * called), just mark this rpc as committed to distinguish these 2
 629          * cases, see mdc_close() for details. The open request reference will
 630          * be put along with freeing \var mod.
 631          */
 632         ptlrpc_request_addref(req);
 633         spin_lock(&req->rq_lock);
 634         req->rq_committed = 1;
 635         spin_unlock(&req->rq_lock);
 636         req->rq_cb_data = NULL;
 637         obd_mod_put(mod);
 638 }
 639
 640 int mdc_set_open_replay_data(struct obd_export *exp,
 641                              struct obd_client_handle *och,
 642                              struct lookup_intent *it)
 643 {
 644         struct md_open_data     *mod;
 645         struct mdt_rec_create   *rec;
 646         struct mdt_body         *body;
 647         struct ptlrpc_request   *open_req = it->it_request;
 648         struct obd_import       *imp = open_req->rq_import;
 649         ENTRY;
 650
 651         if (!open_req->rq_replay)
 652                 RETURN(0);
 653
 654         rec = req_capsule_client_get(&open_req->rq_pill, &RMF_REC_REINT);
 655         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
 656         LASSERT(rec != NULL);
 657         /* Incoming message in my byte order (it's been swabbed). */
 658         /* Outgoing messages always in my byte order. */
 659         LASSERT(body != NULL);
 660
 661         /* Only if the import is replayable, we set replay_open data */
 662         if (och && imp->imp_replayable) {
 663                 mod = obd_mod_alloc();
 664                 if (mod == NULL) {
 665                         DEBUG_REQ(D_ERROR, open_req,
 666                                   "Can't allocate md_open_data");
 667                         RETURN(0);
 668                 }
 669
 670                 /**
 671                  * Take a reference on \var mod, to be freed on mdc_close().
 672                  * It protects \var mod from being freed on eviction (commit
 673                  * callback is called despite rq_replay flag).
 674                  * Another reference for \var och.
 675                  */
 676                 obd_mod_get(mod);
 677                 obd_mod_get(mod);
 678
 679                 spin_lock(&open_req->rq_lock);
 680                 och->och_mod = mod;
 681                 mod->mod_och = och;
 682                 mod->mod_is_create = it_disposition(it, DISP_OPEN_CREATE) ||
 683                                      it_disposition(it, DISP_OPEN_STRIPE);
 684                 mod->mod_open_req = open_req;
 685                 open_req->rq_cb_data = mod;
 686                 open_req->rq_commit_cb = mdc_commit_open;
 687                 open_req->rq_early_free_repbuf = 1;
 688                 spin_unlock(&open_req->rq_lock);
 689         }
 690
 691         rec->cr_fid2 = body->mbo_fid1;
 692         rec->cr_ioepoch = body->mbo_ioepoch;
 693         rec->cr_old_handle.cookie = body->mbo_handle.cookie;
 694         open_req->rq_replay_cb = mdc_replay_open;
 695         if (!fid_is_sane(&body->mbo_fid1)) {
 696                 DEBUG_REQ(D_ERROR, open_req, "Saving replay request with "
 697                           "insane fid");
 698                 LBUG();
 699         }
 700
 701         DEBUG_REQ(D_RPCTRACE, open_req, "Set up open replay data");
 702         RETURN(0);
 703 }
 704
 705 static void mdc_free_open(struct md_open_data *mod)
 706 {
 707         int committed = 0;
 708
 709         if (mod->mod_is_create == 0 &&
 710             imp_connect_disp_stripe(mod->mod_open_req->rq_import))
 711                 committed = 1;
 712
 713         /**
 714          * No reason to asssert here if the open request has
 715          * rq_replay == 1. It means that mdc_close failed, and
 716          * close request wasn`t sent. It is not fatal to client.
 717          * The worst thing is eviction if the client gets open lock
 718          **/
 719
 720         DEBUG_REQ(D_RPCTRACE, mod->mod_open_req, "free open request rq_replay"
 721                   "= %d\n", mod->mod_open_req->rq_replay);
 722
 723         ptlrpc_request_committed(mod->mod_open_req, committed);
 724         if (mod->mod_close_req)
 725                 ptlrpc_request_committed(mod->mod_close_req, committed);
 726 }
 727
 728 int mdc_clear_open_replay_data(struct obd_export *exp,
 729                                struct obd_client_handle *och)
 730 {
 731         struct md_open_data *mod = och->och_mod;
 732         ENTRY;
 733
 734         /**
 735          * It is possible to not have \var mod in a case of eviction between
 736          * lookup and ll_file_open().
 737          **/
 738         if (mod == NULL)
 739                 RETURN(0);
 740
 741         LASSERT(mod != LP_POISON);
 742         LASSERT(mod->mod_open_req != NULL);
 743
 744         spin_lock(&mod->mod_open_req->rq_lock);
 745         if (mod->mod_och)
 746                 mod->mod_och->och_fh.cookie = 0;
 747         mod->mod_open_req->rq_early_free_repbuf = 0;
 748         spin_unlock(&mod->mod_open_req->rq_lock);
 749         mdc_free_open(mod);
 750
 751         mod->mod_och = NULL;
 752         och->och_mod = NULL;
 753         obd_mod_put(mod);
 754
 755         RETURN(0);
 756 }
 757
 758 static int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
 759                      struct md_open_data *mod, struct ptlrpc_request **request)
 760 {
 761         struct obd_device     *obd = class_exp2obd(exp);
 762         struct ptlrpc_request *req;
 763         struct req_format     *req_fmt;
 764         int                    rc;
 765         int                    saved_rc = 0;
 766         ENTRY;
 767
 768         if (op_data->op_bias & MDS_HSM_RELEASE) {
 769                 req_fmt = &RQF_MDS_INTENT_CLOSE;
 770
 771                 /* allocate a FID for volatile file */
 772                 rc = mdc_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
 773                 if (rc < 0) {
 774                         CERROR("%s: "DFID" failed to allocate FID: %d\n",
 775                                obd->obd_name, PFID(&op_data->op_fid1), rc);
 776                         /* save the errcode and proceed to close */
 777                         saved_rc = rc;
 778                 }
 779         } else if (op_data->op_bias & MDS_CLOSE_LAYOUT_SWAP) {
 780                 req_fmt = &RQF_MDS_INTENT_CLOSE;
 781         } else {
 782                 req_fmt = &RQF_MDS_CLOSE;
 783         }
 784
 785         *request = NULL;
 786         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_CLOSE))
 787                 req = NULL;
 788         else
 789                 req = ptlrpc_request_alloc(class_exp2cliimp(exp), req_fmt);
 790
 791         /* Ensure that this close's handle is fixed up during replay. */
 792         if (likely(mod != NULL)) {
 793                 LASSERTF(mod->mod_open_req != NULL &&
 794                          mod->mod_open_req->rq_type != LI_POISON,
 795                          "POISONED open %p!\n", mod->mod_open_req);
 796
 797                 mod->mod_close_req = req;
 798
 799                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
 800                 /* We no longer want to preserve this open for replay even
 801                  * though the open was committed. b=3632, b=3633 */
 802                 spin_lock(&mod->mod_open_req->rq_lock);
 803                 mod->mod_open_req->rq_replay = 0;
 804                 spin_unlock(&mod->mod_open_req->rq_lock);
 805         } else {
 806                 CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
 807         }
 808         if (req == NULL) {
 809                 /**
 810                  * TODO: repeat close after errors
 811                  */
 812                 CWARN("%s: close of FID "DFID" failed, file reference will be "
 813                       "dropped when this client unmounts or is evicted\n",
 814                       obd->obd_name, PFID(&op_data->op_fid1));
 815                 GOTO(out, rc = -ENOMEM);
 816         }
 817
 818         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
 819         if (rc) {
 820                 ptlrpc_request_free(req);
 821                 req = NULL;
 822                 GOTO(out, rc);
 823         }
 824
 825         /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
 826          * portal whose threads are not taking any DLM locks and are therefore
 827          * always progressing */
 828         req->rq_request_portal = MDS_READPAGE_PORTAL;
 829         ptlrpc_at_set_req_timeout(req);
 830
 831
 832         mdc_close_pack(req, op_data);
 833
 834         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 835                              obd->u.cli.cl_default_mds_easize);
 836
 837         ptlrpc_request_set_replen(req);
 838
 839         mdc_get_mod_rpc_slot(req, NULL);
 840         rc = ptlrpc_queue_wait(req);
 841         mdc_put_mod_rpc_slot(req, NULL);
 842
 843         if (req->rq_repmsg == NULL) {
 844                 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
 845                        req->rq_status);
 846                 if (rc == 0)
 847                         rc = req->rq_status ?: -EIO;
 848         } else if (rc == 0 || rc == -EAGAIN) {
 849                 struct mdt_body *body;
 850
 851                 rc = lustre_msg_get_status(req->rq_repmsg);
 852                 if (lustre_msg_get_type(req->rq_repmsg) == PTL_RPC_MSG_ERR) {
 853                         DEBUG_REQ(D_ERROR, req, "type == PTL_RPC_MSG_ERR, err "
 854                                   "= %d", rc);
 855                         if (rc > 0)
 856                                 rc = -rc;
 857                 }
 858                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 859                 if (body == NULL)
 860                         rc = -EPROTO;
 861         } else if (rc == -ESTALE) {
 862                 /**
 863                  * it can be allowed error after 3633 if open was committed and
 864                  * server failed before close was sent. Let's check if mod
 865                  * exists and return no error in that case
 866                  */
 867                 if (mod) {
 868                         DEBUG_REQ(D_HA, req, "Reset ESTALE = %d", rc);
 869                         LASSERT(mod->mod_open_req != NULL);
 870                         if (mod->mod_open_req->rq_committed)
