lustre/mdt/mdt_handler.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.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2010, 2014, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/mdt/mdt_handler.c
  37  *
  38  * Lustre Metadata Target (mdt) request handler
  39  *
  40  * Author: Peter Braam <braam@clusterfs.com>
  41  * Author: Andreas Dilger <adilger@clusterfs.com>
  42  * Author: Phil Schwan <phil@clusterfs.com>
  43  * Author: Mike Shaver <shaver@clusterfs.com>
  44  * Author: Nikita Danilov <nikita@clusterfs.com>
  45  * Author: Huang Hua <huanghua@clusterfs.com>
  46  * Author: Yury Umanets <umka@clusterfs.com>
  47  */
  48
  49 #define DEBUG_SUBSYSTEM S_MDS
  50
  51 #include <linux/module.h>
  52 /*
  53  * struct OBD_{ALLOC,FREE}*()
  54  */
  55 #include <obd_support.h>
  56 #include <lustre_ioctl.h>
  57 /* struct ptlrpc_request */
  58 #include <lustre_net.h>
  59 /* struct obd_export */
  60 #include <lustre_export.h>
  61 /* struct obd_device */
  62 #include <obd.h>
  63 /* lu2dt_dev() */
  64 #include <dt_object.h>
  65 #include <lustre_mds.h>
  66 #include <lustre_log.h>
  67 #include "mdt_internal.h"
  68 #include <lustre_acl.h>
  69 #include <lustre_param.h>
  70 #include <lustre_quota.h>
  71 #include <lustre_lfsck.h>
  72 #include <lustre_nodemap.h>
  73
  74 static unsigned int max_mod_rpcs_per_client = 8;
  75 CFS_MODULE_PARM(max_mod_rpcs_per_client, "i", uint, 0644,
  76                 "maximum number of modify RPCs in flight allowed per client");
  77
  78 mdl_mode_t mdt_mdl_lock_modes[] = {
  79         [LCK_MINMODE] = MDL_MINMODE,
  80         [LCK_EX]      = MDL_EX,
  81         [LCK_PW]      = MDL_PW,
  82         [LCK_PR]      = MDL_PR,
  83         [LCK_CW]      = MDL_CW,
  84         [LCK_CR]      = MDL_CR,
  85         [LCK_NL]      = MDL_NL,
  86         [LCK_GROUP]   = MDL_GROUP
  87 };
  88
  89 ldlm_mode_t mdt_dlm_lock_modes[] = {
  90         [MDL_MINMODE] = LCK_MINMODE,
  91         [MDL_EX]      = LCK_EX,
  92         [MDL_PW]      = LCK_PW,
  93         [MDL_PR]      = LCK_PR,
  94         [MDL_CW]      = LCK_CW,
  95         [MDL_CR]      = LCK_CR,
  96         [MDL_NL]      = LCK_NL,
  97         [MDL_GROUP]   = LCK_GROUP
  98 };
  99
 100 static struct mdt_device *mdt_dev(struct lu_device *d);
 101 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
 102
 103 static const struct lu_object_operations mdt_obj_ops;
 104
 105 /* Slab for MDT object allocation */
 106 static struct kmem_cache *mdt_object_kmem;
 107
 108 /* For HSM restore handles */
 109 struct kmem_cache *mdt_hsm_cdt_kmem;
 110
 111 /* For HSM request handles */
 112 struct kmem_cache *mdt_hsm_car_kmem;
 113
 114 static struct lu_kmem_descr mdt_caches[] = {
 115         {
 116                 .ckd_cache = &mdt_object_kmem,
 117                 .ckd_name  = "mdt_obj",
 118                 .ckd_size  = sizeof(struct mdt_object)
 119         },
 120         {
 121                 .ckd_cache      = &mdt_hsm_cdt_kmem,
 122                 .ckd_name       = "mdt_cdt_restore_handle",
 123                 .ckd_size       = sizeof(struct cdt_restore_handle)
 124         },
 125         {
 126                 .ckd_cache      = &mdt_hsm_car_kmem,
 127                 .ckd_name       = "mdt_cdt_agent_req",
 128                 .ckd_size       = sizeof(struct cdt_agent_req)
 129         },
 130         {
 131                 .ckd_cache = NULL
 132         }
 133 };
 134
 135 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag)
 136 {
 137         if (!rep)
 138                 return 0;
 139         return rep->lock_policy_res1 & op_flag;
 140 }
 141
 142 void mdt_clear_disposition(struct mdt_thread_info *info,
 143                            struct ldlm_reply *rep, __u64 op_flag)
 144 {
 145         if (info) {
 146                 info->mti_opdata &= ~op_flag;
 147                 tgt_opdata_clear(info->mti_env, op_flag);
 148         }
 149         if (rep)
 150                 rep->lock_policy_res1 &= ~op_flag;
 151 }
 152
 153 void mdt_set_disposition(struct mdt_thread_info *info,
 154                          struct ldlm_reply *rep, __u64 op_flag)
 155 {
 156         if (info) {
 157                 info->mti_opdata |= op_flag;
 158                 tgt_opdata_set(info->mti_env, op_flag);
 159         }
 160         if (rep)
 161                 rep->lock_policy_res1 |= op_flag;
 162 }
 163
 164 void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
 165 {
 166         lh->mlh_pdo_hash = 0;
 167         lh->mlh_reg_mode = lm;
 168         lh->mlh_rreg_mode = lm;
 169         lh->mlh_type = MDT_REG_LOCK;
 170 }
 171
 172 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lock_mode,
 173                        const struct lu_name *lname)
 174 {
 175         lh->mlh_reg_mode = lock_mode;
 176         lh->mlh_rreg_mode = lock_mode;
 177         lh->mlh_type = MDT_PDO_LOCK;
 178
 179         if (lu_name_is_valid(lname)) {
 180                 lh->mlh_pdo_hash = full_name_hash(lname->ln_name,
 181                                                   lname->ln_namelen);
 182                 /* XXX Workaround for LU-2856
 183                  *
 184                  * Zero is a valid return value of full_name_hash, but
 185                  * several users of mlh_pdo_hash assume a non-zero
 186                  * hash value. We therefore map zero onto an
 187                  * arbitrary, but consistent value (1) to avoid
 188                  * problems further down the road. */
 189                 if (unlikely(lh->mlh_pdo_hash == 0))
 190                         lh->mlh_pdo_hash = 1;
 191         } else {
 192                 lh->mlh_pdo_hash = 0;
 193         }
 194 }
 195
 196 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 197                               struct mdt_lock_handle *lh)
 198 {
 199         mdl_mode_t mode;
 200         ENTRY;
 201
 202         /*
 203          * Any dir access needs couple of locks:
 204          *
 205          * 1) on part of dir we gonna take lookup/modify;
 206          *
 207          * 2) on whole dir to protect it from concurrent splitting and/or to
 208          * flush client's cache for readdir().
 209          *
 210          * so, for a given mode and object this routine decides what lock mode
 211          * to use for lock #2:
 212          *
 213          * 1) if caller's gonna lookup in dir then we need to protect dir from
 214          * being splitted only - LCK_CR
 215          *
 216          * 2) if caller's gonna modify dir then we need to protect dir from
 217          * being splitted and to flush cache - LCK_CW
 218          *
 219          * 3) if caller's gonna modify dir and that dir seems ready for
 220          * splitting then we need to protect it from any type of access
 221          * (lookup/modify/split) - LCK_EX --bzzz
 222          */
 223
 224         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 225         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 226
 227         /*
 228          * Ask underlaying level its opinion about preferable PDO lock mode
 229          * having access type passed as regular lock mode:
 230          *
 231          * - MDL_MINMODE means that lower layer does not want to specify lock
 232          * mode;
 233          *
 234          * - MDL_NL means that no PDO lock should be taken. This is used in some
 235          * cases. Say, for non-splittable directories no need to use PDO locks
 236          * at all.
 237          */
 238         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 239                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 240
 241         if (mode != MDL_MINMODE) {
 242                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 243         } else {
 244                 /*
 245                  * Lower layer does not want to specify locking mode. We do it
 246                  * our selves. No special protection is needed, just flush
 247                  * client's cache on modification and allow concurrent
 248                  * mondification.
 249                  */
 250                 switch (lh->mlh_reg_mode) {
 251                 case LCK_EX:
 252                         lh->mlh_pdo_mode = LCK_EX;
 253                         break;
 254                 case LCK_PR:
 255                         lh->mlh_pdo_mode = LCK_CR;
 256                         break;
 257                 case LCK_PW:
 258                         lh->mlh_pdo_mode = LCK_CW;
 259                         break;
 260                 default:
 261                         CERROR("Not expected lock type (0x%x)\n",
 262                                (int)lh->mlh_reg_mode);
 263                         LBUG();
 264                 }
 265         }
 266
 267         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 268         EXIT;
 269 }
 270
 271 static int mdt_getstatus(struct tgt_session_info *tsi)
 272 {
 273         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
 274         struct mdt_device       *mdt = info->mti_mdt;
 275         struct mdt_body         *repbody;
 276         int                      rc;
 277
 278         ENTRY;
 279
 280         rc = mdt_check_ucred(info);
 281         if (rc)
 282                 GOTO(out, rc = err_serious(rc));
 283
 284         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
 285                 GOTO(out, rc = err_serious(-ENOMEM));
 286
 287         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 288         repbody->mbo_fid1 = mdt->mdt_md_root_fid;
 289         repbody->mbo_valid |= OBD_MD_FLID;
 290
 291         EXIT;
 292 out:
 293         mdt_thread_info_fini(info);
 294         return rc;
 295 }
 296
 297 static int mdt_statfs(struct tgt_session_info *tsi)
 298 {
 299         struct ptlrpc_request           *req = tgt_ses_req(tsi);
 300         struct mdt_thread_info          *info = tsi2mdt_info(tsi);
 301         struct md_device                *next = info->mti_mdt->mdt_child;
 302         struct ptlrpc_service_part      *svcpt;
 303         struct obd_statfs               *osfs;
 304         int                             rc;
 305
 306         ENTRY;
 307
 308         svcpt = req->rq_rqbd->rqbd_svcpt;
 309
 310         /* This will trigger a watchdog timeout */
 311         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 312                          (MDT_SERVICE_WATCHDOG_FACTOR *
 313                           at_get(&svcpt->scp_at_estimate)) + 1);
 314
 315         rc = mdt_check_ucred(info);
 316         if (rc)
 317                 GOTO(out, rc = err_serious(rc));
 318
 319         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
 320                 GOTO(out, rc = err_serious(-ENOMEM));
 321
 322         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 323         if (!osfs)
 324                 GOTO(out, rc = -EPROTO);
 325
 326         /** statfs information are cached in the mdt_device */
 327         if (cfs_time_before_64(info->mti_mdt->mdt_osfs_age,
 328                                cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS))) {
 329                 /** statfs data is too old, get up-to-date one */
 330                 rc = next->md_ops->mdo_statfs(info->mti_env, next, osfs);
 331                 if (rc)
 332                         GOTO(out, rc);
 333                 spin_lock(&info->mti_mdt->mdt_osfs_lock);
 334                 info->mti_mdt->mdt_osfs = *osfs;
 335                 info->mti_mdt->mdt_osfs_age = cfs_time_current_64();
 336                 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
 337         } else {
 338                 /** use cached statfs data */
 339                 spin_lock(&info->mti_mdt->mdt_osfs_lock);
 340                 *osfs = info->mti_mdt->mdt_osfs;
 341                 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
 342         }
 343
 344         if (rc == 0)
 345                 mdt_counter_incr(req, LPROC_MDT_STATFS);
 346 out:
 347         mdt_thread_info_fini(info);
 348         RETURN(rc);
 349 }
 350
 351 #ifdef CONFIG_FS_POSIX_ACL
 352 /*
 353  * Pack ACL data into the reply. UIDs/GIDs are mapped and filtered by nodemap.
 354  *
 355  * \param       info    thread info object
 356  * \param       repbody reply to pack ACLs into
 357  * \param       o       mdt object of file to examine
 358  * \param       nodemap nodemap of client to reply to
 359  * \retval      0       success
 360  * \retval      -errno  error getting or parsing ACL from disk
 361  */
 362 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
 363                       struct mdt_object *o, struct lu_nodemap *nodemap)
 364 {
 365         const struct lu_env     *env = info->mti_env;
 366         struct md_object        *next = mdt_object_child(o);
 367         struct lu_buf           *buf = &info->mti_buf;
 368         int rc;
 369
 370         buf->lb_buf = req_capsule_server_get(info->mti_pill, &RMF_ACL);
 371         buf->lb_len = req_capsule_get_size(info->mti_pill, &RMF_ACL,
 372                                            RCL_SERVER);
 373         if (buf->lb_len == 0)
 374                 return 0;
 375
 376         rc = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_ACCESS);
 377         if (rc < 0) {
 378                 if (rc == -ENODATA) {
 379                         repbody->mbo_aclsize = 0;
 380                         repbody->mbo_valid |= OBD_MD_FLACL;
 381                         rc = 0;
 382                 } else if (rc == -EOPNOTSUPP) {
 383                         rc = 0;
 384                 } else {
 385                         CERROR("%s: unable to read "DFID" ACL: rc = %d\n",
 386                                mdt_obd_name(info->mti_mdt),
 387                                PFID(mdt_object_fid(o)), rc);
 388                 }
 389         } else {
 390                 rc = nodemap_map_acl(nodemap, buf->lb_buf,
 391                                      rc, NODEMAP_FS_TO_CLIENT);
 392                 /* if all ACLs mapped out, rc is still >= 0 */
 393                 if (rc < 0) {
 394                         CERROR("%s: nodemap_map_acl unable to parse "DFID
 395                                " ACL: rc = %d\n", mdt_obd_name(info->mti_mdt),
 396                                PFID(mdt_object_fid(o)), rc);
 397                 } else {
 398                         repbody->mbo_aclsize = rc;
 399                         repbody->mbo_valid |= OBD_MD_FLACL;
 400                         rc = 0;
 401                 }
 402         }
 403         return rc;
 404 }
 405 #endif
 406
 407 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 408                         const struct lu_attr *attr, const struct lu_fid *fid)
 409 {
 410         struct md_attr          *ma = &info->mti_attr;
 411         struct obd_export       *exp = info->mti_exp;
 412         struct lu_nodemap       *nodemap = exp->exp_target_data.ted_nodemap;
 413
 414         LASSERT(ma->ma_valid & MA_INODE);
 415
 416         if (attr->la_valid & LA_ATIME) {
 417                 b->mbo_atime = attr->la_atime;
 418                 b->mbo_valid |= OBD_MD_FLATIME;
 419         }
 420         if (attr->la_valid & LA_MTIME) {
 421                 b->mbo_mtime = attr->la_mtime;
 422                 b->mbo_valid |= OBD_MD_FLMTIME;
 423         }
 424         if (attr->la_valid & LA_CTIME) {
 425                 b->mbo_ctime = attr->la_ctime;
 426                 b->mbo_valid |= OBD_MD_FLCTIME;
 427         }
 428         if (attr->la_valid & LA_FLAGS) {
 429                 b->mbo_flags = attr->la_flags;
 430                 b->mbo_valid |= OBD_MD_FLFLAGS;
 431         }
 432         if (attr->la_valid & LA_NLINK) {
 433                 b->mbo_nlink = attr->la_nlink;
 434                 b->mbo_valid |= OBD_MD_FLNLINK;
 435         }
 436         if (attr->la_valid & LA_UID) {
 437                 b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 438                                             NODEMAP_FS_TO_CLIENT,
 439                                             attr->la_uid);
 440                 b->mbo_valid |= OBD_MD_FLUID;
 441         }
 442         if (attr->la_valid & LA_GID) {
 443                 b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 444                                             NODEMAP_FS_TO_CLIENT,
 445                                             attr->la_gid);
 446                 b->mbo_valid |= OBD_MD_FLGID;
 447         }
 448         b->mbo_mode = attr->la_mode;
 449         if (attr->la_valid & LA_MODE)
 450                 b->mbo_valid |= OBD_MD_FLMODE;
 451         if (attr->la_valid & LA_TYPE)
 452                 b->mbo_valid |= OBD_MD_FLTYPE;
 453
 454         if (fid != NULL) {
 455                 b->mbo_fid1 = *fid;
 456                 b->mbo_valid |= OBD_MD_FLID;
 457                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid="LPX64"\n",
 458                        PFID(fid), b->mbo_nlink, b->mbo_mode, b->mbo_valid);
 459         }
 460
 461         if (info != NULL)
 462                 mdt_body_reverse_idmap(info, b);
 463
 464         if (!(attr->la_valid & LA_TYPE))
 465                 return;
 466
 467         b->mbo_rdev   = attr->la_rdev;
 468         b->mbo_size   = attr->la_size;
 469         b->mbo_blocks = attr->la_blocks;
 470
 471         if (!S_ISREG(attr->la_mode)) {
 472                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 473         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
 474                 /* means no objects are allocated on osts. */
 475                 LASSERT(!(ma->ma_valid & MA_LOV));
 476                 /* just ignore blocks occupied by extend attributes on MDS */
 477                 b->mbo_blocks = 0;
 478                 /* if no object is allocated on osts, the size on mds is valid.
 479                  * b=22272 */
 480                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 481         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL &&
 482                    ma->ma_lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) {
 483                 /* A released file stores its size on MDS. */
 484                 /* But return 1 block for released file, unless tools like tar
 485                  * will consider it fully sparse. (LU-3864)
 486                  */
 487                 if (unlikely(b->mbo_size == 0))
 488                         b->mbo_blocks = 0;
 489                 else
 490                         b->mbo_blocks = 1;
 491                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 492         }
 493
 494         if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE))
 495                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 496                        PFID(fid), (unsigned long long)b->mbo_size);
 497 }
 498
 499 static inline int mdt_body_has_lov(const struct lu_attr *la,
 500                                    const struct mdt_body *body)
 501 {
 502         return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
 503                (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
 504 }
 505
 506 void mdt_client_compatibility(struct mdt_thread_info *info)
 507 {
 508         struct mdt_body       *body;
 509         struct ptlrpc_request *req = mdt_info_req(info);
 510         struct obd_export     *exp = req->rq_export;
 511         struct md_attr        *ma = &info->mti_attr;
 512         struct lu_attr        *la = &ma->ma_attr;
 513         ENTRY;
 514
 515         if (exp_connect_layout(exp))
 516                 /* the client can deal with 16-bit lmm_stripe_count */
 517                 RETURN_EXIT;
 518
 519         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 520
 521         if (!mdt_body_has_lov(la, body))
 522                 RETURN_EXIT;
 523
 524         /* now we have a reply with a lov for a client not compatible with the
 525          * layout lock so we have to clean the layout generation number */
 526         if (S_ISREG(la->la_mode))
 527                 ma->ma_lmm->lmm_layout_gen = 0;
 528         EXIT;
 529 }
 530
 531 static int mdt_attr_get_eabuf_size(struct mdt_thread_info *info,
 532                                    struct mdt_object *o)
 533 {
 534         const struct lu_env *env = info->mti_env;
 535         int rc, rc2;
 536
 537         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 538                           XATTR_NAME_LOV);
 539
 540         if (rc == -ENODATA)
 541                 rc = 0;
 542
 543         if (rc < 0)
 544                 goto out;
 545
 546         /* Is it a directory? Let's check for the LMV as well */
 547         if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
 548                 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 549                                    XATTR_NAME_LMV);
 550
 551                 if (rc2 == -ENODATA)
 552                         rc2 = mo_xattr_get(env, mdt_object_child(o),
 553                                            &LU_BUF_NULL,
 554                                            XATTR_NAME_DEFAULT_LMV);
 555
 556                 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
 557                         rc = rc2;
 558         }
 559
 560 out:
 561         return rc;
 562 }
 563
 564 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
 565                       const char *name)
 566 {
 567         const struct lu_env *env = info->mti_env;
 568         int rc;
 569         ENTRY;
 570
 571         LASSERT(info->mti_big_lmm_used == 0);
 572         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
 573         if (rc < 0)
 574                 RETURN(rc);
 575
 576         /* big_lmm may need to be grown */
 577         if (info->mti_big_lmmsize < rc) {
 578                 int size = size_roundup_power2(rc);
 579
 580                 if (info->mti_big_lmmsize > 0) {
 581                         /* free old buffer */
 582                         LASSERT(info->mti_big_lmm);
 583                         OBD_FREE_LARGE(info->mti_big_lmm,
 584                                        info->mti_big_lmmsize);
 585                         info->mti_big_lmm = NULL;
 586                         info->mti_big_lmmsize = 0;
 587                 }
 588
 589                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
 590                 if (info->mti_big_lmm == NULL)
 591                         RETURN(-ENOMEM);
 592                 info->mti_big_lmmsize = size;
 593         }
 594         LASSERT(info->mti_big_lmmsize >= rc);
 595
 596         info->mti_buf.lb_buf = info->mti_big_lmm;
 597         info->mti_buf.lb_len = info->mti_big_lmmsize;
 598         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
 599
 600         RETURN(rc);
 601 }
 602
 603 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
 604                    struct md_attr *ma, const char *name)
 605 {
 606         struct md_object *next = mdt_object_child(o);
 607         struct lu_buf    *buf = &info->mti_buf;
 608         int rc;
 609
 610         if (strcmp(name, XATTR_NAME_LOV) == 0) {
 611                 buf->lb_buf = ma->ma_lmm;
 612                 buf->lb_len = ma->ma_lmm_size;
 613                 LASSERT(!(ma->ma_valid & MA_LOV));
 614         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 615                 buf->lb_buf = ma->ma_lmv;
 616                 buf->lb_len = ma->ma_lmv_size;
 617                 LASSERT(!(ma->ma_valid & MA_LMV));
 618         } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 619                 buf->lb_buf = ma->ma_lmv;
 620                 buf->lb_len = ma->ma_lmv_size;
 621                 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
 622         } else {
 623                 return -EINVAL;
 624         }
 625
 626         rc = mo_xattr_get(info->mti_env, next, buf, name);
 627         if (rc > 0) {
 628
 629 got:
 630                 if (strcmp(name, XATTR_NAME_LOV) == 0) {
 631                         if (info->mti_big_lmm_used)
 632                                 ma->ma_lmm = info->mti_big_lmm;
 633
 634                         /* NOT return LOV EA with hole to old client. */
 635                         if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
 636                                      LOV_PATTERN_F_HOLE) &&
 637                             !(exp_connect_flags(info->mti_exp) &
 638                               OBD_CONNECT_LFSCK)) {
 639                                 return -EIO;
 640                         } else {
 641                                 ma->ma_lmm_size = rc;
 642                                 ma->ma_valid |= MA_LOV;
 643                         }
 644                 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 645                         if (info->mti_big_lmm_used)
 646                                 ma->ma_lmv = info->mti_big_lmm;
 647
 648                         ma->ma_lmv_size = rc;
 649                         ma->ma_valid |= MA_LMV;
 650                 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 651                         ma->ma_lmv_size = rc;
 652                         ma->ma_valid |= MA_LMV_DEF;
 653                 }
 654
 655                 /* Update mdt_max_mdsize so all clients will be aware that */
 656                 if (info->mti_mdt->mdt_max_mdsize < rc)
 657                         info->mti_mdt->mdt_max_mdsize = rc;
 658
 659                 rc = 0;
 660         } else if (rc == -ENODATA) {
 661                 /* no LOV EA */
 662                 rc = 0;
 663         } else if (rc == -ERANGE) {
 664                 /* Default LMV has fixed size, so it must be able to fit
 665                  * in the original buffer */
 666                 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
 667                         return rc;
 668                 rc = mdt_big_xattr_get(info, o, name);
 669                 if (rc > 0) {
 670                         info->mti_big_lmm_used = 1;
 671                         goto got;
 672                 }
 673         }
 674
 675         return rc;
 676 }
 677
 678 static int mdt_attr_get_pfid(struct mdt_thread_info *info,
 679                              struct mdt_object *o, struct lu_fid *pfid)
 680 {
 681         struct lu_buf           *buf = &info->mti_buf;
 682         struct link_ea_header   *leh;
 683         struct link_ea_entry    *lee;
 684         int                      rc;
 685         ENTRY;
 686
 687         buf->lb_buf = info->mti_big_lmm;
 688         buf->lb_len = info->mti_big_lmmsize;
 689         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
 690                           buf, XATTR_NAME_LINK);
 691         /* ignore errors, MA_PFID won't be set and it is
 692          * up to the caller to treat this as an error */
 693         if (rc == -ERANGE || buf->lb_len == 0) {
 694                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
 695                 buf->lb_buf = info->mti_big_lmm;
 696                 buf->lb_len = info->mti_big_lmmsize;
 697         }
 698
 699         if (rc < 0)
 700                 RETURN(rc);
 701         if (rc < sizeof(*leh)) {
 702                 CERROR("short LinkEA on "DFID": rc = %d\n",
 703                        PFID(mdt_object_fid(o)), rc);
 704                 RETURN(-ENODATA);
 705         }
 706
 707         leh = (struct link_ea_header *) buf->lb_buf;
 708         lee = (struct link_ea_entry *)(leh + 1);
 709         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
 710                 leh->leh_magic = LINK_EA_MAGIC;
 711                 leh->leh_reccount = __swab32(leh->leh_reccount);
 712                 leh->leh_len = __swab64(leh->leh_len);
 713         }
 714         if (leh->leh_magic != LINK_EA_MAGIC)
 715                 RETURN(-EINVAL);
 716         if (leh->leh_reccount == 0)
 717                 RETURN(-ENODATA);
 718
 719         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
 720         fid_be_to_cpu(pfid, pfid);
 721
 722         RETURN(0);
 723 }
 724
 725 int mdt_attr_get_complex(struct mdt_thread_info *info,
 726                          struct mdt_object *o, struct md_attr *ma)
 727 {
 728         const struct lu_env *env = info->mti_env;
 729         struct md_object    *next = mdt_object_child(o);
 730         struct lu_buf       *buf = &info->mti_buf;
 731         int                  need = ma->ma_need;
 732         int                  rc = 0, rc2;
 733         u32                  mode;
 734         ENTRY;
 735
 736         ma->ma_valid = 0;
 737
 738         if (mdt_object_exists(o) == 0)
 739                 GOTO(out, rc = -ENOENT);
 740         mode = lu_object_attr(&next->mo_lu);
 741
 742         if (need & MA_INODE) {
 743                 ma->ma_need = MA_INODE;
