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, 2015, 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 enum ldlm_mode 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, enum ldlm_mode 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, enum ldlm_mode 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         if (!mdt_object_exists(child)) {
1440                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1441                                 &child->mot_obj,
1442                                 "Object doesn't exist!\n");
1443                 GOTO(out_child, rc = -ENOENT);
1444         }
1445
1446         rc = mdt_check_resent_lock(info, child, lhc);
1447         if (rc < 0) {
1448                 GOTO(out_child, rc);
1449         } else if (rc > 0) {
1450                 mdt_lock_handle_init(lhc);
1451                 mdt_lock_reg_init(lhc, LCK_PR);
1452                 try_layout = false;
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 if (KEY_IS(KEY_EVICT_BY_NID)) {
1634                 if (vallen > 0)
1635                         obd_export_evict_by_nid(req->rq_export->exp_obd, val);
1636         } else {
1637                 RETURN(-EINVAL);
1638         }
1639         RETURN(rc);
1640 }
1641
1642 static int mdt_readpage(struct tgt_session_info *tsi)
1643 {
1644         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
1645         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
1646         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
1647         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
1648         struct mdt_body         *repbody;
1649         int                      rc;
1650         int                      i;
1651
1652         ENTRY;
1653
1654         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1655                 RETURN(err_serious(-ENOMEM));
1656
1657         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
1658         if (repbody == NULL || reqbody == NULL)
1659                 RETURN(err_serious(-EFAULT));
1660
1661         /*
1662          * prepare @rdpg before calling lower layers and transfer itself. Here
1663          * reqbody->size contains offset of where to start to read and
1664          * reqbody->nlink contains number bytes to read.
1665          */
1666         rdpg->rp_hash = reqbody->mbo_size;
1667         if (rdpg->rp_hash != reqbody->mbo_size) {
1668                 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1669                        rdpg->rp_hash, reqbody->mbo_size);
1670                 RETURN(-EFAULT);
1671         }
1672
1673         rdpg->rp_attrs = reqbody->mbo_mode;
1674         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
1675                 rdpg->rp_attrs |= LUDA_64BITHASH;
1676         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
1677                                 exp_max_brw_size(tsi->tsi_exp));
1678         rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE - 1) >>
1679                           PAGE_CACHE_SHIFT;
1680         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1681         if (rdpg->rp_pages == NULL)
1682                 RETURN(-ENOMEM);
1683
1684         for (i = 0; i < rdpg->rp_npages; ++i) {
1685                 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1686                 if (rdpg->rp_pages[i] == NULL)
1687                         GOTO(free_rdpg, rc = -ENOMEM);
1688         }
1689
1690         /* call lower layers to fill allocated pages with directory data */
1691         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
1692         if (rc < 0)
1693                 GOTO(free_rdpg, rc);
1694
1695         /* send pages to client */
1696         rc = tgt_sendpage(tsi, rdpg, rc);
1697
1698         EXIT;
1699 free_rdpg:
1700
1701         for (i = 0; i < rdpg->rp_npages; i++)
1702                 if (rdpg->rp_pages[i] != NULL)
1703                         __free_page(rdpg->rp_pages[i]);
1704         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1705
1706         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1707                 RETURN(0);
1708
1709         return rc;
1710 }
1711
1712 static int mdt_reint_internal(struct mdt_thread_info *info,
1713                               struct mdt_lock_handle *lhc,
1714                               __u32 op)
1715 {
1716         struct req_capsule      *pill = info->mti_pill;
1717         struct mdt_body         *repbody;
1718         int                      rc = 0, rc2;
1719
1720         ENTRY;
1721
1722         rc = mdt_reint_unpack(info, op);
1723         if (rc != 0) {
1724                 CERROR("Can't unpack reint, rc %d\n", rc);
1725                 RETURN(err_serious(rc));
1726         }
1727
1728         /* for replay (no_create) lmm is not needed, client has it already */
1729         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1730                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1731                                      DEF_REP_MD_SIZE);
1732
1733         /* llog cookies are always 0, the field is kept for compatibility */
1734         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1735                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
1736
1737         rc = req_capsule_server_pack(pill);
1738         if (rc != 0) {
1739                 CERROR("Can't pack response, rc %d\n", rc);
1740                 RETURN(err_serious(rc));
1741         }
1742
1743         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1744                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1745                 LASSERT(repbody);
1746                 repbody->mbo_eadatasize = 0;
1747                 repbody->mbo_aclsize = 0;
1748         }
1749
1750         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1751
1752         /* for replay no cookkie / lmm need, because client have this already */
1753         if (info->mti_spec.no_create)
1754                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1755                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1756
1757         rc = mdt_init_ucred_reint(info);
1758         if (rc)
1759                 GOTO(out_shrink, rc);
1760
1761         rc = mdt_fix_attr_ucred(info, op);
1762         if (rc != 0)
1763                 GOTO(out_ucred, rc = err_serious(rc));
1764
1765         rc = mdt_check_resent(info, mdt_reconstruct, lhc);
1766         if (rc < 0) {
1767                 GOTO(out_ucred, rc);
1768         } else if (rc == 1) {
1769                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
1770                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1771                 GOTO(out_ucred, rc);
1772         }
1773         rc = mdt_reint_rec(info, lhc);
1774         EXIT;
1775 out_ucred:
1776         mdt_exit_ucred(info);
1777 out_shrink:
1778         mdt_client_compatibility(info);
1779         rc2 = mdt_fix_reply(info);
1780         if (rc == 0)
1781                 rc = rc2;
1782         return rc;
1783 }
1784
1785 static long mdt_reint_opcode(struct ptlrpc_request *req,
1786                              const struct req_format **fmt)
1787 {
1788         struct mdt_device       *mdt;
1789         struct mdt_rec_reint    *rec;
1790         long                     opc;
1791
1792         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
1793         if (rec != NULL) {
1794                 opc = rec->rr_opcode;
1795                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
1796                 if (opc < REINT_MAX && fmt[opc] != NULL)
1797                         req_capsule_extend(&req->rq_pill, fmt[opc]);
1798                 else {
1799                         mdt = mdt_exp2dev(req->rq_export);
1800                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
1801                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
1802                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
1803                         opc = err_serious(-EFAULT);
1804                 }
1805         } else {
1806                 opc = err_serious(-EFAULT);
1807         }
1808         return opc;
1809 }
1810
1811 static int mdt_reint(struct tgt_session_info *tsi)
1812 {
1813         long opc;
1814         int  rc;
1815         static const struct req_format *reint_fmts[REINT_MAX] = {
1816                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
1817                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
1818                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
1819                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
1820                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
1821                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
1822                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
1823                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
1824                 [REINT_MIGRATE]  = &RQF_MDS_REINT_RENAME
1825         };
1826
1827         ENTRY;
1828
1829         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
1830         if (opc >= 0) {
1831                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1832                 /*
1833                  * No lock possible here from client to pass it to reint code
1834                  * path.
1835                  */
1836                 rc = mdt_reint_internal(info, NULL, opc);
1837                 mdt_thread_info_fini(info);
1838         } else {
1839                 rc = opc;
1840         }
1841
1842         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1843         RETURN(rc);
1844 }
1845
1846 /* this should sync the whole device */
1847 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1848 {
1849         struct dt_device *dt = mdt->mdt_bottom;
1850         int rc;
1851         ENTRY;
1852
1853         rc = dt->dd_ops->dt_sync(env, dt);
1854         RETURN(rc);
1855 }
1856
1857 /* this should sync this object */
1858 static int mdt_object_sync(struct mdt_thread_info *info)
1859 {
1860         struct md_object *next;
1861         int rc;
1862         ENTRY;
1863
1864         if (!mdt_object_exists(info->mti_object)) {
1865                 CWARN("%s: non existing object "DFID": rc = %d\n",
1866                       mdt_obd_name(info->mti_mdt),
1867                       PFID(mdt_object_fid(info->mti_object)), -ESTALE);
1868                 RETURN(-ESTALE);
1869         }
1870         next = mdt_object_child(info->mti_object);
1871         rc = mo_object_sync(info->mti_env, next);
1872
1873         RETURN(rc);
1874 }
1875
1876 static int mdt_sync(struct tgt_session_info *tsi)
1877 {
1878         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1879         struct req_capsule      *pill = tsi->tsi_pill;
1880         struct mdt_body         *body;
1881         int                      rc;
1882
1883         ENTRY;
1884
1885         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1886                 RETURN(err_serious(-ENOMEM));
1887
1888         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
1889                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
1890         } else {
1891                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1892
1893                 /* sync an object */
1894                 rc = mdt_object_sync(info);
1895                 if (rc == 0) {
1896                         const struct lu_fid *fid;
1897                         struct lu_attr *la = &info->mti_attr.ma_attr;
1898
1899                         info->mti_attr.ma_need = MA_INODE;
1900                         info->mti_attr.ma_valid = 0;
1901                         rc = mdt_attr_get_complex(info, info->mti_object,
1902                                                   &info->mti_attr);
1903                         if (rc == 0) {
1904                                 body = req_capsule_server_get(pill,
1905                                                               &RMF_MDT_BODY);
1906                                 fid = mdt_object_fid(info->mti_object);
1907                                 mdt_pack_attr2body(info, body, la, fid);
1908                         }
1909                 }
1910                 mdt_thread_info_fini(info);
1911         }
1912         if (rc == 0)
1913                 mdt_counter_incr(req, LPROC_MDT_SYNC);
1914
1915         RETURN(rc);
1916 }
1917
1918 /*
1919  * Handle quota control requests to consult current usage/limit, but also
1920  * to configure quota enforcement
1921  */
1922 static int mdt_quotactl(struct tgt_session_info *tsi)
1923 {
1924         struct obd_export       *exp  = tsi->tsi_exp;
1925         struct req_capsule      *pill = tsi->tsi_pill;
1926         struct obd_quotactl     *oqctl, *repoqc;
1927         int                      id, rc;
1928         struct mdt_device       *mdt = mdt_exp2dev(exp);
1929         struct lu_device        *qmt = mdt->mdt_qmt_dev;
1930         struct lu_nodemap       *nodemap = exp->exp_target_data.ted_nodemap;
1931         ENTRY;
1932
1933         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1934         if (oqctl == NULL)
1935                 RETURN(err_serious(-EPROTO));
1936
1937         rc = req_capsule_server_pack(pill);
1938         if (rc)
1939                 RETURN(err_serious(rc));
1940
1941         switch (oqctl->qc_cmd) {
1942                 /* master quotactl */
1943         case Q_SETINFO:
1944         case Q_SETQUOTA:
1945                 if (!nodemap_can_setquota(nodemap))
1946                         RETURN(-EPERM);
1947         case Q_GETINFO:
1948         case Q_GETQUOTA:
1949                 if (qmt == NULL)
1950                         RETURN(-EOPNOTSUPP);
1951                 /* slave quotactl */
1952         case Q_GETOINFO:
1953         case Q_GETOQUOTA:
1954                 break;
1955         default:
1956                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
1957                 RETURN(-EFAULT);
1958         }
1959
1960         /* map uid/gid for remote client */
1961         id = oqctl->qc_id;
1962         if (exp_connect_rmtclient(exp)) {
1963                 struct lustre_idmap_table *idmap;
1964
1965                 idmap = exp->exp_mdt_data.med_idmap;
1966
1967                 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
1968                              oqctl->qc_cmd != Q_GETINFO))
1969                         RETURN(-EPERM);
1970
1971                 if (oqctl->qc_type == USRQUOTA)
1972                         id = lustre_idmap_lookup_uid(NULL, idmap, 0,
1973                                                      oqctl->qc_id);
1974                 else if (oqctl->qc_type == GRPQUOTA)
1975                         id = lustre_idmap_lookup_gid(NULL, idmap, 0,
1976                                                      oqctl->qc_id);
1977                 else
1978                         RETURN(-EINVAL);
1979
1980                 if (id == CFS_IDMAP_NOTFOUND) {
1981                         CDEBUG(D_QUOTA, "no mapping for id %u\n", oqctl->qc_id);
1982                         RETURN(-EACCES);
1983                 }
1984         }
1985
1986         if (oqctl->qc_type == USRQUOTA)
1987                 id = nodemap_map_id(nodemap, NODEMAP_UID,
1988                                     NODEMAP_CLIENT_TO_FS, id);
1989         else if (oqctl->qc_type == GRPQUOTA)
1990                 id = nodemap_map_id(nodemap, NODEMAP_UID,
1991                                     NODEMAP_CLIENT_TO_FS, id);
1992
1993         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
1994         if (repoqc == NULL)
1995                 RETURN(err_serious(-EFAULT));
1996
1997         if (oqctl->qc_id != id)
1998                 swap(oqctl->qc_id, id);
1999
2000         switch (oqctl->qc_cmd) {
2001
2002         case Q_GETINFO:
2003         case Q_SETINFO:
2004         case Q_SETQUOTA:
2005         case Q_GETQUOTA:
2006                 /* forward quotactl request to QMT */
2007                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2008                 break;
2009
2010         case Q_GETOINFO:
2011         case Q_GETOQUOTA:
2012                 /* slave quotactl */
2013                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2014                                    oqctl);
2015                 break;
2016
2017         default:
2018                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2019                 RETURN(-EFAULT);
2020         }
2021
2022         if (oqctl->qc_id != id)
2023                 swap(oqctl->qc_id, id);
2024
2025         *repoqc = *oqctl;
2026         RETURN(rc);
2027 }
2028
2029 /** clone llog ctxt from child (mdd)
2030  * This allows remote llog (replicator) access.
2031  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2032  * context was originally set up, or we can handle them directly.
2033  * I choose the latter, but that means I need any llog
2034  * contexts set up by child to be accessable by the mdt.  So we clone the
2035  * context into our context list here.
2036  */
2037 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2038                                int idx)
2039 {
2040         struct md_device  *next = mdt->mdt_child;
2041         struct llog_ctxt *ctxt;
2042         int rc;
2043
2044         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2045                 return 0;
2046
2047         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2048         if (rc || ctxt == NULL) {
2049                 return 0;
2050         }
2051
2052         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2053         if (rc)
2054                 CERROR("Can't set mdt ctxt %d\n", rc);
2055
2056         return rc;
2057 }
2058
2059 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2060                                  struct mdt_device *mdt, int idx)
2061 {
2062         struct llog_ctxt *ctxt;
2063
2064         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2065         if (ctxt == NULL)
2066                 return 0;
2067         /* Put once for the get we just did, and once for the clone */
2068         llog_ctxt_put(ctxt);
2069         llog_ctxt_put(ctxt);
2070         return 0;
2071 }
2072
2073 /*
2074  * sec context handlers
2075  */
2076 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2077 {
2078         int rc;
2079
2080         rc = mdt_handle_idmap(tsi);
2081         if (unlikely(rc)) {
2082                 struct ptlrpc_request   *req = tgt_ses_req(tsi);
2083                 __u32                    opc;
2084
2085                 opc = lustre_msg_get_opc(req->rq_reqmsg);
2086                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2087                         sptlrpc_svc_ctx_invalidate(req);
2088         }
2089
2090         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2091
2092         return rc;
2093 }
2094
2095 /*
2096  * quota request handlers
2097  */
2098 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2099 {
2100         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
2101         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2102         int                      rc;
2103         ENTRY;
2104
2105         if (qmt == NULL)
2106                 RETURN(err_serious(-EOPNOTSUPP));
2107
2108         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2109         RETURN(rc);
2110 }
2111
2112 struct mdt_object *mdt_object_new(const struct lu_env *env,
2113                                   struct mdt_device *d,
2114                                   const struct lu_fid *f)
2115 {
2116         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2117         struct lu_object *o;
2118         struct mdt_object *m;
2119         ENTRY;
2120
2121         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2122         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2123         if (unlikely(IS_ERR(o)))
2124                 m = (struct mdt_object *)o;
2125         else
2126                 m = mdt_obj(o);
2127         RETURN(m);
2128 }
2129
2130 struct mdt_object *mdt_object_find(const struct lu_env *env,
2131                                    struct mdt_device *d,
2132                                    const struct lu_fid *f)
2133 {
2134         struct lu_object *o;
2135         struct mdt_object *m;
2136         ENTRY;
2137
2138         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2139         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2140         if (unlikely(IS_ERR(o)))
2141                 m = (struct mdt_object *)o;
2142         else
2143                 m = mdt_obj(o);
2144
2145         RETURN(m);
2146 }
2147
2148 /**
2149  * Asyncronous commit for mdt device.
2150  *
2151  * Pass asynchonous commit call down the MDS stack.
2152  *
2153  * \param env environment
2154  * \param mdt the mdt device
2155  */
2156 static void mdt_device_commit_async(const struct lu_env *env,
2157                                     struct mdt_device *mdt)
2158 {
2159         struct dt_device *dt = mdt->mdt_bottom;
2160         int rc;
2161
2162         rc = dt->dd_ops->dt_commit_async(env, dt);
2163         if (unlikely(rc != 0))
2164                 CWARN("%s: async commit start failed: rc = %d\n",
2165                       mdt_obd_name(mdt), rc);
2166 }
2167
2168 /**
2169  * Mark the lock as "synchonous".
2170  *
2171  * Mark the lock to deffer transaction commit to the unlock time.
2172  *
2173  * \param lock the lock to mark as "synchonous"
2174  *
2175  * \see mdt_is_lock_sync
2176  * \see mdt_save_lock
2177  */
2178 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2179 {
2180         lock->l_ast_data = (void*)1;
2181 }
2182
2183 /**
2184  * Check whehter the lock "synchonous" or not.
2185  *
2186  * \param lock the lock to check
2187  * \retval 1 the lock is "synchonous"
2188  * \retval 0 the lock isn't "synchronous"
2189  *
2190  * \see mdt_set_lock_sync
2191  * \see mdt_save_lock
2192  */
2193 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2194 {
2195         return lock->l_ast_data != NULL;
2196 }
2197
2198 /**
2199  * Blocking AST for mdt locks.