 871                                 rc = 0;
 872                 }
 873         }
 874
 875 out:
 876         if (mod) {
 877                 if (rc != 0)
 878                         mod->mod_close_req = NULL;
 879                 /* Since now, mod is accessed through open_req only,
 880                  * thus close req does not keep a reference on mod anymore. */
 881                 obd_mod_put(mod);
 882         }
 883         *request = req;
 884
 885         RETURN(rc < 0 ? rc : saved_rc);
 886 }
 887
 888 static int mdc_getpage(struct obd_export *exp, const struct lu_fid *fid,
 889                        u64 offset, struct page **pages, int npages,
 890                        struct ptlrpc_request **request)
 891 {
 892         struct ptlrpc_request   *req;
 893         struct ptlrpc_bulk_desc *desc;
 894         int                      i;
 895         wait_queue_head_t        waitq;
 896         int                      resends = 0;
 897         struct l_wait_info       lwi;
 898         int                      rc;
 899         ENTRY;
 900
 901         *request = NULL;
 902         init_waitqueue_head(&waitq);
 903
 904 restart_bulk:
 905         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
 906         if (req == NULL)
 907                 RETURN(-ENOMEM);
 908
 909         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
 910         if (rc) {
 911                 ptlrpc_request_free(req);
 912                 RETURN(rc);
 913         }
 914
 915         req->rq_request_portal = MDS_READPAGE_PORTAL;
 916         ptlrpc_at_set_req_timeout(req);
 917
 918         desc = ptlrpc_prep_bulk_imp(req, npages, 1,
 919                                     PTLRPC_BULK_PUT_SINK | PTLRPC_BULK_BUF_KIOV,
 920                                     MDS_BULK_PORTAL,
 921                                     &ptlrpc_bulk_kiov_pin_ops);
 922         if (desc == NULL) {
 923                 ptlrpc_req_finished(req);
 924                 RETURN(-ENOMEM);
 925         }
 926
 927         /* NB req now owns desc and will free it when it gets freed */
 928         for (i = 0; i < npages; i++)
 929                 desc->bd_frag_ops->add_kiov_frag(desc, pages[i], 0,
 930                                                  PAGE_SIZE);
 931
 932         mdc_readdir_pack(req, offset, PAGE_SIZE * npages, fid);
 933
 934         ptlrpc_request_set_replen(req);
 935         rc = ptlrpc_queue_wait(req);
 936         if (rc) {
 937                 ptlrpc_req_finished(req);
 938                 if (rc != -ETIMEDOUT)
 939                         RETURN(rc);
 940
 941                 resends++;
 942                 if (!client_should_resend(resends, &exp->exp_obd->u.cli)) {
 943                         CERROR("%s: too many resend retries: rc = %d\n",
 944                                exp->exp_obd->obd_name, -EIO);
 945                         RETURN(-EIO);
 946                 }
 947                 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(resends), NULL, NULL,
 948                                        NULL);
 949                 l_wait_event(waitq, 0, &lwi);
 950
 951                 goto restart_bulk;
 952         }
 953
 954         rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
 955                                           req->rq_bulk->bd_nob_transferred);
 956         if (rc < 0) {
 957                 ptlrpc_req_finished(req);
 958                 RETURN(rc);
 959         }
 960
 961         if (req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK) {
 962                 CERROR("%s: unexpected bytes transferred: %d (%ld expected)\n",
 963                        exp->exp_obd->obd_name, req->rq_bulk->bd_nob_transferred,
 964                        PAGE_SIZE * npages);
 965                 ptlrpc_req_finished(req);
 966                 RETURN(-EPROTO);
 967         }
 968
 969         *request = req;
 970         RETURN(0);
 971 }
 972
 973 static void mdc_release_page(struct page *page, int remove)
 974 {
 975         if (remove) {
 976                 lock_page(page);
 977                 if (likely(page->mapping != NULL))
 978                         truncate_complete_page(page->mapping, page);
 979                 unlock_page(page);
 980         }
 981         put_page(page);
 982 }
 983
 984 static struct page *mdc_page_locate(struct address_space *mapping, __u64 *hash,
 985                                     __u64 *start, __u64 *end, int hash64)
 986 {
 987         /*
 988          * Complement of hash is used as an index so that
 989          * radix_tree_gang_lookup() can be used to find a page with starting
 990          * hash _smaller_ than one we are looking for.
 991          */
 992         unsigned long offset = hash_x_index(*hash, hash64);
 993         struct page *page;
 994         int found;
 995
 996         spin_lock_irq(&mapping->tree_lock);
 997         found = radix_tree_gang_lookup(&mapping->page_tree,
 998                                        (void **)&page, offset, 1);
 999         if (found > 0 && !radix_tree_exceptional_entry(page)) {
1000                 struct lu_dirpage *dp;
1001
1002                 get_page(page);
1003                 spin_unlock_irq(&mapping->tree_lock);
1004                 /*
1005                  * In contrast to find_lock_page() we are sure that directory
1006                  * page cannot be truncated (while DLM lock is held) and,
1007                  * hence, can avoid restart.
1008                  *
1009                  * In fact, page cannot be locked here at all, because
1010                  * mdc_read_page_remote does synchronous io.
1011                  */
1012                 wait_on_page_locked(page);
1013                 if (PageUptodate(page)) {
1014                         dp = kmap(page);
1015                         if (BITS_PER_LONG == 32 && hash64) {
1016                                 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1017                                 *end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
1018                                 *hash  = *hash >> 32;
1019                         } else {
1020                                 *start = le64_to_cpu(dp->ldp_hash_start);
1021                                 *end   = le64_to_cpu(dp->ldp_hash_end);
1022                         }
1023                         if (unlikely(*start == 1 && *hash == 0))
1024                                 *hash = *start;
1025                         else
1026                                 LASSERTF(*start <= *hash, "start = %#llx"
1027                                          ",end = %#llx,hash = %#llx\n",
1028                                          *start, *end, *hash);
1029                         CDEBUG(D_VFSTRACE, "offset %lx [%#llx %#llx],"
1030                               " hash %#llx\n", offset, *start, *end, *hash);
1031                         if (*hash > *end) {
1032                                 kunmap(page);
1033                                 mdc_release_page(page, 0);
1034                                 page = NULL;
1035                         } else if (*end != *start && *hash == *end) {
1036                                 /*
1037                                  * upon hash collision, remove this page,
1038                                  * otherwise put page reference, and
1039                                  * mdc_read_page_remote() will issue RPC to
1040                                  * fetch the page we want.
1041                                  */
1042                                 kunmap(page);
1043                                 mdc_release_page(page,
1044                                     le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
1045                                 page = NULL;
1046                         }
1047                 } else {
1048                         put_page(page);
1049                         page = ERR_PTR(-EIO);
1050                 }
1051         } else {
1052                 spin_unlock_irq(&mapping->tree_lock);
1053                 page = NULL;
1054         }
1055         return page;
1056 }
1057
1058 /*
1059  * Adjust a set of pages, each page containing an array of lu_dirpages,
1060  * so that each page can be used as a single logical lu_dirpage.
1061  *
1062  * A lu_dirpage is laid out as follows, where s = ldp_hash_start,
1063  * e = ldp_hash_end, f = ldp_flags, p = padding, and each "ent" is a
1064  * struct lu_dirent.  It has size up to LU_PAGE_SIZE. The ldp_hash_end
1065  * value is used as a cookie to request the next lu_dirpage in a
1066  * directory listing that spans multiple pages (two in this example):
1067  *   ________
1068  *  |        |
1069  * .|--------v-------   -----.
1070  * |s|e|f|p|ent|ent| ... |ent|
1071  * '--|--------------   -----'   Each PAGE contains a single
1072  *    '------.                   lu_dirpage.
1073  * .---------v-------   -----.
1074  * |s|e|f|p|ent| 0 | ... | 0 |
1075  * '-----------------   -----'
1076  *
1077  * However, on hosts where the native VM page size (PAGE_SIZE) is
1078  * larger than LU_PAGE_SIZE, a single host page may contain multiple
1079  * lu_dirpages. After reading the lu_dirpages from the MDS, the
1080  * ldp_hash_end of the first lu_dirpage refers to the one immediately
1081  * after it in the same PAGE (arrows simplified for brevity, but
1082  * in general e0==s1, e1==s2, etc.):
1083  *
1084  * .--------------------   -----.
1085  * |s0|e0|f0|p|ent|ent| ... |ent|
1086  * |---v----------------   -----|
1087  * |s1|e1|f1|p|ent|ent| ... |ent|
1088  * |---v----------------   -----|  Here, each PAGE contains
1089  *             ...                 multiple lu_dirpages.
1090  * |---v----------------   -----|
1091  * |s'|e'|f'|p|ent|ent| ... |ent|
1092  * '---|----------------   -----'
1093  *     v
1094  * .----------------------------.
1095  * |        next PAGE           |
1096  *
1097  * This structure is transformed into a single logical lu_dirpage as follows:
1098  *
1099  * - Replace e0 with e' so the request for the next lu_dirpage gets the page
1100  *   labeled 'next PAGE'.
1101  *
1102  * - Copy the LDF_COLLIDE flag from f' to f0 to correctly reflect whether
1103  *   a hash collision with the next page exists.
1104  *
1105  * - Adjust the lde_reclen of the ending entry of each lu_dirpage to span
1106  *   to the first entry of the next lu_dirpage.