 744                 rc = mo_attr_get(env, next, ma);
 745                 if (rc)
 746                         GOTO(out, rc);
 747                 ma->ma_valid |= MA_INODE;
 748         }
 749
 750         if (need & MA_PFID) {
 751                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
 752                 if (rc == 0)
 753                         ma->ma_valid |= MA_PFID;
 754                 /* ignore this error, parent fid is not mandatory */
 755                 rc = 0;
 756         }
 757
 758         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
 759                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
 760                 if (rc)
 761                         GOTO(out, rc);
 762         }
 763
 764         if (need & MA_LMV && S_ISDIR(mode)) {
 765                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
 766                 if (rc != 0)
 767                         GOTO(out, rc);
 768         }
 769
 770         if (need & MA_LMV_DEF && S_ISDIR(mode)) {
 771                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
 772                 if (rc != 0)
 773                         GOTO(out, rc);
 774         }
 775
 776         if (need & MA_HSM && S_ISREG(mode)) {
 777                 buf->lb_buf = info->mti_xattr_buf;
 778                 buf->lb_len = sizeof(info->mti_xattr_buf);
 779                 CLASSERT(sizeof(struct hsm_attrs) <=
 780                          sizeof(info->mti_xattr_buf));
 781                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
 782                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
 783                 if (rc2 == 0)
 784                         ma->ma_valid |= MA_HSM;
 785                 else if (rc2 < 0 && rc2 != -ENODATA)
 786                         GOTO(out, rc = rc2);
 787         }
 788
 789 #ifdef CONFIG_FS_POSIX_ACL
 790         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
 791                 buf->lb_buf = ma->ma_acl;
 792                 buf->lb_len = ma->ma_acl_size;
 793                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
 794                 if (rc2 > 0) {
 795                         ma->ma_acl_size = rc2;
 796                         ma->ma_valid |= MA_ACL_DEF;
 797                 } else if (rc2 == -ENODATA) {
 798                         /* no ACLs */
 799                         ma->ma_acl_size = 0;
 800                 } else
 801                         GOTO(out, rc = rc2);
 802         }
 803 #endif
 804 out:
 805         ma->ma_need = need;
 806         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
 807                rc, ma->ma_valid, ma->ma_lmm);
 808         RETURN(rc);
 809 }
 810
 811 static int mdt_getattr_internal(struct mdt_thread_info *info,
 812                                 struct mdt_object *o, int ma_need)
 813 {
 814         struct md_object        *next = mdt_object_child(o);
 815         const struct mdt_body   *reqbody = info->mti_body;
 816         struct ptlrpc_request   *req = mdt_info_req(info);
 817         struct md_attr          *ma = &info->mti_attr;
 818         struct lu_attr          *la = &ma->ma_attr;
 819         struct req_capsule      *pill = info->mti_pill;
 820         const struct lu_env     *env = info->mti_env;
 821         struct mdt_body         *repbody;
 822         struct lu_buf           *buffer = &info->mti_buf;
 823         struct obd_export       *exp = info->mti_exp;
 824         struct lu_nodemap       *nodemap = exp->exp_target_data.ted_nodemap;
 825         int                      rc;
 826         int                      is_root;
 827         ENTRY;
 828
 829         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
 830                 RETURN(err_serious(-ENOMEM));
 831
 832         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 833
 834         ma->ma_valid = 0;
 835
 836         if (mdt_object_remote(o)) {
 837                 /* This object is located on remote node.*/
 838                 /* Return -ENOTSUPP for old client */
 839                 if (!mdt_is_dne_client(req->rq_export))
 840                         GOTO(out, rc = -ENOTSUPP);
 841
 842                 repbody->mbo_fid1 = *mdt_object_fid(o);
 843                 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
 844                 GOTO(out, rc = 0);
 845         }
 846
 847         if (reqbody->mbo_eadatasize > 0) {
 848                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
 849                 if (buffer->lb_buf == NULL)
 850                         GOTO(out, rc = -EPROTO);
 851                 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
 852                                                       RCL_SERVER);
 853         } else {
 854                 buffer->lb_buf = NULL;
 855                 buffer->lb_len = 0;
 856                 ma_need &= ~(MA_LOV | MA_LMV);
 857                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
 858                        mdt_obd_name(info->mti_mdt),
 859                        req->rq_export->exp_client_uuid.uuid);
 860         }
 861
 862         /* If it is dir object and client require MEA, then we got MEA */
 863         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 864             (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
 865                 /* Assumption: MDT_MD size is enough for lmv size. */
 866                 ma->ma_lmv = buffer->lb_buf;
 867                 ma->ma_lmv_size = buffer->lb_len;
 868                 ma->ma_need = MA_INODE;
 869                 if (ma->ma_lmv_size > 0) {
 870                         if (reqbody->mbo_valid & OBD_MD_MEA)
 871                                 ma->ma_need |= MA_LMV;
 872                         else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA)
 873                                 ma->ma_need |= MA_LMV_DEF;
 874                 }
 875         } else {
 876                 ma->ma_lmm = buffer->lb_buf;
 877                 ma->ma_lmm_size = buffer->lb_len;
 878                 ma->ma_need = MA_INODE | MA_HSM;
 879                 if (ma->ma_lmm_size > 0)
 880                         ma->ma_need |= MA_LOV;
 881         }
 882
 883         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 884             reqbody->mbo_valid & OBD_MD_FLDIREA  &&
 885             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
 886                 /* get default stripe info for this dir. */
 887                 ma->ma_need |= MA_LOV_DEF;
 888         }
 889         ma->ma_need |= ma_need;
 890
 891         rc = mdt_attr_get_complex(info, o, ma);
 892         if (unlikely(rc)) {
 893                 CERROR("%s: getattr error for "DFID": rc = %d\n",
 894                        mdt_obd_name(info->mti_mdt),
 895                        PFID(mdt_object_fid(o)), rc);
 896                 RETURN(rc);
 897         }
 898
 899         /* if file is released, check if a restore is running */
 900         if (ma->ma_valid & MA_HSM) {
 901                 repbody->mbo_valid |= OBD_MD_TSTATE;
 902                 if ((ma->ma_hsm.mh_flags & HS_RELEASED) &&
 903                     mdt_hsm_restore_is_running(info, mdt_object_fid(o)))
 904                         repbody->mbo_t_state = MS_RESTORE;
 905         }
 906
 907         is_root = lu_fid_eq(mdt_object_fid(o), &info->mti_mdt->mdt_md_root_fid);
 908
 909         /* the Lustre protocol supposes to return default striping
 910          * on the user-visible root if explicitly requested */
 911         if ((ma->ma_valid & MA_LOV) == 0 && S_ISDIR(la->la_mode) &&
 912             (ma->ma_need & MA_LOV_DEF && is_root) && ma->ma_need & MA_LOV) {
 913                 struct lu_fid      rootfid;
 914                 struct mdt_object *root;
 915                 struct mdt_device *mdt = info->mti_mdt;
 916
 917                 rc = dt_root_get(env, mdt->mdt_bottom, &rootfid);
 918                 if (rc)
 919                         RETURN(rc);
 920                 root = mdt_object_find(env, mdt, &rootfid);
 921                 if (IS_ERR(root))
 922                         RETURN(PTR_ERR(root));
 923                 rc = mdt_stripe_get(info, root, ma, XATTR_NAME_LOV);
 924                 mdt_object_put(info->mti_env, root);
 925                 if (unlikely(rc)) {
 926                         CERROR("%s: getattr error for "DFID": rc = %d\n",
 927                                mdt_obd_name(info->mti_mdt),
 928                                PFID(mdt_object_fid(o)), rc);
 929                         RETURN(rc);
 930                 }
 931         }
 932
 933         if (likely(ma->ma_valid & MA_INODE))
 934                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
 935         else
 936                 RETURN(-EFAULT);
 937
 938         if (mdt_body_has_lov(la, reqbody)) {
 939                 if (ma->ma_valid & MA_LOV) {
 940                         LASSERT(ma->ma_lmm_size);
 941                         repbody->mbo_eadatasize = ma->ma_lmm_size;
 942                         if (S_ISDIR(la->la_mode))
 943                                 repbody->mbo_valid |= OBD_MD_FLDIREA;
 944                         else
 945                                 repbody->mbo_valid |= OBD_MD_FLEASIZE;
 946                         mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
 947                 }
 948                 if (ma->ma_valid & MA_LMV) {
 949                         /* Return -ENOTSUPP for old client */
 950                         if (!mdt_is_striped_client(req->rq_export))
 951                                 RETURN(-ENOTSUPP);
 952
 953                         LASSERT(S_ISDIR(la->la_mode));
 954                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
 955                         repbody->mbo_eadatasize = ma->ma_lmv_size;
 956                         repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
 957                 }
 958                 if (ma->ma_valid & MA_LMV_DEF) {
 959                         /* Return -ENOTSUPP for old client */
 960                         if (!mdt_is_striped_client(req->rq_export))
 961                                 RETURN(-ENOTSUPP);
 962                         LASSERT(S_ISDIR(la->la_mode));
 963                         repbody->mbo_eadatasize = ma->ma_lmv_size;
 964                         repbody->mbo_valid |= (OBD_MD_FLDIREA |
 965                                                OBD_MD_DEFAULT_MEA);
 966                 }
 967         } else if (S_ISLNK(la->la_mode) &&
 968                    reqbody->mbo_valid & OBD_MD_LINKNAME) {
 969                 buffer->lb_buf = ma->ma_lmm;
 970                 /* eadatasize from client includes NULL-terminator, so
 971                  * there is no need to read it */
 972                 buffer->lb_len = reqbody->mbo_eadatasize - 1;
 973                 rc = mo_readlink(env, next, buffer);
 974                 if (unlikely(rc <= 0)) {
 975                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
 976                                mdt_obd_name(info->mti_mdt),
 977                                PFID(mdt_object_fid(o)), rc);
 978                         rc = -EFAULT;
 979                 } else {
 980                         int print_limit = min_t(int, PAGE_CACHE_SIZE - 128, rc);
 981
 982                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
 983                                 rc -= 2;
 984                         repbody->mbo_valid |= OBD_MD_LINKNAME;
 985                         /* we need to report back size with NULL-terminator
 986                          * because client expects that */
 987                         repbody->mbo_eadatasize = rc + 1;
 988                         if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
 989                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
 990                                        "on "DFID ", expected %d\n",
 991                                        mdt_obd_name(info->mti_mdt),
 992                                        rc, PFID(mdt_object_fid(o)),
 993                                        reqbody->mbo_eadatasize - 1);
 994                         /* NULL terminate */
 995                         ((char *)ma->ma_lmm)[rc] = 0;
 996
 997                         /* If the total CDEBUG() size is larger than a page, it
 998                          * will print a warning to the console, avoid this by
 999                          * printing just the last part of the symlink. */
1000                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
1001                                print_limit < rc ? "..." : "", print_limit,
1002                                (char *)ma->ma_lmm + rc - print_limit, rc);
1003                         rc = 0;
1004                 }
1005         }
1006
1007         if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
1008                 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1009                 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1010                 CDEBUG(D_INODE, "changing the max MD size to %u\n",
1011                        repbody->mbo_max_mdsize);
1012         }
1013
1014         if (exp_connect_rmtclient(info->mti_exp) &&
1015             reqbody->mbo_valid & OBD_MD_FLRMTPERM) {
1016                 void *buf = req_capsule_server_get(pill, &RMF_ACL);
1017
1018                 /* mdt_getattr_lock only */
1019                 rc = mdt_pack_remote_perm(info, o, buf);
1020                 if (rc) {
1021                         repbody->mbo_valid &= ~OBD_MD_FLRMTPERM;
1022                         repbody->mbo_aclsize = 0;
1023                         RETURN(rc);
1024                 } else {
1025                         repbody->mbo_valid |= OBD_MD_FLRMTPERM;
1026                         repbody->mbo_aclsize = sizeof(struct mdt_remote_perm);
1027                 }
1028         }
1029 #ifdef CONFIG_FS_POSIX_ACL
1030         else if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1031                  (reqbody->mbo_valid & OBD_MD_FLACL))
1032                 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
1033 #endif
1034
1035 out:
1036         if (rc == 0)
1037                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
1038
1039         RETURN(rc);
1040 }
1041
1042 static int mdt_getattr(struct tgt_session_info *tsi)
1043 {
1044         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1045         struct mdt_object       *obj = info->mti_object;
1046         struct req_capsule      *pill = info->mti_pill;
1047         struct mdt_body         *reqbody;
1048         struct mdt_body         *repbody;
1049         int rc, rc2;
1050         ENTRY;
1051
1052         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1053         LASSERT(reqbody);
1054         LASSERT(obj != NULL);
1055         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1056
1057         /* Unlike intent case where we need to pre-fill out buffers early on
1058          * in intent policy for ldlm reasons, here we can have a much better
1059          * guess at EA size by just reading it from disk.
1060          * Exceptions are readdir and (missing) directory striping */
1061         /* Readlink */
1062         if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1063                 /* No easy way to know how long is the symlink, but it cannot
1064                  * be more than PATH_MAX, so we allocate +1 */
1065                 rc = PATH_MAX + 1;
1066
1067         /* A special case for fs ROOT: getattr there might fetch
1068          * default EA for entire fs, not just for this dir!
1069          */
1070         } else if (lu_fid_eq(mdt_object_fid(obj),
1071                              &info->mti_mdt->mdt_md_root_fid) &&
1072                    (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1073                    (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1074                                                                  MDS_GETATTR)) {
1075                 /* Should the default strping be bigger, mdt_fix_reply
1076                  * will reallocate */
1077                 rc = DEF_REP_MD_SIZE;
1078         } else {
1079                 /* Read the actual EA size from disk */
1080                 rc = mdt_attr_get_eabuf_size(info, obj);
1081         }
1082
1083         if (rc < 0)
1084                 GOTO(out, rc = err_serious(rc));
1085
1086         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1087
1088         rc = req_capsule_server_pack(pill);
1089         if (unlikely(rc != 0))
1090                 GOTO(out, rc = err_serious(rc));
1091
1092         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1093         LASSERT(repbody != NULL);
1094         repbody->mbo_eadatasize = 0;
1095         repbody->mbo_aclsize = 0;
1096
1097         if (reqbody->mbo_valid & OBD_MD_FLRMTPERM)
1098                 rc = mdt_init_ucred(info, reqbody);
1099         else
1100                 rc = mdt_check_ucred(info);
1101         if (unlikely(rc))
1102                 GOTO(out_shrink, rc);
1103
1104         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1105
1106         rc = mdt_getattr_internal(info, obj, 0);
1107         if (reqbody->mbo_valid & OBD_MD_FLRMTPERM)
1108                 mdt_exit_ucred(info);
1109         EXIT;
1110 out_shrink:
1111         mdt_client_compatibility(info);
1112         rc2 = mdt_fix_reply(info);
1113         if (rc == 0)
1114                 rc = rc2;
1115 out:
1116         mdt_thread_info_fini(info);
1117         return rc;
1118 }
1119
1120 /**
1121  * Exchange MOF_LOV_CREATED flags between two objects after a
1122  * layout swap. No assumption is made on whether o1 or o2 have
1123  * created objects or not.
1124  *
1125  * \param[in,out] o1    First swap layout object
1126  * \param[in,out] o2    Second swap layout object
1127  */
1128 static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2)
1129 {
1130         __u64   o1_flags;
1131
1132         mutex_lock(&o1->mot_lov_mutex);
1133         mutex_lock(&o2->mot_lov_mutex);
1134
1135         o1_flags = o1->mot_flags;
1136         o1->mot_flags = (o1->mot_flags & ~MOF_LOV_CREATED) |
1137                         (o2->mot_flags & MOF_LOV_CREATED);
1138
1139         o2->mot_flags = (o2->mot_flags & ~MOF_LOV_CREATED) |
1140                         (o1_flags & MOF_LOV_CREATED);
1141
1142         mutex_unlock(&o2->mot_lov_mutex);
1143         mutex_unlock(&o1->mot_lov_mutex);
1144 }
1145
1146 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1147 {
1148         struct mdt_thread_info  *info;
1149         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1150         struct obd_export       *exp = req->rq_export;
1151         struct mdt_object       *o1, *o2, *o;
1152         struct mdt_lock_handle  *lh1, *lh2;
1153         struct mdc_swap_layouts *msl;
1154         int                      rc;
1155         ENTRY;
1156
1157         /* client does not support layout lock, so layout swaping
1158          * is disabled.
1159          * FIXME: there is a problem for old clients which don't support
1160          * layout lock yet. If those clients have already opened the file
1161          * they won't be notified at all so that old layout may still be
1162          * used to do IO. This can be fixed after file release is landed by
1163          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1164         if (!exp_connect_layout(exp))
1165                 RETURN(-EOPNOTSUPP);
1166
1167         info = tsi2mdt_info(tsi);
1168
1169         if (info->mti_dlm_req != NULL)
1170                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1171
1172         o1 = info->mti_object;
1173         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1174                                 &info->mti_body->mbo_fid2);
1175         if (IS_ERR(o))
1176                 GOTO(out, rc = PTR_ERR(o));
1177
1178         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1179                 GOTO(put, rc = -ENOENT);
1180
1181         rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1182         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1183                 GOTO(put, rc);
1184
1185         if (rc < 0)
1186                 swap(o1, o2);
1187
1188         /* permission check. Make sure the calling process having permission
1189          * to write both files. */
1190         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1191                                 MAY_WRITE);
1192         if (rc < 0)
1193                 GOTO(put, rc);
1194
1195         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1196                                 MAY_WRITE);
1197         if (rc < 0)
1198                 GOTO(put, rc);
1199
1200         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1201         if (msl == NULL)
1202                 GOTO(put, rc = -EPROTO);
1203
1204         lh1 = &info->mti_lh[MDT_LH_NEW];
1205         mdt_lock_reg_init(lh1, LCK_EX);
1206         lh2 = &info->mti_lh[MDT_LH_OLD];
1207         mdt_lock_reg_init(lh2, LCK_EX);
1208
1209         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1210                              MDS_INODELOCK_XATTR);
1211         if (rc < 0)
1212                 GOTO(put, rc);
1213
1214         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1215                              MDS_INODELOCK_XATTR);
1216         if (rc < 0)
1217                 GOTO(unlock1, rc);
1218
1219         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1220                              mdt_object_child(o2), msl->msl_flags);
1221         if (rc < 0)
1222                 GOTO(unlock2, rc);
1223
1224         mdt_swap_lov_flag(o1, o2);
1225
1226 unlock2:
1227         mdt_object_unlock(info, o2, lh2, rc);
1228 unlock1:
1229         mdt_object_unlock(info, o1, lh1, rc);
1230 put:
1231         mdt_object_put(info->mti_env, o);
1232 out:
1233         mdt_thread_info_fini(info);
1234         RETURN(rc);
1235 }
1236
1237 static int mdt_raw_lookup(struct mdt_thread_info *info,
1238                           struct mdt_object *parent,
1239                           const struct lu_name *lname,
1240                           struct ldlm_reply *ldlm_rep)
1241 {
1242         struct lu_fid   *child_fid = &info->mti_tmp_fid1;
1243         int              rc;
1244         ENTRY;
1245
1246         LASSERT(!info->mti_cross_ref);
1247
1248         /* Only got the fid of this obj by name */
1249         fid_zero(child_fid);
1250         rc = mdo_lookup(info->mti_env, mdt_object_child(info->mti_object),
1251                         lname, child_fid, &info->mti_spec);
1252         if (rc == 0) {
1253                 struct mdt_body *repbody;
1254
1255                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1256                 repbody->mbo_fid1 = *child_fid;
1257                 repbody->mbo_valid = OBD_MD_FLID;
1258                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1259         } else if (rc == -ENOENT) {
1260                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1261         }
1262
1263         RETURN(rc);
1264 }
1265
1266 /*
1267  * UPDATE lock should be taken against parent, and be released before exit;
1268  * child_bits lock should be taken against child, and be returned back:
1269  *            (1)normal request should release the child lock;
1270  *            (2)intent request will grant the lock to client.
1271  */
1272 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1273                                  struct mdt_lock_handle *lhc,
1274                                  __u64 child_bits,
1275                                  struct ldlm_reply *ldlm_rep)
1276 {
1277         struct ptlrpc_request  *req       = mdt_info_req(info);
1278         struct mdt_body        *reqbody   = NULL;
1279         struct mdt_object      *parent    = info->mti_object;
1280         struct mdt_object      *child;
1281         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1282         struct lu_name         *lname     = NULL;
1283         struct mdt_lock_handle *lhp       = NULL;
1284         struct ldlm_lock       *lock;
1285         bool                    is_resent;
1286         bool                    try_layout;
1287         int                     ma_need = 0;
1288         int                     rc;
1289         ENTRY;
1290
1291         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1292         LASSERT(ergo(is_resent,
1293                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1294
1295         if (parent == NULL)
1296                 RETURN(-ENOENT);
1297
1298         if (info->mti_cross_ref) {
1299                 /* Only getattr on the child. Parent is on another node. */
1300                 mdt_set_disposition(info, ldlm_rep,
1301                                     DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1302                 child = parent;
1303                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1304                        "ldlm_rep = %p\n",
1305                        PFID(mdt_object_fid(child)), ldlm_rep);
1306
1307                 rc = mdt_check_resent_lock(info, child, lhc);
1308                 if (rc < 0) {
1309                         RETURN(rc);
1310                 } else if (rc > 0) {
1311                         mdt_lock_handle_init(lhc);
1312                         mdt_lock_reg_init(lhc, LCK_PR);
1313
1314                         /*
1315                          * Object's name is on another MDS, no lookup or layout
1316                          * lock is needed here but update lock is.
1317                          */
1318                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
1319                                         MDS_INODELOCK_LAYOUT);
1320                         child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1321
1322                         rc = mdt_object_lock(info, child, lhc, child_bits);
1323                         if (rc < 0)
1324                                 RETURN(rc);
1325                 }
1326
1327                 /* Finally, we can get attr for child. */
1328                 if (!mdt_object_exists(child)) {
1329                         LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1330                                         &child->mot_obj,
1331                                         "remote object doesn't exist.\n");
1332                         mdt_object_unlock(info, child, lhc, 1);
1333                         RETURN(-ENOENT);
1334                 }
1335
1336                 rc = mdt_getattr_internal(info, child, 0);
1337                 if (unlikely(rc != 0))
1338                         mdt_object_unlock(info, child, lhc, 1);
1339
1340                 RETURN(rc);
1341         }
1342
1343         lname = &info->mti_name;
1344         mdt_name_unpack(info->mti_pill, &RMF_NAME, lname, MNF_FIX_ANON);
1345
1346         if (lu_name_is_valid(lname)) {
1347                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1348                        "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1349                        PNAME(lname), ldlm_rep);
1350         } else {
1351                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1352                 if (unlikely(reqbody == NULL))
1353                         RETURN(err_serious(-EPROTO));
1354
1355                 *child_fid = reqbody->mbo_fid2;
1356
1357                 if (unlikely(!fid_is_sane(child_fid)))
1358                         RETURN(err_serious(-EINVAL));
1359
1360                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1361                        "ldlm_rep = %p\n",
1362                        PFID(mdt_object_fid(parent)),
1363                        PFID(&reqbody->mbo_fid2), ldlm_rep);
1364         }
1365
1366         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1367
1368         if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1369                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1370                                 &parent->mot_obj,
1371                                 "Parent doesn't exist!\n");
1372                 RETURN(-ESTALE);
1373         }
1374
1375         if (mdt_object_remote(parent)) {
1376                 CERROR("%s: parent "DFID" is on remote target\n",
1377                        mdt_obd_name(info->mti_mdt),
1378                        PFID(mdt_object_fid(parent)));
1379                 RETURN(-EIO);
1380         }
1381
1382         if (lu_name_is_valid(lname)) {
1383                 /* Always allow to lookup ".." */
1384                 if (unlikely(lname->ln_namelen == 2 &&
1385                              lname->ln_name[0] == '.' &&
1386                              lname->ln_name[1] == '.'))