2200  *
2201  * Starts transaction commit if in case of COS lock conflict or
2202  * deffers such a commit to the mdt_save_lock.
2203  *
2204  * \param lock the lock which blocks a request or cancelling lock
2205  * \param desc unused
2206  * \param data unused
2207  * \param flag indicates whether this cancelling or blocking callback
2208  * \retval 0
2209  * \see ldlm_blocking_ast_nocheck
2210  */
2211 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2212                      void *data, int flag)
2213 {
2214         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2215         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2216         int rc;
2217         ENTRY;
2218
2219         if (flag == LDLM_CB_CANCELING)
2220                 RETURN(0);
2221         lock_res_and_lock(lock);
2222         if (lock->l_blocking_ast != mdt_blocking_ast) {
2223                 unlock_res_and_lock(lock);
2224                 RETURN(0);
2225         }
2226         if (mdt_cos_is_enabled(mdt) &&
2227             lock->l_req_mode & (LCK_PW | LCK_EX) &&
2228             lock->l_blocking_lock != NULL &&
2229             lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
2230                 mdt_set_lock_sync(lock);
2231         }
2232         rc = ldlm_blocking_ast_nocheck(lock);
2233
2234         /* There is no lock conflict if l_blocking_lock == NULL,
2235          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2236          * when the last reference to a local lock was released */
2237         if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2238                 struct lu_env env;
2239
2240                 rc = lu_env_init(&env, LCT_LOCAL);
2241                 if (unlikely(rc != 0))
2242                         CWARN("%s: lu_env initialization failed, cannot "
2243                               "start asynchronous commit: rc = %d\n",
2244                               obd->obd_name, rc);
2245                 else
2246                         mdt_device_commit_async(&env, mdt);
2247                 lu_env_fini(&env);
2248         }
2249         RETURN(rc);
2250 }
2251
2252 /* Used for cross-MDT lock */
2253 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2254                             void *data, int flag)
2255 {
2256         struct lustre_handle lockh;
2257         int               rc;
2258
2259         switch (flag) {
2260         case LDLM_CB_BLOCKING:
2261                 ldlm_lock2handle(lock, &lockh);
2262                 rc = ldlm_cli_cancel(&lockh,
2263                         ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
2264                 if (rc < 0) {
2265                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2266                         RETURN(rc);
2267                 }
2268                 break;
2269         case LDLM_CB_CANCELING:
2270                 LDLM_DEBUG(lock, "Revoke remote lock\n");
2271                 break;
2272         default:
2273                 LBUG();
2274         }
2275         RETURN(0);
2276 }
2277
2278 int mdt_check_resent_lock(struct mdt_thread_info *info,
2279                           struct mdt_object *mo,
2280                           struct mdt_lock_handle *lhc)
2281 {
2282         /* the lock might already be gotten in ldlm_handle_enqueue() */
2283         if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
2284                 struct ptlrpc_request *req = mdt_info_req(info);
2285                 struct ldlm_lock      *lock;
2286
2287                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2288                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
2289                 if (lock == NULL) {
2290                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
2291                          * a resend case, however, it could be already destroyed
2292                          * due to client eviction or a raced cancel RPC. */
2293                         LDLM_DEBUG_NOLOCK("Invalid lock handle "LPX64"\n",
2294                                           lhc->mlh_reg_lh.cookie);
2295                         RETURN(-ESTALE);
2296                 }
2297
2298                 if (!fid_res_name_eq(mdt_object_fid(mo),
2299                                      &lock->l_resource->lr_name)) {
2300                         CWARN("%s: Although resent, but still not "
2301                               "get child lock:"DFID"\n",
2302                               info->mti_exp->exp_obd->obd_name,
2303                               PFID(mdt_object_fid(mo)));
2304                         LDLM_LOCK_PUT(lock);
2305                         RETURN(-EPROTO);
2306                 }
2307                 LDLM_LOCK_PUT(lock);
2308                 return 0;
2309         }
2310         return 1;
2311 }
2312
2313 int mdt_remote_object_lock(struct mdt_thread_info *mti, struct mdt_object *o,
2314                            const struct lu_fid *fid, struct lustre_handle *lh,
2315                            enum ldlm_mode mode, __u64 ibits, bool nonblock)
2316 {
2317         struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
2318         union ldlm_policy_data *policy = &mti->mti_policy;
2319         struct ldlm_res_id *res_id = &mti->mti_res_id;
2320         int rc = 0;
2321         ENTRY;
2322
2323         LASSERT(mdt_object_remote(o));
2324
2325         fid_build_reg_res_name(fid, res_id);
2326
2327         memset(einfo, 0, sizeof(*einfo));
2328         einfo->ei_type = LDLM_IBITS;
2329         einfo->ei_mode = mode;
2330         einfo->ei_cb_bl = mdt_remote_blocking_ast;
2331         einfo->ei_cb_cp = ldlm_completion_ast;
2332         einfo->ei_enq_slave = 0;
2333         einfo->ei_res_id = res_id;
2334         if (nonblock)
2335                 einfo->ei_nonblock = 1;
2336
2337         memset(policy, 0, sizeof(*policy));
2338         policy->l_inodebits.bits = ibits;
2339
2340         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2341                             policy);
2342         RETURN(rc);
2343 }
2344
2345 static int mdt_object_local_lock(struct mdt_thread_info *info,
2346                                  struct mdt_object *o,
2347                                  struct mdt_lock_handle *lh, __u64 ibits,
2348                                  bool nonblock)
2349 {
2350         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2351         union ldlm_policy_data *policy = &info->mti_policy;
2352         struct ldlm_res_id *res_id = &info->mti_res_id;
2353         __u64 dlmflags;
2354         int rc;
2355         ENTRY;
2356
2357         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2358         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2359         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2360         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2361
2362         /* Only enqueue LOOKUP lock for remote object */
2363         if (mdt_object_remote(o))
2364                 LASSERT(ibits == MDS_INODELOCK_LOOKUP);
2365
2366         if (lh->mlh_type == MDT_PDO_LOCK) {
2367                 /* check for exists after object is locked */
2368                 if (mdt_object_exists(o) == 0) {
2369                         /* Non-existent object shouldn't have PDO lock */
2370                         RETURN(-ESTALE);
2371                 } else {
2372                         /* Non-dir object shouldn't have PDO lock */
2373                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2374                                 RETURN(-ENOTDIR);
2375                 }
2376         }
2377
2378         memset(policy, 0, sizeof(*policy));
2379         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2380
2381         dlmflags = LDLM_FL_ATOMIC_CB;
2382         if (nonblock)
2383                 dlmflags |= LDLM_FL_BLOCK_NOWAIT;
2384
2385         /*
2386          * Take PDO lock on whole directory and build correct @res_id for lock
2387          * on part of directory.
2388          */
2389         if (lh->mlh_pdo_hash != 0) {
2390                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2391                 mdt_lock_pdo_mode(info, o, lh);
2392                 if (lh->mlh_pdo_mode != LCK_NL) {
2393                         /*
2394                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2395                          * is never going to be sent to client and we do not
2396                          * want it slowed down due to possible cancels.
2397                          */
2398                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2399                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2400                                           policy, res_id, dlmflags,
2401                                           info->mti_exp == NULL ? NULL :
2402                                           &info->mti_exp->exp_handle.h_cookie);
2403                         if (unlikely(rc != 0))
2404                                 GOTO(out_unlock, rc);
2405                 }
2406
2407                 /*
2408                  * Finish res_id initializing by name hash marking part of
2409                  * directory which is taking modification.
2410                  */
2411                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2412         }
2413
2414         policy->l_inodebits.bits = ibits;
2415
2416         /*
2417          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2418          * going to be sent to client. If it is - mdt_intent_policy() path will
2419          * fix it up and turn FL_LOCAL flag off.
2420          */
2421         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2422                           res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2423                           info->mti_exp == NULL ? NULL :
2424                           &info->mti_exp->exp_handle.h_cookie);
2425 out_unlock:
2426         if (rc != 0)
2427                 mdt_object_unlock(info, o, lh, 1);
2428         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2429                    lh->mlh_pdo_hash != 0 &&
2430                    (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
2431                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2432
2433         RETURN(rc);
2434 }
2435
2436 static int
2437 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
2438                          struct mdt_lock_handle *lh, __u64 ibits,
2439                          bool nonblock)
2440 {
2441         struct mdt_lock_handle *local_lh = NULL;
2442         int rc;
2443         ENTRY;
2444
2445         if (!mdt_object_remote(o))
2446                 return mdt_object_local_lock(info, o, lh, ibits, nonblock);
2447
2448         /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
2449         ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
2450                    MDS_INODELOCK_XATTR);
2451
2452         /* Only enqueue LOOKUP lock for remote object */
2453         if (ibits & MDS_INODELOCK_LOOKUP) {
2454                 rc = mdt_object_local_lock(info, o, lh,
2455                                            MDS_INODELOCK_LOOKUP,
2456                                            nonblock);
2457                 if (rc != ELDLM_OK)
2458                         RETURN(rc);
2459
2460                 local_lh = lh;
2461         }
2462
2463         if (ibits & MDS_INODELOCK_UPDATE) {
2464                 /* Sigh, PDO needs to enqueue 2 locks right now, but
2465                  * enqueue RPC can only request 1 lock, to avoid extra
2466                  * RPC, so it will instead enqueue EX lock for remote
2467                  * object anyway XXX*/
2468                 if (lh->mlh_type == MDT_PDO_LOCK &&
2469                     lh->mlh_pdo_hash != 0) {
2470                         CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
2471                                "EX lock.\n", mdt_obd_name(info->mti_mdt),
2472                                PFID(mdt_object_fid(o)));
2473                         lh->mlh_pdo_hash = 0;
2474                         lh->mlh_rreg_mode = LCK_EX;
2475                         lh->mlh_type = MDT_REG_LOCK;
2476                 }
2477                 rc = mdt_remote_object_lock(info, o, mdt_object_fid(o),
2478                                             &lh->mlh_rreg_lh,
2479                                             lh->mlh_rreg_mode,
2480                                             MDS_INODELOCK_UPDATE, nonblock);
2481                 if (rc != ELDLM_OK) {
2482                         if (local_lh != NULL)
2483                                 mdt_object_unlock(info, o, local_lh, rc);
2484                         RETURN(rc);
2485                 }
2486         }
2487
2488         RETURN(0);
2489 }
2490
2491 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2492                     struct mdt_lock_handle *lh, __u64 ibits)
2493 {
2494         return mdt_object_lock_internal(info, o, lh, ibits, false);
2495 }
2496
2497 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
2498                         struct mdt_lock_handle *lh, __u64 ibits)
2499 {
2500         struct mdt_lock_handle tmp = *lh;
2501         int rc;
2502
2503         rc = mdt_object_lock_internal(info, o, &tmp, ibits, true);
2504         if (rc == 0)
2505                 *lh = tmp;
2506
2507         return rc == 0;
2508 }
2509
2510 /**
2511  * Save a lock within request object.
2512  *
2513  * Keep the lock referenced until whether client ACK or transaction
2514  * commit happens or release the lock immediately depending on input
2515  * parameters. If COS is ON, a write lock is converted to COS lock
2516  * before saving.
2517  *
2518  * \param info thead info object
2519  * \param h lock handle
2520  * \param mode lock mode
2521  * \param decref force immediate lock releasing
2522  */
2523 static void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2524                           enum ldlm_mode mode, int decref)
2525 {
2526         ENTRY;
2527
2528         if (lustre_handle_is_used(h)) {
2529                 if (decref || !info->mti_has_trans ||
2530                     !(mode & (LCK_PW | LCK_EX))) {
2531                         mdt_fid_unlock(h, mode);
2532                 } else {
2533                         struct mdt_device *mdt = info->mti_mdt;
2534                         struct ldlm_lock *lock = ldlm_handle2lock(h);
2535                         struct ptlrpc_request *req = mdt_info_req(info);
2536                         int no_ack = 0;
2537
2538                         LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2539                                  h->cookie);
2540                         /* there is no request if mdt_object_unlock() is called
2541                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
2542                         if (likely(req != NULL)) {
2543                                 CDEBUG(D_HA, "request = %p reply state = %p"
2544                                        " transno = "LPD64"\n", req,
2545                                        req->rq_reply_state, req->rq_transno);
2546                                 if (mdt_cos_is_enabled(mdt)) {
2547                                         no_ack = 1;
2548                                         ldlm_lock_downgrade(lock, LCK_COS);
2549                                         mode = LCK_COS;
2550                                 }
2551                                 ptlrpc_save_lock(req, h, mode, no_ack);
2552                         } else {
2553                                 ldlm_lock_decref(h, mode);
2554                         }
2555                         if (mdt_is_lock_sync(lock)) {
2556                                 CDEBUG(D_HA, "found sync-lock,"
2557                                        " async commit started\n");
2558                                 mdt_device_commit_async(info->mti_env,
2559                                                         mdt);
2560                         }
2561                         LDLM_LOCK_PUT(lock);
2562                 }
2563                 h->cookie = 0ull;
2564         }
2565
2566         EXIT;
2567 }
2568
2569 /**
2570  * Unlock mdt object.
2571  *
2572  * Immeditely release the regular lock and the PDO lock or save the
2573  * lock in request and keep them referenced until client ACK or
2574  * transaction commit.
2575  *
2576  * \param info thread info object
2577  * \param o mdt object
2578  * \param lh mdt lock handle referencing regular and PDO locks
2579  * \param decref force immediate lock releasing
2580  */
2581 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2582                        struct mdt_lock_handle *lh, int decref)
2583 {
2584         ENTRY;
2585
2586         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2587         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2588
2589         if (lustre_handle_is_used(&lh->mlh_rreg_lh))
2590                 ldlm_lock_decref(&lh->mlh_rreg_lh, lh->mlh_rreg_mode);
2591
2592         EXIT;
2593 }
2594
2595 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2596                                         const struct lu_fid *f,
2597                                         struct mdt_lock_handle *lh,
2598                                         __u64 ibits)
2599 {
2600         struct mdt_object *o;
2601
2602         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2603         if (!IS_ERR(o)) {
2604                 int rc;
2605
2606                 rc = mdt_object_lock(info, o, lh, ibits);
2607                 if (rc != 0) {
2608                         mdt_object_put(info->mti_env, o);
2609                         o = ERR_PTR(rc);
2610                 }
2611         }
2612         return o;
2613 }
2614
2615 void mdt_object_unlock_put(struct mdt_thread_info * info,
2616                            struct mdt_object * o,
2617                            struct mdt_lock_handle *lh,
2618                            int decref)
2619 {
2620         mdt_object_unlock(info, o, lh, decref);
2621         mdt_object_put(info->mti_env, o);
2622 }
2623
2624 /*
2625  * Generic code handling requests that have struct mdt_body passed in:
2626  *
2627  *  - extract mdt_body from request and save it in @info, if present;
2628  *
2629  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
2630  *  @info;
2631  *
2632  *  - if HABEO_CORPUS flag is set for this request type check whether object
2633  *  actually exists on storage (lu_object_exists()).
2634  *
2635  */
2636 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2637 {
2638         const struct mdt_body    *body;
2639         struct mdt_object        *obj;
2640         const struct lu_env      *env;
2641         struct req_capsule       *pill;
2642         int                       rc;
2643         ENTRY;
2644
2645         env = info->mti_env;
2646         pill = info->mti_pill;
2647
2648         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2649         if (body == NULL)
2650                 RETURN(-EFAULT);
2651
2652         if (!(body->mbo_valid & OBD_MD_FLID))
2653                 RETURN(0);
2654
2655         if (!fid_is_sane(&body->mbo_fid1)) {
2656                 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
2657                 RETURN(-EINVAL);
2658         }
2659
2660         obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
2661         if (!IS_ERR(obj)) {
2662                 if ((flags & HABEO_CORPUS) && !mdt_object_exists(obj)) {
2663                         mdt_object_put(env, obj);
2664                         rc = -ENOENT;
2665                 } else {
2666                         info->mti_object = obj;
2667                         rc = 0;
2668                 }
2669         } else
2670                 rc = PTR_ERR(obj);
2671
2672         RETURN(rc);
2673 }
2674
2675 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2676 {
2677         struct req_capsule *pill = info->mti_pill;
2678         int rc;
2679         ENTRY;
2680
2681         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2682                 rc = mdt_body_unpack(info, flags);
2683         else
2684                 rc = 0;
2685
2686         if (rc == 0 && (flags & HABEO_REFERO)) {
2687                 /* Pack reply. */
2688                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2689                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2690                                              DEF_REP_MD_SIZE);
2691                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2692                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
2693                                              RCL_SERVER, 0);
2694
2695                 rc = req_capsule_server_pack(pill);
2696         }
2697         RETURN(rc);
2698 }
2699
2700 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2701 {
2702         lh->mlh_type = MDT_NUL_LOCK;
2703         lh->mlh_reg_lh.cookie = 0ull;
2704         lh->mlh_reg_mode = LCK_MINMODE;
2705         lh->mlh_pdo_lh.cookie = 0ull;
2706         lh->mlh_pdo_mode = LCK_MINMODE;
2707         lh->mlh_rreg_lh.cookie = 0ull;
2708         lh->mlh_rreg_mode = LCK_MINMODE;
2709 }
2710
2711 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2712 {
2713         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2714         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2715 }
2716
2717 /*
2718  * Initialize fields of struct mdt_thread_info. Other fields are left in
2719  * uninitialized state, because it's too expensive to zero out whole
2720  * mdt_thread_info (> 1K) on each request arrival.