1107  */
1108 #if PAGE_SIZE > LU_PAGE_SIZE
1109 static void mdc_adjust_dirpages(struct page **pages, int cfs_pgs, int lu_pgs)
1110 {
1111         int i;
1112
1113         for (i = 0; i < cfs_pgs; i++) {
1114                 struct lu_dirpage       *dp = kmap(pages[i]);
1115                 struct lu_dirpage       *first = dp;
1116                 struct lu_dirent        *end_dirent = NULL;
1117                 struct lu_dirent        *ent;
1118                 __u64           hash_end = le64_to_cpu(dp->ldp_hash_end);
1119                 __u32           flags = le32_to_cpu(dp->ldp_flags);
1120
1121                 while (--lu_pgs > 0) {
1122                         ent = lu_dirent_start(dp);
1123                         for (end_dirent = ent; ent != NULL;
1124                              end_dirent = ent, ent = lu_dirent_next(ent));
1125
1126                         /* Advance dp to next lu_dirpage. */
1127                         dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE);
1128
1129                         /* Check if we've reached the end of the PAGE. */
1130                         if (!((unsigned long)dp & ~PAGE_MASK))
1131                                 break;
1132
1133                         /* Save the hash and flags of this lu_dirpage. */
1134                         hash_end = le64_to_cpu(dp->ldp_hash_end);
1135                         flags = le32_to_cpu(dp->ldp_flags);
1136
1137                         /* Check if lu_dirpage contains no entries. */
1138                         if (end_dirent == NULL)
1139                                 break;
1140
1141                         /* Enlarge the end entry lde_reclen from 0 to
1142                          * first entry of next lu_dirpage. */
1143                         LASSERT(le16_to_cpu(end_dirent->lde_reclen) == 0);
1144                         end_dirent->lde_reclen =
1145                                 cpu_to_le16((char *)(dp->ldp_entries) -
1146                                             (char *)end_dirent);
1147                 }
1148
1149                 first->ldp_hash_end = hash_end;
1150                 first->ldp_flags &= ~cpu_to_le32(LDF_COLLIDE);
1151                 first->ldp_flags |= flags & cpu_to_le32(LDF_COLLIDE);
1152
1153                 kunmap(pages[i]);
1154         }
1155         LASSERTF(lu_pgs == 0, "left = %d\n", lu_pgs);
1156 }
1157 #else
1158 #define mdc_adjust_dirpages(pages, cfs_pgs, lu_pgs) do {} while (0)
1159 #endif  /* PAGE_SIZE > LU_PAGE_SIZE */
1160
1161 /* parameters for readdir page */
1162 struct readpage_param {
1163         struct md_op_data       *rp_mod;
1164         __u64                   rp_off;
1165         int                     rp_hash64;
1166         struct obd_export       *rp_exp;
1167         struct md_callback      *rp_cb;
1168 };
1169
1170 #ifndef HAVE_DELETE_FROM_PAGE_CACHE
1171 static inline void delete_from_page_cache(struct page *page)
1172 {
1173         remove_from_page_cache(page);
1174         put_page(page);
1175 }
1176 #endif
1177
1178 /**
1179  * Read pages from server.
1180  *
1181  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
1182  * a header lu_dirpage which describes the start/end hash, and whether this
1183  * page is empty (contains no dir entry) or hash collide with next page.
1184  * After client receives reply, several pages will be integrated into dir page
1185  * in PAGE_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the
1186  * lu_dirpage for this integrated page will be adjusted.
1187  **/
1188 static int mdc_read_page_remote(void *data, struct page *page0)
1189 {
1190         struct readpage_param *rp = data;
1191         struct page **page_pool;
1192         struct page *page;
1193         struct lu_dirpage *dp;
1194         struct md_op_data *op_data = rp->rp_mod;
1195         struct ptlrpc_request *req;
1196         int max_pages;
1197         struct inode *inode;
1198         struct lu_fid *fid;
1199         int rd_pgs = 0; /* number of pages actually read */
1200         int npages;
1201         int i;
1202         int rc;
1203         ENTRY;
1204
1205         max_pages = rp->rp_exp->exp_obd->u.cli.cl_max_pages_per_rpc;
1206         inode = op_data->op_data;
1207         fid = &op_data->op_fid1;
1208         LASSERT(inode != NULL);
1209
1210         OBD_ALLOC(page_pool, sizeof(page_pool[0]) * max_pages);
1211         if (page_pool != NULL) {
1212                 page_pool[0] = page0;
1213         } else {
1214                 page_pool = &page0;
1215                 max_pages = 1;
1216         }
1217
1218         for (npages = 1; npages < max_pages; npages++) {
1219                 page = page_cache_alloc_cold(inode->i_mapping);
1220                 if (page == NULL)
1221                         break;
1222                 page_pool[npages] = page;
1223         }
1224
1225         rc = mdc_getpage(rp->rp_exp, fid, rp->rp_off, page_pool, npages, &req);
1226         if (rc < 0) {
1227                 /* page0 is special, which was added into page cache early */
1228                 delete_from_page_cache(page0);
1229         } else {
1230                 int lu_pgs;
1231
1232                 rd_pgs = (req->rq_bulk->bd_nob_transferred + PAGE_SIZE - 1) >>
1233                         PAGE_SHIFT;
1234                 lu_pgs = req->rq_bulk->bd_nob_transferred >> LU_PAGE_SHIFT;
1235                 LASSERT(!(req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK));
1236
1237                 CDEBUG(D_INODE, "read %d(%d) pages\n", rd_pgs, lu_pgs);
1238
1239                 mdc_adjust_dirpages(page_pool, rd_pgs, lu_pgs);
1240
1241                 SetPageUptodate(page0);
1242         }
1243         unlock_page(page0);
1244
1245         ptlrpc_req_finished(req);
1246         CDEBUG(D_CACHE, "read %d/%d pages\n", rd_pgs, npages);
1247         for (i = 1; i < npages; i++) {
1248                 unsigned long   offset;
1249                 __u64           hash;
1250                 int ret;
1251
1252                 page = page_pool[i];
1253
1254                 if (rc < 0 || i >= rd_pgs) {
1255                         put_page(page);
1256                         continue;
1257                 }
1258
1259                 SetPageUptodate(page);
1260
1261                 dp = kmap(page);
1262                 hash = le64_to_cpu(dp->ldp_hash_start);
1263                 kunmap(page);
1264
1265                 offset = hash_x_index(hash, rp->rp_hash64);
1266
1267                 prefetchw(&page->flags);
1268                 ret = add_to_page_cache_lru(page, inode->i_mapping, offset,
1269                                             GFP_KERNEL);
1270                 if (ret == 0)
1271                         unlock_page(page);
1272                 else
1273                         CDEBUG(D_VFSTRACE, "page %lu add to page cache failed:"
1274                                " rc = %d\n", offset, ret);
1275                 put_page(page);
1276         }
1277
1278         if (page_pool != &page0)
1279                 OBD_FREE(page_pool, sizeof(page_pool[0]) * max_pages);
1280
1281         RETURN(rc);
1282 }
1283
1284 /**
1285  * Read dir page from cache first, if it can not find it, read it from
1286  * server and add into the cache.
1287  *
1288  * \param[in] exp       MDC export
1289  * \param[in] op_data   client MD stack parameters, transfering parameters
1290  *                      between different layers on client MD stack.
1291  * \param[in] cb_op     callback required for ldlm lock enqueue during
1292  *                      read page
1293  * \param[in] hash_offset the hash offset of the page to be read
1294  * \param[in] ppage     the page to be read
1295  *
1296  * retval               = 0 get the page successfully
1297  *                      errno(<0) get the page failed
1298  */
1299 static int mdc_read_page(struct obd_export *exp, struct md_op_data *op_data,
1300                          struct md_callback *cb_op, __u64 hash_offset,
1301                          struct page **ppage)
1302 {
1303         struct lookup_intent    it = { .it_op = IT_READDIR };
1304         struct page             *page;
1305         struct inode            *dir = op_data->op_data;
1306         struct address_space    *mapping;
1307         struct lu_dirpage       *dp;
1308         __u64                   start = 0;
1309         __u64                   end = 0;
1310         struct lustre_handle    lockh;
1311         struct ptlrpc_request   *enq_req = NULL;
1312         struct readpage_param   rp_param;
1313         int rc;
1314
1315         ENTRY;
1316
1317         *ppage = NULL;
1318
1319         LASSERT(dir != NULL);
1320         mapping = dir->i_mapping;
1321
1322         rc = mdc_intent_lock(exp, op_data, &it, &enq_req,
1323                              cb_op->md_blocking_ast, 0);
1324         if (enq_req != NULL)
1325                 ptlrpc_req_finished(enq_req);
1326
1327         if (rc < 0) {
1328                 CERROR("%s: "DFID" lock enqueue fails: rc = %d\n",
1329                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1), rc);
1330                 RETURN(rc);
1331         }
1332
1333         rc = 0;
1334         lockh.cookie = it.it_lock_handle;
1335         mdc_set_lock_data(exp, &lockh, dir, NULL);
1336
1337         rp_param.rp_off = hash_offset;
1338         rp_param.rp_hash64 = op_data->op_cli_flags & CLI_HASH64;
1339         page = mdc_page_locate(mapping, &rp_param.rp_off, &start, &end,
1340                                rp_param.rp_hash64);
1341         if (IS_ERR(page)) {
1342                 CERROR("%s: dir page locate: "DFID" at %llu: rc %ld\n",
1343                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1344                        rp_param.rp_off, PTR_ERR(page));
1345                 GOTO(out_unlock, rc = PTR_ERR(page));
1346         } else if (page != NULL) {
1347                 /*
1348                  * XXX nikita: not entirely correct handling of a corner case:
1349                  * suppose hash chain of entries with hash value HASH crosses
1350                  * border between pages P0 and P1. First both P0 and P1 are
1351                  * cached, seekdir() is called for some entry from the P0 part
1352                  * of the chain. Later P0 goes out of cache. telldir(HASH)
1353                  * happens and finds P1, as it starts with matching hash
1354                  * value. Remaining entries from P0 part of the chain are
1355                  * skipped. (Is that really a bug?)
1356                  *
1357                  * Possible solutions: 0. don't cache P1 is such case, handle
1358                  * it as an "overflow" page. 1. invalidate all pages at
1359                  * once. 2. use HASH|1 as an index for P1.