1387                         info->mti_spec.sp_permitted = 1;
1388
1389                 if (info->mti_body->mbo_valid == OBD_MD_FLID) {
1390                         rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1391
1392                         RETURN(rc);
1393                 }
1394
1395                 /* step 1: lock parent only if parent is a directory */
1396                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1397                         lhp = &info->mti_lh[MDT_LH_PARENT];
1398                         mdt_lock_pdo_init(lhp, LCK_PR, lname);
1399                         rc = mdt_object_lock(info, parent, lhp,
1400                                              MDS_INODELOCK_UPDATE);
1401                         if (unlikely(rc != 0))
1402                                 RETURN(rc);
1403                 }
1404
1405                 /* step 2: lookup child's fid by name */
1406                 fid_zero(child_fid);
1407                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
1408                                 child_fid, &info->mti_spec);
1409                 if (rc == -ENOENT)
1410                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1411
1412                 if (rc != 0)
1413                         GOTO(out_parent, rc);
1414         }
1415
1416         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1417
1418         /*
1419          *step 3: find the child object by fid & lock it.
1420          *        regardless if it is local or remote.
1421          *
1422          *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1423          *      set parent dir fid the same as child fid in getattr by fid case
1424          *      we should not lu_object_find() the object again, could lead
1425          *      to hung if there is a concurrent unlink destroyed the object.
1426          */
1427         if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1428                 mdt_object_get(info->mti_env, parent);
1429                 child = parent;
1430         } else {
1431                 child = mdt_object_find(info->mti_env, info->mti_mdt,
1432                                         child_fid);
1433         }
1434
1435         if (unlikely(IS_ERR(child)))
1436                 GOTO(out_parent, rc = PTR_ERR(child));
1437
1438         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
1439         rc = mdt_check_resent_lock(info, child, lhc);
1440         if (rc < 0) {
1441                 GOTO(out_child, rc);
1442         } else if (rc > 0) {
1443                 mdt_lock_handle_init(lhc);
1444                 mdt_lock_reg_init(lhc, LCK_PR);
1445                 try_layout = false;
1446
1447                 if (!mdt_object_exists(child)) {
1448                         LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1449                                         &child->mot_obj,
1450                                         "Object doesn't exist!\n");
1451                         GOTO(out_child, rc = -ENOENT);
1452                 }
1453
1454                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1455                       mdt_object_exists(child) && !mdt_object_remote(child)) {
1456                         struct md_attr *ma = &info->mti_attr;
1457
1458                         ma->ma_valid = 0;
1459                         ma->ma_need = MA_INODE;
1460                         rc = mdt_attr_get_complex(info, child, ma);
1461                         if (unlikely(rc != 0))
1462                                 GOTO(out_child, rc);
1463
1464                         /* If the file has not been changed for some time, we
1465                          * return not only a LOOKUP lock, but also an UPDATE
1466                          * lock and this might save us RPC on later STAT. For
1467                          * directories, it also let negative dentry cache start
1468                          * working for this dir. */
1469                         if (ma->ma_valid & MA_INODE &&
1470                             ma->ma_attr.la_valid & LA_CTIME &&
1471                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1472                                 ma->ma_attr.la_ctime < cfs_time_current_sec())
1473                                 child_bits |= MDS_INODELOCK_UPDATE;
1474                 }
1475
1476                 /* layout lock must be granted in a best-effort way
1477                  * for IT operations */
1478                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1479                 if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1480                     exp_connect_layout(info->mti_exp) &&
1481                     S_ISREG(lu_object_attr(&child->mot_obj)) &&
1482                     !mdt_object_remote(child) && ldlm_rep != NULL) {
1483                         /* try to grant layout lock for regular file. */
1484                         try_layout = true;
1485                 }
1486
1487                 rc = 0;
1488                 if (try_layout) {
1489                         child_bits |= MDS_INODELOCK_LAYOUT;
1490                         /* try layout lock, it may fail to be granted due to
1491                          * contention at LOOKUP or UPDATE */
1492                         if (!mdt_object_lock_try(info, child, lhc,
1493                                                  child_bits)) {
1494                                 child_bits &= ~MDS_INODELOCK_LAYOUT;
1495                                 LASSERT(child_bits != 0);
1496                                 rc = mdt_object_lock(info, child, lhc,
1497                                                      child_bits);
1498                         } else {
1499                                 ma_need |= MA_LOV;
1500                         }
1501                 } else {
1502                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1503                          * client will enqueue the lock to the remote MDT */
1504                         if (mdt_object_remote(child))
1505                                 child_bits &= ~MDS_INODELOCK_UPDATE;
1506                         rc = mdt_object_lock(info, child, lhc, child_bits);
1507                 }
1508                 if (unlikely(rc != 0))
1509                         GOTO(out_child, rc);
1510         }
1511
1512         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1513
1514         /* finally, we can get attr for child. */
1515         rc = mdt_getattr_internal(info, child, ma_need);
1516         if (unlikely(rc != 0)) {
1517                 mdt_object_unlock(info, child, lhc, 1);
1518         } else if (lock) {
1519                 /* Debugging code. */
1520                 LDLM_DEBUG(lock, "Returning lock to client");
1521                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1522                                          &lock->l_resource->lr_name),
1523                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1524                          PLDLMRES(lock->l_resource),
1525                          PFID(mdt_object_fid(child)));
1526         }
1527         if (lock)
1528                 LDLM_LOCK_PUT(lock);
1529
1530         EXIT;
1531 out_child:
1532         mdt_object_put(info->mti_env, child);
1533 out_parent:
1534         if (lhp)
1535                 mdt_object_unlock(info, parent, lhp, 1);
1536         return rc;
1537 }
1538
1539 /* normal handler: should release the child lock */
1540 static int mdt_getattr_name(struct tgt_session_info *tsi)
1541 {
1542         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1543         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1544         struct mdt_body        *reqbody;
1545         struct mdt_body        *repbody;
1546         int rc, rc2;
1547         ENTRY;
1548
1549         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1550         LASSERT(reqbody != NULL);
1551         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1552         LASSERT(repbody != NULL);
1553
1554         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1555         repbody->mbo_eadatasize = 0;
1556         repbody->mbo_aclsize = 0;
1557
1558         rc = mdt_init_ucred(info, reqbody);
1559         if (unlikely(rc))
1560                 GOTO(out_shrink, rc);
1561
1562         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1563         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1564                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1565                 lhc->mlh_reg_lh.cookie = 0;
1566         }
1567         mdt_exit_ucred(info);
1568         EXIT;
1569 out_shrink:
1570         mdt_client_compatibility(info);
1571         rc2 = mdt_fix_reply(info);
1572         if (rc == 0)
1573                 rc = rc2;
1574         mdt_thread_info_fini(info);
1575         return rc;
1576 }
1577
1578 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1579                          void *karg, void *uarg);
1580
1581 static int mdt_set_info(struct tgt_session_info *tsi)
1582 {
1583         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1584         char                    *key;
1585         void                    *val;
1586         int                      keylen, vallen, rc = 0;
1587
1588         ENTRY;
1589
1590         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
1591         if (key == NULL) {
1592                 DEBUG_REQ(D_HA, req, "no set_info key");
1593                 RETURN(err_serious(-EFAULT));
1594         }
1595
1596         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
1597                                       RCL_CLIENT);
1598
1599         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
1600         if (val == NULL) {
1601                 DEBUG_REQ(D_HA, req, "no set_info val");
1602                 RETURN(err_serious(-EFAULT));
1603         }
1604
1605         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
1606                                       RCL_CLIENT);
1607
1608         /* Swab any part of val you need to here */
1609         if (KEY_IS(KEY_READ_ONLY)) {
1610                 spin_lock(&req->rq_export->exp_lock);
1611                 if (*(__u32 *)val)
1612                         *exp_connect_flags_ptr(req->rq_export) |=
1613                                 OBD_CONNECT_RDONLY;
1614                 else
1615                         *exp_connect_flags_ptr(req->rq_export) &=
1616                                 ~OBD_CONNECT_RDONLY;
1617                 spin_unlock(&req->rq_export->exp_lock);
1618         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1619                 struct changelog_setinfo *cs = val;
1620
1621                 if (vallen != sizeof(*cs)) {
1622                         CERROR("%s: bad changelog_clear setinfo size %d\n",
1623                                tgt_name(tsi->tsi_tgt), vallen);
1624                         RETURN(-EINVAL);
1625                 }
1626                 if (ptlrpc_req_need_swab(req)) {
1627                         __swab64s(&cs->cs_recno);
1628                         __swab32s(&cs->cs_id);
1629                 }
1630
1631                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
1632                                    vallen, val, NULL);
1633         } else {
1634                 RETURN(-EINVAL);
1635         }
1636         RETURN(rc);
1637 }
1638
1639 static int mdt_readpage(struct tgt_session_info *tsi)
1640 {
1641         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
1642         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
1643         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
1644         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
1645         struct mdt_body         *repbody;
1646         int                      rc;
1647         int                      i;
1648
1649         ENTRY;
1650
1651         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1652                 RETURN(err_serious(-ENOMEM));
1653
1654         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
1655         if (repbody == NULL || reqbody == NULL)
1656                 RETURN(err_serious(-EFAULT));
1657
1658         /*
1659          * prepare @rdpg before calling lower layers and transfer itself. Here
1660          * reqbody->size contains offset of where to start to read and
1661          * reqbody->nlink contains number bytes to read.
1662          */
1663         rdpg->rp_hash = reqbody->mbo_size;
1664         if (rdpg->rp_hash != reqbody->mbo_size) {
1665                 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1666                        rdpg->rp_hash, reqbody->mbo_size);
1667                 RETURN(-EFAULT);
1668         }
1669
1670         rdpg->rp_attrs = reqbody->mbo_mode;
1671         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
1672                 rdpg->rp_attrs |= LUDA_64BITHASH;
1673         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
1674                                 exp_max_brw_size(tsi->tsi_exp));
1675         rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE - 1) >>
1676                           PAGE_CACHE_SHIFT;
1677         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1678         if (rdpg->rp_pages == NULL)
1679                 RETURN(-ENOMEM);
1680
1681         for (i = 0; i < rdpg->rp_npages; ++i) {
1682                 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1683                 if (rdpg->rp_pages[i] == NULL)
1684                         GOTO(free_rdpg, rc = -ENOMEM);
1685         }
1686
1687         /* call lower layers to fill allocated pages with directory data */
1688         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
1689         if (rc < 0)
1690                 GOTO(free_rdpg, rc);
1691
1692         /* send pages to client */
1693         rc = tgt_sendpage(tsi, rdpg, rc);
1694
1695         EXIT;
1696 free_rdpg:
1697
1698         for (i = 0; i < rdpg->rp_npages; i++)
1699                 if (rdpg->rp_pages[i] != NULL)
1700                         __free_page(rdpg->rp_pages[i]);
1701         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1702
1703         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1704                 RETURN(0);
1705
1706         return rc;
1707 }
1708
1709 static int mdt_reint_internal(struct mdt_thread_info *info,
1710                               struct mdt_lock_handle *lhc,
1711                               __u32 op)
1712 {
1713         struct req_capsule      *pill = info->mti_pill;
1714         struct mdt_body         *repbody;
1715         int                      rc = 0, rc2;
1716
1717         ENTRY;
1718
1719         rc = mdt_reint_unpack(info, op);
1720         if (rc != 0) {
1721                 CERROR("Can't unpack reint, rc %d\n", rc);
1722                 RETURN(err_serious(rc));
1723         }
1724
1725         /* for replay (no_create) lmm is not needed, client has it already */
1726         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1727                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1728                                      DEF_REP_MD_SIZE);
1729
1730         /* llog cookies are always 0, the field is kept for compatibility */
1731         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1732                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
1733
1734         rc = req_capsule_server_pack(pill);
1735         if (rc != 0) {
1736                 CERROR("Can't pack response, rc %d\n", rc);
1737                 RETURN(err_serious(rc));
1738         }
1739
1740         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1741                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1742                 LASSERT(repbody);
1743                 repbody->mbo_eadatasize = 0;
1744                 repbody->mbo_aclsize = 0;
1745         }
1746
1747         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1748
1749         /* for replay no cookkie / lmm need, because client have this already */
1750         if (info->mti_spec.no_create)
1751                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1752                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1753
1754         rc = mdt_init_ucred_reint(info);
1755         if (rc)
1756                 GOTO(out_shrink, rc);
1757
1758         rc = mdt_fix_attr_ucred(info, op);
1759         if (rc != 0)
1760                 GOTO(out_ucred, rc = err_serious(rc));
1761
1762         rc = mdt_check_resent(info, mdt_reconstruct, lhc);
1763         if (rc < 0) {
1764                 GOTO(out_ucred, rc);
1765         } else if (rc == 1) {
1766                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
1767                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1768                 GOTO(out_ucred, rc);
1769         }
1770         rc = mdt_reint_rec(info, lhc);
1771         EXIT;
1772 out_ucred:
1773         mdt_exit_ucred(info);
1774 out_shrink:
1775         mdt_client_compatibility(info);
1776         rc2 = mdt_fix_reply(info);
1777         if (rc == 0)
1778                 rc = rc2;
1779         return rc;
1780 }
1781
1782 static long mdt_reint_opcode(struct ptlrpc_request *req,
1783                              const struct req_format **fmt)
1784 {
1785         struct mdt_device       *mdt;
1786         struct mdt_rec_reint    *rec;
1787         long                     opc;
1788
1789         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
1790         if (rec != NULL) {
1791                 opc = rec->rr_opcode;
1792                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
1793                 if (opc < REINT_MAX && fmt[opc] != NULL)
1794                         req_capsule_extend(&req->rq_pill, fmt[opc]);
1795                 else {
1796                         mdt = mdt_exp2dev(req->rq_export);
1797                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
1798                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
1799                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
1800                         opc = err_serious(-EFAULT);
1801                 }
1802         } else {
1803                 opc = err_serious(-EFAULT);
1804         }
1805         return opc;
1806 }
1807
1808 static int mdt_reint(struct tgt_session_info *tsi)
1809 {
1810         long opc;
1811         int  rc;
1812         static const struct req_format *reint_fmts[REINT_MAX] = {
1813                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
1814                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
1815                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
1816                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
1817                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
1818                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
1819                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
1820                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
1821                 [REINT_MIGRATE]  = &RQF_MDS_REINT_RENAME
1822         };
1823
1824         ENTRY;
1825
1826         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
1827         if (opc >= 0) {
1828                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1829                 /*
1830                  * No lock possible here from client to pass it to reint code
1831                  * path.
1832                  */
1833                 rc = mdt_reint_internal(info, NULL, opc);
1834                 mdt_thread_info_fini(info);
1835         } else {
1836                 rc = opc;
1837         }
1838
1839         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1840         RETURN(rc);
1841 }
1842
1843 /* this should sync the whole device */
1844 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1845 {
1846         struct dt_device *dt = mdt->mdt_bottom;
1847         int rc;
1848         ENTRY;
1849
1850         rc = dt->dd_ops->dt_sync(env, dt);
1851         RETURN(rc);
1852 }
1853
1854 /* this should sync this object */
1855 static int mdt_object_sync(struct mdt_thread_info *info)
1856 {
1857         struct md_object *next;
1858         int rc;
1859         ENTRY;
1860
1861         if (!mdt_object_exists(info->mti_object)) {
1862                 CWARN("%s: non existing object "DFID": rc = %d\n",
1863                       mdt_obd_name(info->mti_mdt),
1864                       PFID(mdt_object_fid(info->mti_object)), -ESTALE);
1865                 RETURN(-ESTALE);
1866         }
1867         next = mdt_object_child(info->mti_object);
1868         rc = mo_object_sync(info->mti_env, next);
1869
1870         RETURN(rc);
1871 }
1872
1873 static int mdt_sync(struct tgt_session_info *tsi)
1874 {
1875         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1876         struct req_capsule      *pill = tsi->tsi_pill;
1877         struct mdt_body         *body;
1878         int                      rc;
1879
1880         ENTRY;
1881
1882         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1883                 RETURN(err_serious(-ENOMEM));
1884
1885         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
1886                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
1887         } else {
1888                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1889
1890                 /* sync an object */
1891                 rc = mdt_object_sync(info);
1892                 if (rc == 0) {
1893                         const struct lu_fid *fid;
1894                         struct lu_attr *la = &info->mti_attr.ma_attr;
1895
1896                         info->mti_attr.ma_need = MA_INODE;
1897                         info->mti_attr.ma_valid = 0;
1898                         rc = mdt_attr_get_complex(info, info->mti_object,
1899                                                   &info->mti_attr);
1900                         if (rc == 0) {
1901                                 body = req_capsule_server_get(pill,
1902                                                               &RMF_MDT_BODY);
1903                                 fid = mdt_object_fid(info->mti_object);
1904                                 mdt_pack_attr2body(info, body, la, fid);
1905                         }
1906                 }
1907                 mdt_thread_info_fini(info);
1908         }
1909         if (rc == 0)
1910                 mdt_counter_incr(req, LPROC_MDT_SYNC);
1911
1912         RETURN(rc);
1913 }
1914
1915 /*
1916  * Handle quota control requests to consult current usage/limit, but also
1917  * to configure quota enforcement
1918  */
1919 static int mdt_quotactl(struct tgt_session_info *tsi)
1920 {
1921         struct obd_export       *exp  = tsi->tsi_exp;
1922         struct req_capsule      *pill = tsi->tsi_pill;
1923         struct obd_quotactl     *oqctl, *repoqc;
1924         int                      id, rc;
1925         struct mdt_device       *mdt = mdt_exp2dev(exp);
1926         struct lu_device        *qmt = mdt->mdt_qmt_dev;
1927         struct lu_nodemap       *nodemap = exp->exp_target_data.ted_nodemap;
1928         ENTRY;
1929
1930         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1931         if (oqctl == NULL)
1932                 RETURN(err_serious(-EPROTO));
1933
1934         rc = req_capsule_server_pack(pill);
1935         if (rc)
1936                 RETURN(err_serious(rc));
1937
1938         switch (oqctl->qc_cmd) {
1939                 /* master quotactl */
1940         case Q_SETINFO:
1941         case Q_SETQUOTA:
1942                 if (!nodemap_can_setquota(nodemap))
1943                         RETURN(-EPERM);
1944         case Q_GETINFO:
1945         case Q_GETQUOTA:
1946                 if (qmt == NULL)
1947                         RETURN(-EOPNOTSUPP);
1948                 /* slave quotactl */
1949         case Q_GETOINFO:
1950         case Q_GETOQUOTA:
1951                 break;
1952         default:
1953                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
1954                 RETURN(-EFAULT);
1955         }
1956
1957         /* map uid/gid for remote client */
1958         id = oqctl->qc_id;
1959         if (exp_connect_rmtclient(exp)) {
1960                 struct lustre_idmap_table *idmap;
1961
1962                 idmap = exp->exp_mdt_data.med_idmap;
1963
1964                 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
1965                              oqctl->qc_cmd != Q_GETINFO))
1966                         RETURN(-EPERM);
1967
1968                 if (oqctl->qc_type == USRQUOTA)
1969                         id = lustre_idmap_lookup_uid(NULL, idmap, 0,
1970                                                      oqctl->qc_id);
1971                 else if (oqctl->qc_type == GRPQUOTA)
1972                         id = lustre_idmap_lookup_gid(NULL, idmap, 0,
1973                                                      oqctl->qc_id);
1974                 else
1975                         RETURN(-EINVAL);
1976
1977                 if (id == CFS_IDMAP_NOTFOUND) {
1978                         CDEBUG(D_QUOTA, "no mapping for id %u\n", oqctl->qc_id);
1979                         RETURN(-EACCES);
1980                 }
1981         }
1982
1983         if (oqctl->qc_type == USRQUOTA)
1984                 id = nodemap_map_id(nodemap, NODEMAP_UID,
1985                                     NODEMAP_CLIENT_TO_FS, id);
1986         else if (oqctl->qc_type == GRPQUOTA)
1987                 id = nodemap_map_id(nodemap, NODEMAP_UID,
1988                                     NODEMAP_CLIENT_TO_FS, id);
1989
1990         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
1991         if (repoqc == NULL)
1992                 RETURN(err_serious(-EFAULT));
1993
1994         if (oqctl->qc_id != id)
1995                 swap(oqctl->qc_id, id);
1996
1997         switch (oqctl->qc_cmd) {
1998
1999         case Q_GETINFO:
2000         case Q_SETINFO:
2001         case Q_SETQUOTA:
2002         case Q_GETQUOTA:
2003                 /* forward quotactl request to QMT */
2004                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2005                 break;
2006
2007         case Q_GETOINFO:
2008         case Q_GETOQUOTA:
2009                 /* slave quotactl */
2010                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2011                                    oqctl);
2012                 break;
2013
2014         default:
2015                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2016                 RETURN(-EFAULT);
2017         }
2018
2019         if (oqctl->qc_id != id)
2020                 swap(oqctl->qc_id, id);
2021
2022         *repoqc = *oqctl;
2023         RETURN(rc);
2024 }
2025
2026 /** clone llog ctxt from child (mdd)
2027  * This allows remote llog (replicator) access.
2028  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2029  * context was originally set up, or we can handle them directly.
2030  * I choose the latter, but that means I need any llog
2031  * contexts set up by child to be accessable by the mdt.  So we clone the
2032  * context into our context list here.
2033  */
2034 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2035                                int idx)
2036 {
2037         struct md_device  *next = mdt->mdt_child;
2038         struct llog_ctxt *ctxt;
2039         int rc;
2040
2041         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2042                 return 0;
2043
2044         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2045         if (rc || ctxt == NULL) {
2046                 return 0;
2047         }
2048
2049         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2050         if (rc)
2051                 CERROR("Can't set mdt ctxt %d\n", rc);
2052
2053         return rc;
2054 }
2055
2056 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2057                                  struct mdt_device *mdt, int idx)
2058 {
2059         struct llog_ctxt *ctxt;
2060
2061         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2062         if (ctxt == NULL)
2063                 return 0;
2064         /* Put once for the get we just did, and once for the clone */
2065         llog_ctxt_put(ctxt);
2066         llog_ctxt_put(ctxt);
2067         return 0;
2068 }
2069
2070 /*
2071  * sec context handlers
2072  */
2073 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2074 {
2075         int rc;
2076
2077         rc = mdt_handle_idmap(tsi);
2078         if (unlikely(rc)) {
2079                 struct ptlrpc_request   *req = tgt_ses_req(tsi);
2080                 __u32                    opc;
2081
2082                 opc = lustre_msg_get_opc(req->rq_reqmsg);
2083                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2084                         sptlrpc_svc_ctx_invalidate(req);
2085         }
2086
2087         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2088
2089         return rc;
2090 }
2091
2092 /*
2093  * quota request handlers
2094  */
2095 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2096 {
2097         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
2098         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2099         int                      rc;
2100         ENTRY;
2101
2102         if (qmt == NULL)
2103                 RETURN(err_serious(-EOPNOTSUPP));
2104
2105         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2106         RETURN(rc);
2107 }
2108
2109 struct mdt_object *mdt_object_new(const struct lu_env *env,
2110                                   struct mdt_device *d,
2111                                   const struct lu_fid *f)
2112 {
2113         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2114         struct lu_object *o;
2115         struct mdt_object *m;
2116         ENTRY;
2117
2118         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2119         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2120         if (unlikely(IS_ERR(o)))
2121                 m = (struct mdt_object *)o;
2122         else
2123                 m = mdt_obj(o);
2124         RETURN(m);
2125 }
2126
2127 struct mdt_object *mdt_object_find(const struct lu_env *env,
2128                                    struct mdt_device *d,
2129                                    const struct lu_fid *f)
2130 {
2131         struct lu_object *o;
2132         struct mdt_object *m;
2133         ENTRY;
2134
2135         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2136         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2137         if (unlikely(IS_ERR(o)))
2138                 m = (struct mdt_object *)o;
2139         else
2140                 m = mdt_obj(o);
2141
2142         RETURN(m);
2143 }
2144
2145 /**
2146  * Asyncronous commit for mdt device.
2147  *
2148  * Pass asynchonous commit call down the MDS stack.
2149  *
2150  * \param env environment
2151  * \param mdt the mdt device
2152  */
2153 static void mdt_device_commit_async(const struct lu_env *env,
2154                                     struct mdt_device *mdt)
2155 {
2156         struct dt_device *dt = mdt->mdt_bottom;
2157         int rc;
2158
2159         rc = dt->dd_ops->dt_commit_async(env, dt);
2160         if (unlikely(rc != 0))
2161                 CWARN("%s: async commit start failed: rc = %d\n",
2162                       mdt_obd_name(mdt), rc);
2163 }
2164
2165 /**
2166  * Mark the lock as "synchonous".