2721  */
2722 void mdt_thread_info_init(struct ptlrpc_request *req,
2723                           struct mdt_thread_info *info)
2724 {
2725         int i;
2726
2727         info->mti_pill = &req->rq_pill;
2728
2729         /* lock handle */
2730         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2731                 mdt_lock_handle_init(&info->mti_lh[i]);
2732
2733         /* mdt device: it can be NULL while CONNECT */
2734         if (req->rq_export) {
2735                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2736                 info->mti_exp = req->rq_export;
2737         } else
2738                 info->mti_mdt = NULL;
2739         info->mti_env = req->rq_svc_thread->t_env;
2740         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2741
2742         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2743         info->mti_big_buf = LU_BUF_NULL;
2744         info->mti_body = NULL;
2745         info->mti_object = NULL;
2746         info->mti_dlm_req = NULL;
2747         info->mti_has_trans = 0;
2748         info->mti_cross_ref = 0;
2749         info->mti_opdata = 0;
2750         info->mti_big_lmm_used = 0;
2751
2752         info->mti_spec.no_create = 0;
2753         info->mti_spec.sp_rm_entry = 0;
2754         info->mti_spec.sp_permitted = 0;
2755         info->mti_spec.sp_migrate_close = 0;
2756
2757         info->mti_spec.u.sp_ea.eadata = NULL;
2758         info->mti_spec.u.sp_ea.eadatalen = 0;
2759 }
2760
2761 void mdt_thread_info_fini(struct mdt_thread_info *info)
2762 {
2763         int i;
2764
2765         if (info->mti_object != NULL) {
2766                 mdt_object_put(info->mti_env, info->mti_object);
2767                 info->mti_object = NULL;
2768         }
2769
2770         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2771                 mdt_lock_handle_fini(&info->mti_lh[i]);
2772         info->mti_env = NULL;
2773         info->mti_pill = NULL;
2774         info->mti_exp = NULL;
2775
2776         if (unlikely(info->mti_big_buf.lb_buf != NULL))
2777                 lu_buf_free(&info->mti_big_buf);
2778 }
2779
2780 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
2781 {
2782         struct mdt_thread_info  *mti;
2783
2784         mti = mdt_th_info(tsi->tsi_env);
2785         LASSERT(mti != NULL);
2786
2787         mdt_thread_info_init(tgt_ses_req(tsi), mti);
2788         if (tsi->tsi_corpus != NULL) {
2789                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
2790                 lu_object_get(tsi->tsi_corpus);
2791         }
2792         mti->mti_body = tsi->tsi_mdt_body;
2793         mti->mti_dlm_req = tsi->tsi_dlm_req;
2794
2795         return mti;
2796 }
2797
2798 static int mdt_tgt_connect(struct tgt_session_info *tsi)
2799 {
2800         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2801         int                      rc;
2802
2803         ENTRY;
2804
2805         if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
2806             cfs_fail_val ==
2807             tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id) {
2808                 set_current_state(TASK_UNINTERRUPTIBLE);
2809                 schedule_timeout(msecs_to_jiffies(3 * MSEC_PER_SEC));
2810         }
2811
2812         rc = tgt_connect(tsi);
2813         if (rc != 0)
2814                 RETURN(rc);
2815
2816         rc = mdt_init_idmap(tsi);
2817         if (rc != 0)
2818                 GOTO(err, rc);
2819         RETURN(0);
2820 err:
2821         obd_disconnect(class_export_get(req->rq_export));
2822         return rc;
2823 }
2824
2825 enum mdt_it_code {
2826         MDT_IT_OPEN,
2827         MDT_IT_OCREAT,
2828         MDT_IT_CREATE,
2829         MDT_IT_GETATTR,
2830         MDT_IT_READDIR,
2831         MDT_IT_LOOKUP,
2832         MDT_IT_UNLINK,
2833         MDT_IT_TRUNC,
2834         MDT_IT_GETXATTR,
2835         MDT_IT_LAYOUT,
2836         MDT_IT_QUOTA,
2837         MDT_IT_NR
2838 };
2839
2840 static int mdt_intent_getattr(enum mdt_it_code opcode,
2841                               struct mdt_thread_info *info,
2842                               struct ldlm_lock **,
2843                               __u64);
2844
2845 static int mdt_intent_getxattr(enum mdt_it_code opcode,
2846                                 struct mdt_thread_info *info,
2847                                 struct ldlm_lock **lockp,
2848                                 __u64 flags);
2849
2850 static int mdt_intent_layout(enum mdt_it_code opcode,
2851                              struct mdt_thread_info *info,
2852                              struct ldlm_lock **,
2853                              __u64);
2854 static int mdt_intent_reint(enum mdt_it_code opcode,
2855                             struct mdt_thread_info *info,
2856                             struct ldlm_lock **,
2857                             __u64);
2858
2859 static struct mdt_it_flavor {
2860         const struct req_format *it_fmt;
2861         __u32                    it_flags;
2862         int                    (*it_act)(enum mdt_it_code ,
2863                                          struct mdt_thread_info *,
2864                                          struct ldlm_lock **,
2865                                          __u64);
2866         long                     it_reint;
2867 } mdt_it_flavor[] = {
2868         [MDT_IT_OPEN]     = {
2869                 .it_fmt   = &RQF_LDLM_INTENT,
2870                 /*.it_flags = HABEO_REFERO,*/
2871                 .it_flags = 0,
2872                 .it_act   = mdt_intent_reint,
2873                 .it_reint = REINT_OPEN
2874         },
2875         [MDT_IT_OCREAT]   = {
2876                 .it_fmt   = &RQF_LDLM_INTENT,
2877                 /*
2878                  * OCREAT is not a MUTABOR request as if the file
2879                  * already exists.
2880                  * We do the extra check of OBD_CONNECT_RDONLY in
2881                  * mdt_reint_open() when we really need to create
2882                  * the object.
2883                  */
2884                 .it_flags = 0,
2885                 .it_act   = mdt_intent_reint,
2886                 .it_reint = REINT_OPEN
2887         },
2888         [MDT_IT_CREATE]   = {
2889                 .it_fmt   = &RQF_LDLM_INTENT,
2890                 .it_flags = MUTABOR,
2891                 .it_act   = mdt_intent_reint,
2892                 .it_reint = REINT_CREATE
2893         },
2894         [MDT_IT_GETATTR]  = {
2895                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2896                 .it_flags = HABEO_REFERO,
2897                 .it_act   = mdt_intent_getattr
2898         },
2899         [MDT_IT_READDIR]  = {
2900                 .it_fmt   = NULL,
2901                 .it_flags = 0,
2902                 .it_act   = NULL
2903         },
2904         [MDT_IT_LOOKUP]   = {
2905                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2906                 .it_flags = HABEO_REFERO,
2907                 .it_act   = mdt_intent_getattr
2908         },
2909         [MDT_IT_UNLINK]   = {
2910                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
2911                 .it_flags = MUTABOR,
2912                 .it_act   = NULL,
2913                 .it_reint = REINT_UNLINK
2914         },
2915         [MDT_IT_TRUNC]    = {
2916                 .it_fmt   = NULL,
2917                 .it_flags = MUTABOR,
2918                 .it_act   = NULL
2919         },
2920         [MDT_IT_GETXATTR] = {
2921                 .it_fmt   = &RQF_LDLM_INTENT_GETXATTR,
2922                 .it_flags = HABEO_CORPUS,
2923                 .it_act   = mdt_intent_getxattr
2924         },
2925         [MDT_IT_LAYOUT] = {
2926                 .it_fmt   = &RQF_LDLM_INTENT_LAYOUT,
2927                 .it_flags = 0,
2928                 .it_act   = mdt_intent_layout
2929         }
2930 };
2931
2932 static int
2933 mdt_intent_lock_replace(struct mdt_thread_info *info,
2934                         struct ldlm_lock **lockp,
2935                         struct mdt_lock_handle *lh,
2936                         __u64 flags)
2937 {
2938         struct ptlrpc_request  *req = mdt_info_req(info);
2939         struct ldlm_lock       *lock = *lockp;
2940         struct ldlm_lock       *new_lock;
2941
2942         /* If possible resent found a lock, @lh is set to its handle */
2943         new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
2944
2945         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
2946                 lh->mlh_reg_lh.cookie = 0;
2947                 RETURN(0);
2948         }
2949
2950         LASSERTF(new_lock != NULL,
2951                  "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
2952
2953         /*
2954          * If we've already given this lock to a client once, then we should
2955          * have no readers or writers.  Otherwise, we should have one reader
2956          * _or_ writer ref (which will be zeroed below) before returning the
2957          * lock to a client.
2958          */
2959         if (new_lock->l_export == req->rq_export) {
2960                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
2961         } else {
2962                 LASSERT(new_lock->l_export == NULL);
2963                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
2964         }
2965
2966         *lockp = new_lock;
2967
2968         if (new_lock->l_export == req->rq_export) {
2969                 /*
2970                  * Already gave this to the client, which means that we
2971                  * reconstructed a reply.
2972                  */
2973                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
2974                         MSG_RESENT);
2975
2976                 LDLM_LOCK_RELEASE(new_lock);
2977                 lh->mlh_reg_lh.cookie = 0;
2978                 RETURN(ELDLM_LOCK_REPLACED);
2979         }
2980
2981         /*
2982          * Fixup the lock to be given to the client.
2983          */
2984         lock_res_and_lock(new_lock);
2985         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
2986          * possible blocking AST. */
2987         while (new_lock->l_readers > 0) {
2988                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
2989                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
2990                 new_lock->l_readers--;
2991         }
2992         while (new_lock->l_writers > 0) {
2993                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
2994                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
2995                 new_lock->l_writers--;
2996         }
2997
2998         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
2999         new_lock->l_blocking_ast = lock->l_blocking_ast;
3000         new_lock->l_completion_ast = lock->l_completion_ast;
3001         new_lock->l_remote_handle = lock->l_remote_handle;
3002         new_lock->l_flags &= ~LDLM_FL_LOCAL;
3003
3004         unlock_res_and_lock(new_lock);
3005
3006         cfs_hash_add(new_lock->l_export->exp_lock_hash,
3007                      &new_lock->l_remote_handle,
3008                      &new_lock->l_exp_hash);
3009
3010         LDLM_LOCK_RELEASE(new_lock);
3011         lh->mlh_reg_lh.cookie = 0;
3012
3013         RETURN(ELDLM_LOCK_REPLACED);
3014 }
3015
3016 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3017                                     struct ldlm_lock *new_lock,
3018                                     struct mdt_lock_handle *lh,
3019                                     __u64 flags)
3020 {
3021         struct ptlrpc_request  *req = mdt_info_req(info);
3022         struct ldlm_request    *dlmreq;
3023
3024         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3025                 return;
3026
3027         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3028
3029         /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
3030          * lock was found by ldlm_handle_enqueue(); if so @lh must be
3031          * initialized. */
3032         if (flags & LDLM_FL_RESENT) {
3033                 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
3034                 lh->mlh_reg_mode = new_lock->l_granted_mode;
3035
3036                 LDLM_DEBUG(new_lock, "Restoring lock cookie");
3037                 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie "LPX64,
3038                           lh->mlh_reg_lh.cookie);
3039                 return;
3040         }
3041
3042         /*
3043          * If the xid matches, then we know this is a resent request, and allow
3044          * it. (It's probably an OPEN, for which we don't send a lock.
3045          */
3046         if (req_can_reconstruct(req, NULL))
3047                 return;
3048
3049         /*
3050          * This remote handle isn't enqueued, so we never received or processed
3051          * this request.  Clear MSG_RESENT, because it can be handled like any
3052          * normal request now.
3053          */
3054         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3055
3056         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
3057                   dlmreq->lock_handle[0].cookie);
3058 }
3059
3060 static int mdt_intent_getxattr(enum mdt_it_code opcode,
3061                                 struct mdt_thread_info *info,
3062                                 struct ldlm_lock **lockp,
3063                                 __u64 flags)
3064 {
3065         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3066         struct ldlm_reply      *ldlm_rep = NULL;
3067         int rc, grc;
3068
3069         /*
3070          * Initialize lhc->mlh_reg_lh either from a previously granted lock
3071          * (for the resend case) or a new lock. Below we will use it to
3072          * replace the original lock.
3073          */
3074         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3075         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3076                 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3077                 rc = mdt_object_lock(info, info->mti_object, lhc,
3078                                      MDS_INODELOCK_XATTR);
3079                 if (rc)
3080                         return rc;
3081         }
3082
3083         grc = mdt_getxattr(info);
3084
3085         rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3086
3087         if (mdt_info_req(info)->rq_repmsg != NULL)
3088                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3089         if (ldlm_rep == NULL)
3090                 RETURN(err_serious(-EFAULT));
3091
3092         ldlm_rep->lock_policy_res2 = grc;
3093
3094         return rc;
3095 }
3096
3097 static int mdt_intent_getattr(enum mdt_it_code opcode,
3098                               struct mdt_thread_info *info,
3099                               struct ldlm_lock **lockp,
3100                               __u64 flags)
3101 {
3102         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3103         __u64                   child_bits;
3104         struct ldlm_reply      *ldlm_rep;
3105         struct mdt_body        *reqbody;
3106         struct mdt_body        *repbody;
3107         int                     rc, rc2;
3108         ENTRY;
3109
3110         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3111         LASSERT(reqbody);
3112
3113         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3114         LASSERT(repbody);
3115
3116         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
3117         repbody->mbo_eadatasize = 0;
3118         repbody->mbo_aclsize = 0;
3119
3120         switch (opcode) {
3121         case MDT_IT_LOOKUP:
3122                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3123                 break;
3124         case MDT_IT_GETATTR:
3125                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3126                              MDS_INODELOCK_PERM;
3127                 break;
3128         default:
3129                 CERROR("Unsupported intent (%d)\n", opcode);
3130                 GOTO(out_shrink, rc = -EINVAL);
3131         }
3132
3133         rc = mdt_init_ucred_intent_getattr(info, reqbody);
3134         if (rc)
3135                 GOTO(out_shrink, rc);
3136
3137         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3138         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3139
3140         /* Get lock from request for possible resent case. */
3141         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3142
3143         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3144         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3145
3146         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3147                 ldlm_rep->lock_policy_res2 = 0;
3148         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3149             ldlm_rep->lock_policy_res2) {
3150                 lhc->mlh_reg_lh.cookie = 0ull;
3151                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3152         }
3153
3154         rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3155         EXIT;
3156 out_ucred:
3157         mdt_exit_ucred(info);
3158 out_shrink:
3159         mdt_client_compatibility(info);
3160         rc2 = mdt_fix_reply(info);
3161         if (rc == 0)
3162                 rc = rc2;
3163         return rc;
3164 }
3165
3166 static int mdt_intent_layout(enum mdt_it_code opcode,
3167                              struct mdt_thread_info *info,
3168                              struct ldlm_lock **lockp,
3169                              __u64 flags)
3170 {
3171         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_LAYOUT];
3172         struct layout_intent *layout;
3173         struct lu_fid *fid;
3174         struct mdt_object *obj = NULL;
3175         int layout_size = 0;
3176         int rc = 0;
3177         ENTRY;
3178
3179         if (opcode != MDT_IT_LAYOUT) {
3180                 CERROR("%s: Unknown intent (%d)\n", mdt_obd_name(info->mti_mdt),
3181                         opcode);
3182                 RETURN(-EINVAL);
3183         }
3184
3185         layout = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3186         if (layout == NULL)
3187                 RETURN(-EPROTO);
3188
3189         if (layout->li_opc != LAYOUT_INTENT_ACCESS) {
3190                 CERROR("%s: Unsupported layout intent opc %d\n",
3191                        mdt_obd_name(info->mti_mdt), layout->li_opc);
3192                 RETURN(-EINVAL);
3193         }
3194
3195         fid = &info->mti_tmp_fid2;
3196         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3197
3198         /* Get lock from request for possible resent case. */
3199         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3200
3201         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3202         if (IS_ERR(obj))
3203                 GOTO(out, rc = PTR_ERR(obj));
3204
3205         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3206                 layout_size = mdt_attr_get_eabuf_size(info, obj);
3207                 if (layout_size < 0)
3208                         GOTO(out_obj, rc = layout_size);
3209
3210                 if (layout_size > info->mti_mdt->mdt_max_mdsize)
3211                         info->mti_mdt->mdt_max_mdsize = layout_size;
3212         }
3213
3214         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3215         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
3216                              layout_size);
3217         rc = req_capsule_server_pack(info->mti_pill);
3218         GOTO(out_obj, rc);
3219
3220 out_obj:
3221         mdt_object_put(info->mti_env, obj);
3222
3223         if (rc == 0 && lustre_handle_is_used(&lhc->mlh_reg_lh))
3224                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3225
3226 out:
3227         lhc->mlh_reg_lh.cookie = 0;
3228
3229         return rc;
3230 }
3231
3232 static int mdt_intent_reint(enum mdt_it_code opcode,
3233                             struct mdt_thread_info *info,
3234                             struct ldlm_lock **lockp,
3235                             __u64 flags)
3236 {
3237         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3238         struct ldlm_reply      *rep = NULL;
3239         long                    opc;
3240         int                     rc;
3241
3242         static const struct req_format *intent_fmts[REINT_MAX] = {
3243                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3244                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3245         };
3246
3247         ENTRY;
3248
3249         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
3250         if (opc < 0)
3251                 RETURN(opc);
3252
3253         if (mdt_it_flavor[opcode].it_reint != opc) {
3254                 CERROR("Reint code %ld doesn't match intent: %d\n",
3255                        opc, opcode);
3256                 RETURN(err_serious(-EPROTO));
3257         }
3258
3259         /* Get lock from request for possible resent case. */
3260         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3261
3262         rc = mdt_reint_internal(info, lhc, opc);
3263
3264         /* Check whether the reply has been packed successfully. */
3265         if (mdt_info_req(info)->rq_repmsg != NULL)
3266                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3267         if (rep == NULL)
3268                 RETURN(err_serious(-EFAULT));
3269
3270         /* MDC expects this in any case */
3271         if (rc != 0)
3272                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3273
3274         /* the open lock or the lock for cross-ref object should be
3275          * returned to the client */
3276         if (lustre_handle_is_used(&lhc->mlh_reg_lh) &&
3277             (rc == 0 || rc == -MDT_EREMOTE_OPEN)) {
3278                 rep->lock_policy_res2 = 0;
3279                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3280                 RETURN(rc);
3281         }
3282
3283         rep->lock_policy_res2 = clear_serious(rc);
3284
3285         if (rep->lock_policy_res2 == -ENOENT &&
3286             mdt_get_disposition(rep, DISP_LOOKUP_NEG) &&
3287             !mdt_get_disposition(rep, DISP_OPEN_CREATE))
3288                 rep->lock_policy_res2 = 0;
3289
3290         lhc->mlh_reg_lh.cookie = 0ull;
3291         if (rc == -ENOTCONN || rc == -ENODEV ||
3292             rc == -EOVERFLOW) { /**< if VBR failure then return error */
3293                 /*
3294                  * If it is the disconnect error (ENODEV & ENOCONN), the error
3295                  * will be returned by rq_status, and client at ptlrpc layer
3296                  * will detect this, then disconnect, reconnect the import
3297                  * immediately, instead of impacting the following the rpc.