1360                  */
1361                 GOTO(hash_collision, page);
1362         }
1363
1364         rp_param.rp_exp = exp;
1365         rp_param.rp_mod = op_data;
1366         page = read_cache_page(mapping,
1367                                hash_x_index(rp_param.rp_off,
1368                                             rp_param.rp_hash64),
1369                                mdc_read_page_remote, &rp_param);
1370         if (IS_ERR(page)) {
1371                 CDEBUG(D_INFO, "%s: read cache page: "DFID" at %llu: %ld\n",
1372                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1373                        rp_param.rp_off, PTR_ERR(page));
1374                 GOTO(out_unlock, rc = PTR_ERR(page));
1375         }
1376
1377         wait_on_page_locked(page);
1378         (void)kmap(page);
1379         if (!PageUptodate(page)) {
1380                 CERROR("%s: page not updated: "DFID" at %llu: rc %d\n",
1381                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1382                        rp_param.rp_off, -5);
1383                 goto fail;
1384         }
1385         if (!PageChecked(page))
1386                 SetPageChecked(page);
1387         if (PageError(page)) {
1388                 CERROR("%s: page error: "DFID" at %llu: rc %d\n",
1389                        exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
1390                        rp_param.rp_off, -5);
1391                 goto fail;
1392         }
1393
1394 hash_collision:
1395         dp = page_address(page);
1396         if (BITS_PER_LONG == 32 && rp_param.rp_hash64) {
1397                 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
1398                 end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
1399                 rp_param.rp_off = hash_offset >> 32;
1400         } else {
1401                 start = le64_to_cpu(dp->ldp_hash_start);
1402                 end   = le64_to_cpu(dp->ldp_hash_end);
1403                 rp_param.rp_off = hash_offset;
1404         }
1405         if (end == start) {
1406                 LASSERT(start == rp_param.rp_off);
1407                 CWARN("Page-wide hash collision: %#lx\n", (unsigned long)end);
1408 #if BITS_PER_LONG == 32
1409                 CWARN("Real page-wide hash collision at [%llu %llu] with "
1410                       "hash %llu\n", le64_to_cpu(dp->ldp_hash_start),
1411                       le64_to_cpu(dp->ldp_hash_end), hash_offset);
1412 #endif
1413
1414                 /*
1415                  * Fetch whole overflow chain...
1416                  *
1417                  * XXX not yet.
1418                  */
1419                 goto fail;
1420         }
1421         *ppage = page;
1422 out_unlock:
1423         ldlm_lock_decref(&lockh, it.it_lock_mode);
1424         return rc;
1425 fail:
1426         kunmap(page);
1427         mdc_release_page(page, 1);
1428         rc = -EIO;
1429         goto out_unlock;
1430 }
1431
1432
1433 static int mdc_statfs(const struct lu_env *env,
1434                       struct obd_export *exp, struct obd_statfs *osfs,
1435                       __u64 max_age, __u32 flags)
1436 {
1437         struct obd_device     *obd = class_exp2obd(exp);
1438         struct ptlrpc_request *req;
1439         struct obd_statfs     *msfs;
1440         struct obd_import     *imp = NULL;
1441         int                    rc;
1442         ENTRY;
1443
1444         /*
1445          * Since the request might also come from lprocfs, so we need
1446          * sync this with client_disconnect_export Bug15684
1447          */
1448         down_read(&obd->u.cli.cl_sem);
1449         if (obd->u.cli.cl_import)
1450                 imp = class_import_get(obd->u.cli.cl_import);
1451         up_read(&obd->u.cli.cl_sem);
1452         if (!imp)
1453                 RETURN(-ENODEV);
1454
1455         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_STATFS,
1456                                         LUSTRE_MDS_VERSION, MDS_STATFS);
1457         if (req == NULL)
1458                 GOTO(output, rc = -ENOMEM);
1459
1460         ptlrpc_request_set_replen(req);
1461
1462         if (flags & OBD_STATFS_NODELAY) {
1463                 /* procfs requests not want stay in wait for avoid deadlock */
1464                 req->rq_no_resend = 1;
1465                 req->rq_no_delay = 1;
1466         }
1467
1468         rc = ptlrpc_queue_wait(req);
1469         if (rc) {
1470                 /* check connection error first */
1471                 if (imp->imp_connect_error)
1472                         rc = imp->imp_connect_error;
1473                 GOTO(out, rc);
1474         }
1475
1476         msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
1477         if (msfs == NULL)
1478                 GOTO(out, rc = -EPROTO);
1479
1480         *osfs = *msfs;
1481         EXIT;
1482 out:
1483         ptlrpc_req_finished(req);
1484 output:
1485         class_import_put(imp);
1486         return rc;
1487 }
1488
1489 static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
1490 {
1491         __u32 keylen, vallen;
1492         void *key;
1493         int rc;
1494
1495         if (gf->gf_pathlen > PATH_MAX)
1496                 RETURN(-ENAMETOOLONG);
1497         if (gf->gf_pathlen < 2)
1498                 RETURN(-EOVERFLOW);
1499
1500         /* Key is KEY_FID2PATH + getinfo_fid2path description */
1501         keylen = cfs_size_round(sizeof(KEY_FID2PATH) + sizeof(*gf) +
1502                                 sizeof(struct lu_fid));
1503         OBD_ALLOC(key, keylen);
1504         if (key == NULL)
1505                 RETURN(-ENOMEM);
1506         memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
1507         memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf));
1508         memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf),
1509                gf->gf_u.gf_root_fid, sizeof(struct lu_fid));
1510         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
1511                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
1512
1513         if (!fid_is_sane(&gf->gf_fid))
1514                 GOTO(out, rc = -EINVAL);
1515
1516         /* Val is struct getinfo_fid2path result plus path */
1517         vallen = sizeof(*gf) + gf->gf_pathlen;
1518
1519         rc = obd_get_info(NULL, exp, keylen, key, &vallen, gf);
1520         if (rc != 0 && rc != -EREMOTE)
1521                 GOTO(out, rc);
1522
1523         if (vallen <= sizeof(*gf))
1524                 GOTO(out, rc = -EPROTO);
1525         if (vallen > sizeof(*gf) + gf->gf_pathlen)
1526                 GOTO(out, rc = -EOVERFLOW);
1527
1528         CDEBUG(D_IOCTL, "path got "DFID" from %llu #%d: %s\n",
1529                PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno,
1530                gf->gf_pathlen < 512 ? gf->gf_u.gf_path :
1531                /* only log the last 512 characters of the path */
1532                gf->gf_u.gf_path + gf->gf_pathlen - 512);
1533
1534 out:
1535         OBD_FREE(key, keylen);
1536         return rc;
1537 }
1538
1539 static int mdc_ioc_hsm_progress(struct obd_export *exp,
1540                                 struct hsm_progress_kernel *hpk)
1541 {
1542         struct obd_import               *imp = class_exp2cliimp(exp);
1543         struct hsm_progress_kernel      *req_hpk;
1544         struct ptlrpc_request           *req;
1545         int                              rc;
1546         ENTRY;
1547
1548         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_PROGRESS,
1549                                         LUSTRE_MDS_VERSION, MDS_HSM_PROGRESS);
1550         if (req == NULL)
1551                 GOTO(out, rc = -ENOMEM);
1552
1553         mdc_pack_body(req, NULL, 0, 0, -1, 0);
1554
1555         /* Copy hsm_progress struct */
1556         req_hpk = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_PROGRESS);
1557         if (req_hpk == NULL)
1558                 GOTO(out, rc = -EPROTO);
1559
1560         *req_hpk = *hpk;
1561         req_hpk->hpk_errval = lustre_errno_hton(hpk->hpk_errval);
1562
1563         ptlrpc_request_set_replen(req);
1564
1565         mdc_get_mod_rpc_slot(req, NULL);
1566         rc = ptlrpc_queue_wait(req);
1567         mdc_put_mod_rpc_slot(req, NULL);
1568
1569         GOTO(out, rc);
1570 out:
1571         ptlrpc_req_finished(req);
1572         return rc;
1573 }
1574
1575 static int mdc_ioc_hsm_ct_register(struct obd_import *imp, __u32 archives)
1576 {
1577         __u32                   *archive_mask;
1578         struct ptlrpc_request   *req;
1579         int                      rc;
1580         ENTRY;
1581
1582         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_REGISTER,
1583                                         LUSTRE_MDS_VERSION,
1584                                         MDS_HSM_CT_REGISTER);
1585         if (req == NULL)
1586                 GOTO(out, rc = -ENOMEM);
1587
1588         mdc_pack_body(req, NULL, 0, 0, -1, 0);
1589
1590         /* Copy hsm_progress struct */
1591         archive_mask = req_capsule_client_get(&req->rq_pill,
1592                                               &RMF_MDS_HSM_ARCHIVE);
1593         if (archive_mask == NULL)
1594                 GOTO(out, rc = -EPROTO);
1595
1596         *archive_mask = archives;
1597
1598         ptlrpc_request_set_replen(req);
1599
1600         rc = mdc_queue_wait(req);
1601         GOTO(out, rc);
1602 out:
1603         ptlrpc_req_finished(req);
1604         return rc;
1605 }
1606