2167  *
2168  * Mark the lock to deffer transaction commit to the unlock time.
2169  *
2170  * \param lock the lock to mark as "synchonous"
2171  *
2172  * \see mdt_is_lock_sync
2173  * \see mdt_save_lock
2174  */
2175 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2176 {
2177         lock->l_ast_data = (void*)1;
2178 }
2179
2180 /**
2181  * Check whehter the lock "synchonous" or not.
2182  *
2183  * \param lock the lock to check
2184  * \retval 1 the lock is "synchonous"
2185  * \retval 0 the lock isn't "synchronous"
2186  *
2187  * \see mdt_set_lock_sync
2188  * \see mdt_save_lock
2189  */
2190 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2191 {
2192         return lock->l_ast_data != NULL;
2193 }
2194
2195 /**
2196  * Blocking AST for mdt locks.
2197  *
2198  * Starts transaction commit if in case of COS lock conflict or
2199  * deffers such a commit to the mdt_save_lock.
2200  *
2201  * \param lock the lock which blocks a request or cancelling lock
2202  * \param desc unused
2203  * \param data unused
2204  * \param flag indicates whether this cancelling or blocking callback
2205  * \retval 0
2206  * \see ldlm_blocking_ast_nocheck
2207  */
2208 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2209                      void *data, int flag)
2210 {
2211         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2212         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2213         int rc;
2214         ENTRY;
2215
2216         if (flag == LDLM_CB_CANCELING)
2217                 RETURN(0);
2218         lock_res_and_lock(lock);
2219         if (lock->l_blocking_ast != mdt_blocking_ast) {
2220                 unlock_res_and_lock(lock);
2221                 RETURN(0);
2222         }
2223         if (mdt_cos_is_enabled(mdt) &&
2224             lock->l_req_mode & (LCK_PW | LCK_EX) &&
2225             lock->l_blocking_lock != NULL &&
2226             lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
2227                 mdt_set_lock_sync(lock);
2228         }
2229         rc = ldlm_blocking_ast_nocheck(lock);
2230
2231         /* There is no lock conflict if l_blocking_lock == NULL,
2232          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2233          * when the last reference to a local lock was released */
2234         if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2235                 struct lu_env env;
2236
2237                 rc = lu_env_init(&env, LCT_LOCAL);
2238                 if (unlikely(rc != 0))
2239                         CWARN("%s: lu_env initialization failed, cannot "
2240                               "start asynchronous commit: rc = %d\n",
2241                               obd->obd_name, rc);
2242                 else
2243                         mdt_device_commit_async(&env, mdt);
2244                 lu_env_fini(&env);
2245         }
2246         RETURN(rc);
2247 }
2248
2249 /* Used for cross-MDT lock */
2250 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2251                             void *data, int flag)
2252 {
2253         struct lustre_handle lockh;
2254         int               rc;
2255
2256         switch (flag) {
2257         case LDLM_CB_BLOCKING:
2258                 ldlm_lock2handle(lock, &lockh);
2259                 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
2260                 if (rc < 0) {
2261                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2262                         RETURN(rc);
2263                 }
2264                 break;
2265         case LDLM_CB_CANCELING:
2266                 LDLM_DEBUG(lock, "Revoke remote lock\n");
2267                 break;
2268         default:
2269                 LBUG();
2270         }
2271         RETURN(0);
2272 }
2273
2274 int mdt_check_resent_lock(struct mdt_thread_info *info,
2275                           struct mdt_object *mo,
2276                           struct mdt_lock_handle *lhc)
2277 {
2278         /* the lock might already be gotten in ldlm_handle_enqueue() */
2279         if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
2280                 struct ptlrpc_request *req = mdt_info_req(info);
2281                 struct ldlm_lock      *lock;
2282
2283                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2284                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
2285                 if (lock == NULL) {
2286                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
2287                          * a resend case, however, it could be already destroyed
2288                          * due to client eviction or a raced cancel RPC. */
2289                         LDLM_DEBUG_NOLOCK("Invalid lock handle "LPX64"\n",
2290                                           lhc->mlh_reg_lh.cookie);
2291                         RETURN(-ESTALE);
2292                 }
2293
2294                 if (!fid_res_name_eq(mdt_object_fid(mo),
2295                                      &lock->l_resource->lr_name)) {
2296                         CWARN("%s: Although resent, but still not "
2297                               "get child lock:"DFID"\n",
2298                               info->mti_exp->exp_obd->obd_name,
2299                               PFID(mdt_object_fid(mo)));
2300                         LDLM_LOCK_PUT(lock);
2301                         RETURN(-EPROTO);
2302                 }
2303                 LDLM_LOCK_PUT(lock);
2304                 return 0;
2305         }
2306         return 1;
2307 }
2308
2309 int mdt_remote_object_lock(struct mdt_thread_info *mti,
2310                            struct mdt_object *o, const struct lu_fid *fid,
2311                            struct lustre_handle *lh, ldlm_mode_t mode,
2312                            __u64 ibits, bool nonblock)
2313 {
2314         struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
2315         ldlm_policy_data_t *policy = &mti->mti_policy;
2316         struct ldlm_res_id *res_id = &mti->mti_res_id;
2317         int rc = 0;
2318         ENTRY;
2319
2320         LASSERT(mdt_object_remote(o));
2321
2322         fid_build_reg_res_name(fid, res_id);
2323
2324         memset(einfo, 0, sizeof(*einfo));
2325         einfo->ei_type = LDLM_IBITS;
2326         einfo->ei_mode = mode;
2327         einfo->ei_cb_bl = mdt_remote_blocking_ast;
2328         einfo->ei_cb_cp = ldlm_completion_ast;
2329         einfo->ei_enq_slave = 0;
2330         einfo->ei_res_id = res_id;
2331         if (nonblock)
2332                 einfo->ei_nonblock = 1;
2333
2334         memset(policy, 0, sizeof(*policy));
2335         policy->l_inodebits.bits = ibits;
2336
2337         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2338                             policy);
2339         RETURN(rc);
2340 }
2341
2342 static int mdt_object_local_lock(struct mdt_thread_info *info,
2343                                  struct mdt_object *o,
2344                                  struct mdt_lock_handle *lh, __u64 ibits,
2345                                  bool nonblock)
2346 {
2347         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2348         ldlm_policy_data_t *policy = &info->mti_policy;
2349         struct ldlm_res_id *res_id = &info->mti_res_id;
2350         __u64 dlmflags;
2351         int rc;
2352         ENTRY;
2353
2354         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2355         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2356         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2357         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2358
2359         /* Only enqueue LOOKUP lock for remote object */
2360         if (mdt_object_remote(o))
2361                 LASSERT(ibits == MDS_INODELOCK_LOOKUP);
2362
2363         if (lh->mlh_type == MDT_PDO_LOCK) {
2364                 /* check for exists after object is locked */
2365                 if (mdt_object_exists(o) == 0) {
2366                         /* Non-existent object shouldn't have PDO lock */
2367                         RETURN(-ESTALE);
2368                 } else {
2369                         /* Non-dir object shouldn't have PDO lock */
2370                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2371                                 RETURN(-ENOTDIR);
2372                 }
2373         }
2374
2375         memset(policy, 0, sizeof(*policy));
2376         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2377
2378         dlmflags = LDLM_FL_ATOMIC_CB;
2379         if (nonblock)
2380                 dlmflags |= LDLM_FL_BLOCK_NOWAIT;
2381
2382         /*
2383          * Take PDO lock on whole directory and build correct @res_id for lock
2384          * on part of directory.
2385          */
2386         if (lh->mlh_pdo_hash != 0) {
2387                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2388                 mdt_lock_pdo_mode(info, o, lh);
2389                 if (lh->mlh_pdo_mode != LCK_NL) {
2390                         /*
2391                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2392                          * is never going to be sent to client and we do not
2393                          * want it slowed down due to possible cancels.
2394                          */
2395                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2396                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2397                                           policy, res_id, dlmflags,
2398                                           info->mti_exp == NULL ? NULL :
2399                                           &info->mti_exp->exp_handle.h_cookie);
2400                         if (unlikely(rc != 0))
2401                                 GOTO(out_unlock, rc);
2402                 }
2403
2404                 /*
2405                  * Finish res_id initializing by name hash marking part of
2406                  * directory which is taking modification.
2407                  */
2408                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2409         }
2410
2411         policy->l_inodebits.bits = ibits;
2412
2413         /*
2414          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2415          * going to be sent to client. If it is - mdt_intent_policy() path will
2416          * fix it up and turn FL_LOCAL flag off.
2417          */
2418         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2419                           res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2420                           info->mti_exp == NULL ? NULL :
2421                           &info->mti_exp->exp_handle.h_cookie);
2422 out_unlock:
2423         if (rc != 0)
2424                 mdt_object_unlock(info, o, lh, 1);
2425         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2426                    lh->mlh_pdo_hash != 0 &&
2427                    (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
2428                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2429
2430         RETURN(rc);
2431 }
2432
2433 static int
2434 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
2435                          struct mdt_lock_handle *lh, __u64 ibits,
2436                          bool nonblock)
2437 {
2438         struct mdt_lock_handle *local_lh = NULL;
2439         int rc;
2440         ENTRY;
2441
2442         if (!mdt_object_remote(o))
2443                 return mdt_object_local_lock(info, o, lh, ibits, nonblock);
2444
2445         /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
2446         ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
2447                    MDS_INODELOCK_XATTR);
2448
2449         /* Only enqueue LOOKUP lock for remote object */
2450         if (ibits & MDS_INODELOCK_LOOKUP) {
2451                 rc = mdt_object_local_lock(info, o, lh,
2452                                            MDS_INODELOCK_LOOKUP,
2453                                            nonblock);
2454                 if (rc != ELDLM_OK)
2455                         RETURN(rc);
2456
2457                 local_lh = lh;
2458         }
2459
2460         if (ibits & MDS_INODELOCK_UPDATE) {
2461                 /* Sigh, PDO needs to enqueue 2 locks right now, but
2462                  * enqueue RPC can only request 1 lock, to avoid extra
2463                  * RPC, so it will instead enqueue EX lock for remote
2464                  * object anyway XXX*/
2465                 if (lh->mlh_type == MDT_PDO_LOCK &&
2466                     lh->mlh_pdo_hash != 0) {
2467                         CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
2468                                "EX lock.\n", mdt_obd_name(info->mti_mdt),
2469                                PFID(mdt_object_fid(o)));
2470                         lh->mlh_pdo_hash = 0;
2471                         lh->mlh_rreg_mode = LCK_EX;
2472                         lh->mlh_type = MDT_REG_LOCK;
2473                 }
2474                 rc = mdt_remote_object_lock(info, o, mdt_object_fid(o),
2475                                             &lh->mlh_rreg_lh,
2476                                             lh->mlh_rreg_mode,
2477                                             MDS_INODELOCK_UPDATE, nonblock);
2478                 if (rc != ELDLM_OK) {
2479                         if (local_lh != NULL)
2480                                 mdt_object_unlock(info, o, local_lh, rc);
2481                         RETURN(rc);
2482                 }
2483         }
2484
2485         RETURN(0);
2486 }
2487
2488 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2489                     struct mdt_lock_handle *lh, __u64 ibits)
2490 {
2491         return mdt_object_lock_internal(info, o, lh, ibits, false);
2492 }
2493
2494 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
2495                         struct mdt_lock_handle *lh, __u64 ibits)
2496 {
2497         struct mdt_lock_handle tmp = *lh;
2498         int rc;
2499
2500         rc = mdt_object_lock_internal(info, o, &tmp, ibits, true);
2501         if (rc == 0)
2502                 *lh = tmp;
2503
2504         return rc == 0;
2505 }
2506
2507 /**
2508  * Save a lock within request object.
2509  *
2510  * Keep the lock referenced until whether client ACK or transaction
2511  * commit happens or release the lock immediately depending on input
2512  * parameters. If COS is ON, a write lock is converted to COS lock
2513  * before saving.
2514  *
2515  * \param info thead info object
2516  * \param h lock handle
2517  * \param mode lock mode
2518  * \param decref force immediate lock releasing
2519  */
2520 static
2521 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2522                    ldlm_mode_t mode, int decref)
2523 {
2524         ENTRY;
2525
2526         if (lustre_handle_is_used(h)) {
2527                 if (decref || !info->mti_has_trans ||
2528                     !(mode & (LCK_PW | LCK_EX))){
2529                         mdt_fid_unlock(h, mode);
2530                 } else {
2531                         struct mdt_device *mdt = info->mti_mdt;
2532                         struct ldlm_lock *lock = ldlm_handle2lock(h);
2533                         struct ptlrpc_request *req = mdt_info_req(info);
2534                         int no_ack = 0;
2535
2536                         LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2537                                  h->cookie);
2538                         /* there is no request if mdt_object_unlock() is called
2539                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
2540                         if (likely(req != NULL)) {
2541                                 CDEBUG(D_HA, "request = %p reply state = %p"
2542                                        " transno = "LPD64"\n", req,
2543                                        req->rq_reply_state, req->rq_transno);
2544                                 if (mdt_cos_is_enabled(mdt)) {
2545                                         no_ack = 1;
2546                                         ldlm_lock_downgrade(lock, LCK_COS);
2547                                         mode = LCK_COS;
2548                                 }
2549                                 ptlrpc_save_lock(req, h, mode, no_ack);
2550                         } else {
2551                                 ldlm_lock_decref(h, mode);
2552                         }
2553                         if (mdt_is_lock_sync(lock)) {
2554                                 CDEBUG(D_HA, "found sync-lock,"
2555                                        " async commit started\n");
2556                                 mdt_device_commit_async(info->mti_env,
2557                                                         mdt);
2558                         }
2559                         LDLM_LOCK_PUT(lock);
2560                 }
2561                 h->cookie = 0ull;
2562         }
2563
2564         EXIT;
2565 }
2566
2567 /**
2568  * Unlock mdt object.
2569  *
2570  * Immeditely release the regular lock and the PDO lock or save the
2571  * lock in request and keep them referenced until client ACK or
2572  * transaction commit.
2573  *
2574  * \param info thread info object
2575  * \param o mdt object
2576  * \param lh mdt lock handle referencing regular and PDO locks
2577  * \param decref force immediate lock releasing
2578  */
2579 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2580                        struct mdt_lock_handle *lh, int decref)
2581 {
2582         ENTRY;
2583
2584         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2585         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2586
2587         if (lustre_handle_is_used(&lh->mlh_rreg_lh))
2588                 ldlm_lock_decref(&lh->mlh_rreg_lh, lh->mlh_rreg_mode);
2589
2590         EXIT;
2591 }
2592
2593 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2594                                         const struct lu_fid *f,
2595                                         struct mdt_lock_handle *lh,
2596                                         __u64 ibits)
2597 {
2598         struct mdt_object *o;
2599
2600         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2601         if (!IS_ERR(o)) {
2602                 int rc;
2603
2604                 rc = mdt_object_lock(info, o, lh, ibits);
2605                 if (rc != 0) {
2606                         mdt_object_put(info->mti_env, o);
2607                         o = ERR_PTR(rc);
2608                 }
2609         }
2610         return o;
2611 }
2612
2613 void mdt_object_unlock_put(struct mdt_thread_info * info,
2614                            struct mdt_object * o,
2615                            struct mdt_lock_handle *lh,
2616                            int decref)
2617 {
2618         mdt_object_unlock(info, o, lh, decref);
2619         mdt_object_put(info->mti_env, o);
2620 }
2621
2622 /*
2623  * Generic code handling requests that have struct mdt_body passed in:
2624  *
2625  *  - extract mdt_body from request and save it in @info, if present;
2626  *
2627  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
2628  *  @info;
2629  *
2630  *  - if HABEO_CORPUS flag is set for this request type check whether object
2631  *  actually exists on storage (lu_object_exists()).
2632  *
2633  */
2634 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2635 {
2636         const struct mdt_body    *body;
2637         struct mdt_object        *obj;
2638         const struct lu_env      *env;
2639         struct req_capsule       *pill;
2640         int                       rc;
2641         ENTRY;
2642
2643         env = info->mti_env;
2644         pill = info->mti_pill;
2645
2646         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2647         if (body == NULL)
2648                 RETURN(-EFAULT);
2649
2650         if (!(body->mbo_valid & OBD_MD_FLID))
2651                 RETURN(0);
2652
2653         if (!fid_is_sane(&body->mbo_fid1)) {
2654                 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
2655                 RETURN(-EINVAL);
2656         }
2657
2658         obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
2659         if (!IS_ERR(obj)) {
2660                 if ((flags & HABEO_CORPUS) && !mdt_object_exists(obj)) {
2661                         mdt_object_put(env, obj);
2662                         rc = -ENOENT;
2663                 } else {
2664                         info->mti_object = obj;
2665                         rc = 0;
2666                 }
2667         } else
2668                 rc = PTR_ERR(obj);
2669
2670         RETURN(rc);
2671 }
2672
2673 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2674 {
2675         struct req_capsule *pill = info->mti_pill;
2676         int rc;
2677         ENTRY;
2678
2679         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2680                 rc = mdt_body_unpack(info, flags);
2681         else
2682                 rc = 0;
2683
2684         if (rc == 0 && (flags & HABEO_REFERO)) {
2685                 /* Pack reply. */
2686                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2687                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2688                                              DEF_REP_MD_SIZE);
2689                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2690                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
2691                                              RCL_SERVER, 0);
2692
2693                 rc = req_capsule_server_pack(pill);
2694         }
2695         RETURN(rc);
2696 }
2697
2698 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2699 {
2700         lh->mlh_type = MDT_NUL_LOCK;
2701         lh->mlh_reg_lh.cookie = 0ull;
2702         lh->mlh_reg_mode = LCK_MINMODE;
2703         lh->mlh_pdo_lh.cookie = 0ull;
2704         lh->mlh_pdo_mode = LCK_MINMODE;
2705         lh->mlh_rreg_lh.cookie = 0ull;
2706         lh->mlh_rreg_mode = LCK_MINMODE;
2707 }
2708
2709 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2710 {
2711         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2712         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2713 }
2714
2715 /*
2716  * Initialize fields of struct mdt_thread_info. Other fields are left in
2717  * uninitialized state, because it's too expensive to zero out whole
2718  * mdt_thread_info (> 1K) on each request arrival.
2719  */
2720 void mdt_thread_info_init(struct ptlrpc_request *req,
2721                           struct mdt_thread_info *info)
2722 {
2723         int i;
2724
2725         info->mti_pill = &req->rq_pill;
2726
2727         /* lock handle */
2728         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2729                 mdt_lock_handle_init(&info->mti_lh[i]);
2730
2731         /* mdt device: it can be NULL while CONNECT */
2732         if (req->rq_export) {
2733                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2734                 info->mti_exp = req->rq_export;
2735         } else
2736                 info->mti_mdt = NULL;
2737         info->mti_env = req->rq_svc_thread->t_env;
2738         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2739
2740         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2741         info->mti_big_buf = LU_BUF_NULL;
2742         info->mti_body = NULL;
2743         info->mti_object = NULL;
2744         info->mti_dlm_req = NULL;
2745         info->mti_has_trans = 0;
2746         info->mti_cross_ref = 0;
2747         info->mti_opdata = 0;
2748         info->mti_big_lmm_used = 0;
2749
2750         info->mti_spec.no_create = 0;
2751         info->mti_spec.sp_rm_entry = 0;
2752         info->mti_spec.sp_permitted = 0;
2753         info->mti_spec.sp_migrate_close = 0;
2754
2755         info->mti_spec.u.sp_ea.eadata = NULL;
2756         info->mti_spec.u.sp_ea.eadatalen = 0;
2757 }
2758
2759 void mdt_thread_info_fini(struct mdt_thread_info *info)
2760 {
2761         int i;
2762
2763         if (info->mti_object != NULL) {
2764                 mdt_object_put(info->mti_env, info->mti_object);
2765                 info->mti_object = NULL;
2766         }
2767
2768         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2769                 mdt_lock_handle_fini(&info->mti_lh[i]);
2770         info->mti_env = NULL;
2771         info->mti_pill = NULL;
2772         info->mti_exp = NULL;
2773
2774         if (unlikely(info->mti_big_buf.lb_buf != NULL))
2775                 lu_buf_free(&info->mti_big_buf);
2776 }
2777
2778 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
2779 {
2780         struct mdt_thread_info  *mti;
2781
2782         mti = mdt_th_info(tsi->tsi_env);
2783         LASSERT(mti != NULL);
2784
2785         mdt_thread_info_init(tgt_ses_req(tsi), mti);
2786         if (tsi->tsi_corpus != NULL) {
2787                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
2788                 lu_object_get(tsi->tsi_corpus);
2789         }
2790         mti->mti_body = tsi->tsi_mdt_body;
2791         mti->mti_dlm_req = tsi->tsi_dlm_req;
2792
2793         return mti;
2794 }
2795
2796 static int mdt_tgt_connect(struct tgt_session_info *tsi)
2797 {
2798         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2799         int                      rc;
2800
2801         ENTRY;
2802
2803         if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
2804             cfs_fail_val ==
2805             tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id) {
2806                 set_current_state(TASK_UNINTERRUPTIBLE);
2807                 schedule_timeout(msecs_to_jiffies(3 * MSEC_PER_SEC));
2808         }
2809
2810         rc = tgt_connect(tsi);
2811         if (rc != 0)
2812                 RETURN(rc);
2813
2814         rc = mdt_init_idmap(tsi);
2815         if (rc != 0)
2816                 GOTO(err, rc);
2817         RETURN(0);
2818 err:
2819         obd_disconnect(class_export_get(req->rq_export));
2820         return rc;
2821 }
2822
2823 enum mdt_it_code {
2824         MDT_IT_OPEN,
2825         MDT_IT_OCREAT,
2826         MDT_IT_CREATE,
2827         MDT_IT_GETATTR,
2828         MDT_IT_READDIR,
2829         MDT_IT_LOOKUP,
2830         MDT_IT_UNLINK,
2831         MDT_IT_TRUNC,
2832         MDT_IT_GETXATTR,
2833         MDT_IT_LAYOUT,
2834         MDT_IT_QUOTA,
2835         MDT_IT_NR
2836 };
2837
2838 static int mdt_intent_getattr(enum mdt_it_code opcode,
2839                               struct mdt_thread_info *info,
2840                               struct ldlm_lock **,
2841                               __u64);
2842
2843 static int mdt_intent_getxattr(enum mdt_it_code opcode,
2844                                 struct mdt_thread_info *info,
2845                                 struct ldlm_lock **lockp,
2846                                 __u64 flags);
2847
2848 static int mdt_intent_layout(enum mdt_it_code opcode,
2849                              struct mdt_thread_info *info,
2850                              struct ldlm_lock **,
2851                              __u64);
2852 static int mdt_intent_reint(enum mdt_it_code opcode,
2853                             struct mdt_thread_info *info,
2854                             struct ldlm_lock **,
2855                             __u64);
2856
2857 static struct mdt_it_flavor {
2858         const struct req_format *it_fmt;
2859         __u32                    it_flags;
2860         int                    (*it_act)(enum mdt_it_code ,
2861                                          struct mdt_thread_info *,
2862                                          struct ldlm_lock **,
2863                                          __u64);
2864         long                     it_reint;
2865 } mdt_it_flavor[] = {
2866         [MDT_IT_OPEN]     = {
2867                 .it_fmt   = &RQF_LDLM_INTENT,
2868                 /*.it_flags = HABEO_REFERO,*/
2869                 .it_flags = 0,
2870                 .it_act   = mdt_intent_reint,
2871                 .it_reint = REINT_OPEN
2872         },
2873         [MDT_IT_OCREAT]   = {
2874                 .it_fmt   = &RQF_LDLM_INTENT,
2875                 /*
2876                  * OCREAT is not a MUTABOR request as if the file
2877                  * already exists.
2878                  * We do the extra check of OBD_CONNECT_RDONLY in
2879                  * mdt_reint_open() when we really need to create
2880                  * the object.