3298                  */
3299                 RETURN(rc);
3300         }
3301         /*
3302          * For other cases, the error will be returned by intent, and client
3303          * will retrieve the result from intent.
3304          */
3305         RETURN(ELDLM_LOCK_ABORTED);
3306 }
3307
3308 static int mdt_intent_code(enum ldlm_intent_flags itcode)
3309 {
3310         int rc;
3311
3312         switch (itcode) {
3313         case IT_OPEN:
3314                 rc = MDT_IT_OPEN;
3315                 break;
3316         case IT_OPEN|IT_CREAT:
3317                 rc = MDT_IT_OCREAT;
3318                 break;
3319         case IT_CREAT:
3320                 rc = MDT_IT_CREATE;
3321                 break;
3322         case IT_READDIR:
3323                 rc = MDT_IT_READDIR;
3324                 break;
3325         case IT_GETATTR:
3326                 rc = MDT_IT_GETATTR;
3327                 break;
3328         case IT_LOOKUP:
3329                 rc = MDT_IT_LOOKUP;
3330                 break;
3331         case IT_UNLINK:
3332                 rc = MDT_IT_UNLINK;
3333                 break;
3334         case IT_TRUNC:
3335                 rc = MDT_IT_TRUNC;
3336                 break;
3337         case IT_GETXATTR:
3338                 rc = MDT_IT_GETXATTR;
3339                 break;
3340         case IT_LAYOUT:
3341                 rc = MDT_IT_LAYOUT;
3342                 break;
3343         case IT_QUOTA_DQACQ:
3344         case IT_QUOTA_CONN:
3345                 rc = MDT_IT_QUOTA;
3346                 break;
3347         default:
3348                 CERROR("Unknown intent opcode: 0x%08x\n", itcode);
3349                 rc = -EINVAL;
3350                 break;
3351         }
3352         return rc;
3353 }
3354
3355 static int mdt_intent_opc(enum ldlm_intent_flags itopc,
3356                           struct mdt_thread_info *info,
3357                           struct ldlm_lock **lockp, __u64 flags)
3358 {
3359         struct req_capsule      *pill = info->mti_pill;
3360         struct ptlrpc_request   *req = mdt_info_req(info);
3361         struct mdt_it_flavor    *flv;
3362         int opc;
3363         int rc;
3364         ENTRY;
3365
3366         opc = mdt_intent_code(itopc);
3367         if (opc < 0)
3368                 RETURN(-EINVAL);
3369
3370         if (opc == MDT_IT_QUOTA) {
3371                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
3372
3373                 if (qmt == NULL)
3374                         RETURN(-EOPNOTSUPP);
3375
3376                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
3377                 /* pass the request to quota master */
3378                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
3379                                                  mdt_info_req(info), lockp,
3380                                                  flags);
3381                 RETURN(rc);
3382         }
3383
3384         flv = &mdt_it_flavor[opc];
3385         if (flv->it_fmt != NULL)
3386                 req_capsule_extend(pill, flv->it_fmt);
3387
3388         rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3389         if (rc < 0)
3390                 RETURN(rc);
3391
3392         if (flv->it_flags & MUTABOR &&
3393             exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
3394                 RETURN(-EROFS);
3395
3396         if (flv->it_act != NULL) {
3397                 struct ldlm_reply *rep;
3398
3399                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_INTENT_DELAY, 10);
3400
3401                 /* execute policy */
3402                 rc = flv->it_act(opc, info, lockp, flags);
3403
3404                 /* Check whether the reply has been packed successfully. */
3405                 if (req->rq_repmsg != NULL) {
3406                         rep = req_capsule_server_get(info->mti_pill,
3407                                                      &RMF_DLM_REP);
3408                         rep->lock_policy_res2 =
3409                                 ptlrpc_status_hton(rep->lock_policy_res2);
3410                 }
3411         }
3412
3413         RETURN(rc);
3414 }
3415
3416 static int mdt_intent_policy(struct ldlm_namespace *ns,
3417                              struct ldlm_lock **lockp, void *req_cookie,
3418                              enum ldlm_mode mode, __u64 flags, void *data)
3419 {
3420         struct tgt_session_info *tsi;
3421         struct mdt_thread_info  *info;
3422         struct ptlrpc_request   *req  =  req_cookie;
3423         struct ldlm_intent      *it;
3424         struct req_capsule      *pill;
3425         int rc;
3426
3427         ENTRY;
3428
3429         LASSERT(req != NULL);
3430
3431         tsi = tgt_ses_info(req->rq_svc_thread->t_env);
3432
3433         info = tsi2mdt_info(tsi);
3434         LASSERT(info != NULL);
3435         pill = info->mti_pill;
3436         LASSERT(pill->rc_req == req);
3437
3438         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3439                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
3440                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3441                 if (it != NULL) {
3442                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
3443                         if (rc == 0)
3444                                 rc = ELDLM_OK;
3445
3446                         /* Lock without inodebits makes no sense and will oops
3447                          * later in ldlm. Let's check it now to see if we have
3448                          * ibits corrupted somewhere in mdt_intent_opc().
3449                          * The case for client miss to set ibits has been
3450                          * processed by others. */
3451                         LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
3452                                         lr_type == LDLM_IBITS,
3453                                      info->mti_dlm_req->lock_desc.\
3454                                         l_policy_data.l_inodebits.bits != 0));
3455                 } else
3456                         rc = err_serious(-EFAULT);
3457         } else {
3458                 /* No intent was provided */
3459                 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3460                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
3461                 rc = req_capsule_server_pack(pill);
3462                 if (rc)
3463                         rc = err_serious(rc);
3464         }
3465         mdt_thread_info_fini(info);
3466         RETURN(rc);
3467 }
3468
3469 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
3470 {
3471         struct seq_server_site  *ss = mdt_seq_site(mdt);
3472
3473         if (ss->ss_node_id == 0)
3474                 return;
3475
3476         if (ss->ss_client_seq != NULL) {
3477                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3478                 ss->ss_client_seq->lcs_exp = NULL;
3479         }
3480
3481         if (ss->ss_server_fld != NULL) {
3482                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
3483                 ss->ss_server_fld->lsf_control_exp = NULL;
3484         }
3485 }
3486
3487 static void mdt_seq_fini_cli(struct mdt_device *mdt)
3488 {
3489         struct seq_server_site *ss = mdt_seq_site(mdt);
3490
3491         if (ss == NULL)
3492                 return;
3493
3494         if (ss->ss_server_seq != NULL)
3495                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
3496 }
3497
3498 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
3499 {
3500         mdt_seq_fini_cli(mdt);
3501         mdt_deregister_seq_exp(mdt);
3502
3503         return seq_site_fini(env, mdt_seq_site(mdt));
3504 }
3505
3506 /**
3507  * It will retrieve its FLDB entries from MDT0, and it only happens
3508  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
3509  * and it needs to refresh FLDB from the MDT0.
3510  **/
3511 static int mdt_register_lwp_callback(void *data)
3512 {
3513         struct lu_env           env;
3514         struct mdt_device       *mdt = data;
3515         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
3516         int                     rc;
3517         ENTRY;
3518
3519         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
3520
3521         rc = lu_env_init(&env, LCT_MD_THREAD);
3522         if (rc < 0) {
3523                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
3524                 RETURN(rc);
3525         }
3526
3527         /* Allocate new sequence now to avoid creating local transaction
3528          * in the normal transaction process */
3529         rc = seq_server_check_and_alloc_super(&env,
3530                                               mdt_seq_site(mdt)->ss_server_seq);
3531         if (rc < 0)
3532                 GOTO(out, rc);
3533
3534         if (fld->lsf_new) {
3535                 rc = fld_update_from_controller(&env, fld);
3536                 if (rc != 0) {
3537                         CERROR("%s: cannot update controller: rc = %d\n",
3538                                mdt_obd_name(mdt), rc);
3539                         GOTO(out, rc);
3540                 }
3541         }
3542 out:
3543         lu_env_fini(&env);
3544         RETURN(rc);
3545 }
3546
3547 static int mdt_register_seq_exp(struct mdt_device *mdt)
3548 {
3549         struct seq_server_site  *ss = mdt_seq_site(mdt);
3550         char                    *lwp_name = NULL;
3551         int                     rc;
3552
3553         if (ss->ss_node_id == 0)
3554                 return 0;
3555
3556         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
3557         if (lwp_name == NULL)
3558                 GOTO(out_free, rc = -ENOMEM);
3559
3560         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
3561         if (rc != 0)
3562                 GOTO(out_free, rc);
3563
3564         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
3565                                       NULL, NULL);
3566         if (rc != 0)
3567                 GOTO(out_free, rc);
3568
3569         rc = lustre_register_lwp_item(lwp_name,
3570                                       &ss->ss_server_fld->lsf_control_exp,
3571                                       mdt_register_lwp_callback, mdt);
3572         if (rc != 0) {
3573                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3574                 ss->ss_client_seq->lcs_exp = NULL;
3575                 GOTO(out_free, rc);
3576         }
3577 out_free:
3578         if (lwp_name != NULL)
3579                 OBD_FREE(lwp_name, MAX_OBD_NAME);
3580
3581         return rc;
3582 }
3583
3584 /*
3585  * Init client sequence manager which is used by local MDS to talk to sequence
3586  * controller on remote node.
3587  */
3588 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
3589 {
3590         struct seq_server_site  *ss = mdt_seq_site(mdt);
3591         int                     rc;
3592         char                    *prefix;
3593         ENTRY;
3594
3595         /* check if this is adding the first MDC and controller is not yet
3596          * initialized. */
3597         OBD_ALLOC_PTR(ss->ss_client_seq);
3598         if (ss->ss_client_seq == NULL)
3599                 RETURN(-ENOMEM);
3600
3601         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3602         if (prefix == NULL) {
3603                 OBD_FREE_PTR(ss->ss_client_seq);
3604                 ss->ss_client_seq = NULL;
3605                 RETURN(-ENOMEM);
3606         }
3607
3608         /* Note: seq_client_fini will be called in seq_site_fini */
3609         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
3610         rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
3611                              prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
3612                                                             NULL);
3613         OBD_FREE(prefix, MAX_OBD_NAME + 5);
3614         if (rc != 0) {
3615                 OBD_FREE_PTR(ss->ss_client_seq);
3616                 ss->ss_client_seq = NULL;
3617                 RETURN(rc);
3618         }
3619
3620         rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
3621
3622         RETURN(rc);
3623 }
3624
3625 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
3626 {
3627         struct seq_server_site  *ss;
3628         int                     rc;
3629         ENTRY;
3630
3631         ss = mdt_seq_site(mdt);
3632         /* init sequence controller server(MDT0) */
3633         if (ss->ss_node_id == 0) {
3634                 OBD_ALLOC_PTR(ss->ss_control_seq);
3635                 if (ss->ss_control_seq == NULL)
3636                         RETURN(-ENOMEM);
3637
3638                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
3639                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
3640                                      ss);
3641                 if (rc)
3642                         GOTO(out_seq_fini, rc);
3643         }
3644
3645         /* Init normal sequence server */
3646         OBD_ALLOC_PTR(ss->ss_server_seq);
3647         if (ss->ss_server_seq == NULL)
3648                 GOTO(out_seq_fini, rc = -ENOMEM);
3649
3650         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
3651                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
3652         if (rc)
3653                 GOTO(out_seq_fini, rc);
3654
3655         /* init seq client for seq server to talk to seq controller(MDT0) */
3656         rc = mdt_seq_init_cli(env, mdt);
3657         if (rc != 0)
3658                 GOTO(out_seq_fini, rc);
3659
3660         if (ss->ss_node_id != 0)
3661                 /* register controller export through lwp */
3662                 rc = mdt_register_seq_exp(mdt);
3663
3664         EXIT;
3665 out_seq_fini:
3666         if (rc)
3667                 mdt_seq_fini(env, mdt);
3668
3669         return rc;
3670 }
3671
3672 /*
3673  * FLD wrappers
3674  */
3675 static int mdt_fld_fini(const struct lu_env *env,
3676                         struct mdt_device *m)
3677 {
3678         struct seq_server_site *ss = mdt_seq_site(m);
3679         ENTRY;
3680
3681         if (ss && ss->ss_server_fld) {
3682                 fld_server_fini(env, ss->ss_server_fld);
3683                 OBD_FREE_PTR(ss->ss_server_fld);
3684                 ss->ss_server_fld = NULL;
3685         }
3686
3687         RETURN(0);
3688 }
3689
3690 static int mdt_fld_init(const struct lu_env *env,
3691                         const char *uuid,
3692                         struct mdt_device *m)
3693 {
3694         struct seq_server_site *ss;
3695         int rc;
3696         ENTRY;
3697
3698         ss = mdt_seq_site(m);
3699
3700         OBD_ALLOC_PTR(ss->ss_server_fld);
3701         if (ss->ss_server_fld == NULL)
3702                 RETURN(rc = -ENOMEM);
3703
3704         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
3705                              LU_SEQ_RANGE_MDT);
3706         if (rc) {
3707                 OBD_FREE_PTR(ss->ss_server_fld);
3708                 ss->ss_server_fld = NULL;
3709                 RETURN(rc);
3710         }
3711
3712         RETURN(0);
3713 }
3714
3715 static void mdt_stack_pre_fini(const struct lu_env *env,
3716                            struct mdt_device *m, struct lu_device *top)
3717 {
3718         struct lustre_cfg_bufs  *bufs;
3719         struct lustre_cfg       *lcfg;
3720         struct mdt_thread_info  *info;
3721         ENTRY;
3722
3723         LASSERT(top);
3724
3725         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3726         LASSERT(info != NULL);
3727
3728         bufs = &info->mti_u.bufs;
3729
3730         LASSERT(m->mdt_child_exp);
3731         LASSERT(m->mdt_child_exp->exp_obd);
3732
3733         /* process cleanup, pass mdt obd name to get obd umount flags */
3734         /* XXX: this is needed because all layers are referenced by
3735          * objects (some of them are pinned by osd, for example *
3736          * the proper solution should be a model where object used
3737          * by osd only doesn't have mdt/mdd slices -bzzz */
3738         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
3739         lustre_cfg_bufs_set_string(bufs, 1, NULL);
3740         lcfg = lustre_cfg_new(LCFG_PRE_CLEANUP, bufs);
3741         if (lcfg == NULL)
3742                 RETURN_EXIT;
3743
3744         top->ld_ops->ldo_process_config(env, top, lcfg);
3745         lustre_cfg_free(lcfg);
3746         EXIT;
3747 }
3748
3749 static void mdt_stack_fini(const struct lu_env *env,
3750                            struct mdt_device *m, struct lu_device *top)
3751 {
3752         struct obd_device       *obd = mdt2obd_dev(m);
3753         struct lustre_cfg_bufs  *bufs;
3754         struct lustre_cfg       *lcfg;
3755         struct mdt_thread_info  *info;
3756         char                     flags[3] = "";
3757         ENTRY;
3758
3759         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3760         LASSERT(info != NULL);
3761
3762         lu_dev_del_linkage(top->ld_site, top);
3763
3764         lu_site_purge(env, top->ld_site, -1);
3765
3766         bufs = &info->mti_u.bufs;
3767         /* process cleanup, pass mdt obd name to get obd umount flags */
3768         /* another purpose is to let all layers to release their objects */
3769         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
3770         if (obd->obd_force)
3771                 strcat(flags, "F");
3772         if (obd->obd_fail)
3773                 strcat(flags, "A");
3774         lustre_cfg_bufs_set_string(bufs, 1, flags);
3775         lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
3776         if (lcfg == NULL)
3777                 RETURN_EXIT;
3778
3779         LASSERT(top);
3780         top->ld_ops->ldo_process_config(env, top, lcfg);
3781         lustre_cfg_free(lcfg);
3782
3783         lu_site_purge(env, top->ld_site, -1);
3784
3785         m->mdt_child = NULL;
3786         m->mdt_bottom = NULL;
3787
3788         obd_disconnect(m->mdt_child_exp);
3789         m->mdt_child_exp = NULL;
3790
3791         obd_disconnect(m->mdt_bottom_exp);
3792         m->mdt_child_exp = NULL;
3793 }
3794
3795 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
3796                                const char *next, struct obd_export **exp)
3797 {
3798         struct obd_connect_data *data = NULL;
3799         struct obd_device       *obd;
3800         int                      rc;
3801         ENTRY;
3802
3803         OBD_ALLOC_PTR(data);
3804         if (data == NULL)
3805                 GOTO(out, rc = -ENOMEM);
3806
3807         obd = class_name2obd(next);
3808         if (obd == NULL) {
3809                 CERROR("%s: can't locate next device: %s\n",
3810                        mdt_obd_name(m), next);
3811                 GOTO(out, rc = -ENOTCONN);
3812         }
3813
3814         data->ocd_connect_flags = OBD_CONNECT_VERSION;
3815         data->ocd_version = LUSTRE_VERSION_CODE;
3816
3817         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
3818         if (rc) {
3819                 CERROR("%s: cannot connect to next dev %s (%d)\n",
3820                        mdt_obd_name(m), next, rc);
3821                 GOTO(out, rc);
3822         }
3823
3824 out:
3825         if (data)
3826                 OBD_FREE_PTR(data);
3827         RETURN(rc);
3828 }
3829
3830 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
3831                           struct lustre_cfg *cfg)
3832 {
3833         char                   *dev = lustre_cfg_string(cfg, 0);
3834         int                     rc, name_size, uuid_size;
3835         char                   *name, *uuid, *p;
3836         struct lustre_cfg_bufs *bufs;
3837         struct lustre_cfg      *lcfg;
3838         struct obd_device      *obd;
3839         struct lustre_profile  *lprof;
3840         struct lu_site         *site;
3841         ENTRY;
3842
3843         /* in 1.8 we had the only device in the stack - MDS.