1607 static int mdc_ioc_hsm_current_action(struct obd_export *exp,
1608                                       struct md_op_data *op_data)
1609 {
1610         struct hsm_current_action       *hca = op_data->op_data;
1611         struct hsm_current_action       *req_hca;
1612         struct ptlrpc_request           *req;
1613         int                              rc;
1614         ENTRY;
1615
1616         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1617                                    &RQF_MDS_HSM_ACTION);
1618         if (req == NULL)
1619                 RETURN(-ENOMEM);
1620
1621         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_ACTION);
1622         if (rc) {
1623                 ptlrpc_request_free(req);
1624                 RETURN(rc);
1625         }
1626
1627         mdc_pack_body(req, &op_data->op_fid1, 0, 0,
1628                       op_data->op_suppgids[0], 0);
1629
1630         ptlrpc_request_set_replen(req);
1631
1632         rc = mdc_queue_wait(req);
1633         if (rc)
1634                 GOTO(out, rc);
1635
1636         req_hca = req_capsule_server_get(&req->rq_pill,
1637                                          &RMF_MDS_HSM_CURRENT_ACTION);
1638         if (req_hca == NULL)
1639                 GOTO(out, rc = -EPROTO);
1640
1641         *hca = *req_hca;
1642
1643         EXIT;
1644 out:
1645         ptlrpc_req_finished(req);
1646         return rc;
1647 }
1648
1649 static int mdc_ioc_hsm_ct_unregister(struct obd_import *imp)
1650 {
1651         struct ptlrpc_request   *req;
1652         int                      rc;
1653         ENTRY;
1654
1655         req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_UNREGISTER,
1656                                         LUSTRE_MDS_VERSION,
1657                                         MDS_HSM_CT_UNREGISTER);
1658         if (req == NULL)
1659                 GOTO(out, rc = -ENOMEM);
1660
1661         mdc_pack_body(req, NULL, 0, 0, -1, 0);
1662
1663         ptlrpc_request_set_replen(req);
1664
1665         rc = mdc_queue_wait(req);
1666         GOTO(out, rc);
1667 out:
1668         ptlrpc_req_finished(req);
1669         return rc;
1670 }
1671
1672 static int mdc_ioc_hsm_state_get(struct obd_export *exp,
1673                                  struct md_op_data *op_data)
1674 {
1675         struct hsm_user_state   *hus = op_data->op_data;
1676         struct hsm_user_state   *req_hus;
1677         struct ptlrpc_request   *req;
1678         int                      rc;
1679         ENTRY;
1680
1681         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1682                                    &RQF_MDS_HSM_STATE_GET);
1683         if (req == NULL)
1684                 RETURN(-ENOMEM);
1685
1686         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_GET);
1687         if (rc != 0) {
1688                 ptlrpc_request_free(req);
1689                 RETURN(rc);
1690         }
1691
1692         mdc_pack_body(req, &op_data->op_fid1, 0, 0,
1693                       op_data->op_suppgids[0], 0);
1694
1695         ptlrpc_request_set_replen(req);
1696
1697         rc = mdc_queue_wait(req);
1698         if (rc)
1699                 GOTO(out, rc);
1700
1701         req_hus = req_capsule_server_get(&req->rq_pill, &RMF_HSM_USER_STATE);
1702         if (req_hus == NULL)
1703                 GOTO(out, rc = -EPROTO);
1704
1705         *hus = *req_hus;
1706
1707         EXIT;
1708 out:
1709         ptlrpc_req_finished(req);
1710         return rc;
1711 }
1712
1713 static int mdc_ioc_hsm_state_set(struct obd_export *exp,
1714                                  struct md_op_data *op_data)
1715 {
1716         struct hsm_state_set    *hss = op_data->op_data;
1717         struct hsm_state_set    *req_hss;
1718         struct ptlrpc_request   *req;
1719         int                      rc;
1720         ENTRY;
1721
1722         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1723                                    &RQF_MDS_HSM_STATE_SET);
1724         if (req == NULL)
1725                 RETURN(-ENOMEM);
1726
1727         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_SET);
1728         if (rc) {
1729                 ptlrpc_request_free(req);
1730                 RETURN(rc);
1731         }
1732
1733         mdc_pack_body(req, &op_data->op_fid1, 0, 0,
1734                       op_data->op_suppgids[0], 0);
1735
1736         /* Copy states */
1737         req_hss = req_capsule_client_get(&req->rq_pill, &RMF_HSM_STATE_SET);
1738         if (req_hss == NULL)
1739                 GOTO(out, rc = -EPROTO);
1740         *req_hss = *hss;
1741
1742         ptlrpc_request_set_replen(req);
1743
1744         mdc_get_mod_rpc_slot(req, NULL);
1745         rc = ptlrpc_queue_wait(req);
1746         mdc_put_mod_rpc_slot(req, NULL);
1747
1748         GOTO(out, rc);
1749 out:
1750         ptlrpc_req_finished(req);
1751         return rc;
1752 }
1753
1754 static int mdc_ioc_hsm_request(struct obd_export *exp,
1755                                struct hsm_user_request *hur)
1756 {
1757         struct obd_import       *imp = class_exp2cliimp(exp);
1758         struct ptlrpc_request   *req;
1759         struct hsm_request      *req_hr;
1760         struct hsm_user_item    *req_hui;
1761         char                    *req_opaque;
1762         int                      rc;
1763         ENTRY;
1764
1765         req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_REQUEST);
1766         if (req == NULL)
1767                 GOTO(out, rc = -ENOMEM);
1768
1769         req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM, RCL_CLIENT,
1770                              hur->hur_request.hr_itemcount
1771                              * sizeof(struct hsm_user_item));
1772         req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA, RCL_CLIENT,
1773                              hur->hur_request.hr_data_len);
1774
1775         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_REQUEST);
1776         if (rc) {
1777                 ptlrpc_request_free(req);
1778                 RETURN(rc);
1779         }
1780
1781         mdc_pack_body(req, NULL, 0, 0, -1, 0);
1782
1783         /* Copy hsm_request struct */
1784         req_hr = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_REQUEST);
1785         if (req_hr == NULL)
1786                 GOTO(out, rc = -EPROTO);
1787         *req_hr = hur->hur_request;
1788
1789         /* Copy hsm_user_item structs */
1790         req_hui = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM);
1791         if (req_hui == NULL)
1792                 GOTO(out, rc = -EPROTO);
1793         memcpy(req_hui, hur->hur_user_item,
1794                hur->hur_request.hr_itemcount * sizeof(struct hsm_user_item));
1795
1796         /* Copy opaque field */
1797         req_opaque = req_capsule_client_get(&req->rq_pill, &RMF_GENERIC_DATA);
1798         if (req_opaque == NULL)
1799                 GOTO(out, rc = -EPROTO);
1800         memcpy(req_opaque, hur_data(hur), hur->hur_request.hr_data_len);
1801
1802         ptlrpc_request_set_replen(req);
1803
1804         mdc_get_mod_rpc_slot(req, NULL);
1805         rc = ptlrpc_queue_wait(req);
1806         mdc_put_mod_rpc_slot(req, NULL);
1807
1808         GOTO(out, rc);
1809
1810 out:
1811         ptlrpc_req_finished(req);
1812         return rc;
1813 }
1814
1815 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
1816                                 struct lustre_kernelcomm *lk);
1817
1818 static int mdc_quotactl(struct obd_device *unused, struct obd_export *exp,
1819                         struct obd_quotactl *oqctl)
1820 {
1821         struct ptlrpc_request   *req;
1822         struct obd_quotactl     *oqc;
1823         int                      rc;
1824         ENTRY;
1825
1826         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
1827                                         &RQF_MDS_QUOTACTL, LUSTRE_MDS_VERSION,
1828                                         MDS_QUOTACTL);
1829         if (req == NULL)
1830                 RETURN(-ENOMEM);
1831
1832         oqc = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1833         *oqc = *oqctl;
1834
1835         ptlrpc_request_set_replen(req);
1836         ptlrpc_at_set_req_timeout(req);
1837         req->rq_no_resend = 1;
1838
1839         rc = ptlrpc_queue_wait(req);
1840         if (rc)
1841                 CERROR("ptlrpc_queue_wait failed, rc: %d\n", rc);
1842
1843         if (req->rq_repmsg &&
1844             (oqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL))) {
1845                 *oqctl = *oqc;
1846         } else if (!rc) {
1847                 CERROR ("Can't unpack obd_quotactl\n");
1848                 rc = -EPROTO;
1849         }
1850         ptlrpc_req_finished(req);
1851
1852         RETURN(rc);
1853 }
1854
1855 static int mdc_ioc_swap_layouts(struct obd_export *exp,
1856                                 struct md_op_data *op_data)
1857 {
1858         struct list_head cancels = LIST_HEAD_INIT(cancels);
1859         struct ptlrpc_request   *req;
1860         int                      rc, count;
1861         struct mdc_swap_layouts *msl, *payload;
1862         ENTRY;
1863
1864         msl = op_data->op_data;
1865
1866         /* When the MDT will get the MDS_SWAP_LAYOUTS RPC the
1867          * first thing it will do is to cancel the 2 layout
1868          * locks held by this client.
1869          * So the client must cancel its layout locks on the 2 fids
1870          * with the request RPC to avoid extra RPC round trips.