2881                  */
2882                 .it_flags = 0,
2883                 .it_act   = mdt_intent_reint,
2884                 .it_reint = REINT_OPEN
2885         },
2886         [MDT_IT_CREATE]   = {
2887                 .it_fmt   = &RQF_LDLM_INTENT,
2888                 .it_flags = MUTABOR,
2889                 .it_act   = mdt_intent_reint,
2890                 .it_reint = REINT_CREATE
2891         },
2892         [MDT_IT_GETATTR]  = {
2893                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2894                 .it_flags = HABEO_REFERO,
2895                 .it_act   = mdt_intent_getattr
2896         },
2897         [MDT_IT_READDIR]  = {
2898                 .it_fmt   = NULL,
2899                 .it_flags = 0,
2900                 .it_act   = NULL
2901         },
2902         [MDT_IT_LOOKUP]   = {
2903                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2904                 .it_flags = HABEO_REFERO,
2905                 .it_act   = mdt_intent_getattr
2906         },
2907         [MDT_IT_UNLINK]   = {
2908                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
2909                 .it_flags = MUTABOR,
2910                 .it_act   = NULL,
2911                 .it_reint = REINT_UNLINK
2912         },
2913         [MDT_IT_TRUNC]    = {
2914                 .it_fmt   = NULL,
2915                 .it_flags = MUTABOR,
2916                 .it_act   = NULL
2917         },
2918         [MDT_IT_GETXATTR] = {
2919                 .it_fmt   = &RQF_LDLM_INTENT_GETXATTR,
2920                 .it_flags = HABEO_CORPUS,
2921                 .it_act   = mdt_intent_getxattr
2922         },
2923         [MDT_IT_LAYOUT] = {
2924                 .it_fmt   = &RQF_LDLM_INTENT_LAYOUT,
2925                 .it_flags = 0,
2926                 .it_act   = mdt_intent_layout
2927         }
2928 };
2929
2930 static int
2931 mdt_intent_lock_replace(struct mdt_thread_info *info,
2932                         struct ldlm_lock **lockp,
2933                         struct mdt_lock_handle *lh,
2934                         __u64 flags)
2935 {
2936         struct ptlrpc_request  *req = mdt_info_req(info);
2937         struct ldlm_lock       *lock = *lockp;
2938         struct ldlm_lock       *new_lock;
2939
2940         /* If possible resent found a lock, @lh is set to its handle */
2941         new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
2942
2943         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
2944                 lh->mlh_reg_lh.cookie = 0;
2945                 RETURN(0);
2946         }
2947
2948         LASSERTF(new_lock != NULL,
2949                  "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
2950
2951         /*
2952          * If we've already given this lock to a client once, then we should
2953          * have no readers or writers.  Otherwise, we should have one reader
2954          * _or_ writer ref (which will be zeroed below) before returning the
2955          * lock to a client.
2956          */
2957         if (new_lock->l_export == req->rq_export) {
2958                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
2959         } else {
2960                 LASSERT(new_lock->l_export == NULL);
2961                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
2962         }
2963
2964         *lockp = new_lock;
2965
2966         if (new_lock->l_export == req->rq_export) {
2967                 /*
2968                  * Already gave this to the client, which means that we
2969                  * reconstructed a reply.
2970                  */
2971                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
2972                         MSG_RESENT);
2973
2974                 LDLM_LOCK_RELEASE(new_lock);
2975                 lh->mlh_reg_lh.cookie = 0;
2976                 RETURN(ELDLM_LOCK_REPLACED);
2977         }
2978
2979         /*
2980          * Fixup the lock to be given to the client.
2981          */
2982         lock_res_and_lock(new_lock);
2983         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
2984          * possible blocking AST. */
2985         while (new_lock->l_readers > 0) {
2986                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
2987                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
2988                 new_lock->l_readers--;
2989         }
2990         while (new_lock->l_writers > 0) {
2991                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
2992                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
2993                 new_lock->l_writers--;
2994         }
2995
2996         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
2997         new_lock->l_blocking_ast = lock->l_blocking_ast;
2998         new_lock->l_completion_ast = lock->l_completion_ast;
2999         new_lock->l_remote_handle = lock->l_remote_handle;
3000         new_lock->l_flags &= ~LDLM_FL_LOCAL;
3001
3002         unlock_res_and_lock(new_lock);
3003
3004         cfs_hash_add(new_lock->l_export->exp_lock_hash,
3005                      &new_lock->l_remote_handle,
3006                      &new_lock->l_exp_hash);
3007
3008         LDLM_LOCK_RELEASE(new_lock);
3009         lh->mlh_reg_lh.cookie = 0;
3010
3011         RETURN(ELDLM_LOCK_REPLACED);
3012 }
3013
3014 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3015                                     struct ldlm_lock *new_lock,
3016                                     struct mdt_lock_handle *lh,
3017                                     __u64 flags)
3018 {
3019         struct ptlrpc_request  *req = mdt_info_req(info);
3020         struct ldlm_request    *dlmreq;
3021
3022         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3023                 return;
3024
3025         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3026
3027         /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
3028          * lock was found by ldlm_handle_enqueue(); if so @lh must be
3029          * initialized. */
3030         if (flags & LDLM_FL_RESENT) {
3031                 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
3032                 lh->mlh_reg_mode = new_lock->l_granted_mode;
3033
3034                 LDLM_DEBUG(new_lock, "Restoring lock cookie");
3035                 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie "LPX64,
3036                           lh->mlh_reg_lh.cookie);
3037                 return;
3038         }
3039
3040         /*
3041          * If the xid matches, then we know this is a resent request, and allow
3042          * it. (It's probably an OPEN, for which we don't send a lock.
3043          */
3044         if (req_can_reconstruct(req, NULL))
3045                 return;
3046
3047         /*
3048          * This remote handle isn't enqueued, so we never received or processed
3049          * this request.  Clear MSG_RESENT, because it can be handled like any
3050          * normal request now.
3051          */
3052         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3053
3054         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
3055                   dlmreq->lock_handle[0].cookie);
3056 }
3057
3058 static int mdt_intent_getxattr(enum mdt_it_code opcode,
3059                                 struct mdt_thread_info *info,
3060                                 struct ldlm_lock **lockp,
3061                                 __u64 flags)
3062 {
3063         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3064         struct ldlm_reply      *ldlm_rep = NULL;
3065         int rc, grc;
3066
3067         /*
3068          * Initialize lhc->mlh_reg_lh either from a previously granted lock
3069          * (for the resend case) or a new lock. Below we will use it to
3070          * replace the original lock.
3071          */
3072         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3073         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3074                 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3075                 rc = mdt_object_lock(info, info->mti_object, lhc,
3076                                      MDS_INODELOCK_XATTR);
3077                 if (rc)
3078                         return rc;
3079         }
3080
3081         grc = mdt_getxattr(info);
3082
3083         rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3084
3085         if (mdt_info_req(info)->rq_repmsg != NULL)
3086                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3087         if (ldlm_rep == NULL)
3088                 RETURN(err_serious(-EFAULT));
3089
3090         ldlm_rep->lock_policy_res2 = grc;
3091
3092         return rc;
3093 }
3094
3095 static int mdt_intent_getattr(enum mdt_it_code opcode,
3096                               struct mdt_thread_info *info,
3097                               struct ldlm_lock **lockp,
3098                               __u64 flags)
3099 {
3100         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3101         __u64                   child_bits;
3102         struct ldlm_reply      *ldlm_rep;
3103         struct mdt_body        *reqbody;
3104         struct mdt_body        *repbody;
3105         int                     rc, rc2;
3106         ENTRY;
3107
3108         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3109         LASSERT(reqbody);
3110
3111         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3112         LASSERT(repbody);
3113
3114         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
3115         repbody->mbo_eadatasize = 0;
3116         repbody->mbo_aclsize = 0;
3117
3118         switch (opcode) {
3119         case MDT_IT_LOOKUP:
3120                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3121                 break;
3122         case MDT_IT_GETATTR:
3123                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3124                              MDS_INODELOCK_PERM;
3125                 break;
3126         default:
3127                 CERROR("Unsupported intent (%d)\n", opcode);
3128                 GOTO(out_shrink, rc = -EINVAL);
3129         }
3130
3131         rc = mdt_init_ucred_intent_getattr(info, reqbody);
3132         if (rc)
3133                 GOTO(out_shrink, rc);
3134
3135         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3136         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3137
3138         /* Get lock from request for possible resent case. */
3139         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3140
3141         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3142         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3143
3144         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3145                 ldlm_rep->lock_policy_res2 = 0;
3146         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3147             ldlm_rep->lock_policy_res2) {
3148                 lhc->mlh_reg_lh.cookie = 0ull;
3149                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3150         }
3151
3152         rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3153         EXIT;
3154 out_ucred:
3155         mdt_exit_ucred(info);
3156 out_shrink:
3157         mdt_client_compatibility(info);
3158         rc2 = mdt_fix_reply(info);
3159         if (rc == 0)
3160                 rc = rc2;
3161         return rc;
3162 }
3163
3164 static int mdt_intent_layout(enum mdt_it_code opcode,
3165                              struct mdt_thread_info *info,
3166                              struct ldlm_lock **lockp,
3167                              __u64 flags)
3168 {
3169         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_LAYOUT];
3170         struct layout_intent *layout;
3171         struct lu_fid *fid;
3172         struct mdt_object *obj = NULL;
3173         int rc = 0;
3174         ENTRY;
3175
3176         if (opcode != MDT_IT_LAYOUT) {
3177                 CERROR("%s: Unknown intent (%d)\n", mdt_obd_name(info->mti_mdt),
3178                         opcode);
3179                 RETURN(-EINVAL);
3180         }
3181
3182         fid = &info->mti_tmp_fid2;
3183         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3184
3185         /* Get lock from request for possible resent case. */
3186         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3187
3188         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3189         if (IS_ERR(obj))
3190                 RETURN(PTR_ERR(obj));
3191
3192         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3193                 /* get the length of lsm */
3194                 rc = mdt_attr_get_eabuf_size(info, obj);
3195                 if (rc < 0) {
3196                         mdt_object_put(info->mti_env, obj);
3197                         RETURN(rc);
3198                 }
3199
3200                 if (rc > info->mti_mdt->mdt_max_mdsize)
3201                         info->mti_mdt->mdt_max_mdsize = rc;
3202         }
3203
3204         mdt_object_put(info->mti_env, obj);
3205
3206         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3207         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER, rc);
3208         rc = req_capsule_server_pack(info->mti_pill);
3209         if (rc != 0)
3210                 RETURN(-EINVAL);
3211
3212         if (lustre_handle_is_used(&lhc->mlh_reg_lh))
3213                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3214
3215         layout = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3216         LASSERT(layout != NULL);
3217         if (layout->li_opc == LAYOUT_INTENT_ACCESS)
3218                 /* return to normal/resent ldlm handling */
3219                 RETURN(rc);
3220
3221         CERROR("%s: Unsupported layout intent (%d)\n",
3222                 mdt_obd_name(info->mti_mdt), layout->li_opc);
3223         RETURN(-EINVAL);
3224 }
3225
3226 static int mdt_intent_reint(enum mdt_it_code opcode,
3227                             struct mdt_thread_info *info,
3228                             struct ldlm_lock **lockp,
3229                             __u64 flags)
3230 {
3231         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3232         struct ldlm_reply      *rep = NULL;
3233         long                    opc;
3234         int                     rc;
3235
3236         static const struct req_format *intent_fmts[REINT_MAX] = {
3237                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3238                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3239         };
3240
3241         ENTRY;
3242
3243         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
3244         if (opc < 0)
3245                 RETURN(opc);
3246
3247         if (mdt_it_flavor[opcode].it_reint != opc) {
3248                 CERROR("Reint code %ld doesn't match intent: %d\n",
3249                        opc, opcode);
3250                 RETURN(err_serious(-EPROTO));
3251         }
3252
3253         /* Get lock from request for possible resent case. */
3254         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3255
3256         rc = mdt_reint_internal(info, lhc, opc);
3257
3258         /* Check whether the reply has been packed successfully. */
3259         if (mdt_info_req(info)->rq_repmsg != NULL)
3260                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3261         if (rep == NULL)
3262                 RETURN(err_serious(-EFAULT));
3263
3264         /* MDC expects this in any case */
3265         if (rc != 0)
3266                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3267
3268         /* the open lock or the lock for cross-ref object should be
3269          * returned to the client */
3270         if (lustre_handle_is_used(&lhc->mlh_reg_lh) &&
3271             (rc == 0 || rc == -MDT_EREMOTE_OPEN)) {
3272                 rep->lock_policy_res2 = 0;
3273                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3274                 RETURN(rc);
3275         }
3276
3277         rep->lock_policy_res2 = clear_serious(rc);
3278
3279         if (rep->lock_policy_res2 == -ENOENT &&
3280             mdt_get_disposition(rep, DISP_LOOKUP_NEG))
3281                 rep->lock_policy_res2 = 0;
3282
3283         lhc->mlh_reg_lh.cookie = 0ull;
3284         if (rc == -ENOTCONN || rc == -ENODEV ||
3285             rc == -EOVERFLOW) { /**< if VBR failure then return error */
3286                 /*
3287                  * If it is the disconnect error (ENODEV & ENOCONN), the error
3288                  * will be returned by rq_status, and client at ptlrpc layer
3289                  * will detect this, then disconnect, reconnect the import
3290                  * immediately, instead of impacting the following the rpc.
3291                  */
3292                 RETURN(rc);
3293         }
3294         /*
3295          * For other cases, the error will be returned by intent, and client
3296          * will retrieve the result from intent.
3297          */
3298         RETURN(ELDLM_LOCK_ABORTED);
3299 }
3300
3301 static int mdt_intent_code(long itcode)
3302 {
3303         int rc;
3304
3305         switch(itcode) {
3306         case IT_OPEN:
3307                 rc = MDT_IT_OPEN;
3308                 break;
3309         case IT_OPEN|IT_CREAT:
3310                 rc = MDT_IT_OCREAT;
3311                 break;
3312         case IT_CREAT:
3313                 rc = MDT_IT_CREATE;
3314                 break;
3315         case IT_READDIR:
3316                 rc = MDT_IT_READDIR;
3317                 break;
3318         case IT_GETATTR:
3319                 rc = MDT_IT_GETATTR;
3320                 break;
3321         case IT_LOOKUP:
3322                 rc = MDT_IT_LOOKUP;
3323                 break;
3324         case IT_UNLINK:
3325                 rc = MDT_IT_UNLINK;
3326                 break;
3327         case IT_TRUNC:
3328                 rc = MDT_IT_TRUNC;
3329                 break;
3330         case IT_GETXATTR:
3331                 rc = MDT_IT_GETXATTR;
3332                 break;
3333         case IT_LAYOUT:
3334                 rc = MDT_IT_LAYOUT;
3335                 break;
3336         case IT_QUOTA_DQACQ:
3337         case IT_QUOTA_CONN:
3338                 rc = MDT_IT_QUOTA;
3339                 break;
3340         default:
3341                 CERROR("Unknown intent opcode: %ld\n", itcode);
3342                 rc = -EINVAL;
3343                 break;
3344         }
3345         return rc;
3346 }
3347
3348 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
3349                           struct ldlm_lock **lockp, __u64 flags)
3350 {
3351         struct req_capsule      *pill = info->mti_pill;
3352         struct ptlrpc_request   *req = mdt_info_req(info);
3353         struct mdt_it_flavor    *flv;
3354         int opc;
3355         int rc;
3356         ENTRY;
3357
3358         opc = mdt_intent_code(itopc);
3359         if (opc < 0)
3360                 RETURN(-EINVAL);
3361
3362         if (opc == MDT_IT_QUOTA) {
3363                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
3364
3365                 if (qmt == NULL)
3366                         RETURN(-EOPNOTSUPP);
3367
3368                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
3369                 /* pass the request to quota master */
3370                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
3371                                                  mdt_info_req(info), lockp,
3372                                                  flags);
3373                 RETURN(rc);
3374         }
3375
3376         flv = &mdt_it_flavor[opc];
3377         if (flv->it_fmt != NULL)
3378                 req_capsule_extend(pill, flv->it_fmt);
3379
3380         rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3381         if (rc < 0)
3382                 RETURN(rc);
3383
3384         if (flv->it_flags & MUTABOR &&
3385             exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
3386                 RETURN(-EROFS);
3387
3388         if (flv->it_act != NULL) {
3389                 struct ldlm_reply *rep;
3390
3391                 /* execute policy */
3392                 rc = flv->it_act(opc, info, lockp, flags);
3393
3394                 /* Check whether the reply has been packed successfully. */
3395                 if (req->rq_repmsg != NULL) {
3396                         rep = req_capsule_server_get(info->mti_pill,
3397                                                      &RMF_DLM_REP);
3398                         rep->lock_policy_res2 =
3399                                 ptlrpc_status_hton(rep->lock_policy_res2);
3400                 }
3401         }
3402
3403         RETURN(rc);
3404 }
3405
3406 static int mdt_intent_policy(struct ldlm_namespace *ns,
3407                              struct ldlm_lock **lockp, void *req_cookie,
3408                              ldlm_mode_t mode, __u64 flags, void *data)
3409 {
3410         struct tgt_session_info *tsi;
3411         struct mdt_thread_info  *info;
3412         struct ptlrpc_request   *req  =  req_cookie;
3413         struct ldlm_intent      *it;
3414         struct req_capsule      *pill;
3415         int rc;
3416
3417         ENTRY;
3418
3419         LASSERT(req != NULL);
3420
3421         tsi = tgt_ses_info(req->rq_svc_thread->t_env);
3422
3423         info = tsi2mdt_info(tsi);
3424         LASSERT(info != NULL);
3425         pill = info->mti_pill;
3426         LASSERT(pill->rc_req == req);
3427
3428         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3429                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
3430                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3431                 if (it != NULL) {
3432                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
3433                         if (rc == 0)
3434                                 rc = ELDLM_OK;
3435
3436                         /* Lock without inodebits makes no sense and will oops
3437                          * later in ldlm. Let's check it now to see if we have
3438                          * ibits corrupted somewhere in mdt_intent_opc().
3439                          * The case for client miss to set ibits has been
3440                          * processed by others. */
3441                         LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
3442                                         lr_type == LDLM_IBITS,
3443                                      info->mti_dlm_req->lock_desc.\
3444                                         l_policy_data.l_inodebits.bits != 0));
3445                 } else
3446                         rc = err_serious(-EFAULT);
3447         } else {
3448                 /* No intent was provided */
3449                 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3450                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
3451                 rc = req_capsule_server_pack(pill);
3452                 if (rc)
3453                         rc = err_serious(rc);
3454         }
3455         mdt_thread_info_fini(info);
3456         RETURN(rc);
3457 }
3458
3459 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
3460 {
3461         struct seq_server_site  *ss = mdt_seq_site(mdt);
3462
3463         if (ss->ss_node_id == 0)
3464                 return;
3465
3466         if (ss->ss_client_seq != NULL) {
3467                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3468                 ss->ss_client_seq->lcs_exp = NULL;
3469         }
3470
3471         if (ss->ss_server_fld != NULL) {
3472                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
3473                 ss->ss_server_fld->lsf_control_exp = NULL;
3474         }
3475 }
3476
3477 static void mdt_seq_fini_cli(struct mdt_device *mdt)
3478 {
3479         struct seq_server_site *ss = mdt_seq_site(mdt);
3480
3481         if (ss == NULL)
3482                 return;
3483
3484         if (ss->ss_server_seq != NULL)
3485                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
3486 }
3487
3488 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
3489 {
3490         mdt_seq_fini_cli(mdt);
3491         mdt_deregister_seq_exp(mdt);
3492
3493         return seq_site_fini(env, mdt_seq_site(mdt));
3494 }
3495
3496 /**
3497  * It will retrieve its FLDB entries from MDT0, and it only happens
3498  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
3499  * and it needs to refresh FLDB from the MDT0.
3500  **/
3501 static int mdt_register_lwp_callback(void *data)
3502 {
3503         struct lu_env           env;
3504         struct mdt_device       *mdt = data;
3505         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
3506         int                     rc;
3507         ENTRY;
3508
3509         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
3510
3511         rc = lu_env_init(&env, LCT_MD_THREAD);
3512         if (rc < 0) {
3513                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
3514                 RETURN(rc);
3515         }
3516
3517         /* Allocate new sequence now to avoid creating local transaction
3518          * in the normal transaction process */
3519         rc = seq_server_check_and_alloc_super(&env,
3520                                               mdt_seq_site(mdt)->ss_server_seq);
3521         if (rc < 0)
3522                 GOTO(out, rc);
3523
3524         if (fld->lsf_new) {
3525                 rc = fld_update_from_controller(&env, fld);
3526                 if (rc != 0) {
3527                         CERROR("%s: cannot update controller: rc = %d\n",
3528                                mdt_obd_name(mdt), rc);
3529                         GOTO(out, rc);
3530                 }
3531         }
3532 out:
3533         lu_env_fini(&env);
3534         RETURN(rc);
3535 }
3536
3537 static int mdt_register_seq_exp(struct mdt_device *mdt)
3538 {
3539         struct seq_server_site  *ss = mdt_seq_site(mdt);
3540         char                    *lwp_name = NULL;
3541         int                     rc;
3542
3543         if (ss->ss_node_id == 0)
3544                 return 0;
3545
3546         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
3547         if (lwp_name == NULL)
3548                 GOTO(out_free, rc = -ENOMEM);
3549
3550         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
3551         if (rc != 0)
3552                 GOTO(out_free, rc);
3553
3554         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
3555                                       NULL, NULL);
3556         if (rc != 0)
3557                 GOTO(out_free, rc);
3558
3559         rc = lustre_register_lwp_item(lwp_name,
3560                                       &ss->ss_server_fld->lsf_control_exp,
3561                                       mdt_register_lwp_callback, mdt);
3562         if (rc != 0) {
3563                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3564                 ss->ss_client_seq->lcs_exp = NULL;
3565                 GOTO(out_free, rc);
3566         }
3567 out_free:
3568         if (lwp_name != NULL)
3569                 OBD_FREE(lwp_name, MAX_OBD_NAME);
3570
3571         return rc;
3572 }
3573
3574 /*
3575  * Init client sequence manager which is used by local MDS to talk to sequence
3576  * controller on remote node.
