lustre/mdt/mdt_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2010, 2014, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/mdt/mdt_handler.c
  37  *
  38  * Lustre Metadata Target (mdt) request handler
  39  *
  40  * Author: Peter Braam <braam@clusterfs.com>
  41  * Author: Andreas Dilger <adilger@clusterfs.com>
  42  * Author: Phil Schwan <phil@clusterfs.com>
  43  * Author: Mike Shaver <shaver@clusterfs.com>
  44  * Author: Nikita Danilov <nikita@clusterfs.com>
  45  * Author: Huang Hua <huanghua@clusterfs.com>
  46  * Author: Yury Umanets <umka@clusterfs.com>
  47  */
  48
  49 #define DEBUG_SUBSYSTEM S_MDS
  50
  51 #include <linux/module.h>
  52 /*
  53  * struct OBD_{ALLOC,FREE}*()
  54  */
  55 #include <obd_support.h>
  56 #include <lustre_ioctl.h>
  57 /* struct ptlrpc_request */
  58 #include <lustre_net.h>
  59 /* struct obd_export */
  60 #include <lustre_export.h>
  61 /* struct obd_device */
  62 #include <obd.h>
  63 /* lu2dt_dev() */
  64 #include <dt_object.h>
  65 #include <lustre_mds.h>
  66 #include <lustre_log.h>
  67 #include "mdt_internal.h"
  68 #include <lustre_acl.h>
  69 #include <lustre_param.h>
  70 #include <lustre_quota.h>
  71 #include <lustre_lfsck.h>
  72 #include <lustre_nodemap.h>
  73
  74 mdl_mode_t mdt_mdl_lock_modes[] = {
  75         [LCK_MINMODE] = MDL_MINMODE,
  76         [LCK_EX]      = MDL_EX,
  77         [LCK_PW]      = MDL_PW,
  78         [LCK_PR]      = MDL_PR,
  79         [LCK_CW]      = MDL_CW,
  80         [LCK_CR]      = MDL_CR,
  81         [LCK_NL]      = MDL_NL,
  82         [LCK_GROUP]   = MDL_GROUP
  83 };
  84
  85 ldlm_mode_t mdt_dlm_lock_modes[] = {
  86         [MDL_MINMODE] = LCK_MINMODE,
  87         [MDL_EX]      = LCK_EX,
  88         [MDL_PW]      = LCK_PW,
  89         [MDL_PR]      = LCK_PR,
  90         [MDL_CW]      = LCK_CW,
  91         [MDL_CR]      = LCK_CR,
  92         [MDL_NL]      = LCK_NL,
  93         [MDL_GROUP]   = LCK_GROUP
  94 };
  95
  96 static struct mdt_device *mdt_dev(struct lu_device *d);
  97 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
  98
  99 static const struct lu_object_operations mdt_obj_ops;
 100
 101 /* Slab for MDT object allocation */
 102 static struct kmem_cache *mdt_object_kmem;
 103
 104 /* For HSM restore handles */
 105 struct kmem_cache *mdt_hsm_cdt_kmem;
 106
 107 /* For HSM request handles */
 108 struct kmem_cache *mdt_hsm_car_kmem;
 109
 110 static struct lu_kmem_descr mdt_caches[] = {
 111         {
 112                 .ckd_cache = &mdt_object_kmem,
 113                 .ckd_name  = "mdt_obj",
 114                 .ckd_size  = sizeof(struct mdt_object)
 115         },
 116         {
 117                 .ckd_cache      = &mdt_hsm_cdt_kmem,
 118                 .ckd_name       = "mdt_cdt_restore_handle",
 119                 .ckd_size       = sizeof(struct cdt_restore_handle)
 120         },
 121         {
 122                 .ckd_cache      = &mdt_hsm_car_kmem,
 123                 .ckd_name       = "mdt_cdt_agent_req",
 124                 .ckd_size       = sizeof(struct cdt_agent_req)
 125         },
 126         {
 127                 .ckd_cache = NULL
 128         }
 129 };
 130
 131 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag)
 132 {
 133         if (!rep)
 134                 return 0;
 135         return rep->lock_policy_res1 & op_flag;
 136 }
 137
 138 void mdt_clear_disposition(struct mdt_thread_info *info,
 139                            struct ldlm_reply *rep, __u64 op_flag)
 140 {
 141         if (info) {
 142                 info->mti_opdata &= ~op_flag;
 143                 tgt_opdata_clear(info->mti_env, op_flag);
 144         }
 145         if (rep)
 146                 rep->lock_policy_res1 &= ~op_flag;
 147 }
 148
 149 void mdt_set_disposition(struct mdt_thread_info *info,
 150                          struct ldlm_reply *rep, __u64 op_flag)
 151 {
 152         if (info) {