3844          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
3845          * in 2.3 OSD is instantiated by obd_mount.c, so we need
3846          * to generate names and setup MDT, MDD. MDT will be using
3847          * generated name to connect to MDD. for MDD the next device
3848          * will be LOD with name taken from so called "profile" which
3849          * is generated by mount_option line
3850          *
3851          * 1.8 MGS generates config. commands like this:
3852          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
3853          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
3854          * 2.0 MGS generates config. commands like this:
3855          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
3856          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
3857          *                    3:lustre-MDT0000-mdtlov  4:f
3858          *
3859          * we generate MDD name from MDT one, just replacing T with D
3860          *
3861          * after all the preparations, the logical equivalent will be
3862          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
3863          *                    3:lustre-MDT0000-mdtlov  4:f
3864          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
3865          *                    3:lustre-MDD0000  4:f
3866          *
3867          *  notice we build the stack from down to top: MDD first, then MDT */
3868
3869         name_size = MAX_OBD_NAME;
3870         uuid_size = MAX_OBD_NAME;
3871
3872         OBD_ALLOC(name, name_size);
3873         OBD_ALLOC(uuid, uuid_size);
3874         if (name == NULL || uuid == NULL)
3875                 GOTO(cleanup_mem, rc = -ENOMEM);
3876
3877         OBD_ALLOC_PTR(bufs);
3878         if (!bufs)
3879                 GOTO(cleanup_mem, rc = -ENOMEM);
3880
3881         strcpy(name, dev);
3882         p = strstr(name, "-MDT");
3883         if (p == NULL)
3884                 GOTO(free_bufs, rc = -ENOMEM);
3885         p[3] = 'D';
3886
3887         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
3888
3889         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
3890         if (lprof == NULL || lprof->lp_dt == NULL) {
3891                 CERROR("can't find the profile: %s\n",
3892                        lustre_cfg_string(cfg, 0));
3893                 GOTO(free_bufs, rc = -EINVAL);
3894         }
3895
3896         lustre_cfg_bufs_reset(bufs, name);
3897         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
3898         lustre_cfg_bufs_set_string(bufs, 2, uuid);
3899         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
3900
3901         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
3902         if (lcfg == NULL)
3903                 GOTO(put_profile, rc = -ENOMEM);
3904
3905         rc = class_attach(lcfg);
3906         if (rc)
3907                 GOTO(lcfg_cleanup, rc);
3908
3909         obd = class_name2obd(name);
3910         if (!obd) {
3911                 CERROR("Can not find obd %s (%s in config)\n",
3912                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
3913                 GOTO(lcfg_cleanup, rc = -EINVAL);
3914         }
3915
3916         lustre_cfg_free(lcfg);
3917
3918         lustre_cfg_bufs_reset(bufs, name);
3919         lustre_cfg_bufs_set_string(bufs, 1, uuid);
3920         lustre_cfg_bufs_set_string(bufs, 2, dev);
3921         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
3922
3923         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
3924         if (lcfg == NULL)
3925                 GOTO(class_detach, rc = -ENOMEM);
3926
3927         rc = class_setup(obd, lcfg);
3928         if (rc)
3929                 GOTO(class_detach, rc);
3930
3931         /* connect to MDD we just setup */
3932         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
3933         if (rc)
3934                 GOTO(class_detach, rc);
3935
3936         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
3937         LASSERT(site);
3938         LASSERT(mdt_lu_site(mdt) == NULL);
3939         mdt->mdt_lu_dev.ld_site = site;
3940         site->ls_top_dev = &mdt->mdt_lu_dev;
3941         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
3942
3943         /* now connect to bottom OSD */
3944         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
3945         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
3946         if (rc)
3947                 GOTO(class_detach, rc);
3948         mdt->mdt_bottom =
3949                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
3950
3951         rc = lu_env_refill((struct lu_env *)env);
3952         if (rc != 0)
3953                 CERROR("Failure to refill session: '%d'\n", rc);
3954
3955         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
3956
3957         EXIT;
3958 class_detach:
3959         if (rc)
3960                 class_detach(obd, lcfg);
3961 lcfg_cleanup:
3962         lustre_cfg_free(lcfg);
3963 put_profile:
3964         class_put_profile(lprof);
3965 free_bufs:
3966         OBD_FREE_PTR(bufs);
3967 cleanup_mem:
3968         if (name)
3969                 OBD_FREE(name, name_size);
3970         if (uuid)
3971                 OBD_FREE(uuid, uuid_size);
3972         RETURN(rc);
3973 }
3974
3975 /* setup quota master target on MDT0 */
3976 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
3977                           struct lustre_cfg *cfg)
3978 {
3979         struct obd_device       *obd;
3980         char                    *dev = lustre_cfg_string(cfg, 0);
3981         char                    *qmtname, *uuid, *p;
3982         struct lustre_cfg_bufs  *bufs;
3983         struct lustre_cfg       *lcfg;
3984         struct lustre_profile   *lprof;
3985         struct obd_connect_data *data;
3986         int                      rc;
3987         ENTRY;
3988
3989         LASSERT(mdt->mdt_qmt_exp == NULL);
3990         LASSERT(mdt->mdt_qmt_dev == NULL);
3991
3992         /* quota master is on MDT0 only for now */
3993         if (mdt->mdt_seq_site.ss_node_id != 0)
3994                 RETURN(0);
3995
3996         /* MGS generates config commands which look as follows:
3997          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
3998          *                    3:lustre-MDT0000-mdtlov  4:f
3999          *
4000          * We generate the QMT name from the MDT one, just replacing MD with QM
4001          * after all the preparations, the logical equivalent will be:
4002          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
4003          *                    3:lustre-MDT0000-osd  4:f */
4004         OBD_ALLOC(qmtname, MAX_OBD_NAME);
4005         OBD_ALLOC(uuid, UUID_MAX);
4006         OBD_ALLOC_PTR(bufs);
4007         OBD_ALLOC_PTR(data);
4008         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
4009                 GOTO(cleanup_mem, rc = -ENOMEM);
4010
4011         strcpy(qmtname, dev);
4012         p = strstr(qmtname, "-MDT");
4013         if (p == NULL)
4014                 GOTO(cleanup_mem, rc = -ENOMEM);
4015         /* replace MD with QM */
4016         p[1] = 'Q';
4017         p[2] = 'M';
4018
4019         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
4020
4021         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4022         if (lprof == NULL || lprof->lp_dt == NULL) {
4023                 CERROR("can't find profile for %s\n",
4024                        lustre_cfg_string(cfg, 0));
4025                 GOTO(cleanup_mem, rc = -EINVAL);
4026         }
4027
4028         lustre_cfg_bufs_reset(bufs, qmtname);
4029         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
4030         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4031         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4032
4033         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4034         if (lcfg == NULL)
4035                 GOTO(put_profile, rc = -ENOMEM);
4036
4037         rc = class_attach(lcfg);
4038         if (rc)
4039                 GOTO(lcfg_cleanup, rc);
4040
4041         obd = class_name2obd(qmtname);
4042         if (!obd) {
4043                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
4044                        lustre_cfg_string(cfg, 0));
4045                 GOTO(lcfg_cleanup, rc = -EINVAL);
4046         }
4047
4048         lustre_cfg_free(lcfg);
4049
4050         lustre_cfg_bufs_reset(bufs, qmtname);
4051         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4052         lustre_cfg_bufs_set_string(bufs, 2, dev);
4053
4054         /* for quota, the next device should be the OSD device */
4055         lustre_cfg_bufs_set_string(bufs, 3,
4056                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
4057
4058         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4059         if (lcfg == NULL)
4060                 GOTO(class_detach, rc = -ENOMEM);
4061
4062         rc = class_setup(obd, lcfg);
4063         if (rc)
4064                 GOTO(class_detach, rc);
4065
4066         mdt->mdt_qmt_dev = obd->obd_lu_dev;
4067
4068         /* configure local quota objects */
4069         rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
4070                                                    &mdt->mdt_lu_dev,
4071                                                    mdt->mdt_qmt_dev);
4072         if (rc)
4073                 GOTO(class_cleanup, rc);
4074
4075         /* connect to quota master target */
4076         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4077         data->ocd_version = LUSTRE_VERSION_CODE;
4078         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
4079                          data, NULL);
4080         if (rc) {
4081                 CERROR("cannot connect to quota master device %s (%d)\n",
4082                        qmtname, rc);
4083                 GOTO(class_cleanup, rc);
4084         }
4085
4086         EXIT;
4087 class_cleanup:
4088         if (rc) {
4089                 class_manual_cleanup(obd);
4090                 mdt->mdt_qmt_dev = NULL;
4091         }
4092 class_detach:
4093         if (rc)
4094                 class_detach(obd, lcfg);
4095 lcfg_cleanup:
4096         lustre_cfg_free(lcfg);
4097 put_profile:
4098         class_put_profile(lprof);
4099 cleanup_mem:
4100         if (bufs)
4101                 OBD_FREE_PTR(bufs);
4102         if (qmtname)
4103                 OBD_FREE(qmtname, MAX_OBD_NAME);
4104         if (uuid)
4105                 OBD_FREE(uuid, UUID_MAX);
4106         if (data)
4107                 OBD_FREE_PTR(data);
4108         return rc;
4109 }
4110
4111 /* Shutdown quota master target associated with mdt */
4112 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
4113 {
4114         ENTRY;
4115
4116         if (mdt->mdt_qmt_exp == NULL)
4117                 RETURN_EXIT;
4118         LASSERT(mdt->mdt_qmt_dev != NULL);
4119
4120         /* the qmt automatically shuts down when the mdt disconnects */
4121         obd_disconnect(mdt->mdt_qmt_exp);
4122         mdt->mdt_qmt_exp = NULL;
4123         mdt->mdt_qmt_dev = NULL;
4124         EXIT;
4125 }
4126
4127 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
4128  * for now. This will be removed along with converting rest of MDT code
4129  * to use tgt_session_info */
4130 static int mdt_tgt_getxattr(struct tgt_session_info *tsi)
4131 {
4132         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
4133         int                      rc;
4134
4135         rc = mdt_getxattr(info);
4136
4137         mdt_thread_info_fini(info);
4138         return rc;
4139 }
4140
4141 static struct tgt_handler mdt_tgt_handlers[] = {
4142 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4143                 0,                      MDS_CONNECT,    mdt_tgt_connect,
4144                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
4145 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4146                 0,                      MDS_DISCONNECT, tgt_disconnect,
4147                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
4148 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4149                 HABEO_REFERO,           MDS_SET_INFO,   mdt_set_info,
4150                 &RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION),
4151 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
4152 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_GETSTATUS,  mdt_getstatus),
4153 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETATTR,    mdt_getattr),
4154 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_GETATTR_NAME,
4155                                                         mdt_getattr_name),
4156 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETXATTR,   mdt_tgt_getxattr),
4157 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_STATFS,     mdt_statfs),
4158 TGT_MDT_HDL(0           | MUTABOR,      MDS_REINT,      mdt_reint),
4159 TGT_MDT_HDL(HABEO_CORPUS,               MDS_CLOSE,      mdt_close),
4160 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_READPAGE,   mdt_readpage),
4161 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_SYNC,       mdt_sync),
4162 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
4163 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_PROGRESS,
4164                                                         mdt_hsm_progress),
4165 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_REGISTER,
4166                                                         mdt_hsm_ct_register),
4167 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_UNREGISTER,
4168                                                         mdt_hsm_ct_unregister),
4169 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_STATE_GET,
4170                                                         mdt_hsm_state_get),
4171 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_STATE_SET,
4172                                                         mdt_hsm_state_set),
4173 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_ACTION, mdt_hsm_action),
4174 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_REQUEST,
4175                                                         mdt_hsm_request),
4176 TGT_MDT_HDL(HABEO_CLAVIS | HABEO_CORPUS | HABEO_REFERO | MUTABOR,
4177             MDS_SWAP_LAYOUTS,
4178             mdt_swap_layouts),
4179 };
4180
4181 static struct tgt_handler mdt_sec_ctx_ops[] = {
4182 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
4183 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
4184 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
4185 };
4186
4187 static struct tgt_handler mdt_quota_ops[] = {
4188 TGT_QUOTA_HDL(HABEO_REFERO,             QUOTA_DQACQ,      mdt_quota_dqacq),
4189 };
4190
4191 static struct tgt_opc_slice mdt_common_slice[] = {
4192         {
4193                 .tos_opc_start  = MDS_FIRST_OPC,
4194                 .tos_opc_end    = MDS_LAST_OPC,
4195                 .tos_hs         = mdt_tgt_handlers
4196         },
4197         {
4198                 .tos_opc_start  = OBD_FIRST_OPC,
4199                 .tos_opc_end    = OBD_LAST_OPC,
4200                 .tos_hs         = tgt_obd_handlers
4201         },
4202         {
4203                 .tos_opc_start  = LDLM_FIRST_OPC,
4204                 .tos_opc_end    = LDLM_LAST_OPC,
4205                 .tos_hs         = tgt_dlm_handlers
4206         },
4207         {
4208                 .tos_opc_start  = SEC_FIRST_OPC,
4209                 .tos_opc_end    = SEC_LAST_OPC,
4210                 .tos_hs         = mdt_sec_ctx_ops
4211         },
4212         {
4213                 .tos_opc_start  = OUT_UPDATE_FIRST_OPC,
4214                 .tos_opc_end    = OUT_UPDATE_LAST_OPC,
4215                 .tos_hs         = tgt_out_handlers
4216         },
4217         {
4218                 .tos_opc_start  = FLD_FIRST_OPC,
4219                 .tos_opc_end    = FLD_LAST_OPC,
4220                 .tos_hs         = fld_handlers
4221         },
4222         {
4223                 .tos_opc_start  = SEQ_FIRST_OPC,
4224                 .tos_opc_end    = SEQ_LAST_OPC,
4225                 .tos_hs         = seq_handlers
4226         },
4227         {
4228                 .tos_opc_start  = QUOTA_DQACQ,
4229                 .tos_opc_end    = QUOTA_LAST_OPC,
4230                 .tos_hs         = mdt_quota_ops
4231         },
4232         {
4233                 .tos_opc_start  = LLOG_FIRST_OPC,
4234                 .tos_opc_end    = LLOG_LAST_OPC,
4235                 .tos_hs         = tgt_llog_handlers
4236         },
4237         {
4238                 .tos_opc_start  = LFSCK_FIRST_OPC,
4239                 .tos_opc_end    = LFSCK_LAST_OPC,
4240                 .tos_hs         = tgt_lfsck_handlers
4241         },
4242
4243         {
4244                 .tos_hs         = NULL
4245         }
4246 };
4247
4248 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4249 {
4250         struct md_device        *next = m->mdt_child;
4251         struct lu_device        *d    = &m->mdt_lu_dev;
4252         struct obd_device       *obd  = mdt2obd_dev(m);
4253         struct lfsck_stop        stop;
4254         ENTRY;
4255
4256         stop.ls_status = LS_PAUSED;
4257         stop.ls_flags = 0;
4258         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
4259
4260         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
4261         target_recovery_fini(obd);
4262         ping_evictor_stop();
4263
4264         if (m->mdt_opts.mo_coordinator)
4265                 mdt_hsm_cdt_stop(m);
4266
4267         mdt_hsm_cdt_fini(m);
4268
4269         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
4270         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4271
4272         if (m->mdt_namespace != NULL)
4273                 ldlm_namespace_free_prior(m->mdt_namespace, NULL,
4274                                           d->ld_obd->obd_force);
4275
4276         obd_exports_barrier(obd);
4277         obd_zombie_barrier();
4278
4279         mdt_procfs_fini(m);
4280
4281         tgt_fini(env, &m->mdt_lut);
4282         mdt_fs_cleanup(env, m);
4283         upcall_cache_cleanup(m->mdt_identity_cache);
4284         m->mdt_identity_cache = NULL;
4285
4286         if (m->mdt_namespace != NULL) {
4287                 ldlm_namespace_free_post(m->mdt_namespace);
4288                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4289         }
4290
4291         mdt_quota_fini(env, m);
4292
4293         cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
4294
4295         mdt_seq_fini(env, m);
4296         mdt_fld_fini(env, m);
4297
4298         /*
4299          * Finish the stack
4300          */
4301         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4302
4303         LASSERT(atomic_read(&d->ld_ref) == 0);
4304
4305         server_put_mount(mdt_obd_name(m), true);
4306
4307         EXIT;
4308 }
4309
4310 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4311
4312 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4313                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
4314 {
4315         struct mdt_thread_info    *info;
4316         struct obd_device         *obd;
4317         const char                *dev = lustre_cfg_string(cfg, 0);
4318         const char                *num = lustre_cfg_string(cfg, 2);
4319         struct lustre_mount_info  *lmi = NULL;
4320         struct lustre_sb_info     *lsi;
4321         struct lu_site            *s;
4322         struct seq_server_site    *ss_site;
4323         const char                *identity_upcall = "NONE";
4324         struct md_device          *next;
4325         int                        rc;
4326         long                       node_id;
4327         mntopt_t                   mntopts;
4328         ENTRY;
4329
4330         lu_device_init(&m->mdt_lu_dev, ldt);
4331         /*
4332          * Environment (env) might be missing mdt_thread_key values at that
4333          * point, if device is allocated when mdt_thread_key is in QUIESCENT
4334          * mode.