3577  */
3578 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
3579 {
3580         struct seq_server_site  *ss = mdt_seq_site(mdt);
3581         int                     rc;
3582         char                    *prefix;
3583         ENTRY;
3584
3585         /* check if this is adding the first MDC and controller is not yet
3586          * initialized. */
3587         OBD_ALLOC_PTR(ss->ss_client_seq);
3588         if (ss->ss_client_seq == NULL)
3589                 RETURN(-ENOMEM);
3590
3591         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3592         if (prefix == NULL) {
3593                 OBD_FREE_PTR(ss->ss_client_seq);
3594                 ss->ss_client_seq = NULL;
3595                 RETURN(-ENOMEM);
3596         }
3597
3598         /* Note: seq_client_fini will be called in seq_site_fini */
3599         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
3600         rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
3601                              prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
3602                                                             NULL);
3603         OBD_FREE(prefix, MAX_OBD_NAME + 5);
3604         if (rc != 0) {
3605                 OBD_FREE_PTR(ss->ss_client_seq);
3606                 ss->ss_client_seq = NULL;
3607                 RETURN(rc);
3608         }
3609
3610         rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
3611
3612         RETURN(rc);
3613 }
3614
3615 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
3616 {
3617         struct seq_server_site  *ss;
3618         int                     rc;
3619         ENTRY;
3620
3621         ss = mdt_seq_site(mdt);
3622         /* init sequence controller server(MDT0) */
3623         if (ss->ss_node_id == 0) {
3624                 OBD_ALLOC_PTR(ss->ss_control_seq);
3625                 if (ss->ss_control_seq == NULL)
3626                         RETURN(-ENOMEM);
3627
3628                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
3629                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
3630                                      ss);
3631                 if (rc)
3632                         GOTO(out_seq_fini, rc);
3633         }
3634
3635         /* Init normal sequence server */
3636         OBD_ALLOC_PTR(ss->ss_server_seq);
3637         if (ss->ss_server_seq == NULL)
3638                 GOTO(out_seq_fini, rc = -ENOMEM);
3639
3640         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
3641                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
3642         if (rc)
3643                 GOTO(out_seq_fini, rc);
3644
3645         /* init seq client for seq server to talk to seq controller(MDT0) */
3646         rc = mdt_seq_init_cli(env, mdt);
3647         if (rc != 0)
3648                 GOTO(out_seq_fini, rc);
3649
3650         if (ss->ss_node_id != 0)
3651                 /* register controller export through lwp */
3652                 rc = mdt_register_seq_exp(mdt);
3653
3654         EXIT;
3655 out_seq_fini:
3656         if (rc)
3657                 mdt_seq_fini(env, mdt);
3658
3659         return rc;
3660 }
3661
3662 /*
3663  * FLD wrappers
3664  */
3665 static int mdt_fld_fini(const struct lu_env *env,
3666                         struct mdt_device *m)
3667 {
3668         struct seq_server_site *ss = mdt_seq_site(m);
3669         ENTRY;
3670
3671         if (ss && ss->ss_server_fld) {
3672                 fld_server_fini(env, ss->ss_server_fld);
3673                 OBD_FREE_PTR(ss->ss_server_fld);
3674                 ss->ss_server_fld = NULL;
3675         }
3676
3677         RETURN(0);
3678 }
3679
3680 static int mdt_fld_init(const struct lu_env *env,
3681                         const char *uuid,
3682                         struct mdt_device *m)
3683 {
3684         struct seq_server_site *ss;
3685         int rc;
3686         ENTRY;
3687
3688         ss = mdt_seq_site(m);
3689
3690         OBD_ALLOC_PTR(ss->ss_server_fld);
3691         if (ss->ss_server_fld == NULL)
3692                 RETURN(rc = -ENOMEM);
3693
3694         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
3695                              LU_SEQ_RANGE_MDT);
3696         if (rc) {
3697                 OBD_FREE_PTR(ss->ss_server_fld);
3698                 ss->ss_server_fld = NULL;
3699                 RETURN(rc);
3700         }
3701
3702         RETURN(0);
3703 }
3704
3705 static void mdt_stack_pre_fini(const struct lu_env *env,
3706                            struct mdt_device *m, struct lu_device *top)
3707 {
3708         struct lustre_cfg_bufs  *bufs;
3709         struct lustre_cfg       *lcfg;
3710         struct mdt_thread_info  *info;
3711         ENTRY;
3712
3713         LASSERT(top);
3714
3715         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3716         LASSERT(info != NULL);
3717
3718         bufs = &info->mti_u.bufs;
3719
3720         LASSERT(m->mdt_child_exp);
3721         LASSERT(m->mdt_child_exp->exp_obd);
3722
3723         /* process cleanup, pass mdt obd name to get obd umount flags */
3724         /* XXX: this is needed because all layers are referenced by
3725          * objects (some of them are pinned by osd, for example *
3726          * the proper solution should be a model where object used
3727          * by osd only doesn't have mdt/mdd slices -bzzz */
3728         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
3729         lustre_cfg_bufs_set_string(bufs, 1, NULL);
3730         lcfg = lustre_cfg_new(LCFG_PRE_CLEANUP, bufs);
3731         if (lcfg == NULL)
3732                 RETURN_EXIT;
3733
3734         top->ld_ops->ldo_process_config(env, top, lcfg);
3735         lustre_cfg_free(lcfg);
3736         EXIT;
3737 }
3738
3739 static void mdt_stack_fini(const struct lu_env *env,
3740                            struct mdt_device *m, struct lu_device *top)
3741 {
3742         struct obd_device       *obd = mdt2obd_dev(m);
3743         struct lustre_cfg_bufs  *bufs;
3744         struct lustre_cfg       *lcfg;
3745         struct mdt_thread_info  *info;
3746         char                     flags[3] = "";
3747         ENTRY;
3748
3749         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3750         LASSERT(info != NULL);
3751
3752         lu_dev_del_linkage(top->ld_site, top);
3753
3754         lu_site_purge(env, top->ld_site, -1);
3755
3756         bufs = &info->mti_u.bufs;
3757         /* process cleanup, pass mdt obd name to get obd umount flags */
3758         /* another purpose is to let all layers to release their objects */
3759         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
3760         if (obd->obd_force)
3761                 strcat(flags, "F");
3762         if (obd->obd_fail)
3763                 strcat(flags, "A");
3764         lustre_cfg_bufs_set_string(bufs, 1, flags);
3765         lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
3766         if (lcfg == NULL)
3767                 RETURN_EXIT;
3768
3769         LASSERT(top);
3770         top->ld_ops->ldo_process_config(env, top, lcfg);
3771         lustre_cfg_free(lcfg);
3772
3773         lu_site_purge(env, top->ld_site, -1);
3774
3775         m->mdt_child = NULL;
3776         m->mdt_bottom = NULL;
3777
3778         obd_disconnect(m->mdt_child_exp);
3779         m->mdt_child_exp = NULL;
3780
3781         obd_disconnect(m->mdt_bottom_exp);
3782         m->mdt_child_exp = NULL;
3783 }
3784
3785 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
3786                                const char *next, struct obd_export **exp)
3787 {
3788         struct obd_connect_data *data = NULL;
3789         struct obd_device       *obd;
3790         int                      rc;
3791         ENTRY;
3792
3793         OBD_ALLOC_PTR(data);
3794         if (data == NULL)
3795                 GOTO(out, rc = -ENOMEM);
3796
3797         obd = class_name2obd(next);
3798         if (obd == NULL) {
3799                 CERROR("%s: can't locate next device: %s\n",
3800                        mdt_obd_name(m), next);
3801                 GOTO(out, rc = -ENOTCONN);
3802         }
3803
3804         data->ocd_connect_flags = OBD_CONNECT_VERSION;
3805         data->ocd_version = LUSTRE_VERSION_CODE;
3806
3807         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
3808         if (rc) {
3809                 CERROR("%s: cannot connect to next dev %s (%d)\n",
3810                        mdt_obd_name(m), next, rc);
3811                 GOTO(out, rc);
3812         }
3813
3814 out:
3815         if (data)
3816                 OBD_FREE_PTR(data);
3817         RETURN(rc);
3818 }
3819
3820 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
3821                           struct lustre_cfg *cfg)
3822 {
3823         char                   *dev = lustre_cfg_string(cfg, 0);
3824         int                     rc, name_size, uuid_size;
3825         char                   *name, *uuid, *p;
3826         struct lustre_cfg_bufs *bufs;
3827         struct lustre_cfg      *lcfg;
3828         struct obd_device      *obd;
3829         struct lustre_profile  *lprof;
3830         struct lu_site         *site;
3831         ENTRY;
3832
3833         /* in 1.8 we had the only device in the stack - MDS.
3834          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
3835          * in 2.3 OSD is instantiated by obd_mount.c, so we need
3836          * to generate names and setup MDT, MDD. MDT will be using
3837          * generated name to connect to MDD. for MDD the next device
3838          * will be LOD with name taken from so called "profile" which
3839          * is generated by mount_option line
3840          *
3841          * 1.8 MGS generates config. commands like this:
3842          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
3843          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
3844          * 2.0 MGS generates config. commands like this:
3845          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
3846          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
3847          *                    3:lustre-MDT0000-mdtlov  4:f
3848          *
3849          * we generate MDD name from MDT one, just replacing T with D
3850          *
3851          * after all the preparations, the logical equivalent will be
3852          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
3853          *                    3:lustre-MDT0000-mdtlov  4:f
3854          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
3855          *                    3:lustre-MDD0000  4:f
3856          *
3857          *  notice we build the stack from down to top: MDD first, then MDT */
3858
3859         name_size = MAX_OBD_NAME;
3860         uuid_size = MAX_OBD_NAME;
3861
3862         OBD_ALLOC(name, name_size);
3863         OBD_ALLOC(uuid, uuid_size);
3864         if (name == NULL || uuid == NULL)
3865                 GOTO(cleanup_mem, rc = -ENOMEM);
3866
3867         OBD_ALLOC_PTR(bufs);
3868         if (!bufs)
3869                 GOTO(cleanup_mem, rc = -ENOMEM);
3870
3871         strcpy(name, dev);
3872         p = strstr(name, "-MDT");
3873         if (p == NULL)
3874                 GOTO(free_bufs, rc = -ENOMEM);
3875         p[3] = 'D';
3876
3877         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
3878
3879         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
3880         if (lprof == NULL || lprof->lp_dt == NULL) {
3881                 CERROR("can't find the profile: %s\n",
3882                        lustre_cfg_string(cfg, 0));
3883                 GOTO(free_bufs, rc = -EINVAL);
3884         }
3885
3886         lustre_cfg_bufs_reset(bufs, name);
3887         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
3888         lustre_cfg_bufs_set_string(bufs, 2, uuid);
3889         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
3890
3891         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
3892         if (lcfg == NULL)
3893                 GOTO(put_profile, rc = -ENOMEM);
3894
3895         rc = class_attach(lcfg);
3896         if (rc)
3897                 GOTO(lcfg_cleanup, rc);
3898
3899         obd = class_name2obd(name);
3900         if (!obd) {
3901                 CERROR("Can not find obd %s (%s in config)\n",
3902                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
3903                 GOTO(lcfg_cleanup, rc = -EINVAL);
3904         }
3905
3906         lustre_cfg_free(lcfg);
3907
3908         lustre_cfg_bufs_reset(bufs, name);
3909         lustre_cfg_bufs_set_string(bufs, 1, uuid);
3910         lustre_cfg_bufs_set_string(bufs, 2, dev);
3911         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
3912
3913         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
3914         if (lcfg == NULL)
3915                 GOTO(class_detach, rc = -ENOMEM);
3916
3917         rc = class_setup(obd, lcfg);
3918         if (rc)
3919                 GOTO(class_detach, rc);
3920
3921         /* connect to MDD we just setup */
3922         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
3923         if (rc)
3924                 GOTO(class_detach, rc);
3925
3926         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
3927         LASSERT(site);
3928         LASSERT(mdt_lu_site(mdt) == NULL);
3929         mdt->mdt_lu_dev.ld_site = site;
3930         site->ls_top_dev = &mdt->mdt_lu_dev;
3931         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
3932
3933         /* now connect to bottom OSD */
3934         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
3935         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
3936         if (rc)
3937                 GOTO(class_detach, rc);
3938         mdt->mdt_bottom =
3939                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
3940
3941         rc = lu_env_refill((struct lu_env *)env);
3942         if (rc != 0)
3943                 CERROR("Failure to refill session: '%d'\n", rc);
3944
3945         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
3946
3947         EXIT;
3948 class_detach:
3949         if (rc)
3950                 class_detach(obd, lcfg);
3951 lcfg_cleanup:
3952         lustre_cfg_free(lcfg);
3953 put_profile:
3954         class_put_profile(lprof);
3955 free_bufs:
3956         OBD_FREE_PTR(bufs);
3957 cleanup_mem:
3958         if (name)
3959                 OBD_FREE(name, name_size);
3960         if (uuid)
3961                 OBD_FREE(uuid, uuid_size);
3962         RETURN(rc);
3963 }
3964
3965 /* setup quota master target on MDT0 */
3966 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
3967                           struct lustre_cfg *cfg)
3968 {
3969         struct obd_device       *obd;
3970         char                    *dev = lustre_cfg_string(cfg, 0);
3971         char                    *qmtname, *uuid, *p;
3972         struct lustre_cfg_bufs  *bufs;
3973         struct lustre_cfg       *lcfg;
3974         struct lustre_profile   *lprof;
3975         struct obd_connect_data *data;
3976         int                      rc;
3977         ENTRY;
3978
3979         LASSERT(mdt->mdt_qmt_exp == NULL);
3980         LASSERT(mdt->mdt_qmt_dev == NULL);
3981
3982         /* quota master is on MDT0 only for now */
3983         if (mdt->mdt_seq_site.ss_node_id != 0)
3984                 RETURN(0);
3985
3986         /* MGS generates config commands which look as follows:
3987          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
3988          *                    3:lustre-MDT0000-mdtlov  4:f
3989          *
3990          * We generate the QMT name from the MDT one, just replacing MD with QM
3991          * after all the preparations, the logical equivalent will be:
3992          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
3993          *                    3:lustre-MDT0000-osd  4:f */
3994         OBD_ALLOC(qmtname, MAX_OBD_NAME);
3995         OBD_ALLOC(uuid, UUID_MAX);
3996         OBD_ALLOC_PTR(bufs);
3997         OBD_ALLOC_PTR(data);
3998         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
3999                 GOTO(cleanup_mem, rc = -ENOMEM);
4000
4001         strcpy(qmtname, dev);
4002         p = strstr(qmtname, "-MDT");
4003         if (p == NULL)
4004                 GOTO(cleanup_mem, rc = -ENOMEM);
4005         /* replace MD with QM */
4006         p[1] = 'Q';
4007         p[2] = 'M';
4008
4009         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
4010
4011         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4012         if (lprof == NULL || lprof->lp_dt == NULL) {
4013                 CERROR("can't find profile for %s\n",
4014                        lustre_cfg_string(cfg, 0));
4015                 GOTO(cleanup_mem, rc = -EINVAL);
4016         }
4017
4018         lustre_cfg_bufs_reset(bufs, qmtname);
4019         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
4020         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4021         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4022
4023         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4024         if (lcfg == NULL)
4025                 GOTO(put_profile, rc = -ENOMEM);
4026
4027         rc = class_attach(lcfg);
4028         if (rc)
4029                 GOTO(lcfg_cleanup, rc);
4030
4031         obd = class_name2obd(qmtname);
4032         if (!obd) {
4033                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
4034                        lustre_cfg_string(cfg, 0));
4035                 GOTO(lcfg_cleanup, rc = -EINVAL);
4036         }
4037
4038         lustre_cfg_free(lcfg);
4039
4040         lustre_cfg_bufs_reset(bufs, qmtname);
4041         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4042         lustre_cfg_bufs_set_string(bufs, 2, dev);
4043
4044         /* for quota, the next device should be the OSD device */
4045         lustre_cfg_bufs_set_string(bufs, 3,
4046                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
4047
4048         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4049         if (lcfg == NULL)
4050                 GOTO(class_detach, rc = -ENOMEM);
4051
4052         rc = class_setup(obd, lcfg);
4053         if (rc)
4054                 GOTO(class_detach, rc);
4055
4056         mdt->mdt_qmt_dev = obd->obd_lu_dev;
4057
4058         /* configure local quota objects */
4059         rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
4060                                                    &mdt->mdt_lu_dev,
4061                                                    mdt->mdt_qmt_dev);
4062         if (rc)
4063                 GOTO(class_cleanup, rc);
4064
4065         /* connect to quota master target */
4066         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4067         data->ocd_version = LUSTRE_VERSION_CODE;
4068         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
4069                          data, NULL);
4070         if (rc) {
4071                 CERROR("cannot connect to quota master device %s (%d)\n",
4072                        qmtname, rc);
4073                 GOTO(class_cleanup, rc);
4074         }
4075
4076         EXIT;
4077 class_cleanup:
4078         if (rc) {
4079                 class_manual_cleanup(obd);
4080                 mdt->mdt_qmt_dev = NULL;
4081         }
4082 class_detach:
4083         if (rc)
4084                 class_detach(obd, lcfg);
4085 lcfg_cleanup:
4086         lustre_cfg_free(lcfg);
4087 put_profile:
4088         class_put_profile(lprof);
4089 cleanup_mem:
4090         if (bufs)
4091                 OBD_FREE_PTR(bufs);
4092         if (qmtname)
4093                 OBD_FREE(qmtname, MAX_OBD_NAME);
4094         if (uuid)
4095                 OBD_FREE(uuid, UUID_MAX);
4096         if (data)
4097                 OBD_FREE_PTR(data);
4098         return rc;
4099 }
4100
4101 /* Shutdown quota master target associated with mdt */
4102 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
4103 {
4104         ENTRY;
4105
4106         if (mdt->mdt_qmt_exp == NULL)
4107                 RETURN_EXIT;
4108         LASSERT(mdt->mdt_qmt_dev != NULL);
4109
4110         /* the qmt automatically shuts down when the mdt disconnects */
4111         obd_disconnect(mdt->mdt_qmt_exp);
4112         mdt->mdt_qmt_exp = NULL;
4113         mdt->mdt_qmt_dev = NULL;
4114         EXIT;
4115 }
4116
4117 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
4118  * for now. This will be removed along with converting rest of MDT code
4119  * to use tgt_session_info */
4120 static int mdt_tgt_getxattr(struct tgt_session_info *tsi)
4121 {
4122         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
4123         int                      rc;
4124
4125         rc = mdt_getxattr(info);
4126
4127         mdt_thread_info_fini(info);
4128         return rc;
4129 }
4130
4131 static struct tgt_handler mdt_tgt_handlers[] = {
4132 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4133                 0,                      MDS_CONNECT,    mdt_tgt_connect,
4134                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
4135 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4136                 0,                      MDS_DISCONNECT, tgt_disconnect,
4137                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
4138 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4139                 HABEO_REFERO,           MDS_SET_INFO,   mdt_set_info,
4140                 &RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION),
4141 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
4142 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_GETSTATUS,  mdt_getstatus),
4143 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETATTR,    mdt_getattr),
4144 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_GETATTR_NAME,
4145                                                         mdt_getattr_name),
4146 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETXATTR,   mdt_tgt_getxattr),
4147 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_STATFS,     mdt_statfs),
4148 TGT_MDT_HDL(0           | MUTABOR,      MDS_REINT,      mdt_reint),
4149 TGT_MDT_HDL(HABEO_CORPUS,               MDS_CLOSE,      mdt_close),
4150 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_READPAGE,   mdt_readpage),
4151 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_SYNC,       mdt_sync),
4152 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
4153 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_PROGRESS,
4154                                                         mdt_hsm_progress),
4155 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_REGISTER,
4156                                                         mdt_hsm_ct_register),
4157 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_UNREGISTER,
4158                                                         mdt_hsm_ct_unregister),
4159 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_STATE_GET,
4160                                                         mdt_hsm_state_get),
4161 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_STATE_SET,
4162                                                         mdt_hsm_state_set),
4163 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_ACTION, mdt_hsm_action),
4164 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_REQUEST,
4165                                                         mdt_hsm_request),
4166 TGT_MDT_HDL(HABEO_CLAVIS | HABEO_CORPUS | HABEO_REFERO | MUTABOR,
4167             MDS_SWAP_LAYOUTS,
4168             mdt_swap_layouts),
4169 };
4170
4171 static struct tgt_handler mdt_sec_ctx_ops[] = {
4172 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
4173 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
4174 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
4175 };
4176
4177 static struct tgt_handler mdt_quota_ops[] = {
4178 TGT_QUOTA_HDL(HABEO_REFERO,             QUOTA_DQACQ,      mdt_quota_dqacq),
4179 };
4180
4181 static struct tgt_opc_slice mdt_common_slice[] = {
4182         {
4183                 .tos_opc_start  = MDS_FIRST_OPC,
4184                 .tos_opc_end    = MDS_LAST_OPC,
4185                 .tos_hs         = mdt_tgt_handlers
4186         },
4187         {
4188                 .tos_opc_start  = OBD_FIRST_OPC,
4189                 .tos_opc_end    = OBD_LAST_OPC,
4190                 .tos_hs         = tgt_obd_handlers
4191         },
4192         {
4193                 .tos_opc_start  = LDLM_FIRST_OPC,
4194                 .tos_opc_end    = LDLM_LAST_OPC,
4195                 .tos_hs         = tgt_dlm_handlers
4196         },
4197         {
4198                 .tos_opc_start  = SEC_FIRST_OPC,
4199                 .tos_opc_end    = SEC_LAST_OPC,
4200                 .tos_hs         = mdt_sec_ctx_ops
4201         },
4202         {
4203                 .tos_opc_start  = OUT_UPDATE_FIRST_OPC,
4204                 .tos_opc_end    = OUT_UPDATE_LAST_OPC,
4205                 .tos_hs         = tgt_out_handlers
4206         },
4207         {
4208                 .tos_opc_start  = FLD_FIRST_OPC,
4209                 .tos_opc_end    = FLD_LAST_OPC,
4210                 .tos_hs         = fld_handlers
4211         },
4212         {
4213                 .tos_opc_start  = SEQ_FIRST_OPC,
4214                 .tos_opc_end    = SEQ_LAST_OPC,
4215                 .tos_hs         = seq_handlers
4216         },
4217         {
4218                 .tos_opc_start  = QUOTA_DQACQ,
4219                 .tos_opc_end    = QUOTA_LAST_OPC,
4220                 .tos_hs         = mdt_quota_ops
4221         },
4222         {
4223                 .tos_opc_start  = LLOG_FIRST_OPC,
4224                 .tos_opc_end    = LLOG_LAST_OPC,
4225                 .tos_hs         = tgt_llog_handlers
4226         },
4227         {
4228                 .tos_opc_start  = LFSCK_FIRST_OPC,
4229                 .tos_opc_end    = LFSCK_LAST_OPC,
4230                 .tos_hs         = tgt_lfsck_handlers
4231         },
4232
4233         {
4234                 .tos_hs         = NULL
4235         }
4236 };
4237
4238 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4239 {
4240         struct md_device        *next = m->mdt_child;
4241         struct lu_device        *d    = &m->mdt_lu_dev;
4242         struct obd_device       *obd  = mdt2obd_dev(m);
4243         struct lfsck_stop        stop;
4244         ENTRY;
4245
4246         stop.ls_status = LS_PAUSED;
4247         stop.ls_flags = 0;
4248         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
4249
4250         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
4251         target_recovery_fini(obd);
4252         ping_evictor_stop();
4253
4254         if (m->mdt_opts.mo_coordinator)
4255                 mdt_hsm_cdt_stop(m);
4256
4257         mdt_hsm_cdt_fini(m);
4258
4259         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
4260         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4261
4262         if (m->mdt_namespace != NULL)
4263                 ldlm_namespace_free_prior(m->mdt_namespace, NULL,
4264                                           d->ld_obd->obd_force);
4265
4266         obd_exports_barrier(obd);
4267         obd_zombie_barrier();
4268
4269         mdt_procfs_fini(m);
4270
4271         tgt_fini(env, &m->mdt_lut);
4272         mdt_fs_cleanup(env, m);
4273         upcall_cache_cleanup(m->mdt_identity_cache);
4274         m->mdt_identity_cache = NULL;
4275
4276         if (m->mdt_namespace != NULL) {
4277                 ldlm_namespace_free_post(m->mdt_namespace);
4278                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4279         }
4280
4281         mdt_quota_fini(env, m);
4282
4283         cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
4284
4285         mdt_seq_fini(env, m);
4286         mdt_fld_fini(env, m);
4287
4288         /*
4289          * Finish the stack
4290          */
4291         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4292
4293         LASSERT(atomic_read(&d->ld_ref) == 0);
4294
4295         server_put_mount(mdt_obd_name(m), true);
4296
4297         EXIT;
4298 }
4299
4300 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4301
4302 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4303                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
4304 {
4305         struct mdt_thread_info    *info;
4306         struct obd_device         *obd;
4307         const char                *dev = lustre_cfg_string(cfg, 0);
4308         const char                *num = lustre_cfg_string(cfg, 2);
4309         struct lustre_mount_info  *lmi = NULL;
4310         struct lustre_sb_info     *lsi;
4311         struct lu_site            *s;
4312         struct seq_server_site    *ss_site;
4313         const char                *identity_upcall = "NONE";
4314         struct md_device          *next;
4315         int                        rc;
4316         long                       node_id;
4317         mntopt_t                   mntopts;
4318         ENTRY;
4319
4320         lu_device_init(&m->mdt_lu_dev, ldt);
4321         /*
4322          * Environment (env) might be missing mdt_thread_key values at that
4323          * point, if device is allocated when mdt_thread_key is in QUIESCENT
4324          * mode.
4325          *
4326          * Usually device allocation path doesn't use module key values, but
4327          * mdt has to do a lot of work here, so allocate key value.
4328          */
4329         rc = lu_env_refill((struct lu_env *)env);
4330         if (rc != 0)
4331                 RETURN(rc);
4332
4333         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4334         LASSERT(info != NULL);
4335
4336         obd = class_name2obd(dev);
4337         LASSERT(obd != NULL);
4338
4339         m->mdt_max_mdsize = MAX_MD_SIZE; /* 4 stripes */
4340
4341         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4342
4343         /* default is coordinator off, it is started through conf_param
4344          * or /proc */
4345         m->mdt_opts.mo_coordinator = 0;
4346
4347         lmi = server_get_mount(dev);
4348         if (lmi == NULL) {
4349                 CERROR("Cannot get mount info for %s!\n", dev);
4350                 RETURN(-EFAULT);
4351         } else {
4352                 lsi = s2lsi(lmi->lmi_sb);
4353                 /* CMD is supported only in IAM mode */
4354                 LASSERT(num);
4355                 node_id = simple_strtol(num, NULL, 10);
4356                 obd->u.obt.obt_magic = OBT_MAGIC;
4357         }
4358
4359         m->mdt_squash.rsi_uid = 0;
4360         m->mdt_squash.rsi_gid = 0;
4361         INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
4362         init_rwsem(&m->mdt_squash.rsi_sem);
4363         spin_lock_init(&m->mdt_osfs_lock);
4364         m->mdt_osfs_age = cfs_time_shift_64(-1000);
4365         m->mdt_enable_remote_dir = 0;
4366         m->mdt_enable_remote_dir_gid = 0;
4367
4368         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
4369         m->mdt_lu_dev.ld_obd = obd;
4370         /* Set this lu_device to obd for error handling purposes. */
4371         obd->obd_lu_dev = &m->mdt_lu_dev;
4372
4373         /* init the stack */
4374         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
4375         if (rc) {
4376                 CERROR("%s: Can't init device stack, rc %d\n",
4377                        mdt_obd_name(m), rc);
4378                 GOTO(err_lmi, rc);
4379         }
4380
4381         s = mdt_lu_site(m);
4382         ss_site = mdt_seq_site(m);
4383         s->ld_seq_site = ss_site;
4384         ss_site->ss_lu = s;
4385
4386         /* set server index */
4387         ss_site->ss_node_id = node_id;
4388
4389         /* failover is the default
4390          * FIXME: we do not failout mds0/mgs, which may cause some problems.