 153                 info->mti_opdata |= op_flag;
 154                 tgt_opdata_set(info->mti_env, op_flag);
 155         }
 156         if (rep)
 157                 rep->lock_policy_res1 |= op_flag;
 158 }
 159
 160 void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
 161 {
 162         lh->mlh_pdo_hash = 0;
 163         lh->mlh_reg_mode = lm;
 164         lh->mlh_rreg_mode = lm;
 165         lh->mlh_type = MDT_REG_LOCK;
 166 }
 167
 168 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lock_mode,
 169                        const struct lu_name *lname)
 170 {
 171         lh->mlh_reg_mode = lock_mode;
 172         lh->mlh_rreg_mode = lock_mode;
 173         lh->mlh_type = MDT_PDO_LOCK;
 174
 175         if (lu_name_is_valid(lname)) {
 176                 lh->mlh_pdo_hash = full_name_hash(lname->ln_name,
 177                                                   lname->ln_namelen);
 178                 /* XXX Workaround for LU-2856
 179                  *
 180                  * Zero is a valid return value of full_name_hash, but
 181                  * several users of mlh_pdo_hash assume a non-zero
 182                  * hash value. We therefore map zero onto an
 183                  * arbitrary, but consistent value (1) to avoid
 184                  * problems further down the road. */
 185                 if (unlikely(lh->mlh_pdo_hash == 0))
 186                         lh->mlh_pdo_hash = 1;
 187         } else {
 188                 lh->mlh_pdo_hash = 0;
 189         }
 190 }
 191
 192 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 193                               struct mdt_lock_handle *lh)
 194 {
 195         mdl_mode_t mode;
 196         ENTRY;
 197
 198         /*
 199          * Any dir access needs couple of locks:
 200          *
 201          * 1) on part of dir we gonna take lookup/modify;
 202          *
 203          * 2) on whole dir to protect it from concurrent splitting and/or to
 204          * flush client's cache for readdir().
 205          *
 206          * so, for a given mode and object this routine decides what lock mode
 207          * to use for lock #2:
 208          *
 209          * 1) if caller's gonna lookup in dir then we need to protect dir from
 210          * being splitted only - LCK_CR
 211          *
 212          * 2) if caller's gonna modify dir then we need to protect dir from
 213          * being splitted and to flush cache - LCK_CW
 214          *
 215          * 3) if caller's gonna modify dir and that dir seems ready for
 216          * splitting then we need to protect it from any type of access
 217          * (lookup/modify/split) - LCK_EX --bzzz
 218          */
 219
 220         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 221         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 222
 223         /*
 224          * Ask underlaying level its opinion about preferable PDO lock mode
 225          * having access type passed as regular lock mode:
 226          *
 227          * - MDL_MINMODE means that lower layer does not want to specify lock
 228          * mode;
 229          *
 230          * - MDL_NL means that no PDO lock should be taken. This is used in some
 231          * cases. Say, for non-splittable directories no need to use PDO locks
 232          * at all.
 233          */
 234         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 235                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 236
 237         if (mode != MDL_MINMODE) {
 238                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 239         } else {
 240                 /*
 241                  * Lower layer does not want to specify locking mode. We do it
 242                  * our selves. No special protection is needed, just flush
 243                  * client's cache on modification and allow concurrent
 244                  * mondification.