4335          *
4336          * Usually device allocation path doesn't use module key values, but
4337          * mdt has to do a lot of work here, so allocate key value.
4338          */
4339         rc = lu_env_refill((struct lu_env *)env);
4340         if (rc != 0)
4341                 RETURN(rc);
4342
4343         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4344         LASSERT(info != NULL);
4345
4346         obd = class_name2obd(dev);
4347         LASSERT(obd != NULL);
4348
4349         m->mdt_max_mdsize = MAX_MD_SIZE; /* 4 stripes */
4350         m->mdt_opts.mo_evict_tgt_nids = 1;
4351         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4352
4353         /* default is coordinator off, it is started through conf_param
4354          * or /proc */
4355         m->mdt_opts.mo_coordinator = 0;
4356
4357         lmi = server_get_mount(dev);
4358         if (lmi == NULL) {
4359                 CERROR("Cannot get mount info for %s!\n", dev);
4360                 RETURN(-EFAULT);
4361         } else {
4362                 lsi = s2lsi(lmi->lmi_sb);
4363                 /* CMD is supported only in IAM mode */
4364                 LASSERT(num);
4365                 node_id = simple_strtol(num, NULL, 10);
4366                 obd->u.obt.obt_magic = OBT_MAGIC;
4367                 if (lsi->lsi_lmd != NULL &&
4368                     lsi->lsi_lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
4369                         m->mdt_skip_lfsck = 1;
4370         }
4371
4372         m->mdt_squash.rsi_uid = 0;
4373         m->mdt_squash.rsi_gid = 0;
4374         INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
4375         init_rwsem(&m->mdt_squash.rsi_sem);
4376         spin_lock_init(&m->mdt_osfs_lock);
4377         m->mdt_osfs_age = cfs_time_shift_64(-1000);
4378         m->mdt_enable_remote_dir = 0;
4379         m->mdt_enable_remote_dir_gid = 0;
4380
4381         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
4382         m->mdt_lu_dev.ld_obd = obd;
4383         /* Set this lu_device to obd for error handling purposes. */
4384         obd->obd_lu_dev = &m->mdt_lu_dev;
4385
4386         /* init the stack */
4387         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
4388         if (rc) {
4389                 CERROR("%s: Can't init device stack, rc %d\n",
4390                        mdt_obd_name(m), rc);
4391                 GOTO(err_lmi, rc);
4392         }
4393
4394         s = mdt_lu_site(m);
4395         ss_site = mdt_seq_site(m);
4396         s->ld_seq_site = ss_site;
4397         ss_site->ss_lu = s;
4398
4399         /* set server index */
4400         ss_site->ss_node_id = node_id;
4401
4402         /* failover is the default
4403          * FIXME: we do not failout mds0/mgs, which may cause some problems.
4404          * assumed whose ss_node_id == 0 XXX
4405          * */
4406         obd->obd_replayable = 1;
4407         /* No connection accepted until configurations will finish */
4408         obd->obd_no_conn = 1;
4409
4410         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4411                 char *str = lustre_cfg_string(cfg, 4);
4412                 if (strchr(str, 'n')) {
4413                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
4414                         obd->obd_replayable = 0;
4415                 }
4416         }
4417
4418         rc = mdt_fld_init(env, mdt_obd_name(m), m);
4419         if (rc)
4420                 GOTO(err_fini_stack, rc);
4421
4422         rc = mdt_seq_init(env, m);
4423         if (rc)
4424                 GOTO(err_fini_fld, rc);
4425
4426         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
4427                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
4428         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4429                                               LDLM_NAMESPACE_SERVER,
4430                                               LDLM_NAMESPACE_GREEDY,
4431                                               LDLM_NS_TYPE_MDT);
4432         if (m->mdt_namespace == NULL)
4433                 GOTO(err_fini_seq, rc = -ENOMEM);
4434
4435         m->mdt_namespace->ns_lvbp = m;
4436         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
4437
4438         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4439         /* set obd_namespace for compatibility with old code */
4440         obd->obd_namespace = m->mdt_namespace;
4441
4442         rc = mdt_hsm_cdt_init(m);
4443         if (rc != 0) {
4444                 CERROR("%s: error initializing coordinator, rc %d\n",
4445                        mdt_obd_name(m), rc);
4446                 GOTO(err_free_ns, rc);
4447         }
4448
4449         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
4450                       OBD_FAIL_MDS_ALL_REQUEST_NET,
4451                       OBD_FAIL_MDS_ALL_REPLY_NET);
4452         if (rc)
4453                 GOTO(err_free_hsm, rc);
4454
4455         rc = mdt_fs_setup(env, m, obd, lsi);
4456         if (rc)
4457                 GOTO(err_tgt, rc);
4458
4459         tgt_adapt_sptlrpc_conf(&m->mdt_lut, 1);
4460
4461         next = m->mdt_child;
4462         rc = next->md_ops->mdo_iocontrol(env, next, OBD_IOC_GET_MNTOPT, 0,
4463                                          &mntopts);
4464         if (rc)
4465                 GOTO(err_fs_cleanup, rc);
4466
4467         if (mntopts & MNTOPT_USERXATTR)
4468                 m->mdt_opts.mo_user_xattr = 1;
4469         else
4470                 m->mdt_opts.mo_user_xattr = 0;
4471
4472         rc = next->md_ops->mdo_maxeasize_get(env, next, &m->mdt_max_ea_size);
4473         if (rc)
4474                 GOTO(err_fs_cleanup, rc);
4475
4476         if (mntopts & MNTOPT_ACL)
4477                 m->mdt_opts.mo_acl = 1;
4478         else
4479                 m->mdt_opts.mo_acl = 0;
4480
4481         /* XXX: to support suppgid for ACL, we enable identity_upcall
4482          * by default, otherwise, maybe got unexpected -EACCESS. */
4483         if (m->mdt_opts.mo_acl)
4484                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4485
4486         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
4487                                                 identity_upcall,
4488                                                 &mdt_identity_upcall_cache_ops);
4489         if (IS_ERR(m->mdt_identity_cache)) {
4490                 rc = PTR_ERR(m->mdt_identity_cache);
4491                 m->mdt_identity_cache = NULL;
4492                 GOTO(err_fs_cleanup, rc);
4493         }
4494
4495         rc = mdt_procfs_init(m, dev);
4496         if (rc) {
4497                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4498                 GOTO(err_recovery, rc);
4499         }
4500
4501         rc = mdt_quota_init(env, m, cfg);
4502         if (rc)
4503                 GOTO(err_procfs, rc);
4504
4505         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
4506         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
4507                            "mdt_ldlm_client", m->mdt_ldlm_client);
4508
4509         ping_evictor_start();
4510
4511         /* recovery will be started upon mdt_prepare()
4512          * when the whole stack is complete and ready
4513          * to serve the requests */
4514
4515         /* Reduce the initial timeout on an MDS because it doesn't need such
4516          * a long timeout as an OST does. Adaptive timeouts will adjust this
4517          * value appropriately. */
4518         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4519                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4520
4521         RETURN(0);
4522 err_procfs:
4523         mdt_procfs_fini(m);
4524 err_recovery:
4525         target_recovery_fini(obd);
4526         upcall_cache_cleanup(m->mdt_identity_cache);
4527         m->mdt_identity_cache = NULL;
4528 err_fs_cleanup:
4529         mdt_fs_cleanup(env, m);
4530 err_tgt:
4531         tgt_fini(env, &m->mdt_lut);
4532 err_free_hsm:
4533         mdt_hsm_cdt_fini(m);
4534 err_free_ns:
4535         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4536         obd->obd_namespace = m->mdt_namespace = NULL;
4537 err_fini_seq:
4538         mdt_seq_fini(env, m);
4539 err_fini_fld:
4540         mdt_fld_fini(env, m);
4541 err_fini_stack:
4542         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4543 err_lmi:
4544         if (lmi)
4545                 server_put_mount(dev, true);
4546         return(rc);
4547 }
4548
4549 /* For interoperability, the left element is old parameter, the right one
4550  * is the new version of the parameter, if some parameter is deprecated,
4551  * the new version should be set as NULL. */
4552 static struct cfg_interop_param mdt_interop_param[] = {
4553         { "mdt.group_upcall",   NULL },
4554         { "mdt.quota_type",     NULL },
4555         { "mdd.quota_type",     NULL },
4556         { "mdt.rootsquash",     "mdt.root_squash" },
4557         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
4558         { NULL }
4559 };
4560
4561 /* used by MGS to process specific configurations */
4562 static int mdt_process_config(const struct lu_env *env,
4563                               struct lu_device *d, struct lustre_cfg *cfg)
4564 {
4565         struct mdt_device *m = mdt_dev(d);
4566         struct md_device *md_next = m->mdt_child;
4567         struct lu_device *next = md2lu_dev(md_next);
4568         int rc;
4569         ENTRY;
4570
4571         switch (cfg->lcfg_command) {
4572         case LCFG_PARAM: {
4573                 struct obd_device          *obd = d->ld_obd;
4574
4575                 /* For interoperability */
4576                 struct cfg_interop_param   *ptr = NULL;
4577                 struct lustre_cfg          *old_cfg = NULL;
4578                 char                       *param = NULL;
4579
4580                 param = lustre_cfg_string(cfg, 1);
4581                 if (param == NULL) {
4582                         CERROR("param is empty\n");
4583                         rc = -EINVAL;
4584                         break;
4585                 }
4586
4587                 ptr = class_find_old_param(param, mdt_interop_param);
4588                 if (ptr != NULL) {
4589                         if (ptr->new_param == NULL) {
4590                                 rc = 0;
4591                                 CWARN("For interoperability, skip this %s."
4592                                       " It is obsolete.\n", ptr->old_param);
4593                                 break;
4594                         }
4595
4596                         CWARN("Found old param %s, changed it to %s.\n",
4597                               ptr->old_param, ptr->new_param);
4598
4599                         old_cfg = cfg;
4600                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
4601                         if (IS_ERR(cfg)) {
4602                                 rc = PTR_ERR(cfg);
4603                                 break;
4604                         }
4605                 }
4606
4607                 rc = class_process_proc_param(PARAM_MDT, obd->obd_vars,
4608                                               cfg, obd);
4609                 if (rc > 0 || rc == -ENOSYS) {
4610                         /* is it an HSM var ? */
4611                         rc = class_process_proc_param(PARAM_HSM,
4612                                                       hsm_cdt_get_proc_vars(),
4613                                                       cfg, obd);
4614                         if (rc > 0 || rc == -ENOSYS)
4615                                 /* we don't understand; pass it on */
4616                                 rc = next->ld_ops->ldo_process_config(env, next,
4617                                                                       cfg);
4618                 }
4619
4620                 if (old_cfg != NULL)
4621                         lustre_cfg_free(cfg);
4622
4623                 break;
4624         }
4625         default:
4626                 /* others are passed further */
4627                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4628                 break;
4629         }
4630         RETURN(rc);
4631 }
4632
4633 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4634                                           const struct lu_object_header *hdr,
4635                                           struct lu_device *d)
4636 {
4637         struct mdt_object *mo;
4638
4639         ENTRY;
4640
4641         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
4642         if (mo != NULL) {
4643                 struct lu_object *o;
4644                 struct lu_object_header *h;
4645
4646                 o = &mo->mot_obj;
4647                 h = &mo->mot_header;
4648                 lu_object_header_init(h);
4649                 lu_object_init(o, h, d);
4650                 lu_object_add_top(h, o);
4651                 o->lo_ops = &mdt_obj_ops;
4652                 spin_lock_init(&mo->mot_write_lock);
4653                 mutex_init(&mo->mot_lov_mutex);
4654                 init_rwsem(&mo->mot_open_sem);
4655                 RETURN(o);
4656         }
4657         RETURN(NULL);
4658 }
4659
4660 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
4661                            const struct lu_object_conf *unused)
4662 {
4663         struct mdt_device *d = mdt_dev(o->lo_dev);
4664         struct lu_device  *under;
4665         struct lu_object  *below;
4666         int                rc = 0;
4667         ENTRY;
4668
4669         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4670                PFID(lu_object_fid(o)));
4671
4672         under = &d->mdt_child->md_lu_dev;
4673         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4674         if (below != NULL) {
4675                 lu_object_add(o, below);
4676         } else
4677                 rc = -ENOMEM;
4678
4679         RETURN(rc);
4680 }
4681
4682 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4683 {
4684         struct mdt_object *mo = mdt_obj(o);
4685         struct lu_object_header *h;
4686         ENTRY;
4687
4688         h = o->lo_header;
4689         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4690                PFID(lu_object_fid(o)));
4691
4692         LASSERT(atomic_read(&mo->mot_open_count) == 0);
4693         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
4694
4695         lu_object_fini(o);
4696         lu_object_header_fini(h);
4697         OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
4698
4699         EXIT;
4700 }
4701
4702 static int mdt_object_print(const struct lu_env *env, void *cookie,
4703                             lu_printer_t p, const struct lu_object *o)
4704 {
4705         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
4706
4707         return (*p)(env, cookie,
4708                     LUSTRE_MDT_NAME"-object@%p(flags=%d, writecount=%d)",
4709                     mdto, mdto->mot_flags, mdto->mot_write_count);
4710 }
4711
4712 static int mdt_prepare(const struct lu_env *env,
4713                 struct lu_device *pdev,
4714                 struct lu_device *cdev)
4715 {
4716         struct mdt_device *mdt = mdt_dev(cdev);
4717         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
4718         struct obd_device *obd = cdev->ld_obd;
4719         int rc;
4720
4721         ENTRY;
4722
4723         LASSERT(obd);
4724
4725         rc = next->ld_ops->ldo_prepare(env, cdev, next);
4726         if (rc)
4727                 RETURN(rc);
4728
4729         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
4730         if (rc)
4731                 RETURN(rc);
4732
4733         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
4734         if (rc)
4735                 RETURN(rc);
4736
4737         rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
4738         /* The LFSCK instance is registered just now, so it must be there when
4739          * register the namespace to such instance. */
4740         LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
4741
4742         if (mdt->mdt_seq_site.ss_node_id == 0) {
4743                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
4744                                                          &mdt->mdt_md_root_fid);
4745                 if (rc)
4746                         RETURN(rc);
4747         }
4748
4749         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
4750
4751         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
4752         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
4753         LASSERT(obd->obd_no_conn);
4754         spin_lock(&obd->obd_dev_lock);
4755         obd->obd_no_conn = 0;
4756         spin_unlock(&obd->obd_dev_lock);
4757
4758         if (obd->obd_recovering == 0)
4759                 mdt_postrecov(env, mdt);
4760
4761         RETURN(rc);
4762 }
4763
4764 const struct lu_device_operations mdt_lu_ops = {
4765         .ldo_object_alloc   = mdt_object_alloc,
4766         .ldo_process_config = mdt_process_config,
4767         .ldo_prepare        = mdt_prepare,
4768 };
4769
4770 static const struct lu_object_operations mdt_obj_ops = {
4771         .loo_object_init    = mdt_object_init,
4772         .loo_object_free    = mdt_object_free,
4773         .loo_object_print   = mdt_object_print
4774 };
4775
4776 static int mdt_obd_set_info_async(const struct lu_env *env,
4777                                   struct obd_export *exp,
4778                                   __u32 keylen, void *key,
4779                                   __u32 vallen, void *val,
4780                                   struct ptlrpc_request_set *set)
4781 {
4782         int rc;
4783
4784         ENTRY;
4785
4786         if (KEY_IS(KEY_SPTLRPC_CONF)) {
4787                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp), 0);
4788                 RETURN(rc);
4789         }
4790
4791         RETURN(0);
4792 }
4793
4794 /**
4795  * Match client and server connection feature flags.
4796  *
4797  * Compute the compatibility flags for a connection request based on
4798  * features mutually supported by client and server.
4799  *
4800  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
4801  * must not be updated here, otherwise a partially initialized value may
4802  * be exposed. After the connection request is successfully processed,
4803  * the top-level MDT connect request handler atomically updates the export
4804  * connect flags from the obd_connect_data::ocd_connect_flags field of the
4805  * reply. \see mdt_connect().