4391          * assumed whose ss_node_id == 0 XXX
4392          * */
4393         obd->obd_replayable = 1;
4394         /* No connection accepted until configurations will finish */
4395         obd->obd_no_conn = 1;
4396
4397         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4398                 char *str = lustre_cfg_string(cfg, 4);
4399                 if (strchr(str, 'n')) {
4400                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
4401                         obd->obd_replayable = 0;
4402                 }
4403         }
4404
4405         rc = mdt_fld_init(env, mdt_obd_name(m), m);
4406         if (rc)
4407                 GOTO(err_fini_stack, rc);
4408
4409         rc = mdt_seq_init(env, m);
4410         if (rc)
4411                 GOTO(err_fini_fld, rc);
4412
4413         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
4414                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
4415         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4416                                               LDLM_NAMESPACE_SERVER,
4417                                               LDLM_NAMESPACE_GREEDY,
4418                                               LDLM_NS_TYPE_MDT);
4419         if (m->mdt_namespace == NULL)
4420                 GOTO(err_fini_seq, rc = -ENOMEM);
4421
4422         m->mdt_namespace->ns_lvbp = m;
4423         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
4424
4425         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4426         /* set obd_namespace for compatibility with old code */
4427         obd->obd_namespace = m->mdt_namespace;
4428
4429         rc = mdt_hsm_cdt_init(m);
4430         if (rc != 0) {
4431                 CERROR("%s: error initializing coordinator, rc %d\n",
4432                        mdt_obd_name(m), rc);
4433                 GOTO(err_free_ns, rc);
4434         }
4435
4436         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
4437                       OBD_FAIL_MDS_ALL_REQUEST_NET,
4438                       OBD_FAIL_MDS_ALL_REPLY_NET);
4439         if (rc)
4440                 GOTO(err_free_hsm, rc);
4441
4442         rc = mdt_fs_setup(env, m, obd, lsi);
4443         if (rc)
4444                 GOTO(err_tgt, rc);
4445
4446         tgt_adapt_sptlrpc_conf(&m->mdt_lut, 1);
4447
4448         next = m->mdt_child;
4449         rc = next->md_ops->mdo_iocontrol(env, next, OBD_IOC_GET_MNTOPT, 0,
4450                                          &mntopts);
4451         if (rc)
4452                 GOTO(err_fs_cleanup, rc);
4453
4454         if (mntopts & MNTOPT_USERXATTR)
4455                 m->mdt_opts.mo_user_xattr = 1;
4456         else
4457                 m->mdt_opts.mo_user_xattr = 0;
4458
4459         rc = next->md_ops->mdo_maxeasize_get(env, next, &m->mdt_max_ea_size);
4460         if (rc)
4461                 GOTO(err_fs_cleanup, rc);
4462
4463         if (mntopts & MNTOPT_ACL)
4464                 m->mdt_opts.mo_acl = 1;
4465         else
4466                 m->mdt_opts.mo_acl = 0;
4467
4468         /* XXX: to support suppgid for ACL, we enable identity_upcall
4469          * by default, otherwise, maybe got unexpected -EACCESS. */
4470         if (m->mdt_opts.mo_acl)
4471                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4472
4473         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
4474                                                 identity_upcall,
4475                                                 &mdt_identity_upcall_cache_ops);
4476         if (IS_ERR(m->mdt_identity_cache)) {
4477                 rc = PTR_ERR(m->mdt_identity_cache);
4478                 m->mdt_identity_cache = NULL;
4479                 GOTO(err_fs_cleanup, rc);
4480         }
4481
4482         rc = mdt_procfs_init(m, dev);
4483         if (rc) {
4484                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4485                 GOTO(err_recovery, rc);
4486         }
4487
4488         rc = mdt_quota_init(env, m, cfg);
4489         if (rc)
4490                 GOTO(err_procfs, rc);
4491
4492         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
4493         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
4494                            "mdt_ldlm_client", m->mdt_ldlm_client);
4495
4496         ping_evictor_start();
4497
4498         /* recovery will be started upon mdt_prepare()
4499          * when the whole stack is complete and ready
4500          * to serve the requests */
4501
4502         /* Reduce the initial timeout on an MDS because it doesn't need such
4503          * a long timeout as an OST does. Adaptive timeouts will adjust this
4504          * value appropriately. */
4505         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4506                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4507
4508         RETURN(0);
4509 err_procfs:
4510         mdt_procfs_fini(m);
4511 err_recovery:
4512         target_recovery_fini(obd);
4513         upcall_cache_cleanup(m->mdt_identity_cache);
4514         m->mdt_identity_cache = NULL;
4515 err_fs_cleanup:
4516         mdt_fs_cleanup(env, m);
4517 err_tgt:
4518         tgt_fini(env, &m->mdt_lut);
4519 err_free_hsm:
4520         mdt_hsm_cdt_fini(m);
4521 err_free_ns:
4522         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4523         obd->obd_namespace = m->mdt_namespace = NULL;
4524 err_fini_seq:
4525         mdt_seq_fini(env, m);
4526 err_fini_fld:
4527         mdt_fld_fini(env, m);
4528 err_fini_stack:
4529         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4530 err_lmi:
4531         if (lmi)
4532                 server_put_mount(dev, true);
4533         return(rc);
4534 }
4535
4536 /* For interoperability, the left element is old parameter, the right one
4537  * is the new version of the parameter, if some parameter is deprecated,
4538  * the new version should be set as NULL. */
4539 static struct cfg_interop_param mdt_interop_param[] = {
4540         { "mdt.group_upcall",   NULL },
4541         { "mdt.quota_type",     NULL },
4542         { "mdd.quota_type",     NULL },
4543         { "mdt.rootsquash",     "mdt.root_squash" },
4544         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
4545         { NULL }
4546 };
4547
4548 /* used by MGS to process specific configurations */
4549 static int mdt_process_config(const struct lu_env *env,
4550                               struct lu_device *d, struct lustre_cfg *cfg)
4551 {
4552         struct mdt_device *m = mdt_dev(d);
4553         struct md_device *md_next = m->mdt_child;
4554         struct lu_device *next = md2lu_dev(md_next);
4555         int rc;
4556         ENTRY;
4557
4558         switch (cfg->lcfg_command) {
4559         case LCFG_PARAM: {
4560                 struct obd_device          *obd = d->ld_obd;
4561
4562                 /* For interoperability */
4563                 struct cfg_interop_param   *ptr = NULL;
4564                 struct lustre_cfg          *old_cfg = NULL;
4565                 char                       *param = NULL;
4566
4567                 param = lustre_cfg_string(cfg, 1);
4568                 if (param == NULL) {
4569                         CERROR("param is empty\n");
4570                         rc = -EINVAL;
4571                         break;
4572                 }
4573
4574                 ptr = class_find_old_param(param, mdt_interop_param);
4575                 if (ptr != NULL) {
4576                         if (ptr->new_param == NULL) {
4577                                 rc = 0;
4578                                 CWARN("For interoperability, skip this %s."
4579                                       " It is obsolete.\n", ptr->old_param);
4580                                 break;
4581                         }
4582
4583                         CWARN("Found old param %s, changed it to %s.\n",
4584                               ptr->old_param, ptr->new_param);
4585
4586                         old_cfg = cfg;
4587                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
4588                         if (IS_ERR(cfg)) {
4589                                 rc = PTR_ERR(cfg);
4590                                 break;
4591                         }
4592                 }
4593
4594                 rc = class_process_proc_param(PARAM_MDT, obd->obd_vars,
4595                                               cfg, obd);
4596                 if (rc > 0 || rc == -ENOSYS) {
4597                         /* is it an HSM var ? */
4598                         rc = class_process_proc_param(PARAM_HSM,
4599                                                       hsm_cdt_get_proc_vars(),
4600                                                       cfg, obd);
4601                         if (rc > 0 || rc == -ENOSYS)
4602                                 /* we don't understand; pass it on */
4603                                 rc = next->ld_ops->ldo_process_config(env, next,
4604                                                                       cfg);
4605                 }
4606
4607                 if (old_cfg != NULL)
4608                         lustre_cfg_free(cfg);
4609
4610                 break;
4611         }
4612         default:
4613                 /* others are passed further */
4614                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4615                 break;
4616         }
4617         RETURN(rc);
4618 }
4619
4620 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4621                                           const struct lu_object_header *hdr,
4622                                           struct lu_device *d)
4623 {
4624         struct mdt_object *mo;
4625
4626         ENTRY;
4627
4628         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
4629         if (mo != NULL) {
4630                 struct lu_object *o;
4631                 struct lu_object_header *h;
4632
4633                 o = &mo->mot_obj;
4634                 h = &mo->mot_header;
4635                 lu_object_header_init(h);
4636                 lu_object_init(o, h, d);
4637                 lu_object_add_top(h, o);
4638                 o->lo_ops = &mdt_obj_ops;
4639                 spin_lock_init(&mo->mot_write_lock);
4640                 mutex_init(&mo->mot_lov_mutex);
4641                 init_rwsem(&mo->mot_open_sem);
4642                 RETURN(o);
4643         }
4644         RETURN(NULL);
4645 }
4646
4647 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
4648                            const struct lu_object_conf *unused)
4649 {
4650         struct mdt_device *d = mdt_dev(o->lo_dev);
4651         struct lu_device  *under;
4652         struct lu_object  *below;
4653         int                rc = 0;
4654         ENTRY;
4655
4656         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4657                PFID(lu_object_fid(o)));
4658
4659         under = &d->mdt_child->md_lu_dev;
4660         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4661         if (below != NULL) {
4662                 lu_object_add(o, below);
4663         } else
4664                 rc = -ENOMEM;
4665
4666         RETURN(rc);
4667 }
4668
4669 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4670 {
4671         struct mdt_object *mo = mdt_obj(o);
4672         struct lu_object_header *h;
4673         ENTRY;
4674
4675         h = o->lo_header;
4676         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4677                PFID(lu_object_fid(o)));
4678
4679         LASSERT(atomic_read(&mo->mot_open_count) == 0);
4680         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
4681
4682         lu_object_fini(o);
4683         lu_object_header_fini(h);
4684         OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
4685
4686         EXIT;
4687 }
4688
4689 static int mdt_object_print(const struct lu_env *env, void *cookie,
4690                             lu_printer_t p, const struct lu_object *o)
4691 {
4692         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
4693
4694         return (*p)(env, cookie,
4695                     LUSTRE_MDT_NAME"-object@%p(flags=%d, writecount=%d)",
4696                     mdto, mdto->mot_flags, mdto->mot_write_count);
4697 }
4698
4699 static int mdt_prepare(const struct lu_env *env,
4700                 struct lu_device *pdev,
4701                 struct lu_device *cdev)
4702 {
4703         struct mdt_device *mdt = mdt_dev(cdev);
4704         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
4705         struct obd_device *obd = cdev->ld_obd;
4706         int rc;
4707
4708         ENTRY;
4709
4710         LASSERT(obd);
4711
4712         rc = next->ld_ops->ldo_prepare(env, cdev, next);
4713         if (rc)
4714                 RETURN(rc);
4715
4716         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
4717         if (rc)
4718                 RETURN(rc);
4719
4720         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
4721         if (rc)
4722                 RETURN(rc);
4723
4724         rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
4725         /* The LFSCK instance is registered just now, so it must be there when
4726          * register the namespace to such instance. */
4727         LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
4728
4729         if (mdt->mdt_seq_site.ss_node_id == 0) {
4730                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
4731                                                          &mdt->mdt_md_root_fid);
4732                 if (rc)
4733                         RETURN(rc);
4734         }
4735
4736         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
4737
4738         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
4739         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
4740         LASSERT(obd->obd_no_conn);
4741         spin_lock(&obd->obd_dev_lock);
4742         obd->obd_no_conn = 0;
4743         spin_unlock(&obd->obd_dev_lock);
4744
4745         if (obd->obd_recovering == 0)
4746                 mdt_postrecov(env, mdt);
4747
4748         RETURN(rc);
4749 }
4750
4751 const struct lu_device_operations mdt_lu_ops = {
4752         .ldo_object_alloc   = mdt_object_alloc,
4753         .ldo_process_config = mdt_process_config,
4754         .ldo_prepare        = mdt_prepare,
4755 };
4756
4757 static const struct lu_object_operations mdt_obj_ops = {
4758         .loo_object_init    = mdt_object_init,
4759         .loo_object_free    = mdt_object_free,
4760         .loo_object_print   = mdt_object_print
4761 };
4762
4763 static int mdt_obd_set_info_async(const struct lu_env *env,
4764                                   struct obd_export *exp,
4765                                   __u32 keylen, void *key,
4766                                   __u32 vallen, void *val,
4767                                   struct ptlrpc_request_set *set)
4768 {
4769         int rc;
4770
4771         ENTRY;
4772
4773         if (KEY_IS(KEY_SPTLRPC_CONF)) {
4774                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp), 0);
4775                 RETURN(rc);
4776         }
4777
4778         RETURN(0);
4779 }
4780
4781 /**
4782  * Match client and server connection feature flags.
4783  *
4784  * Compute the compatibility flags for a connection request based on
4785  * features mutually supported by client and server.
4786  *
4787  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
4788  * must not be updated here, otherwise a partially initialized value may
4789  * be exposed. After the connection request is successfully processed,
4790  * the top-level MDT connect request handler atomically updates the export
4791  * connect flags from the obd_connect_data::ocd_connect_flags field of the
4792  * reply. \see mdt_connect().
4793  *
4794  * \param exp   the obd_export associated with this client/target pair
4795  * \param mdt   the target device for the connection
4796  * \param data  stores data for this connect request
4797  *
4798  * \retval 0       success
4799  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
4800  * \retval -EBADE  client and server feature requirements are incompatible
4801  */
4802 static int mdt_connect_internal(struct obd_export *exp,
4803                                 struct mdt_device *mdt,
4804                                 struct obd_connect_data *data)
4805 {
4806         LASSERT(data != NULL);
4807
4808         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
4809         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
4810
4811         if (!(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
4812             !(data->ocd_connect_flags & OBD_CONNECT_IBITS)) {
4813                 CWARN("%s: client %s does not support ibits lock, either "
4814                       "very old or an invalid client: flags "LPX64"\n",
4815                       mdt_obd_name(mdt), exp->exp_client_uuid.uuid,
4816                       data->ocd_connect_flags);
4817                 return -EBADE;
4818         }
4819
4820         if (!mdt->mdt_opts.mo_acl)
4821                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
4822
4823         if (!mdt->mdt_opts.mo_user_xattr)
4824                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
4825
4826         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
4827                 data->ocd_brw_size = min(data->ocd_brw_size,
4828                                          (__u32)MD_MAX_BRW_SIZE);
4829                 if (data->ocd_brw_size == 0) {
4830                         CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
4831                                " ocd_version: %x ocd_grant: %d "
4832                                "ocd_index: %u ocd_brw_size is "
4833                                "unexpectedly zero, network data "
4834                                "corruption? Refusing connection of this"
4835                                " client\n",
4836                                mdt_obd_name(mdt),
4837                                exp->exp_client_uuid.uuid,
4838                                exp, data->ocd_connect_flags, data->ocd_version,
4839                                data->ocd_grant, data->ocd_index);
4840                         return -EPROTO;
4841                 }
4842         }
4843
4844         /* NB: Disregard the rule against updating
4845          * exp_connect_data.ocd_connect_flags in this case, since
4846          * tgt_client_new() needs to know if this is a lightweight
4847          * connection, and it is safe to expose this flag before
4848          * connection processing completes. */
4849         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
4850                 spin_lock(&exp->exp_lock);
4851                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
4852                 spin_unlock(&exp->exp_lock);
4853         }
4854
4855         data->ocd_version = LUSTRE_VERSION_CODE;
4856
4857         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
4858                 CWARN("%s: MDS requires FID support, but client not\n",
4859                       mdt_obd_name(mdt));
4860                 return -EBADE;
4861         }
4862
4863         if (OCD_HAS_FLAG(data, PINGLESS)) {
4864                 if (ptlrpc_pinger_suppress_pings()) {
4865                         spin_lock(&exp->exp_obd->obd_dev_lock);
4866                         list_del_init(&exp->exp_obd_chain_timed);
4867                         spin_unlock(&exp->exp_obd->obd_dev_lock);
4868                 } else {
4869                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
4870                 }
4871         }
4872
4873         data->ocd_max_easize = mdt->mdt_max_ea_size;
4874
4875         /* NB: Disregard the rule against updating
4876          * exp_connect_data.ocd_connect_flags in this case, since
4877          * tgt_client_new() needs to know if this is client supports
4878          * multiple modify RPCs, and it is safe to expose this flag before
4879          * connection processing completes. */
4880         if (data->ocd_connect_flags & OBD_CONNECT_MULTIMODRPCS) {
4881                 data->ocd_maxmodrpcs = max_mod_rpcs_per_client;
4882                 spin_lock(&exp->exp_lock);
4883                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_MULTIMODRPCS;
4884                 spin_unlock(&exp->exp_lock);
4885         }
4886
4887         return 0;
4888 }
4889
4890 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
4891                                    struct mdt_thread_info *info,
4892                                    struct mdt_file_data *mfd)
4893 {
4894         struct lu_context ses;
4895         int rc;
4896         ENTRY;
4897
4898         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
4899         if (rc)
4900                 RETURN(rc);
4901
4902         env->le_ses = &ses;
4903         lu_context_enter(&ses);
4904
4905         mdt_ucred(info)->uc_valid = UCRED_OLD;
4906         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
4907
4908         lu_context_exit(&ses);
4909         lu_context_fini(&ses);
4910         env->le_ses = NULL;
4911
4912         RETURN(rc);
4913 }
4914
4915 static int mdt_export_cleanup(struct obd_export *exp)
4916 {
4917         struct list_head         closing_list;
4918         struct mdt_export_data  *med = &exp->exp_mdt_data;
4919         struct obd_device       *obd = exp->exp_obd;
4920         struct mdt_device       *mdt;
4921         struct mdt_thread_info  *info;
4922         struct lu_env            env;
4923         struct mdt_file_data    *mfd, *n;
4924         int rc = 0;
4925         ENTRY;
4926
4927         INIT_LIST_HEAD(&closing_list);
4928         spin_lock(&med->med_open_lock);
4929         while (!list_empty(&med->med_open_head)) {
4930                 struct list_head *tmp = med->med_open_head.next;
4931                 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
4932
4933                 /* Remove mfd handle so it can't be found again.
4934                  * We are consuming the mfd_list reference here. */
4935                 class_handle_unhash(&mfd->mfd_handle);
4936                 list_move_tail(&mfd->mfd_list, &closing_list);
4937         }
4938         spin_unlock(&med->med_open_lock);
4939         mdt = mdt_dev(obd->obd_lu_dev);
4940         LASSERT(mdt != NULL);
4941
4942         rc = lu_env_init(&env, LCT_MD_THREAD);
4943         if (rc)
4944                 RETURN(rc);
4945
4946         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
4947         LASSERT(info != NULL);
4948         memset(info, 0, sizeof *info);
4949         info->mti_env = &env;
4950         info->mti_mdt = mdt;
4951         info->mti_exp = exp;
4952
4953         if (!list_empty(&closing_list)) {
4954                 struct md_attr *ma = &info->mti_attr;
4955
4956                 /* Close any open files (which may also cause orphan
4957                  * unlinking). */
4958                 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
4959                         list_del_init(&mfd->mfd_list);
4960                         ma->ma_need = ma->ma_valid = 0;
4961
4962                         /* This file is being closed due to an eviction, it
4963                          * could have been modified and now dirty regarding to
4964                          * HSM archive, check this!
4965                          * The logic here is to mark a file dirty if there's a
4966                          * chance it was dirtied before the client was evicted,
4967                          * so that we don't have to wait for a release attempt
4968                          * before finding out the file was actually dirty and
4969                          * fail the release. Aggressively marking it dirty here
4970                          * will cause the policy engine to attempt to
4971                          * re-archive it; when rearchiving, we can compare the
4972                          * current version to the HSM data_version and make the
4973                          * archive request into a noop if it's not actually
4974                          * dirty.
4975                          */
4976                         if (mfd->mfd_mode & FMODE_WRITE)
4977                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
4978
4979                         /* Don't unlink orphan on failover umount, LU-184 */
4980                         if (exp->exp_flags & OBD_OPT_FAILOVER) {
4981                                 ma->ma_valid = MA_FLAGS;
4982                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
4983                         }
4984                         mdt_mfd_close(info, mfd);
4985                 }
4986         }
4987         info->mti_mdt = NULL;
4988         /* cleanup client slot early */
4989         /* Do not erase record for recoverable client. */
4990         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
4991                 tgt_client_del(&env, exp);
4992         lu_env_fini(&env);
4993
4994         RETURN(rc);
4995 }
4996
4997 static int mdt_obd_disconnect(struct obd_export *exp)
4998 {
4999         int rc;
5000         ENTRY;
5001
5002         LASSERT(exp);
5003         class_export_get(exp);
5004
5005         nodemap_del_member(exp);
5006         rc = server_disconnect_export(exp);
5007         if (rc != 0)
5008                 CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
5009
5010         rc = mdt_export_cleanup(exp);
5011         class_export_put(exp);
5012         RETURN(rc);
5013 }
5014
5015 /* mds_connect copy */
5016 static int mdt_obd_connect(const struct lu_env *env,
5017                            struct obd_export **exp, struct obd_device *obd,
5018                            struct obd_uuid *cluuid,
5019                            struct obd_connect_data *data,
5020                            void *localdata)
5021 {
5022         struct obd_export       *lexp;
5023         struct lustre_handle    conn = { 0 };
5024         struct mdt_device       *mdt;
5025         int                      rc;
5026         lnet_nid_t              *client_nid = localdata;
5027         ENTRY;
5028
5029         LASSERT(env != NULL);
5030         if (!exp || !obd || !cluuid)
5031                 RETURN(-EINVAL);
5032
5033         mdt = mdt_dev(obd->obd_lu_dev);
5034
5035         /*
5036          * first, check whether the stack is ready to handle requests
5037          * XXX: probably not very appropriate method is used now
5038          *      at some point we should find a better one
5039          */
5040         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) && data != NULL &&
5041             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) &&
5042             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
5043                 rc = obd_get_info(env, mdt->mdt_child_exp,
5044                                   sizeof(KEY_OSP_CONNECTED),
5045                                   KEY_OSP_CONNECTED, NULL, NULL);
5046                 if (rc)
5047                         RETURN(-EAGAIN);
5048                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
5049         }
5050
5051         rc = class_connect(&conn, obd, cluuid);
5052         if (rc)
5053                 RETURN(rc);
5054
5055         lexp = class_conn2export(&conn);
5056         LASSERT(lexp != NULL);
5057
5058         rc = mdt_connect_internal(lexp, mdt, data);
5059         if (rc == 0) {
5060                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5061
5062                 LASSERT(lcd);
5063                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5064                 rc = tgt_client_new(env, lexp);
5065                 if (rc == 0) {
5066                         rc = nodemap_add_member(*client_nid, lexp);
5067                         if (rc != 0 && rc != -EEXIST)
5068                                 goto out;
5069
5070                         mdt_export_stats_init(obd, lexp, localdata);
5071                 }
5072         }
5073 out:
5074         if (rc != 0) {
5075                 class_disconnect(lexp);
5076                 *exp = NULL;
5077         } else {
5078                 *exp = lexp;
5079         }
5080
5081         RETURN(rc);
5082 }
5083
5084 static int mdt_obd_reconnect(const struct lu_env *env,
5085                              struct obd_export *exp, struct obd_device *obd,
5086                              struct obd_uuid *cluuid,
5087                              struct obd_connect_data *data,
5088                              void *localdata)
5089 {
5090         lnet_nid_t             *client_nid = localdata;
5091         int                     rc;
5092         ENTRY;
5093
5094         if (exp == NULL || obd == NULL || cluuid == NULL)
5095                 RETURN(-EINVAL);
5096
5097         rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
5098         if (rc == 0) {
5099                 rc = nodemap_add_member(*client_nid, exp);
5100                 if (rc == 0 || rc == -EEXIST)
5101                         mdt_export_stats_init(obd, exp, localdata);
5102         }
5103
5104         RETURN(rc);
5105 }
5106
5107 /* FIXME: Can we avoid using these two interfaces? */
5108 static int mdt_init_export(struct obd_export *exp)
5109 {
5110         struct mdt_export_data *med = &exp->exp_mdt_data;
5111         int                     rc;
5112         ENTRY;
5113
5114         INIT_LIST_HEAD(&med->med_open_head);
5115         spin_lock_init(&med->med_open_lock);
5116         mutex_init(&med->med_idmap_mutex);
5117         med->med_idmap = NULL;
5118         spin_lock(&exp->exp_lock);
5119         exp->exp_connecting = 1;
5120         spin_unlock(&exp->exp_lock);
5121
5122         /* self-export doesn't need client data and ldlm initialization */
5123         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5124                                      &exp->exp_client_uuid)))
5125                 RETURN(0);
5126
5127         rc = tgt_client_alloc(exp);
5128         if (rc)
5129                 GOTO(err, rc);
5130
5131         rc = ldlm_init_export(exp);
5132         if (rc)
5133                 GOTO(err_free, rc);
5134
5135         RETURN(rc);
5136
5137 err_free:
5138         tgt_client_free(exp);
5139 err:
5140         CERROR("%s: Failed to initialize export: rc = %d\n",
5141                exp->exp_obd->obd_name, rc);
5142         return rc;
5143 }
5144
5145 static int mdt_destroy_export(struct obd_export *exp)
5146 {
5147         ENTRY;
5148
5149         if (exp_connect_rmtclient(exp))
5150                 mdt_cleanup_idmap(&exp->exp_mdt_data);
5151
5152         target_destroy_export(exp);
5153         /* destroy can be called from failed obd_setup, so
5154          * checking uuid is safer than obd_self_export */
5155         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5156                                      &exp->exp_client_uuid)))
5157                 RETURN(0);
5158
5159         ldlm_destroy_export(exp);
5160         tgt_client_free(exp);
5161
5162         LASSERT(list_empty(&exp->exp_outstanding_replies));
5163         LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
5164
5165         RETURN(0);
5166 }
5167
5168 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
5169                    struct linkea_data *ldata)
5170 {
5171         int rc;
5172
5173         LASSERT(ldata->ld_buf->lb_buf != NULL);
5174
5175         if (!mdt_object_exists(mdt_obj))
5176                 return -ENODATA;
5177
5178         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5179                           ldata->ld_buf, XATTR_NAME_LINK);
5180         if (rc == -ERANGE) {
5181                 /* Buf was too small, figure out what we need. */
5182                 lu_buf_free(ldata->ld_buf);
5183                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5184                                   ldata->ld_buf, XATTR_NAME_LINK);
5185                 if (rc < 0)
5186                         return rc;
5187                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
5188                 if (ldata->ld_buf->lb_buf == NULL)
5189                         return -ENOMEM;
5190                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5191                                   ldata->ld_buf, XATTR_NAME_LINK);
5192         }
5193         if (rc < 0)
5194                 return rc;
5195
5196         return linkea_init(ldata);
5197 }
5198
5199 /**
5200  * Given an MDT object, try to look up the full path to the object.