 245                  */
 246                 switch (lh->mlh_reg_mode) {
 247                 case LCK_EX:
 248                         lh->mlh_pdo_mode = LCK_EX;
 249                         break;
 250                 case LCK_PR:
 251                         lh->mlh_pdo_mode = LCK_CR;
 252                         break;
 253                 case LCK_PW:
 254                         lh->mlh_pdo_mode = LCK_CW;
 255                         break;
 256                 default:
 257                         CERROR("Not expected lock type (0x%x)\n",
 258                                (int)lh->mlh_reg_mode);
 259                         LBUG();
 260                 }
 261         }
 262
 263         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 264         EXIT;
 265 }
 266
 267 static int mdt_getstatus(struct tgt_session_info *tsi)
 268 {
 269         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
 270         struct mdt_device       *mdt = info->mti_mdt;
 271         struct mdt_body         *repbody;
 272         int                      rc;
 273
 274         ENTRY;
 275
 276         rc = mdt_check_ucred(info);
 277         if (rc)
 278                 GOTO(out, rc = err_serious(rc));
 279
 280         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
 281                 GOTO(out, rc = err_serious(-ENOMEM));
 282
 283         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 284         repbody->mbo_fid1 = mdt->mdt_md_root_fid;
 285         repbody->mbo_valid |= OBD_MD_FLID;
 286
 287         EXIT;
 288 out:
 289         mdt_thread_info_fini(info);
 290         return rc;
 291 }
 292
 293 static int mdt_statfs(struct tgt_session_info *tsi)
 294 {
 295         struct ptlrpc_request           *req = tgt_ses_req(tsi);
 296         struct mdt_thread_info          *info = tsi2mdt_info(tsi);
 297         struct md_device                *next = info->mti_mdt->mdt_child;
 298         struct ptlrpc_service_part      *svcpt;
 299         struct obd_statfs               *osfs;
 300         int                             rc;
 301
 302         ENTRY;
 303
 304         svcpt = req->rq_rqbd->rqbd_svcpt;
 305
 306         /* This will trigger a watchdog timeout */
 307         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 308                          (MDT_SERVICE_WATCHDOG_FACTOR *
 309                           at_get(&svcpt->scp_at_estimate)) + 1);
 310
 311         rc = mdt_check_ucred(info);
 312         if (rc)
 313                 GOTO(out, rc = err_serious(rc));
 314
 315         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
 316                 GOTO(out, rc = err_serious(-ENOMEM));
 317
 318         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 319         if (!osfs)
 320                 GOTO(out, rc = -EPROTO);
 321
 322         /** statfs information are cached in the mdt_device */
 323         if (cfs_time_before_64(info->mti_mdt->mdt_osfs_age,
 324                                cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS))) {
 325                 /** statfs data is too old, get up-to-date one */
 326                 rc = next->md_ops->mdo_statfs(info->mti_env, next, osfs);
 327                 if (rc)
 328                         GOTO(out, rc);
 329                 spin_lock(&info->mti_mdt->mdt_osfs_lock);
 330                 info->mti_mdt->mdt_osfs = *osfs;
 331                 info->mti_mdt->mdt_osfs_age = cfs_time_current_64();
 332                 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
 333         } else {
 334                 /** use cached statfs data */
 335                 spin_lock(&info->mti_mdt->mdt_osfs_lock);
 336                 *osfs = info->mti_mdt->mdt_osfs;
 337                 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
 338         }
 339
 340         if (rc == 0)
 341                 mdt_counter_incr(req, LPROC_MDT_STATFS);
 342 out:
 343         mdt_thread_info_fini(info);
 344         RETURN(rc);
 345 }
 346
 347 #ifdef CONFIG_FS_POSIX_ACL
 348 /*
 349  * Pack ACL data into the reply. UIDs/GIDs are mapped and filtered by nodemap.