1871          */
1872         count = mdc_resource_get_unused(exp, &op_data->op_fid1, &cancels,
1873                                         LCK_EX, MDS_INODELOCK_LAYOUT |
1874                                         MDS_INODELOCK_XATTR);
1875         count += mdc_resource_get_unused(exp, &op_data->op_fid2, &cancels,
1876                                          LCK_EX, MDS_INODELOCK_LAYOUT |
1877                                          MDS_INODELOCK_XATTR);
1878
1879         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
1880                                    &RQF_MDS_SWAP_LAYOUTS);
1881         if (req == NULL) {
1882                 ldlm_lock_list_put(&cancels, l_bl_ast, count);
1883                 RETURN(-ENOMEM);
1884         }
1885
1886         rc = mdc_prep_elc_req(exp, req, MDS_SWAP_LAYOUTS, &cancels, count);
1887         if (rc) {
1888                 ptlrpc_request_free(req);
1889                 RETURN(rc);
1890         }
1891
1892         mdc_swap_layouts_pack(req, op_data);
1893
1894         payload = req_capsule_client_get(&req->rq_pill, &RMF_SWAP_LAYOUTS);
1895         LASSERT(payload);
1896
1897         *payload = *msl;
1898
1899         ptlrpc_request_set_replen(req);
1900
1901         rc = ptlrpc_queue_wait(req);
1902         if (rc)
1903                 GOTO(out, rc);
1904         EXIT;
1905
1906 out:
1907         ptlrpc_req_finished(req);
1908         return rc;
1909 }
1910
1911 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1912                          void *karg, void __user *uarg)
1913 {
1914         struct obd_device *obd = exp->exp_obd;
1915         struct obd_ioctl_data *data = karg;
1916         struct obd_import *imp = obd->u.cli.cl_import;
1917         int rc;
1918         ENTRY;
1919
1920         if (!try_module_get(THIS_MODULE)) {
1921                 CERROR("%s: cannot get module '%s'\n", obd->obd_name,
1922                        module_name(THIS_MODULE));
1923                 return -EINVAL;
1924         }
1925         switch (cmd) {
1926         case OBD_IOC_FID2PATH:
1927                 rc = mdc_ioc_fid2path(exp, karg);
1928                 GOTO(out, rc);
1929         case LL_IOC_HSM_CT_START:
1930                 rc = mdc_ioc_hsm_ct_start(exp, karg);
1931                 /* ignore if it was already registered on this MDS. */
1932                 if (rc == -EEXIST)
1933                         rc = 0;
1934                 GOTO(out, rc);
1935         case LL_IOC_HSM_PROGRESS:
1936                 rc = mdc_ioc_hsm_progress(exp, karg);
1937                 GOTO(out, rc);
1938         case LL_IOC_HSM_STATE_GET:
1939                 rc = mdc_ioc_hsm_state_get(exp, karg);
1940                 GOTO(out, rc);
1941         case LL_IOC_HSM_STATE_SET:
1942                 rc = mdc_ioc_hsm_state_set(exp, karg);
1943                 GOTO(out, rc);
1944         case LL_IOC_HSM_ACTION:
1945                 rc = mdc_ioc_hsm_current_action(exp, karg);
1946                 GOTO(out, rc);
1947         case LL_IOC_HSM_REQUEST:
1948                 rc = mdc_ioc_hsm_request(exp, karg);
1949                 GOTO(out, rc);
1950         case OBD_IOC_CLIENT_RECOVER:
1951                 rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1, 0);
1952                 if (rc < 0)
1953                         GOTO(out, rc);
1954                 GOTO(out, rc = 0);
1955         case IOC_OSC_SET_ACTIVE:
1956                 rc = ptlrpc_set_import_active(imp, data->ioc_offset);
1957                 GOTO(out, rc);
1958         case OBD_IOC_PING_TARGET:
1959                 rc = ptlrpc_obd_ping(obd);
1960                 GOTO(out, rc);
1961         /*
1962          * Normally IOC_OBD_STATFS, OBD_IOC_QUOTACTL iocontrol are handled by
1963          * LMV instead of MDC. But when the cluster is upgraded from 1.8,
1964          * there'd be no LMV layer thus we might be called here. Eventually
1965          * this code should be removed.
1966          * bz20731, LU-592.
1967          */
1968         case IOC_OBD_STATFS: {
1969                 struct obd_statfs stat_buf = {0};
1970
1971                 if (*((__u32 *) data->ioc_inlbuf2) != 0)
1972                         GOTO(out, rc = -ENODEV);
1973
1974                 /* copy UUID */
1975                 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd),
1976                                  min((int)data->ioc_plen2,
1977                                      (int)sizeof(struct obd_uuid))))
1978                         GOTO(out, rc = -EFAULT);
1979
1980                 rc = mdc_statfs(NULL, obd->obd_self_export, &stat_buf,
1981                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1982                                 0);
1983                 if (rc != 0)
1984                         GOTO(out, rc);
1985
1986                 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1987                                      min((int) data->ioc_plen1,
1988                                          (int) sizeof(stat_buf))))
1989                         GOTO(out, rc = -EFAULT);
1990
1991                 GOTO(out, rc = 0);
1992         }
1993         case OBD_IOC_QUOTACTL: {
1994                 struct if_quotactl *qctl = karg;
1995                 struct obd_quotactl *oqctl;
1996
1997                 OBD_ALLOC_PTR(oqctl);
1998                 if (oqctl == NULL)
1999                         GOTO(out, rc = -ENOMEM);
2000
2001                 QCTL_COPY(oqctl, qctl);
2002                 rc = obd_quotactl(exp, oqctl);
2003                 if (rc == 0) {
2004                         QCTL_COPY(qctl, oqctl);
2005                         qctl->qc_valid = QC_MDTIDX;
2006                         qctl->obd_uuid = obd->u.cli.cl_target_uuid;
2007                 }
2008
2009                 OBD_FREE_PTR(oqctl);
2010                 GOTO(out, rc);
2011         }
2012         case LL_IOC_GET_CONNECT_FLAGS:
2013                 if (copy_to_user(uarg, exp_connect_flags_ptr(exp),
2014                                  sizeof(*exp_connect_flags_ptr(exp))))
2015                         GOTO(out, rc = -EFAULT);
2016
2017                 GOTO(out, rc = 0);
2018         case LL_IOC_LOV_SWAP_LAYOUTS:
2019                 rc = mdc_ioc_swap_layouts(exp, karg);
2020                 GOTO(out, rc);
2021         default:
2022                 CERROR("unrecognised ioctl: cmd = %#x\n", cmd);
2023                 GOTO(out, rc = -ENOTTY);
2024         }
2025 out:
2026         module_put(THIS_MODULE);
2027
2028         return rc;
2029 }
2030
2031 static int mdc_get_info_rpc(struct obd_export *exp,
2032                             u32 keylen, void *key,
2033                             u32 vallen, void *val)
2034 {
2035         struct obd_import      *imp = class_exp2cliimp(exp);
2036         struct ptlrpc_request  *req;
2037         char                   *tmp;
2038         int                     rc = -EINVAL;
2039         ENTRY;
2040
2041         req = ptlrpc_request_alloc(imp, &RQF_MDS_GET_INFO);
2042         if (req == NULL)
2043                 RETURN(-ENOMEM);
2044
2045         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
2046                              RCL_CLIENT, keylen);
2047         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
2048                              RCL_CLIENT, sizeof(vallen));
2049
2050         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
2051         if (rc) {
2052                 ptlrpc_request_free(req);
2053                 RETURN(rc);
2054         }
2055
2056         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
2057         memcpy(tmp, key, keylen);
2058         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
2059         memcpy(tmp, &vallen, sizeof(vallen));
2060
2061         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
2062                              RCL_SERVER, vallen);
2063         ptlrpc_request_set_replen(req);
2064
2065         rc = ptlrpc_queue_wait(req);
2066         /* -EREMOTE means the get_info result is partial, and it needs to
2067          * continue on another MDT, see fid2path part in lmv_iocontrol */
2068         if (rc == 0 || rc == -EREMOTE) {
2069                 tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
2070                 memcpy(val, tmp, vallen);
2071                 if (ptlrpc_rep_need_swab(req)) {
2072                         if (KEY_IS(KEY_FID2PATH))
2073                                 lustre_swab_fid2path(val);
2074                 }
2075         }
2076         ptlrpc_req_finished(req);
2077
2078         RETURN(rc);
2079 }
2080
2081 static void lustre_swab_hai(struct hsm_action_item *h)
2082 {
2083         __swab32s(&h->hai_len);
2084         __swab32s(&h->hai_action);
2085         lustre_swab_lu_fid(&h->hai_fid);
2086         lustre_swab_lu_fid(&h->hai_dfid);
2087         __swab64s(&h->hai_cookie);
2088         __swab64s(&h->hai_extent.offset);
2089         __swab64s(&h->hai_extent.length);
2090         __swab64s(&h->hai_gid);
2091 }
2092
2093 static void lustre_swab_hal(struct hsm_action_list *h)
2094 {
2095         struct hsm_action_item  *hai;
2096         __u32                    i;
2097
2098         __swab32s(&h->hal_version);
2099         __swab32s(&h->hal_count);
2100         __swab32s(&h->hal_archive_id);
2101         __swab64s(&h->hal_flags);
2102         hai = hai_first(h);
2103         for (i = 0; i < h->hal_count; i++, hai = hai_next(hai))
2104                 lustre_swab_hai(hai);
2105 }
2106
2107 static void lustre_swab_kuch(struct kuc_hdr *l)
2108 {
2109         __swab16s(&l->kuc_magic);
2110         /* __u8 l->kuc_transport */
2111         __swab16s(&l->kuc_msgtype);
2112         __swab16s(&l->kuc_msglen);
2113 }
2114
2115 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
2116                                 struct lustre_kernelcomm *lk)
2117 {
2118         struct obd_import  *imp = class_exp2cliimp(exp);
2119         __u32               archive = lk->lk_data;
2120         int                 rc = 0;
2121
2122         if (lk->lk_group != KUC_GRP_HSM) {
2123                 CERROR("Bad copytool group %d\n", lk->lk_group);
2124                 return -EINVAL;
2125         }
2126
2127         CDEBUG(D_HSM, "CT start r%d w%d u%d g%d f%#x\n", lk->lk_rfd, lk->lk_wfd,
2128                lk->lk_uid, lk->lk_group, lk->lk_flags);
2129
2130         if (lk->lk_flags & LK_FLG_STOP) {
2131                 /* Unregister with the coordinator */
2132                 rc = mdc_ioc_hsm_ct_unregister(imp);
2133         } else {
2134                 rc = mdc_ioc_hsm_ct_register(imp, archive);
2135         }
2136
2137         return rc;
2138 }
2139
2140 /**
2141  * Send a message to any listening copytools
2142  * @param val KUC message (kuc_hdr + hsm_action_list)
2143  * @param len total length of message
2144  */
2145 static int mdc_hsm_copytool_send(size_t len, void *val)
2146 {
2147         struct kuc_hdr          *lh = (struct kuc_hdr *)val;
2148         struct hsm_action_list  *hal = (struct hsm_action_list *)(lh + 1);
2149         int                      rc;
2150         ENTRY;
2151
2152         if (len < sizeof(*lh) + sizeof(*hal)) {
2153                 CERROR("Short HSM message %zu < %zu\n", len,
2154                        sizeof(*lh) + sizeof(*hal));
2155                 RETURN(-EPROTO);
2156         }
2157         if (lh->kuc_magic == __swab16(KUC_MAGIC)) {
2158                 lustre_swab_kuch(lh);
2159                 lustre_swab_hal(hal);
2160         } else if (lh->kuc_magic != KUC_MAGIC) {
2161                 CERROR("Bad magic %x!=%x\n", lh->kuc_magic, KUC_MAGIC);
2162                 RETURN(-EPROTO);
2163         }
2164
2165         CDEBUG(D_HSM, " Received message mg=%x t=%d m=%d l=%d actions=%d "
2166                "on %s\n",
2167                lh->kuc_magic, lh->kuc_transport, lh->kuc_msgtype,
2168                lh->kuc_msglen, hal->hal_count, hal->hal_fsname);
2169
2170         /* Broadcast to HSM listeners */
2171         rc = libcfs_kkuc_group_put(KUC_GRP_HSM, lh);
2172
2173         RETURN(rc);
2174 }
2175
2176 /**
2177  * callback function passed to kuc for re-registering each HSM copytool
2178  * running on MDC, after MDT shutdown/recovery.