5201  * Part of the MDT layer implementation of lfs fid2path.
5202  *
5203  * \param[in]     info  Per-thread common data shared by MDT level handlers.
5204  * \param[in]     obj   Object to do path lookup of
5205  * \param[in,out] fp    User-provided struct to store path information
5206  *
5207  * \retval 0 Lookup successful, path information stored in fp
5208  * \retval -EAGAIN Lookup failed, usually because object is being moved
5209  * \retval negative errno if there was a problem
5210  */
5211 static int mdt_path_current(struct mdt_thread_info *info,
5212                             struct mdt_object *obj,
5213                             struct getinfo_fid2path *fp)
5214 {
5215         struct mdt_device       *mdt = info->mti_mdt;
5216         struct mdt_object       *mdt_obj;
5217         struct link_ea_header   *leh;
5218         struct link_ea_entry    *lee;
5219         struct lu_name          *tmpname = &info->mti_name;
5220         struct lu_fid           *tmpfid = &info->mti_tmp_fid1;
5221         struct lu_buf           *buf = &info->mti_big_buf;
5222         char                    *ptr;
5223         int                     reclen;
5224         struct linkea_data      ldata = { NULL };
5225         int                     rc = 0;
5226         bool                    first = true;
5227         ENTRY;
5228
5229         /* temp buffer for path element, the buffer will be finally freed
5230          * in mdt_thread_info_fini */
5231         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
5232         if (buf->lb_buf == NULL)
5233                 RETURN(-ENOMEM);
5234
5235         ldata.ld_buf = buf;
5236         ptr = fp->gf_path + fp->gf_pathlen - 1;
5237         *ptr = 0;
5238         --ptr;
5239         *tmpfid = fp->gf_fid = *mdt_object_fid(obj);
5240
5241         /* root FID only exists on MDT0, and fid2path should also ends at MDT0,
5242          * so checking root_fid can only happen on MDT0. */
5243         while (!lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid)) {
5244                 struct lu_buf           lmv_buf;
5245
5246                 mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
5247                 if (IS_ERR(mdt_obj))
5248                         GOTO(out, rc = PTR_ERR(mdt_obj));
5249
5250                 if (!mdt_object_exists(mdt_obj)) {
5251                         mdt_object_put(info->mti_env, mdt_obj);
5252                         GOTO(out, rc = -ENOENT);
5253                 }
5254
5255                 if (mdt_object_remote(mdt_obj)) {
5256                         mdt_object_put(info->mti_env, mdt_obj);
5257                         GOTO(remote_out, rc = -EREMOTE);
5258                 }
5259
5260                 rc = mdt_links_read(info, mdt_obj, &ldata);
5261                 if (rc != 0) {
5262                         mdt_object_put(info->mti_env, mdt_obj);
5263                         GOTO(out, rc);
5264                 }
5265
5266                 leh = buf->lb_buf;
5267                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
5268                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
5269                 /* If set, use link #linkno for path lookup, otherwise use
5270                    link #0.  Only do this for the final path element. */
5271                 if (first && fp->gf_linkno < leh->leh_reccount) {
5272                         int count;
5273                         for (count = 0; count < fp->gf_linkno; count++) {
5274                                 lee = (struct link_ea_entry *)
5275                                      ((char *)lee + reclen);
5276                                 linkea_entry_unpack(lee, &reclen, tmpname,
5277                                                     tmpfid);
5278                         }
5279                         if (fp->gf_linkno < leh->leh_reccount - 1)
5280                                 /* indicate to user there are more links */
5281                                 fp->gf_linkno++;
5282                 }
5283
5284                 lmv_buf.lb_buf = info->mti_xattr_buf;
5285                 lmv_buf.lb_len = sizeof(info->mti_xattr_buf);
5286                 /* Check if it is slave stripes */
5287                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5288                                   &lmv_buf, XATTR_NAME_LMV);
5289                 mdt_object_put(info->mti_env, mdt_obj);
5290                 if (rc > 0) {
5291                         union lmv_mds_md *lmm = lmv_buf.lb_buf;
5292
5293                         /* For slave stripes, get its master */
5294                         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE) {
5295                                 fp->gf_fid = *tmpfid;
5296                                 continue;
5297                         }
5298                 } else if (rc < 0 && rc != -ENODATA) {
5299                         GOTO(out, rc);
5300                 }
5301
5302                 rc = 0;
5303
5304                 /* Pack the name in the end of the buffer */
5305                 ptr -= tmpname->ln_namelen;
5306                 if (ptr - 1 <= fp->gf_path)
5307                         GOTO(out, rc = -EOVERFLOW);
5308                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
5309                 *(--ptr) = '/';
5310
5311                 /* keep the last resolved fid to the client, so the
5312                  * client will build the left path on another MDT for
5313                  * remote object */
5314                 fp->gf_fid = *tmpfid;
5315
5316                 first = false;
5317         }
5318
5319 remote_out:
5320         ptr++; /* skip leading / */
5321         memmove(fp->gf_path, ptr, fp->gf_path + fp->gf_pathlen - ptr);
5322
5323 out:
5324         RETURN(rc);
5325 }
5326
5327 /**
5328  * Given an MDT object, use mdt_path_current to get the path.
5329  * Essentially a wrapper to retry mdt_path_current a set number of times
5330  * if -EAGAIN is returned (usually because an object is being moved).
5331  *
5332  * Part of the MDT layer implementation of lfs fid2path.
5333  *
5334  * \param[in]     info  Per-thread common data shared by mdt level handlers.
5335  * \param[in]     obj   Object to do path lookup of
5336  * \param[in,out] fp    User-provided struct for arguments and to store path
5337  *                      information
5338  *
5339  * \retval 0 Lookup successful, path information stored in fp
5340  * \retval negative errno if there was a problem
5341  */
5342 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
5343                     struct getinfo_fid2path *fp)
5344 {
5345         struct mdt_device       *mdt = info->mti_mdt;
5346         int                     tries = 3;
5347         int                     rc = -EAGAIN;
5348         ENTRY;
5349
5350         if (fp->gf_pathlen < 3)
5351                 RETURN(-EOVERFLOW);
5352
5353         if (lu_fid_eq(&mdt->mdt_md_root_fid, mdt_object_fid(obj))) {
5354                 fp->gf_path[0] = '\0';
5355                 RETURN(0);
5356         }
5357
5358         /* Retry multiple times in case file is being moved */
5359         while (tries-- && rc == -EAGAIN)
5360                 rc = mdt_path_current(info, obj, fp);
5361
5362         RETURN(rc);
5363 }
5364
5365 /**
5366  * Get the full path of the provided FID, as of changelog record recno.
5367  *
5368  * This checks sanity and looks up object for user provided FID
5369  * before calling the actual path lookup code.
5370  *
5371  * Part of the MDT layer implementation of lfs fid2path.
5372  *
5373  * \param[in]     info  Per-thread common data shared by mdt level handlers.
5374  * \param[in,out] fp    User-provided struct for arguments and to store path
5375  *                      information
5376  *
5377  * \retval 0 Lookup successful, path information and recno stored in fp
5378  * \retval -ENOENT, object does not exist
5379  * \retval negative errno if there was a problem
5380  */
5381 static int mdt_fid2path(struct mdt_thread_info *info,
5382                         struct getinfo_fid2path *fp)
5383 {
5384         struct mdt_device *mdt = info->mti_mdt;
5385         struct mdt_object *obj;
5386         int    rc;
5387         ENTRY;
5388
5389         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
5390                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
5391
5392         if (!fid_is_sane(&fp->gf_fid))
5393                 RETURN(-EINVAL);
5394
5395         if (!fid_is_namespace_visible(&fp->gf_fid)) {
5396                 CWARN("%s: "DFID" is invalid, sequence should be "
5397                       ">= "LPX64"\n", mdt_obd_name(mdt),
5398                       PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
5399                 RETURN(-EINVAL);
5400         }
5401
5402         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
5403         if (IS_ERR(obj)) {
5404                 rc = PTR_ERR(obj);
5405                 CDEBUG(D_IOCTL, "cannot find "DFID": rc = %d\n",
5406                        PFID(&fp->gf_fid), rc);
5407                 RETURN(rc);
5408         }
5409
5410         if (mdt_object_remote(obj))
5411                 rc = -EREMOTE;
5412         else if (!mdt_object_exists(obj))
5413                 rc = -ENOENT;
5414         else
5415                 rc = 0;
5416
5417         if (rc < 0) {
5418                 mdt_object_put(info->mti_env, obj);
5419                 CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n",
5420                        PFID(&fp->gf_fid), rc);
5421                 RETURN(rc);
5422         }
5423
5424         rc = mdt_path(info, obj, fp);
5425
5426         CDEBUG(D_INFO, "fid "DFID", path %s recno "LPX64" linkno %u\n",
5427                PFID(&fp->gf_fid), fp->gf_path, fp->gf_recno, fp->gf_linkno);
5428
5429         mdt_object_put(info->mti_env, obj);
5430
5431         RETURN(rc);
5432 }
5433
5434 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key,
5435                             void *val, int vallen)
5436 {
5437         struct getinfo_fid2path *fpout, *fpin;
5438         int rc = 0;
5439
5440         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
5441         fpout = val;
5442
5443         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5444                 lustre_swab_fid2path(fpin);
5445
5446         memcpy(fpout, fpin, sizeof(*fpin));
5447         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
5448                 RETURN(-EINVAL);
5449
5450         rc = mdt_fid2path(info, fpout);
5451         RETURN(rc);
5452 }
5453
5454 int mdt_get_info(struct tgt_session_info *tsi)
5455 {
5456         char    *key;
5457         int      keylen;
5458         __u32   *vallen;
5459         void    *valout;
5460         int      rc;
5461
5462         ENTRY;
5463
5464         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
5465         if (key == NULL) {
5466                 CDEBUG(D_IOCTL, "No GETINFO key\n");
5467                 RETURN(err_serious(-EFAULT));
5468         }
5469         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
5470                                       RCL_CLIENT);
5471
5472         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
5473         if (vallen == NULL) {
5474                 CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n",
5475                                 tgt_name(tsi->tsi_tgt));
5476                 RETURN(err_serious(-EFAULT));
5477         }
5478
5479         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
5480                              *vallen);
5481         rc = req_capsule_server_pack(tsi->tsi_pill);
5482         if (rc)
5483                 RETURN(err_serious(rc));
5484
5485         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
5486         if (valout == NULL) {
5487                 CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n",
5488                                 tgt_name(tsi->tsi_tgt));
5489                 RETURN(err_serious(-EFAULT));
5490         }
5491
5492         if (KEY_IS(KEY_FID2PATH)) {
5493                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
5494
5495                 rc = mdt_rpc_fid2path(info, key, valout, *vallen);
5496                 mdt_thread_info_fini(info);
5497         } else {
5498                 rc = -EINVAL;
5499         }
5500         RETURN(rc);
5501 }
5502
5503 /* Pass the ioc down */
5504 static int mdt_ioc_child(struct lu_env *env, struct mdt_device *mdt,
5505                          unsigned int cmd, int len, void *data)
5506 {
5507         struct lu_context ioctl_session;
5508         struct md_device *next = mdt->mdt_child;
5509         int rc;
5510         ENTRY;
5511
5512         rc = lu_context_init(&ioctl_session, LCT_SERVER_SESSION);
5513         if (rc)
5514                 RETURN(rc);
5515         ioctl_session.lc_thread = (struct ptlrpc_thread *)current;
5516         lu_context_enter(&ioctl_session);
5517         env->le_ses = &ioctl_session;
5518
5519         LASSERT(next->md_ops->mdo_iocontrol);
5520         rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data);
5521
5522         lu_context_exit(&ioctl_session);
5523         lu_context_fini(&ioctl_session);
5524         RETURN(rc);
5525 }
5526
5527 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
5528 {
5529         struct obd_ioctl_data *data = karg;
5530         struct lu_fid *fid;
5531         __u64 version;
5532         struct mdt_object *obj;
5533         struct mdt_lock_handle  *lh;
5534         int rc;
5535         ENTRY;
5536
5537         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
5538             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
5539                 RETURN(-EINVAL);
5540
5541         fid = (struct lu_fid *)data->ioc_inlbuf1;
5542
5543         if (!fid_is_sane(fid))
5544                 RETURN(-EINVAL);
5545
5546         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
5547
5548         lh = &mti->mti_lh[MDT_LH_PARENT];
5549         mdt_lock_reg_init(lh, LCK_CR);
5550
5551         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
5552         if (IS_ERR(obj))
5553                 RETURN(PTR_ERR(obj));
5554
5555         if (mdt_object_remote(obj)) {
5556                 rc = -EREMOTE;
5557                 /**
5558                  * before calling version get the correct MDS should be
5559                  * fid, this is error to find remote object here
5560                  */
5561                 CERROR("nonlocal object "DFID"\n", PFID(fid));
5562         } else if (!mdt_object_exists(obj)) {
5563                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
5564                 rc = -ENOENT;
5565         } else {
5566                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
5567                *(__u64 *)data->ioc_inlbuf2 = version;
5568                 rc = 0;
5569         }
5570         mdt_object_unlock_put(mti, obj, lh, 1);
5571         RETURN(rc);
5572 }
5573
5574 /* ioctls on obd dev */
5575 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
5576                          void *karg, void *uarg)
5577 {
5578         struct lu_env      env;
5579         struct obd_device *obd = exp->exp_obd;
5580         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
5581         struct dt_device  *dt = mdt->mdt_bottom;
5582         int rc;
5583
5584         ENTRY;
5585         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
5586         rc = lu_env_init(&env, LCT_MD_THREAD);
5587         if (rc)
5588                 RETURN(rc);
5589
5590         switch (cmd) {
5591         case OBD_IOC_SYNC:
5592                 rc = mdt_device_sync(&env, mdt);
5593                 break;
5594         case OBD_IOC_SET_READONLY:
5595                 rc = dt->dd_ops->dt_ro(&env, dt);
5596                 break;
5597         case OBD_IOC_ABORT_RECOVERY:
5598                 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
5599                 obd->obd_force_abort_recovery = 1;
5600                 target_stop_recovery_thread(obd);
5601                 rc = 0;
5602                 break;
5603         case OBD_IOC_CHANGELOG_REG:
5604         case OBD_IOC_CHANGELOG_DEREG:
5605         case OBD_IOC_CHANGELOG_CLEAR:
5606                 rc = mdt_ioc_child(&env, mdt, cmd, len, karg);
5607                 break;
5608         case OBD_IOC_START_LFSCK: {
5609                 struct md_device *next = mdt->mdt_child;
5610                 struct obd_ioctl_data *data = karg;
5611                 struct lfsck_start_param lsp;
5612
5613                 if (unlikely(data == NULL)) {
5614                         rc = -EINVAL;
5615                         break;
5616                 }
5617
5618                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
5619                 lsp.lsp_index_valid = 0;
5620                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
5621                 break;
5622         }
5623         case OBD_IOC_STOP_LFSCK: {
5624                 struct md_device        *next = mdt->mdt_child;
5625                 struct obd_ioctl_data   *data = karg;
5626                 struct lfsck_stop        stop;
5627
5628                 stop.ls_status = LS_STOPPED;
5629                 /* Old lfsck utils may pass NULL @stop. */
5630                 if (data->ioc_inlbuf1 == NULL)
5631                         stop.ls_flags = 0;
5632                 else
5633                         stop.ls_flags =
5634                         ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
5635
5636                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
5637                 break;
5638         }
5639         case OBD_IOC_GET_OBJ_VERSION: {
5640                 struct mdt_thread_info *mti;
5641                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5642                 memset(mti, 0, sizeof *mti);
5643                 mti->mti_env = &env;
5644                 mti->mti_mdt = mdt;
5645                 mti->mti_exp = exp;
5646
5647                 rc = mdt_ioc_version_get(mti, karg);
5648                 break;
5649         }
5650         case OBD_IOC_CATLOGLIST: {
5651                 struct mdt_thread_info *mti;
5652
5653                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5654                 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
5655                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
5656                                        &mti->mti_tmp_fid1);
5657                 break;
5658          }
5659         default:
5660                 rc = -EOPNOTSUPP;
5661                 CERROR("%s: Not supported cmd = %d, rc = %d\n",
5662                         mdt_obd_name(mdt), cmd, rc);
5663         }
5664
5665         lu_env_fini(&env);
5666         RETURN(rc);
5667 }
5668
5669 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
5670 {
5671         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
5672         struct lfsck_start_param lsp;
5673         int rc;
5674         ENTRY;
5675
5676         lsp.lsp_start = NULL;
5677         lsp.lsp_index_valid = 0;
5678         rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
5679                                                    OBD_IOC_START_LFSCK,
5680                                                    0, &lsp);
5681         if (rc != 0 && rc != -EALREADY)
5682                 CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
5683                       mdt_obd_name(mdt), rc);
5684
5685         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
5686         RETURN(rc);
5687 }
5688
5689 static int mdt_obd_postrecov(struct obd_device *obd)
5690 {
5691         struct lu_env env;
5692         int rc;
5693
5694         rc = lu_env_init(&env, LCT_MD_THREAD);
5695         if (rc)
5696                 RETURN(rc);
5697         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
5698         lu_env_fini(&env);
5699         return rc;
5700 }
5701
5702 static struct obd_ops mdt_obd_device_ops = {
5703         .o_owner          = THIS_MODULE,
5704         .o_set_info_async = mdt_obd_set_info_async,
5705         .o_connect        = mdt_obd_connect,
5706         .o_reconnect      = mdt_obd_reconnect,
5707         .o_disconnect     = mdt_obd_disconnect,
5708         .o_init_export    = mdt_init_export,
5709         .o_destroy_export = mdt_destroy_export,
5710         .o_iocontrol      = mdt_iocontrol,
5711         .o_postrecov      = mdt_obd_postrecov,
5712 };
5713
5714 static struct lu_device* mdt_device_fini(const struct lu_env *env,
5715                                          struct lu_device *d)
5716 {
5717         struct mdt_device *m = mdt_dev(d);
5718         ENTRY;
5719
5720         mdt_fini(env, m);
5721         RETURN(NULL);
5722 }
5723
5724 static struct lu_device *mdt_device_free(const struct lu_env *env,
5725                                          struct lu_device *d)
5726 {
5727         struct mdt_device *m = mdt_dev(d);
5728         ENTRY;
5729
5730         lu_device_fini(&m->mdt_lu_dev);
5731         OBD_FREE_PTR(m);
5732
5733         RETURN(NULL);
5734 }
5735
5736 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
5737                                           struct lu_device_type *t,
5738                                           struct lustre_cfg *cfg)
5739 {
5740         struct lu_device  *l;
5741         struct mdt_device *m;
5742
5743         OBD_ALLOC_PTR(m);
5744         if (m != NULL) {
5745                 int rc;
5746
5747                 l = &m->mdt_lu_dev;
5748                 rc = mdt_init0(env, m, t, cfg);
5749                 if (rc != 0) {
5750                         mdt_device_free(env, l);
5751                         l = ERR_PTR(rc);
5752                         return l;
5753                 }
5754         } else
5755                 l = ERR_PTR(-ENOMEM);
5756         return l;
5757 }
5758
5759 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
5760 LU_KEY_INIT(mdt, struct mdt_thread_info);
5761
5762 static void mdt_key_fini(const struct lu_context *ctx,
5763                          struct lu_context_key *key, void* data)
5764 {
5765         struct mdt_thread_info *info = data;
5766
5767         if (info->mti_big_lmm) {
5768                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
5769                 info->mti_big_lmm = NULL;
5770                 info->mti_big_lmmsize = 0;
5771         }
5772         OBD_FREE_PTR(info);
5773 }
5774
5775 /* context key: mdt_thread_key */
5776 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
5777
5778 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
5779 {
5780         return lu_ucred(info->mti_env);
5781 }
5782
5783 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
5784 {
5785         return lu_ucred_check(info->mti_env);
5786 }
5787
5788 /**
5789  * Enable/disable COS (Commit On Sharing).
5790  *
5791  * Set/Clear the COS flag in mdt options.
5792  *
5793  * \param mdt mdt device
5794  * \param val 0 disables COS, other values enable COS
5795  */
5796 void mdt_enable_cos(struct mdt_device *mdt, int val)
5797 {
5798         struct lu_env env;
5799         int rc;
5800
5801         mdt->mdt_opts.mo_cos = !!val;
5802         rc = lu_env_init(&env, LCT_LOCAL);
5803         if (unlikely(rc != 0)) {
5804                 CWARN("%s: lu_env initialization failed, cannot "
5805                       "sync: rc = %d\n", mdt_obd_name(mdt), rc);
5806                 return;
5807         }
5808         mdt_device_sync(&env, mdt);
5809         lu_env_fini(&env);
5810 }
5811
5812 /**
5813  * Check COS (Commit On Sharing) status.
5814  *
5815  * Return COS flag status.
5816  *
5817  * \param mdt mdt device
5818  */
5819 int mdt_cos_is_enabled(struct mdt_device *mdt)
5820 {
5821         return mdt->mdt_opts.mo_cos != 0;
5822 }
5823
5824 static struct lu_device_type_operations mdt_device_type_ops = {
5825         .ldto_device_alloc = mdt_device_alloc,
5826         .ldto_device_free  = mdt_device_free,
5827         .ldto_device_fini  = mdt_device_fini
5828 };
5829
5830 static struct lu_device_type mdt_device_type = {
5831         .ldt_tags     = LU_DEVICE_MD,
5832         .ldt_name     = LUSTRE_MDT_NAME,
5833         .ldt_ops      = &mdt_device_type_ops,
5834         .ldt_ctx_tags = LCT_MD_THREAD
5835 };
5836
5837 static int __init mdt_mod_init(void)
5838 {
5839         int rc;
5840
5841         CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") ==
5842                  FID_NOBRACE_LEN + 1);
5843         CLASSERT(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") ==
5844                  FID_LEN + 1);
5845         rc = lu_kmem_init(mdt_caches);
5846         if (rc)
5847                 return rc;
5848
5849         rc = mds_mod_init();
5850         if (rc)
5851                 GOTO(lu_fini, rc);
5852
5853         rc = class_register_type(&mdt_obd_device_ops, NULL, true, NULL,
5854                                  LUSTRE_MDT_NAME, &mdt_device_type);
5855         if (rc)
5856                 GOTO(mds_fini, rc);
5857 lu_fini:
5858         if (rc)
5859                 lu_kmem_fini(mdt_caches);
5860 mds_fini:
5861         if (rc)
5862                 mds_mod_exit();
5863         return rc;
5864 }
5865
5866 static void __exit mdt_mod_exit(void)
5867 {
5868         class_unregister_type(LUSTRE_MDT_NAME);
5869         mds_mod_exit();
5870         lu_kmem_fini(mdt_caches);
5871 }
5872
5873 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
5874 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
5875 MODULE_VERSION(LUSTRE_VERSION_STRING);
5876 MODULE_LICENSE("GPL");
5877
5878 module_init(mdt_mod_init);
5879 module_exit(mdt_mod_exit);