 350  *
 351  * \param       info    thread info object
 352  * \param       repbody reply to pack ACLs into
 353  * \param       o       mdt object of file to examine
 354  * \param       nodemap nodemap of client to reply to
 355  * \retval      0       success
 356  * \retval      -errno  error getting or parsing ACL from disk
 357  */
 358 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
 359                       struct mdt_object *o, struct lu_nodemap *nodemap)
 360 {
 361         const struct lu_env     *env = info->mti_env;
 362         struct md_object        *next = mdt_object_child(o);
 363         struct lu_buf           *buf = &info->mti_buf;
 364         int rc;
 365
 366         buf->lb_buf = req_capsule_server_get(info->mti_pill, &RMF_ACL);
 367         buf->lb_len = req_capsule_get_size(info->mti_pill, &RMF_ACL,
 368                                            RCL_SERVER);
 369         if (buf->lb_len == 0)
 370                 return 0;
 371
 372         rc = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_ACCESS);
 373         if (rc < 0) {
 374                 if (rc == -ENODATA) {
 375                         repbody->mbo_aclsize = 0;
 376                         repbody->mbo_valid |= OBD_MD_FLACL;
 377                         rc = 0;
 378                 } else if (rc == -EOPNOTSUPP) {
 379                         rc = 0;
 380                 } else {
 381                         CERROR("%s: unable to read "DFID" ACL: rc = %d\n",
 382                                mdt_obd_name(info->mti_mdt),
 383                                PFID(mdt_object_fid(o)), rc);
 384                 }
 385         } else {
 386                 rc = nodemap_map_acl(nodemap, buf->lb_buf,
 387                                      rc, NODEMAP_FS_TO_CLIENT);
 388                 /* if all ACLs mapped out, rc is still >= 0 */
 389                 if (rc < 0) {
 390                         CERROR("%s: nodemap_map_acl unable to parse "DFID
 391                                " ACL: rc = %d\n", mdt_obd_name(info->mti_mdt),
 392                                PFID(mdt_object_fid(o)), rc);
 393                 } else {
 394                         repbody->mbo_aclsize = rc;
 395                         repbody->mbo_valid |= OBD_MD_FLACL;
 396                         rc = 0;
 397                 }
 398         }
 399         return rc;
 400 }
 401 #endif
 402
 403 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 404                         const struct lu_attr *attr, const struct lu_fid *fid)
 405 {
 406         struct md_attr          *ma = &info->mti_attr;
 407         struct obd_export       *exp = info->mti_exp;
 408         struct lu_nodemap       *nodemap = exp->exp_target_data.ted_nodemap;
 409
 410         LASSERT(ma->ma_valid & MA_INODE);
 411
 412         if (attr->la_valid & LA_ATIME) {
 413                 b->mbo_atime = attr->la_atime;
 414                 b->mbo_valid |= OBD_MD_FLATIME;
 415         }
 416         if (attr->la_valid & LA_MTIME) {
 417                 b->mbo_mtime = attr->la_mtime;
 418                 b->mbo_valid |= OBD_MD_FLMTIME;
 419         }
 420         if (attr->la_valid & LA_CTIME) {
 421                 b->mbo_ctime = attr->la_ctime;
 422                 b->mbo_valid |= OBD_MD_FLCTIME;
 423         }
 424         if (attr->la_valid & LA_FLAGS) {
 425                 b->mbo_flags = attr->la_flags;
 426                 b->mbo_valid |= OBD_MD_FLFLAGS;
 427         }
 428         if (attr->la_valid & LA_NLINK) {
 429                 b->mbo_nlink = attr->la_nlink;
 430                 b->mbo_valid |= OBD_MD_FLNLINK;
 431         }
 432         if (attr->la_valid & LA_UID) {
 433                 b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 434                                             NODEMAP_FS_TO_CLIENT,
 435                                             attr->la_uid);
 436                 b->mbo_valid |= OBD_MD_FLUID;
 437         }
 438         if (attr->la_valid & LA_GID) {
 439                 b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 440                                             NODEMAP_FS_TO_CLIENT,
 441                                             attr->la_gid);
 442                 b->mbo_valid |= OBD_MD_FLGID;
 443         }
 444         b->mbo_mode = attr->la_mode;
 445         if (attr->la_valid & LA_MODE)
 446                 b->mbo_valid |= OBD_MD_FLMODE;
 447         if (attr->la_valid & LA_TYPE)
 448                 b->mbo_valid |= OBD_MD_FLTYPE;
 449
 450         if (fid != NULL) {
 451                 b->mbo_fid1 = *fid;
 452                 b->mbo_valid |= OBD_MD_FLID;
 453                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid="LPX64"\n",
 454                        PFID(fid), b->mbo_nlink, b->mbo_mode, b->mbo_valid);
 455         }
 456
 457         if (info != NULL)
 458                 mdt_body_reverse_idmap(info, b);
 459
 460         if (!(attr->la_valid & LA_TYPE))
 461                 return;
 462
 463         b->mbo_rdev   = attr->la_rdev;
 464         b->mbo_size   = attr->la_size;
 465         b->mbo_blocks = attr->la_blocks;
 466
 467         if (!S_ISREG(attr->la_mode)) {
 468                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 469         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
 470                 /* means no objects are allocated on osts. */
 471                 LASSERT(!(ma->ma_valid & MA_LOV));
 472                 /* just ignore blocks occupied by extend attributes on MDS */
 473                 b->mbo_blocks = 0;
 474                 /* if no object is allocated on osts, the size on mds is valid.