2179  * @param data copytool registration data
2180  * @param cb_arg callback argument (obd_import)
2181  */
2182 static int mdc_hsm_ct_reregister(void *data, void *cb_arg)
2183 {
2184         struct kkuc_ct_data     *kcd = data;
2185         struct obd_import       *imp = (struct obd_import *)cb_arg;
2186         int                      rc;
2187
2188         if (kcd == NULL || kcd->kcd_magic != KKUC_CT_DATA_MAGIC)
2189                 return -EPROTO;
2190
2191         if (!obd_uuid_equals(&kcd->kcd_uuid, &imp->imp_obd->obd_uuid))
2192                 return 0;
2193
2194         CDEBUG(D_HA, "%s: recover copytool registration to MDT (archive=%#x)\n",
2195                imp->imp_obd->obd_name, kcd->kcd_archive);
2196         rc = mdc_ioc_hsm_ct_register(imp, kcd->kcd_archive);
2197
2198         /* ignore error if the copytool is already registered */
2199         return (rc == -EEXIST) ? 0 : rc;
2200 }
2201
2202 /**
2203  * Re-establish all kuc contexts with MDT
2204  * after MDT shutdown/recovery.
2205  */
2206 static int mdc_kuc_reregister(struct obd_import *imp)
2207 {
2208         /* re-register HSM agents */
2209         return libcfs_kkuc_group_foreach(KUC_GRP_HSM, mdc_hsm_ct_reregister,
2210                                          (void *)imp);
2211 }
2212
2213 static int mdc_set_info_async(const struct lu_env *env,
2214                               struct obd_export *exp,
2215                               u32 keylen, void *key,
2216                               u32 vallen, void *val,
2217                               struct ptlrpc_request_set *set)
2218 {
2219         struct obd_import       *imp = class_exp2cliimp(exp);
2220         int                      rc;
2221         ENTRY;
2222
2223         if (KEY_IS(KEY_READ_ONLY)) {
2224                 if (vallen != sizeof(int))
2225                         RETURN(-EINVAL);
2226
2227                 spin_lock(&imp->imp_lock);
2228                 if (*((int *)val)) {
2229                         imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
2230                         imp->imp_connect_data.ocd_connect_flags |=
2231                                                         OBD_CONNECT_RDONLY;
2232                 } else {
2233                         imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
2234                         imp->imp_connect_data.ocd_connect_flags &=
2235                                                         ~OBD_CONNECT_RDONLY;
2236                 }
2237                 spin_unlock(&imp->imp_lock);
2238
2239                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2240                                        keylen, key, vallen, val, set);
2241                 RETURN(rc);
2242         }
2243         if (KEY_IS(KEY_SPTLRPC_CONF)) {
2244                 sptlrpc_conf_client_adapt(exp->exp_obd);
2245                 RETURN(0);
2246         }
2247         if (KEY_IS(KEY_FLUSH_CTX)) {
2248                 sptlrpc_import_flush_my_ctx(imp);
2249                 RETURN(0);
2250         }
2251         if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2252                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
2253                                        keylen, key, vallen, val, set);
2254                 RETURN(rc);
2255         }
2256         if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) {
2257                 rc = mdc_hsm_copytool_send(vallen, val);
2258                 RETURN(rc);
2259         }
2260
2261         if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2262                 __u32 *default_easize = val;
2263
2264                 exp->exp_obd->u.cli.cl_default_mds_easize = *default_easize;
2265                 RETURN(0);
2266         }
2267
2268         /* TODO: these OSC-related keys are ignored for now */
2269         if (KEY_IS(KEY_CHECKSUM) || KEY_IS(KEY_CACHE_SET) ||
2270             KEY_IS(KEY_CACHE_LRU_SHRINK) || KEY_IS(KEY_GRANT_SHRINK))
2271                 RETURN(0);
2272
2273         CERROR("%s: Unknown key %s\n", exp->exp_obd->obd_name, (char *)key);
2274         RETURN(-EINVAL);
2275 }
2276
2277 static int mdc_get_info(const struct lu_env *env, struct obd_export *exp,
2278                         __u32 keylen, void *key, __u32 *vallen, void *val)
2279 {
2280         int rc = -EINVAL;
2281
2282         if (KEY_IS(KEY_MAX_EASIZE)) {
2283                 __u32 mdsize, *max_easize;
2284
2285                 if (*vallen != sizeof(int))
2286                         RETURN(-EINVAL);
2287                 mdsize = *(__u32 *)val;
2288                 if (mdsize > exp->exp_obd->u.cli.cl_max_mds_easize)
2289                         exp->exp_obd->u.cli.cl_max_mds_easize = mdsize;
2290                 max_easize = val;
2291                 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
2292                 RETURN(0);
2293         } else if (KEY_IS(KEY_DEFAULT_EASIZE)) {
2294                 __u32 *default_easize;
2295
2296                 if (*vallen != sizeof(int))
2297                         RETURN(-EINVAL);
2298                 default_easize = val;
2299                 *default_easize = exp->exp_obd->u.cli.cl_default_mds_easize;
2300                 RETURN(0);
2301         } else if (KEY_IS(KEY_CONN_DATA)) {
2302                 struct obd_import *imp = class_exp2cliimp(exp);
2303                 struct obd_connect_data *data = val;
2304
2305                 if (*vallen != sizeof(*data))
2306                         RETURN(-EINVAL);
2307
2308                 *data = imp->imp_connect_data;
2309                 RETURN(0);
2310         } else if (KEY_IS(KEY_TGT_COUNT)) {
2311                 *((__u32 *)val) = 1;
2312                 RETURN(0);
2313         }
2314
2315         rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
2316
2317         RETURN(rc);
2318 }
2319
2320 static int mdc_fsync(struct obd_export *exp, const struct lu_fid *fid,
2321                      struct ptlrpc_request **request)
2322 {
2323         struct ptlrpc_request *req;
2324         int                    rc;
2325         ENTRY;
2326
2327         *request = NULL;
2328         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_SYNC);
2329         if (req == NULL)
2330                 RETURN(-ENOMEM);
2331
2332         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
2333         if (rc) {
2334                 ptlrpc_request_free(req);
2335                 RETURN(rc);
2336         }
2337
2338         mdc_pack_body(req, fid, 0, 0, -1, 0);
2339
2340         ptlrpc_request_set_replen(req);
2341
2342         rc = ptlrpc_queue_wait(req);
2343         if (rc)
2344                 ptlrpc_req_finished(req);
2345         else
2346                 *request = req;
2347         RETURN(rc);
2348 }
2349
2350 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
2351                             enum obd_import_event event)
2352 {
2353         int rc = 0;
2354
2355         LASSERT(imp->imp_obd == obd);
2356
2357         switch (event) {
2358         case IMP_EVENT_DISCON: {
2359                 struct client_obd *cli = &obd->u.cli;
2360
2361                 spin_lock(&cli->cl_loi_list_lock);
2362                 cli->cl_avail_grant = 0;
2363                 cli->cl_lost_grant = 0;
2364                 spin_unlock(&cli->cl_loi_list_lock);
2365                 break;
2366         }
2367         case IMP_EVENT_INACTIVE: {
2368                 struct client_obd *cli = &obd->u.cli;
2369                 /*
2370                  * Flush current sequence to make client obtain new one
2371                  * from server in case of disconnect/reconnect.