4806  *
4807  * \param exp   the obd_export associated with this client/target pair
4808  * \param mdt   the target device for the connection
4809  * \param data  stores data for this connect request
4810  *
4811  * \retval 0       success
4812  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
4813  * \retval -EBADE  client and server feature requirements are incompatible
4814  */
4815 static int mdt_connect_internal(struct obd_export *exp,
4816                                 struct mdt_device *mdt,
4817                                 struct obd_connect_data *data)
4818 {
4819         LASSERT(data != NULL);
4820
4821         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
4822         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
4823
4824         if (!(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
4825             !(data->ocd_connect_flags & OBD_CONNECT_IBITS)) {
4826                 CWARN("%s: client %s does not support ibits lock, either "
4827                       "very old or an invalid client: flags "LPX64"\n",
4828                       mdt_obd_name(mdt), exp->exp_client_uuid.uuid,
4829                       data->ocd_connect_flags);
4830                 return -EBADE;
4831         }
4832
4833         if (!mdt->mdt_opts.mo_acl)
4834                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
4835
4836         if (!mdt->mdt_opts.mo_user_xattr)
4837                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
4838
4839         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
4840                 data->ocd_brw_size = min(data->ocd_brw_size,
4841                                          (__u32)MD_MAX_BRW_SIZE);
4842                 if (data->ocd_brw_size == 0) {
4843                         CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
4844                                " ocd_version: %x ocd_grant: %d "
4845                                "ocd_index: %u ocd_brw_size is "
4846                                "unexpectedly zero, network data "
4847                                "corruption? Refusing connection of this"
4848                                " client\n",
4849                                mdt_obd_name(mdt),
4850                                exp->exp_client_uuid.uuid,
4851                                exp, data->ocd_connect_flags, data->ocd_version,
4852                                data->ocd_grant, data->ocd_index);
4853                         return -EPROTO;
4854                 }
4855         }
4856
4857         /* NB: Disregard the rule against updating
4858          * exp_connect_data.ocd_connect_flags in this case, since
4859          * tgt_client_new() needs to know if this is a lightweight
4860          * connection, and it is safe to expose this flag before
4861          * connection processing completes. */
4862         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
4863                 spin_lock(&exp->exp_lock);
4864                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
4865                 spin_unlock(&exp->exp_lock);
4866         }
4867
4868         data->ocd_version = LUSTRE_VERSION_CODE;
4869
4870         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
4871                 CWARN("%s: MDS requires FID support, but client not\n",
4872                       mdt_obd_name(mdt));
4873                 return -EBADE;
4874         }
4875
4876         if (OCD_HAS_FLAG(data, PINGLESS)) {
4877                 if (ptlrpc_pinger_suppress_pings()) {
4878                         spin_lock(&exp->exp_obd->obd_dev_lock);
4879                         list_del_init(&exp->exp_obd_chain_timed);
4880                         spin_unlock(&exp->exp_obd->obd_dev_lock);
4881                 } else {
4882                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
4883                 }
4884         }
4885
4886         data->ocd_max_easize = mdt->mdt_max_ea_size;
4887
4888         /* NB: Disregard the rule against updating
4889          * exp_connect_data.ocd_connect_flags in this case, since
4890          * tgt_client_new() needs to know if this is client supports
4891          * multiple modify RPCs, and it is safe to expose this flag before
4892          * connection processing completes. */
4893         if (data->ocd_connect_flags & OBD_CONNECT_MULTIMODRPCS) {
4894                 data->ocd_maxmodrpcs = max_mod_rpcs_per_client;
4895                 spin_lock(&exp->exp_lock);
4896                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_MULTIMODRPCS;
4897                 spin_unlock(&exp->exp_lock);
4898         }
4899
4900         return 0;
4901 }
4902
4903 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
4904                                    struct mdt_thread_info *info,
4905                                    struct mdt_file_data *mfd)
4906 {
4907         struct lu_context ses;
4908         int rc;
4909         ENTRY;
4910
4911         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
4912         if (rc)
4913                 RETURN(rc);
4914
4915         env->le_ses = &ses;
4916         lu_context_enter(&ses);
4917
4918         mdt_ucred(info)->uc_valid = UCRED_OLD;
4919         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
4920
4921         lu_context_exit(&ses);
4922         lu_context_fini(&ses);
4923         env->le_ses = NULL;
4924
4925         RETURN(rc);
4926 }
4927
4928 static int mdt_export_cleanup(struct obd_export *exp)
4929 {
4930         struct list_head         closing_list;
4931         struct mdt_export_data  *med = &exp->exp_mdt_data;
4932         struct obd_device       *obd = exp->exp_obd;
4933         struct mdt_device       *mdt;
4934         struct mdt_thread_info  *info;
4935         struct lu_env            env;
4936         struct mdt_file_data    *mfd, *n;
4937         int rc = 0;
4938         ENTRY;
4939
4940         INIT_LIST_HEAD(&closing_list);
4941         spin_lock(&med->med_open_lock);
4942         while (!list_empty(&med->med_open_head)) {
4943                 struct list_head *tmp = med->med_open_head.next;
4944                 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
4945
4946                 /* Remove mfd handle so it can't be found again.
4947                  * We are consuming the mfd_list reference here. */
4948                 class_handle_unhash(&mfd->mfd_handle);
4949                 list_move_tail(&mfd->mfd_list, &closing_list);
4950         }
4951         spin_unlock(&med->med_open_lock);
4952         mdt = mdt_dev(obd->obd_lu_dev);
4953         LASSERT(mdt != NULL);
4954
4955         rc = lu_env_init(&env, LCT_MD_THREAD);
4956         if (rc)
4957                 RETURN(rc);
4958
4959         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
4960         LASSERT(info != NULL);
4961         memset(info, 0, sizeof *info);
4962         info->mti_env = &env;
4963         info->mti_mdt = mdt;
4964         info->mti_exp = exp;
4965
4966         if (!list_empty(&closing_list)) {
4967                 struct md_attr *ma = &info->mti_attr;
4968
4969                 /* Close any open files (which may also cause orphan
4970                  * unlinking). */
4971                 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
4972                         list_del_init(&mfd->mfd_list);
4973                         ma->ma_need = ma->ma_valid = 0;
4974
4975                         /* This file is being closed due to an eviction, it
4976                          * could have been modified and now dirty regarding to
4977                          * HSM archive, check this!
4978                          * The logic here is to mark a file dirty if there's a
4979                          * chance it was dirtied before the client was evicted,
4980                          * so that we don't have to wait for a release attempt
4981                          * before finding out the file was actually dirty and
4982                          * fail the release. Aggressively marking it dirty here
4983                          * will cause the policy engine to attempt to
4984                          * re-archive it; when rearchiving, we can compare the
4985                          * current version to the HSM data_version and make the
4986                          * archive request into a noop if it's not actually
4987                          * dirty.
4988                          */
4989                         if (mfd->mfd_mode & FMODE_WRITE)
4990                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
4991
4992                         /* Don't unlink orphan on failover umount, LU-184 */
4993                         if (exp->exp_flags & OBD_OPT_FAILOVER) {
4994                                 ma->ma_valid = MA_FLAGS;
4995                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
4996                         }
4997                         mdt_mfd_close(info, mfd);
4998                 }
4999         }
5000         info->mti_mdt = NULL;
5001         /* cleanup client slot early */
5002         /* Do not erase record for recoverable client. */
5003         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5004                 tgt_client_del(&env, exp);
5005         lu_env_fini(&env);
5006
5007         RETURN(rc);
5008 }
5009
5010 static int mdt_obd_disconnect(struct obd_export *exp)
5011 {
5012         int rc;
5013         ENTRY;
5014
5015         LASSERT(exp);
5016         class_export_get(exp);
5017
5018         rc = server_disconnect_export(exp);
5019         if (rc != 0)
5020                 CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
5021
5022         rc = mdt_export_cleanup(exp);
5023         nodemap_del_member(exp);
5024         class_export_put(exp);
5025         RETURN(rc);
5026 }
5027
5028 /* mds_connect copy */
5029 static int mdt_obd_connect(const struct lu_env *env,
5030                            struct obd_export **exp, struct obd_device *obd,
5031                            struct obd_uuid *cluuid,
5032                            struct obd_connect_data *data,
5033                            void *localdata)
5034 {
5035         struct obd_export       *lexp;
5036         struct lustre_handle    conn = { 0 };
5037         struct mdt_device       *mdt;
5038         int                      rc;
5039         lnet_nid_t              *client_nid = localdata;
5040         ENTRY;
5041
5042         LASSERT(env != NULL);
5043         if (!exp || !obd || !cluuid)
5044                 RETURN(-EINVAL);
5045
5046         mdt = mdt_dev(obd->obd_lu_dev);
5047
5048         /*
5049          * first, check whether the stack is ready to handle requests
5050          * XXX: probably not very appropriate method is used now
5051          *      at some point we should find a better one
5052          */
5053         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) && data != NULL &&
5054             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) &&
5055             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
5056                 rc = obd_get_info(env, mdt->mdt_child_exp,
5057                                   sizeof(KEY_OSP_CONNECTED),
5058                                   KEY_OSP_CONNECTED, NULL, NULL);
5059                 if (rc)
5060                         RETURN(-EAGAIN);
5061                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
5062         }
5063
5064         rc = class_connect(&conn, obd, cluuid);
5065         if (rc)
5066                 RETURN(rc);
5067
5068         lexp = class_conn2export(&conn);
5069         LASSERT(lexp != NULL);
5070
5071         rc = nodemap_add_member(*client_nid, lexp);
5072         if (rc != 0 && rc != -EEXIST)
5073                 GOTO(out, rc);
5074
5075         rc = mdt_connect_internal(lexp, mdt, data);
5076         if (rc == 0) {
5077                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5078
5079                 LASSERT(lcd);
5080                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5081                 rc = tgt_client_new(env, lexp);
5082                 if (rc == 0)
5083                         mdt_export_stats_init(obd, lexp, localdata);
5084         }
5085 out:
5086         if (rc != 0) {
5087                 class_disconnect(lexp);
5088                 nodemap_del_member(lexp);
5089                 *exp = NULL;
5090         } else {
5091                 *exp = lexp;
5092         }
5093
5094         RETURN(rc);
5095 }
5096
5097 static int mdt_obd_reconnect(const struct lu_env *env,
5098                              struct obd_export *exp, struct obd_device *obd,
5099                              struct obd_uuid *cluuid,
5100                              struct obd_connect_data *data,
5101                              void *localdata)
5102 {
5103         lnet_nid_t             *client_nid = localdata;
5104         int                     rc;
5105         ENTRY;
5106
5107         if (exp == NULL || obd == NULL || cluuid == NULL)
5108                 RETURN(-EINVAL);
5109
5110         rc = nodemap_add_member(*client_nid, exp);
5111         if (rc != 0 && rc != -EEXIST)
5112                 RETURN(rc);
5113
5114         rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
5115         if (rc == 0)
5116                 mdt_export_stats_init(obd, exp, localdata);
5117         else
5118                 nodemap_del_member(exp);
5119
5120         RETURN(rc);
5121 }
5122
5123 /* FIXME: Can we avoid using these two interfaces? */
5124 static int mdt_init_export(struct obd_export *exp)
5125 {
5126         struct mdt_export_data *med = &exp->exp_mdt_data;
5127         int                     rc;
5128         ENTRY;
5129
5130         INIT_LIST_HEAD(&med->med_open_head);
5131         spin_lock_init(&med->med_open_lock);
5132         mutex_init(&med->med_idmap_mutex);
5133         med->med_idmap = NULL;
5134         spin_lock(&exp->exp_lock);
5135         exp->exp_connecting = 1;
5136         spin_unlock(&exp->exp_lock);
5137
5138         /* self-export doesn't need client data and ldlm initialization */
5139         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5140                                      &exp->exp_client_uuid)))
5141                 RETURN(0);
5142
5143         rc = tgt_client_alloc(exp);
5144         if (rc)
5145                 GOTO(err, rc);
5146
5147         rc = ldlm_init_export(exp);
5148         if (rc)
5149                 GOTO(err_free, rc);
5150
5151         RETURN(rc);
5152
5153 err_free:
5154         tgt_client_free(exp);
5155 err:
5156         CERROR("%s: Failed to initialize export: rc = %d\n",
5157                exp->exp_obd->obd_name, rc);
5158         return rc;
5159 }
5160
5161 static int mdt_destroy_export(struct obd_export *exp)
5162 {
5163         ENTRY;
5164
5165         if (exp_connect_rmtclient(exp))
5166                 mdt_cleanup_idmap(&exp->exp_mdt_data);
5167
5168         target_destroy_export(exp);
5169         /* destroy can be called from failed obd_setup, so
5170          * checking uuid is safer than obd_self_export */
5171         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5172                                      &exp->exp_client_uuid)))
5173                 RETURN(0);
5174
5175         ldlm_destroy_export(exp);
5176         tgt_client_free(exp);
5177
5178         LASSERT(list_empty(&exp->exp_outstanding_replies));
5179         LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
5180
5181         RETURN(0);
5182 }
5183
5184 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
5185                    struct linkea_data *ldata)
5186 {
5187         int rc;
5188
5189         LASSERT(ldata->ld_buf->lb_buf != NULL);
5190
5191         if (!mdt_object_exists(mdt_obj))
5192                 return -ENODATA;
5193
5194         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5195                           ldata->ld_buf, XATTR_NAME_LINK);
5196         if (rc == -ERANGE) {
5197                 /* Buf was too small, figure out what we need. */
5198                 lu_buf_free(ldata->ld_buf);
5199                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5200                                   ldata->ld_buf, XATTR_NAME_LINK);
5201                 if (rc < 0)
5202                         return rc;
5203                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
5204                 if (ldata->ld_buf->lb_buf == NULL)
5205                         return -ENOMEM;
5206                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5207                                   ldata->ld_buf, XATTR_NAME_LINK);
5208         }
5209         if (rc < 0)
5210                 return rc;
5211
5212         return linkea_init(ldata);
5213 }
5214
5215 /**
5216  * Given an MDT object, try to look up the full path to the object.
5217  * Part of the MDT layer implementation of lfs fid2path.
5218  *
5219  * \param[in]     info  Per-thread common data shared by MDT level handlers.
5220  * \param[in]     obj   Object to do path lookup of
5221  * \param[in,out] fp    User-provided struct to store path information
5222  *
5223  * \retval 0 Lookup successful, path information stored in fp
5224  * \retval -EAGAIN Lookup failed, usually because object is being moved
5225  * \retval negative errno if there was a problem
5226  */
5227 static int mdt_path_current(struct mdt_thread_info *info,
5228                             struct mdt_object *obj,
5229                             struct getinfo_fid2path *fp)
5230 {
5231         struct mdt_device       *mdt = info->mti_mdt;
5232         struct mdt_object       *mdt_obj;
5233         struct link_ea_header   *leh;
5234         struct link_ea_entry    *lee;
5235         struct lu_name          *tmpname = &info->mti_name;
5236         struct lu_fid           *tmpfid = &info->mti_tmp_fid1;
5237         struct lu_buf           *buf = &info->mti_big_buf;
5238         char                    *ptr;
5239         int                     reclen;
5240         struct linkea_data      ldata = { NULL };
5241         int                     rc = 0;
5242         bool                    first = true;
5243         ENTRY;
5244
5245         /* temp buffer for path element, the buffer will be finally freed
5246          * in mdt_thread_info_fini */
5247         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
5248         if (buf->lb_buf == NULL)
5249                 RETURN(-ENOMEM);
5250
5251         ldata.ld_buf = buf;
5252         ptr = fp->gf_path + fp->gf_pathlen - 1;
5253         *ptr = 0;
5254         --ptr;
5255         *tmpfid = fp->gf_fid = *mdt_object_fid(obj);
5256
5257         /* root FID only exists on MDT0, and fid2path should also ends at MDT0,
5258          * so checking root_fid can only happen on MDT0. */
5259         while (!lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid)) {
5260                 struct lu_buf           lmv_buf;
5261
5262                 mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
5263                 if (IS_ERR(mdt_obj))
5264                         GOTO(out, rc = PTR_ERR(mdt_obj));
5265
5266                 if (!mdt_object_exists(mdt_obj)) {
5267                         mdt_object_put(info->mti_env, mdt_obj);
5268                         GOTO(out, rc = -ENOENT);
5269                 }
5270
5271                 if (mdt_object_remote(mdt_obj)) {
5272                         mdt_object_put(info->mti_env, mdt_obj);
5273                         GOTO(remote_out, rc = -EREMOTE);
5274                 }
5275
5276                 rc = mdt_links_read(info, mdt_obj, &ldata);
5277                 if (rc != 0) {
5278                         mdt_object_put(info->mti_env, mdt_obj);
5279                         GOTO(out, rc);
5280                 }
5281
5282                 leh = buf->lb_buf;
5283                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
5284                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
5285                 /* If set, use link #linkno for path lookup, otherwise use
5286                    link #0.  Only do this for the final path element. */
5287                 if (first && fp->gf_linkno < leh->leh_reccount) {
5288                         int count;
5289                         for (count = 0; count < fp->gf_linkno; count++) {
5290                                 lee = (struct link_ea_entry *)
5291                                      ((char *)lee + reclen);
5292                                 linkea_entry_unpack(lee, &reclen, tmpname,
5293                                                     tmpfid);
5294                         }
5295                         if (fp->gf_linkno < leh->leh_reccount - 1)
5296                                 /* indicate to user there are more links */
5297                                 fp->gf_linkno++;
5298                 }
5299
5300                 lmv_buf.lb_buf = info->mti_xattr_buf;
5301                 lmv_buf.lb_len = sizeof(info->mti_xattr_buf);
5302                 /* Check if it is slave stripes */
5303                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5304                                   &lmv_buf, XATTR_NAME_LMV);
5305                 mdt_object_put(info->mti_env, mdt_obj);
5306                 if (rc > 0) {
5307                         union lmv_mds_md *lmm = lmv_buf.lb_buf;
5308
5309                         /* For slave stripes, get its master */
5310                         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE) {
5311                                 fp->gf_fid = *tmpfid;
5312                                 continue;
5313                         }
5314                 } else if (rc < 0 && rc != -ENODATA) {
5315                         GOTO(out, rc);
5316                 }
5317
5318                 rc = 0;
5319
5320                 /* Pack the name in the end of the buffer */
5321                 ptr -= tmpname->ln_namelen;
5322                 if (ptr - 1 <= fp->gf_path)
5323                         GOTO(out, rc = -EOVERFLOW);
5324                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
5325                 *(--ptr) = '/';
5326
5327                 /* keep the last resolved fid to the client, so the
5328                  * client will build the left path on another MDT for
5329                  * remote object */
5330                 fp->gf_fid = *tmpfid;
5331
5332                 first = false;
5333         }
5334
5335 remote_out:
5336         ptr++; /* skip leading / */
5337         memmove(fp->gf_path, ptr, fp->gf_path + fp->gf_pathlen - ptr);
5338
5339 out:
5340         RETURN(rc);
5341 }
5342
5343 /**
5344  * Given an MDT object, use mdt_path_current to get the path.
5345  * Essentially a wrapper to retry mdt_path_current a set number of times
5346  * if -EAGAIN is returned (usually because an object is being moved).