 475                  * b=22272 */
 476                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 477         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL &&
 478                    ma->ma_lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) {
 479                 /* A released file stores its size on MDS. */
 480                 /* But return 1 block for released file, unless tools like tar
 481                  * will consider it fully sparse. (LU-3864)
 482                  */
 483                 if (unlikely(b->mbo_size == 0))
 484                         b->mbo_blocks = 0;
 485                 else
 486                         b->mbo_blocks = 1;
 487                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 488         }
 489
 490         if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE))
 491                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 492                        PFID(fid), (unsigned long long)b->mbo_size);
 493 }
 494
 495 static inline int mdt_body_has_lov(const struct lu_attr *la,
 496                                    const struct mdt_body *body)
 497 {
 498         return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
 499                (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
 500 }
 501
 502 void mdt_client_compatibility(struct mdt_thread_info *info)
 503 {
 504         struct mdt_body       *body;
 505         struct ptlrpc_request *req = mdt_info_req(info);
 506         struct obd_export     *exp = req->rq_export;
 507         struct md_attr        *ma = &info->mti_attr;
 508         struct lu_attr        *la = &ma->ma_attr;
 509         ENTRY;
 510
 511         if (exp_connect_layout(exp))
 512                 /* the client can deal with 16-bit lmm_stripe_count */
 513                 RETURN_EXIT;
 514
 515         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 516
 517         if (!mdt_body_has_lov(la, body))
 518                 RETURN_EXIT;
 519
 520         /* now we have a reply with a lov for a client not compatible with the
 521          * layout lock so we have to clean the layout generation number */
 522         if (S_ISREG(la->la_mode))
 523                 ma->ma_lmm->lmm_layout_gen = 0;
 524         EXIT;
 525 }
 526
 527 static int mdt_attr_get_eabuf_size(struct mdt_thread_info *info,
 528                                    struct mdt_object *o)
 529 {
 530         const struct lu_env *env = info->mti_env;
 531         int rc, rc2;
 532
 533         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 534                           XATTR_NAME_LOV);
 535
 536         if (rc == -ENODATA)
 537                 rc = 0;
 538
 539         if (rc < 0)
 540                 goto out;
 541
 542         /* Is it a directory? Let's check for the LMV as well */
 543         if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
 544                 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 545                                    XATTR_NAME_LMV);
 546
 547                 if (rc2 == -ENODATA)
 548                         rc2 = mo_xattr_get(env, mdt_object_child(o),
 549                                            &LU_BUF_NULL,
 550                                            XATTR_NAME_DEFAULT_LMV);
 551
 552                 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
 553                         rc = rc2;
 554         }
 555
 556 out:
 557         return rc;
 558 }
 559
 560 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
 561                       const char *name)
 562 {
 563         const struct lu_env *env = info->mti_env;
 564         int rc;
 565         ENTRY;
 566
 567         LASSERT(info->mti_big_lmm_used == 0);
 568         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
 569         if (rc < 0)
 570                 RETURN(rc);
 571
 572         /* big_lmm may need to be grown */
 573         if (info->mti_big_lmmsize < rc) {
 574                 int size = size_roundup_power2(rc);
 575
 576                 if (info->mti_big_lmmsize > 0) {
 577                         /* free old buffer */
 578                         LASSERT(info->mti_big_lmm);
 579                         OBD_FREE_LARGE(info->mti_big_lmm,
 580                                        info->mti_big_lmmsize);
 581                         info->mti_big_lmm = NULL;
 582                         info->mti_big_lmmsize = 0;
 583                 }
 584