2372                  */
2373                 down_read(&cli->cl_seq_rwsem);
2374                 if (cli->cl_seq)
2375                         seq_client_flush(cli->cl_seq);
2376                 up_read(&cli->cl_seq_rwsem);
2377
2378                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE);
2379                 break;
2380         }
2381         case IMP_EVENT_INVALIDATE: {
2382                 struct ldlm_namespace *ns = obd->obd_namespace;
2383
2384                 ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
2385
2386                 break;
2387         }
2388         case IMP_EVENT_ACTIVE:
2389                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE);
2390                 /* redo the kuc registration after reconnecting */
2391                 if (rc == 0)
2392                         rc = mdc_kuc_reregister(imp);
2393                 break;
2394         case IMP_EVENT_OCD: {
2395                 struct obd_connect_data *ocd = &imp->imp_connect_data;
2396
2397                 if (OCD_HAS_FLAG(ocd, GRANT))
2398                         osc_init_grant(&obd->u.cli, ocd);
2399
2400                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD);
2401                 break;
2402         }
2403         case IMP_EVENT_DEACTIVATE:
2404         case IMP_EVENT_ACTIVATE:
2405                 break;
2406         default:
2407                 CERROR("Unknown import event %x\n", event);
2408                 LBUG();
2409         }
2410         RETURN(rc);
2411 }
2412
2413 int mdc_fid_alloc(const struct lu_env *env, struct obd_export *exp,
2414                   struct lu_fid *fid, struct md_op_data *op_data)
2415 {
2416         struct client_obd *cli = &exp->exp_obd->u.cli;
2417         int rc = -EIO;
2418
2419         ENTRY;
2420
2421         down_read(&cli->cl_seq_rwsem);
2422         if (cli->cl_seq)
2423                 rc = seq_client_alloc_fid(env, cli->cl_seq, fid);
2424         up_read(&cli->cl_seq_rwsem);
2425
2426         RETURN(rc);
2427 }
2428
2429 static struct obd_uuid *mdc_get_uuid(struct obd_export *exp)
2430 {
2431         struct client_obd *cli = &exp->exp_obd->u.cli;
2432         return &cli->cl_target_uuid;
2433 }
2434
2435 /**
2436  * Determine whether the lock can be canceled before replaying it during
2437  * recovery, non zero value will be return if the lock can be canceled,
2438  * or zero returned for not
2439  */
2440 static int mdc_cancel_weight(struct ldlm_lock *lock)
2441 {
2442         if (lock->l_resource->lr_type != LDLM_IBITS)
2443                 RETURN(0);
2444
2445         /* FIXME: if we ever get into a situation where there are too many
2446          * opened files with open locks on a single node, then we really
2447          * should replay these open locks to reget it */
2448         if (lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_OPEN)
2449                 RETURN(0);
2450
2451         RETURN(1);
2452 }
2453
2454 static int mdc_resource_inode_free(struct ldlm_resource *res)
2455 {
2456         if (res->lr_lvb_inode)
2457                 res->lr_lvb_inode = NULL;
2458
2459         return 0;
2460 }
2461
2462 static struct ldlm_valblock_ops inode_lvbo = {
2463         .lvbo_free = mdc_resource_inode_free
2464 };
2465
2466 static int mdc_llog_init(struct obd_device *obd)
2467 {
2468         struct obd_llog_group   *olg = &obd->obd_olg;
2469         struct llog_ctxt        *ctxt;
2470         int                      rc;
2471
2472         ENTRY;
2473
2474         rc = llog_setup(NULL, obd, olg, LLOG_CHANGELOG_REPL_CTXT, obd,
2475                         &llog_client_ops);
2476         if (rc < 0)
2477                 RETURN(rc);
2478
2479         ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
2480         llog_initiator_connect(ctxt);
2481         llog_ctxt_put(ctxt);
2482
2483         RETURN(0);
2484 }
2485
2486 static void mdc_llog_finish(struct obd_device *obd)
2487 {
2488         struct llog_ctxt *ctxt;
2489
2490         ENTRY;
2491
2492         ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
2493         if (ctxt != NULL)
2494                 llog_cleanup(NULL, ctxt);
2495
2496         EXIT;
2497 }
2498
2499 int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
2500 {
2501         int                             rc;
2502         ENTRY;
2503
2504         rc = ptlrpcd_addref();
2505         if (rc < 0)
2506                 RETURN(rc);
2507
2508         rc = client_obd_setup(obd, cfg);
2509         if (rc)
2510                 GOTO(err_ptlrpcd_decref, rc);
2511 #ifdef CONFIG_PROC_FS
2512         obd->obd_vars = lprocfs_mdc_obd_vars;
2513         lprocfs_obd_setup(obd, false);
2514         lprocfs_alloc_md_stats(obd, 0);
2515 #endif
2516         sptlrpc_lprocfs_cliobd_attach(obd);
2517         ptlrpc_lprocfs_register_obd(obd);
2518
2519         ns_register_cancel(obd->obd_namespace, mdc_cancel_weight);
2520
2521         obd->obd_namespace->ns_lvbo = &inode_lvbo;
2522
2523         rc = mdc_llog_init(obd);
2524         if (rc) {
2525                 CERROR("%s: failed to setup llogging subsystems: rc = %d\n",
2526                        obd->obd_name, rc);
2527                 GOTO(err_mdc_cleanup, rc);
2528         }
2529
2530         rc = mdc_changelog_cdev_init(obd);
2531         if (rc) {
2532                 CERROR("%s: failed to setup changelog char device: rc = %d\n",
2533                        obd->obd_name, rc);
2534                 GOTO(err_mdc_cleanup, rc);
2535         }
2536
2537         EXIT;
2538 err_mdc_cleanup:
2539         if (rc)
2540                 client_obd_cleanup(obd);
2541
2542 err_ptlrpcd_decref:
2543         if (rc)
2544                 ptlrpcd_decref();
2545
2546         return rc;
2547 }
2548
2549 /* Initialize the default and maximum LOV EA sizes.  This allows
2550  * us to make MDS RPCs with large enough reply buffers to hold a default
2551  * sized EA without having to calculate this (via a call into the
2552  * LOV + OSCs) each time we make an RPC.  The maximum size is also tracked
2553  * but not used to avoid wastefully vmalloc()'ing large reply buffers when
2554  * a large number of stripes is possible.  If a larger reply buffer is
2555  * required it will be reallocated in the ptlrpc layer due to overflow.
2556  */
2557 static int mdc_init_ea_size(struct obd_export *exp, __u32 easize,
2558                             __u32 def_easize)
2559 {
2560         struct obd_device *obd = exp->exp_obd;
2561         struct client_obd *cli = &obd->u.cli;
2562         ENTRY;
2563
2564         if (cli->cl_max_mds_easize < easize)
2565                 cli->cl_max_mds_easize = easize;
2566
2567         if (cli->cl_default_mds_easize < def_easize)
2568                 cli->cl_default_mds_easize = def_easize;
2569
2570         RETURN(0);
2571 }
2572
2573 static int mdc_precleanup(struct obd_device *obd)
2574 {
2575         ENTRY;
2576
2577         /* Failsafe, ok if racy */
2578         if (obd->obd_type->typ_refcnt <= 1)
2579                 libcfs_kkuc_group_rem(0, KUC_GRP_HSM);
2580
2581         mdc_changelog_cdev_finish(obd);
2582
2583         obd_cleanup_client_import(obd);
2584         ptlrpc_lprocfs_unregister_obd(obd);
2585         lprocfs_obd_cleanup(obd);
2586         lprocfs_free_md_stats(obd);
2587         mdc_llog_finish(obd);
2588         RETURN(0);
2589 }
2590
2591 static int mdc_cleanup(struct obd_device *obd)
2592 {
2593         ptlrpcd_decref();
2594
2595         return client_obd_cleanup(obd);
2596 }
2597
2598 int mdc_process_config(struct obd_device *obd, size_t len, void *buf)
2599 {
2600         struct lustre_cfg *lcfg = buf;
2601         int rc;
2602
2603         rc = class_process_proc_param(PARAM_MDC, obd->obd_vars, lcfg, obd);
2604         return (rc > 0 ? 0: rc);
2605 }
2606
2607 static struct obd_ops mdc_obd_ops = {
2608         .o_owner            = THIS_MODULE,
2609         .o_setup            = mdc_setup,
2610         .o_precleanup       = mdc_precleanup,
2611         .o_cleanup          = mdc_cleanup,
2612         .o_add_conn         = client_import_add_conn,
2613         .o_del_conn         = client_import_del_conn,
2614         .o_connect          = client_connect_import,
2615         .o_disconnect       = client_disconnect_export,
2616         .o_iocontrol        = mdc_iocontrol,
2617         .o_set_info_async   = mdc_set_info_async,
2618         .o_statfs           = mdc_statfs,
2619         .o_fid_init         = client_fid_init,
2620         .o_fid_fini         = client_fid_fini,
2621         .o_fid_alloc        = mdc_fid_alloc,
2622         .o_import_event     = mdc_import_event,
2623         .o_get_info         = mdc_get_info,
2624         .o_process_config   = mdc_process_config,
2625         .o_get_uuid         = mdc_get_uuid,
2626         .o_quotactl         = mdc_quotactl,
2627 };
2628
2629 static struct md_ops mdc_md_ops = {
2630         .m_get_root         = mdc_get_root,
2631         .m_null_inode       = mdc_null_inode,
2632         .m_close            = mdc_close,
2633         .m_create           = mdc_create,
2634         .m_enqueue          = mdc_enqueue,
2635         .m_getattr          = mdc_getattr,
2636         .m_getattr_name     = mdc_getattr_name,
2637         .m_intent_lock      = mdc_intent_lock,
2638         .m_link             = mdc_link,
2639         .m_rename           = mdc_rename,
2640         .m_setattr          = mdc_setattr,
2641         .m_setxattr         = mdc_setxattr,
2642         .m_getxattr         = mdc_getxattr,
2643         .m_fsync                = mdc_fsync,
2644         .m_read_page            = mdc_read_page,
2645         .m_unlink           = mdc_unlink,
2646         .m_cancel_unused    = mdc_cancel_unused,
2647         .m_init_ea_size     = mdc_init_ea_size,
2648         .m_set_lock_data    = mdc_set_lock_data,
2649         .m_lock_match       = mdc_lock_match,
2650         .m_get_lustre_md    = mdc_get_lustre_md,
2651         .m_free_lustre_md   = mdc_free_lustre_md,
2652         .m_set_open_replay_data = mdc_set_open_replay_data,
2653         .m_clear_open_replay_data = mdc_clear_open_replay_data,
2654         .m_intent_getattr_async = mdc_intent_getattr_async,
2655         .m_revalidate_lock      = mdc_revalidate_lock
2656 };
2657
2658 static int __init mdc_init(void)
2659 {
2660         return class_register_type(&mdc_obd_ops, &mdc_md_ops, true, NULL,
2661                                    LUSTRE_MDC_NAME, &mdc_device_type);
2662 }
2663
2664 static void __exit mdc_exit(void)
2665 {
2666         class_unregister_type(LUSTRE_MDC_NAME);
2667 }
2668
2669 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
2670 MODULE_DESCRIPTION("Lustre Metadata Client");
2671 MODULE_VERSION(LUSTRE_VERSION_STRING);
2672 MODULE_LICENSE("GPL");
2673
2674 module_init(mdc_init);
2675 module_exit(mdc_exit);