5347  *
5348  * Part of the MDT layer implementation of lfs fid2path.
5349  *
5350  * \param[in]     info  Per-thread common data shared by mdt level handlers.
5351  * \param[in]     obj   Object to do path lookup of
5352  * \param[in,out] fp    User-provided struct for arguments and to store path
5353  *                      information
5354  *
5355  * \retval 0 Lookup successful, path information stored in fp
5356  * \retval negative errno if there was a problem
5357  */
5358 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
5359                     struct getinfo_fid2path *fp)
5360 {
5361         struct mdt_device       *mdt = info->mti_mdt;
5362         int                     tries = 3;
5363         int                     rc = -EAGAIN;
5364         ENTRY;
5365
5366         if (fp->gf_pathlen < 3)
5367                 RETURN(-EOVERFLOW);
5368
5369         if (lu_fid_eq(&mdt->mdt_md_root_fid, mdt_object_fid(obj))) {
5370                 fp->gf_path[0] = '\0';
5371                 RETURN(0);
5372         }
5373
5374         /* Retry multiple times in case file is being moved */
5375         while (tries-- && rc == -EAGAIN)
5376                 rc = mdt_path_current(info, obj, fp);
5377
5378         RETURN(rc);
5379 }
5380
5381 /**
5382  * Get the full path of the provided FID, as of changelog record recno.
5383  *
5384  * This checks sanity and looks up object for user provided FID
5385  * before calling the actual path lookup code.
5386  *
5387  * Part of the MDT layer implementation of lfs fid2path.
5388  *
5389  * \param[in]     info  Per-thread common data shared by mdt level handlers.
5390  * \param[in,out] fp    User-provided struct for arguments and to store path
5391  *                      information
5392  *
5393  * \retval 0 Lookup successful, path information and recno stored in fp
5394  * \retval -ENOENT, object does not exist
5395  * \retval negative errno if there was a problem
5396  */
5397 static int mdt_fid2path(struct mdt_thread_info *info,
5398                         struct getinfo_fid2path *fp)
5399 {
5400         struct mdt_device *mdt = info->mti_mdt;
5401         struct mdt_object *obj;
5402         int    rc;
5403         ENTRY;
5404
5405         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
5406                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
5407
5408         if (!fid_is_sane(&fp->gf_fid))
5409                 RETURN(-EINVAL);
5410
5411         if (!fid_is_namespace_visible(&fp->gf_fid)) {
5412                 CWARN("%s: "DFID" is invalid, sequence should be "
5413                       ">= "LPX64"\n", mdt_obd_name(mdt),
5414                       PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
5415                 RETURN(-EINVAL);
5416         }
5417
5418         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
5419         if (IS_ERR(obj)) {
5420                 rc = PTR_ERR(obj);
5421                 CDEBUG(D_IOCTL, "cannot find "DFID": rc = %d\n",
5422                        PFID(&fp->gf_fid), rc);
5423                 RETURN(rc);
5424         }
5425
5426         if (mdt_object_remote(obj))
5427                 rc = -EREMOTE;
5428         else if (!mdt_object_exists(obj))
5429                 rc = -ENOENT;
5430         else
5431                 rc = 0;
5432
5433         if (rc < 0) {
5434                 mdt_object_put(info->mti_env, obj);
5435                 CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n",
5436                        PFID(&fp->gf_fid), rc);
5437                 RETURN(rc);
5438         }
5439
5440         rc = mdt_path(info, obj, fp);
5441
5442         CDEBUG(D_INFO, "fid "DFID", path %s recno "LPX64" linkno %u\n",
5443                PFID(&fp->gf_fid), fp->gf_path, fp->gf_recno, fp->gf_linkno);
5444
5445         mdt_object_put(info->mti_env, obj);
5446
5447         RETURN(rc);
5448 }
5449
5450 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key,
5451                             void *val, int vallen)
5452 {
5453         struct getinfo_fid2path *fpout, *fpin;
5454         int rc = 0;
5455
5456         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
5457         fpout = val;
5458
5459         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5460                 lustre_swab_fid2path(fpin);
5461
5462         memcpy(fpout, fpin, sizeof(*fpin));
5463         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
5464                 RETURN(-EINVAL);
5465
5466         rc = mdt_fid2path(info, fpout);
5467         RETURN(rc);
5468 }
5469
5470 int mdt_get_info(struct tgt_session_info *tsi)
5471 {
5472         char    *key;
5473         int      keylen;
5474         __u32   *vallen;
5475         void    *valout;
5476         int      rc;
5477
5478         ENTRY;
5479
5480         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
5481         if (key == NULL) {
5482                 CDEBUG(D_IOCTL, "No GETINFO key\n");
5483                 RETURN(err_serious(-EFAULT));
5484         }
5485         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
5486                                       RCL_CLIENT);
5487
5488         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
5489         if (vallen == NULL) {
5490                 CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n",
5491                                 tgt_name(tsi->tsi_tgt));
5492                 RETURN(err_serious(-EFAULT));
5493         }
5494
5495         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
5496                              *vallen);
5497         rc = req_capsule_server_pack(tsi->tsi_pill);
5498         if (rc)
5499                 RETURN(err_serious(rc));
5500
5501         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
5502         if (valout == NULL) {
5503                 CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n",
5504                                 tgt_name(tsi->tsi_tgt));
5505                 RETURN(err_serious(-EFAULT));
5506         }
5507
5508         if (KEY_IS(KEY_FID2PATH)) {
5509                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
5510
5511                 rc = mdt_rpc_fid2path(info, key, valout, *vallen);
5512                 mdt_thread_info_fini(info);
5513         } else {
5514                 rc = -EINVAL;
5515         }
5516         RETURN(rc);
5517 }
5518
5519 /* Pass the ioc down */
5520 static int mdt_ioc_child(struct lu_env *env, struct mdt_device *mdt,
5521                          unsigned int cmd, int len, void *data)
5522 {
5523         struct lu_context ioctl_session;
5524         struct md_device *next = mdt->mdt_child;
5525         int rc;
5526         ENTRY;
5527
5528         rc = lu_context_init(&ioctl_session, LCT_SERVER_SESSION);
5529         if (rc)
5530                 RETURN(rc);
5531         ioctl_session.lc_thread = (struct ptlrpc_thread *)current;
5532         lu_context_enter(&ioctl_session);
5533         env->le_ses = &ioctl_session;
5534
5535         LASSERT(next->md_ops->mdo_iocontrol);
5536         rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data);
5537
5538         lu_context_exit(&ioctl_session);
5539         lu_context_fini(&ioctl_session);
5540         RETURN(rc);
5541 }
5542
5543 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
5544 {
5545         struct obd_ioctl_data *data = karg;
5546         struct lu_fid *fid;
5547         __u64 version;
5548         struct mdt_object *obj;
5549         struct mdt_lock_handle  *lh;
5550         int rc;
5551         ENTRY;
5552
5553         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
5554             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
5555                 RETURN(-EINVAL);
5556
5557         fid = (struct lu_fid *)data->ioc_inlbuf1;
5558
5559         if (!fid_is_sane(fid))
5560                 RETURN(-EINVAL);
5561
5562         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
5563
5564         lh = &mti->mti_lh[MDT_LH_PARENT];
5565         mdt_lock_reg_init(lh, LCK_CR);
5566
5567         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
5568         if (IS_ERR(obj))
5569                 RETURN(PTR_ERR(obj));
5570
5571         if (mdt_object_remote(obj)) {
5572                 rc = -EREMOTE;
5573                 /**
5574                  * before calling version get the correct MDS should be
5575                  * fid, this is error to find remote object here
5576                  */
5577                 CERROR("nonlocal object "DFID"\n", PFID(fid));
5578         } else if (!mdt_object_exists(obj)) {
5579                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
5580                 rc = -ENOENT;
5581         } else {
5582                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
5583                *(__u64 *)data->ioc_inlbuf2 = version;
5584                 rc = 0;
5585         }
5586         mdt_object_unlock_put(mti, obj, lh, 1);
5587         RETURN(rc);
5588 }
5589
5590 /* ioctls on obd dev */
5591 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
5592                          void *karg, void *uarg)
5593 {
5594         struct lu_env      env;
5595         struct obd_device *obd = exp->exp_obd;
5596         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
5597         struct dt_device  *dt = mdt->mdt_bottom;
5598         int rc;
5599
5600         ENTRY;
5601         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
5602         rc = lu_env_init(&env, LCT_MD_THREAD);
5603         if (rc)
5604                 RETURN(rc);
5605
5606         switch (cmd) {
5607         case OBD_IOC_SYNC:
5608                 rc = mdt_device_sync(&env, mdt);
5609                 break;
5610         case OBD_IOC_SET_READONLY:
5611                 rc = dt->dd_ops->dt_ro(&env, dt);
5612                 break;
5613         case OBD_IOC_ABORT_RECOVERY:
5614                 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
5615                 obd->obd_force_abort_recovery = 1;
5616                 target_stop_recovery_thread(obd);
5617                 rc = 0;
5618                 break;
5619         case OBD_IOC_CHANGELOG_REG:
5620         case OBD_IOC_CHANGELOG_DEREG:
5621         case OBD_IOC_CHANGELOG_CLEAR:
5622                 rc = mdt_ioc_child(&env, mdt, cmd, len, karg);
5623                 break;
5624         case OBD_IOC_START_LFSCK: {
5625                 struct md_device *next = mdt->mdt_child;
5626                 struct obd_ioctl_data *data = karg;
5627                 struct lfsck_start_param lsp;
5628
5629                 if (unlikely(data == NULL)) {
5630                         rc = -EINVAL;
5631                         break;
5632                 }
5633
5634                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
5635                 lsp.lsp_index_valid = 0;
5636                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
5637                 break;
5638         }
5639         case OBD_IOC_STOP_LFSCK: {
5640                 struct md_device        *next = mdt->mdt_child;
5641                 struct obd_ioctl_data   *data = karg;
5642                 struct lfsck_stop        stop;
5643
5644                 stop.ls_status = LS_STOPPED;
5645                 /* Old lfsck utils may pass NULL @stop. */
5646                 if (data->ioc_inlbuf1 == NULL)
5647                         stop.ls_flags = 0;
5648                 else
5649                         stop.ls_flags =
5650                         ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
5651
5652                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
5653                 break;
5654         }
5655         case OBD_IOC_GET_OBJ_VERSION: {
5656                 struct mdt_thread_info *mti;
5657                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5658                 memset(mti, 0, sizeof *mti);
5659                 mti->mti_env = &env;
5660                 mti->mti_mdt = mdt;
5661                 mti->mti_exp = exp;
5662
5663                 rc = mdt_ioc_version_get(mti, karg);
5664                 break;
5665         }
5666         case OBD_IOC_CATLOGLIST: {
5667                 struct mdt_thread_info *mti;
5668
5669                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5670                 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
5671                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
5672                                        &mti->mti_tmp_fid1);
5673                 break;
5674          }
5675         default:
5676                 rc = -EOPNOTSUPP;
5677                 CERROR("%s: Not supported cmd = %d, rc = %d\n",
5678                         mdt_obd_name(mdt), cmd, rc);
5679         }
5680
5681         lu_env_fini(&env);
5682         RETURN(rc);
5683 }
5684
5685 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
5686 {
5687         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
5688         int rc;
5689         ENTRY;
5690
5691         if (!mdt->mdt_skip_lfsck) {
5692                 struct lfsck_start_param lsp;
5693
5694                 lsp.lsp_start = NULL;
5695                 lsp.lsp_index_valid = 0;
5696                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
5697                                                            OBD_IOC_START_LFSCK,
5698                                                            0, &lsp);
5699                 if (rc != 0 && rc != -EALREADY)
5700                         CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
5701                               mdt_obd_name(mdt), rc);
5702         }
5703
5704         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
5705         RETURN(rc);
5706 }
5707
5708 static int mdt_obd_postrecov(struct obd_device *obd)
5709 {
5710         struct lu_env env;
5711         int rc;
5712
5713         rc = lu_env_init(&env, LCT_MD_THREAD);
5714         if (rc)
5715                 RETURN(rc);
5716         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
5717         lu_env_fini(&env);
5718         return rc;
5719 }
5720
5721 static struct obd_ops mdt_obd_device_ops = {
5722         .o_owner          = THIS_MODULE,
5723         .o_set_info_async = mdt_obd_set_info_async,
5724         .o_connect        = mdt_obd_connect,
5725         .o_reconnect      = mdt_obd_reconnect,
5726         .o_disconnect     = mdt_obd_disconnect,
5727         .o_init_export    = mdt_init_export,
5728         .o_destroy_export = mdt_destroy_export,
5729         .o_iocontrol      = mdt_iocontrol,
5730         .o_postrecov      = mdt_obd_postrecov,
5731 };
5732
5733 static struct lu_device* mdt_device_fini(const struct lu_env *env,
5734                                          struct lu_device *d)
5735 {
5736         struct mdt_device *m = mdt_dev(d);
5737         ENTRY;
5738
5739         mdt_fini(env, m);
5740         RETURN(NULL);
5741 }
5742
5743 static struct lu_device *mdt_device_free(const struct lu_env *env,
5744                                          struct lu_device *d)
5745 {
5746         struct mdt_device *m = mdt_dev(d);
5747         ENTRY;
5748
5749         lu_device_fini(&m->mdt_lu_dev);
5750         OBD_FREE_PTR(m);
5751
5752         RETURN(NULL);
5753 }
5754
5755 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
5756                                           struct lu_device_type *t,
5757                                           struct lustre_cfg *cfg)
5758 {
5759         struct lu_device  *l;
5760         struct mdt_device *m;
5761
5762         OBD_ALLOC_PTR(m);
5763         if (m != NULL) {
5764                 int rc;
5765
5766                 l = &m->mdt_lu_dev;
5767                 rc = mdt_init0(env, m, t, cfg);
5768                 if (rc != 0) {
5769                         mdt_device_free(env, l);
5770                         l = ERR_PTR(rc);
5771                         return l;
5772                 }
5773         } else
5774                 l = ERR_PTR(-ENOMEM);
5775         return l;
5776 }
5777
5778 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
5779 LU_KEY_INIT(mdt, struct mdt_thread_info);
5780
5781 static void mdt_key_fini(const struct lu_context *ctx,
5782                          struct lu_context_key *key, void* data)
5783 {
5784         struct mdt_thread_info *info = data;
5785
5786         if (info->mti_big_lmm) {
5787                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
5788                 info->mti_big_lmm = NULL;
5789                 info->mti_big_lmmsize = 0;
5790         }
5791         OBD_FREE_PTR(info);
5792 }
5793
5794 /* context key: mdt_thread_key */
5795 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
5796
5797 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
5798 {
5799         return lu_ucred(info->mti_env);
5800 }
5801
5802 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
5803 {
5804         return lu_ucred_check(info->mti_env);
5805 }
5806
5807 /**
5808  * Enable/disable COS (Commit On Sharing).
5809  *
5810  * Set/Clear the COS flag in mdt options.
5811  *
5812  * \param mdt mdt device
5813  * \param val 0 disables COS, other values enable COS
5814  */
5815 void mdt_enable_cos(struct mdt_device *mdt, int val)
5816 {
5817         struct lu_env env;
5818         int rc;
5819
5820         mdt->mdt_opts.mo_cos = !!val;
5821         rc = lu_env_init(&env, LCT_LOCAL);
5822         if (unlikely(rc != 0)) {
5823                 CWARN("%s: lu_env initialization failed, cannot "
5824                       "sync: rc = %d\n", mdt_obd_name(mdt), rc);
5825                 return;
5826         }
5827         mdt_device_sync(&env, mdt);
5828         lu_env_fini(&env);
5829 }
5830
5831 /**
5832  * Check COS (Commit On Sharing) status.
5833  *
5834  * Return COS flag status.
5835  *
5836  * \param mdt mdt device
5837  */
5838 int mdt_cos_is_enabled(struct mdt_device *mdt)
5839 {
5840         return mdt->mdt_opts.mo_cos != 0;
5841 }
5842
5843 static struct lu_device_type_operations mdt_device_type_ops = {
5844         .ldto_device_alloc = mdt_device_alloc,
5845         .ldto_device_free  = mdt_device_free,
5846         .ldto_device_fini  = mdt_device_fini
5847 };
5848
5849 static struct lu_device_type mdt_device_type = {
5850         .ldt_tags     = LU_DEVICE_MD,
5851         .ldt_name     = LUSTRE_MDT_NAME,
5852         .ldt_ops      = &mdt_device_type_ops,
5853         .ldt_ctx_tags = LCT_MD_THREAD
5854 };
5855
5856 static int __init mdt_init(void)
5857 {
5858         int rc;
5859
5860         CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") ==
5861                  FID_NOBRACE_LEN + 1);
5862         CLASSERT(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") ==
5863                  FID_LEN + 1);
5864         rc = lu_kmem_init(mdt_caches);
5865         if (rc)
5866                 return rc;
5867
5868         rc = mds_mod_init();
5869         if (rc)
5870                 GOTO(lu_fini, rc);
5871
5872         rc = class_register_type(&mdt_obd_device_ops, NULL, true, NULL,
5873                                  LUSTRE_MDT_NAME, &mdt_device_type);
5874         if (rc)
5875                 GOTO(mds_fini, rc);
5876 lu_fini:
5877         if (rc)
5878                 lu_kmem_fini(mdt_caches);
5879 mds_fini:
5880         if (rc)
5881                 mds_mod_exit();
5882         return rc;
5883 }
5884
5885 static void __exit mdt_exit(void)
5886 {
5887         class_unregister_type(LUSTRE_MDT_NAME);
5888         mds_mod_exit();
5889         lu_kmem_fini(mdt_caches);
5890 }
5891
5892 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
5893 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
5894 MODULE_VERSION(LUSTRE_VERSION_STRING);
5895 MODULE_LICENSE("GPL");
5896
5897 module_init(mdt_init);
5898 module_exit(mdt_exit);