 585                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
 586                 if (info->mti_big_lmm == NULL)
 587                         RETURN(-ENOMEM);
 588                 info->mti_big_lmmsize = size;
 589         }
 590         LASSERT(info->mti_big_lmmsize >= rc);
 591
 592         info->mti_buf.lb_buf = info->mti_big_lmm;
 593         info->mti_buf.lb_len = info->mti_big_lmmsize;
 594         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
 595
 596         RETURN(rc);
 597 }
 598
 599 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
 600                    struct md_attr *ma, const char *name)
 601 {
 602         struct md_object *next = mdt_object_child(o);
 603         struct lu_buf    *buf = &info->mti_buf;
 604         int rc;
 605
 606         if (strcmp(name, XATTR_NAME_LOV) == 0) {
 607                 buf->lb_buf = ma->ma_lmm;
 608                 buf->lb_len = ma->ma_lmm_size;
 609                 LASSERT(!(ma->ma_valid & MA_LOV));
 610         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 611                 buf->lb_buf = ma->ma_lmv;
 612                 buf->lb_len = ma->ma_lmv_size;
 613                 LASSERT(!(ma->ma_valid & MA_LMV));
 614         } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 615                 buf->lb_buf = ma->ma_lmv;
 616                 buf->lb_len = ma->ma_lmv_size;
 617                 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
 618         } else {
 619                 return -EINVAL;
 620         }
 621
 622         rc = mo_xattr_get(info->mti_env, next, buf, name);
 623         if (rc > 0) {
 624
 625 got:
 626                 if (strcmp(name, XATTR_NAME_LOV) == 0) {
 627                         if (info->mti_big_lmm_used)
 628                                 ma->ma_lmm = info->mti_big_lmm;
 629
 630                         /* NOT return LOV EA with hole to old client. */
 631                         if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
 632                                      LOV_PATTERN_F_HOLE) &&
 633                             !(exp_connect_flags(info->mti_exp) &
 634                               OBD_CONNECT_LFSCK)) {
 635                                 return -EIO;
 636                         } else {
 637                                 ma->ma_lmm_size = rc;
 638                                 ma->ma_valid |= MA_LOV;
 639                         }
 640                 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 641                         if (info->mti_big_lmm_used)
 642                                 ma->ma_lmv = info->mti_big_lmm;
 643
 644                         ma->ma_lmv_size = rc;
 645                         ma->ma_valid |= MA_LMV;
 646                 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 647                         ma->ma_lmv_size = rc;
 648                         ma->ma_valid |= MA_LMV_DEF;
 649                 }
 650
 651                 /* Update mdt_max_mdsize so all clients will be aware that */
 652                 if (info->mti_mdt->mdt_max_mdsize < rc)
 653                         info->mti_mdt->mdt_max_mdsize = rc;
 654
 655                 rc = 0;
 656         } else if (rc == -ENODATA) {
 657                 /* no LOV EA */
 658                 rc = 0;
 659         } else if (rc == -ERANGE) {
 660                 /* Default LMV has fixed size, so it must be able to fit
 661                  * in the original buffer */
 662                 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
 663                         return rc;
 664                 rc = mdt_big_xattr_get(info, o, name);
 665                 if (rc > 0) {
 666                         info->mti_big_lmm_used = 1;
 667                         goto got;
 668                 }
 669         }
 670
 671         return rc;
 672 }
 673
 674 static int mdt_attr_get_pfid(struct mdt_thread_info *info,
 675                              struct mdt_object *o, struct lu_fid *pfid)
 676 {
 677         struct lu_buf           *buf = &info->mti_buf;
 678         struct link_ea_header   *leh;
 679         struct link_ea_entry    *lee;
 680         int                      rc;
 681         ENTRY;
 682
 683         buf->lb_buf = info->mti_big_lmm;
 684         buf->lb_len = info->mti_big_lmmsize;
 685         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
 686                           buf, XATTR_NAME_LINK);
 687         /* ignore errors, MA_PFID won't be set and it is
 688          * up to the caller to treat this as an error */
 689         if (rc == -ERANGE || buf->lb_len == 0) {
 690                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
 691                 buf->lb_buf = info->mti_big_lmm;
 692                 buf->lb_len = info->mti_big_lmmsize;
 693         }
 694
 695         if (rc < 0)
 696                 RETURN(rc);
 697         if (rc < sizeof(*leh)) {
 698                 CERROR("short LinkEA on "DFID": rc = %d\n",
 699                        PFID(mdt_object_fid(o)), rc);
 700                 RETURN(-ENODATA);
 701         }
 702
 703         leh = (struct link_ea_header *) buf->lb_buf;
 704         lee = (struct link_ea_entry *)(leh + 1);
 705         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
 706                 leh->leh_magic = LINK_EA_MAGIC;
 707                 leh->leh_reccount = __swab32(leh->leh_reccount);
 708                 leh->leh_len = __swab64(leh->leh_len);
 709         }
 710         if (leh->leh_magic != LINK_EA_MAGIC)
 711                 RETURN(-EINVAL);
 712         if (leh->leh_reccount == 0)
 713                 RETURN(-ENODATA);
 714
 715         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
 716         fid_be_to_cpu(pfid, pfid);
 717
 718         RETURN(0);
 719 }
 720
 721 int mdt_attr_get_complex(struct mdt_thread_info *info,
 722                          struct mdt_object *o, struct md_attr *ma)
 723 {
 724         const struct lu_env *env = info->mti_env;
 725         struct md_object    *next = mdt_object_child(o);
 726         struct lu_buf       *buf = &info->mti_buf;
 727         int                  need = ma->ma_need;
 728         int                  rc = 0, rc2;
 729         u32                  mode;
 730         ENTRY;
 731
 732         ma->ma_valid = 0;
 733
 734         if (mdt_object_exists(o) == 0)
 735                 GOTO(out, rc = -ENOENT);
 736         mode = lu_object_attr(&next->mo_lu);
 737
 738         if (need & MA_INODE) {
 739                 ma->ma_need = MA_INODE;
 740                 rc = mo_attr_get(env, next, ma);
 741                 if (rc)
 742                         GOTO(out, rc);
 743                 ma->ma_valid |= MA_INODE;
 744         }
 745
 746         if (need & MA_PFID) {
 747                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
 748                 if (rc == 0)
 749                         ma->ma_valid |= MA_PFID;
 750                 /* ignore this error, parent fid is not mandatory */
 751                 rc = 0;
 752         }
 753
 754         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
 755                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
 756                 if (rc)
 757                         GOTO(out, rc);
 758         }
 759
 760         if (need & MA_LMV && S_ISDIR(mode)) {
 761                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
 762                 if (rc != 0)
 763                         GOTO(out, rc);
 764         }
 765
 766         if (need & MA_LMV_DEF && S_ISDIR(mode)) {
 767                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
 768                 if (rc != 0)
 769                         GOTO(out, rc);
 770         }
 771
 772         if (need & MA_HSM && S_ISREG(mode)) {
 773                 buf->lb_buf = info->mti_xattr_buf;
 774                 buf->lb_len = sizeof(info->mti_xattr_buf);
 775                 CLASSERT(sizeof(struct hsm_attrs) <=
 776                          sizeof(info->mti_xattr_buf));
 777                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
 778                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
 779                 if (rc2 == 0)
 780                         ma->ma_valid |= MA_HSM;
 781                 else if (rc2 < 0 && rc2 != -ENODATA)
 782                         GOTO(out, rc = rc2);
 783         }
 784
 785 #ifdef CONFIG_FS_POSIX_ACL
 786         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
 787                 buf->lb_buf = ma->ma_acl;
 788                 buf->lb_len = ma->ma_acl_size;
 789                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
 790                 if (rc2 > 0) {
 791                         ma->ma_acl_size = rc2;
 792                         ma->ma_valid |= MA_ACL_DEF;
 793                 } else if (rc2 == -ENODATA) {
 794                         /* no ACLs */
 795                         ma->ma_acl_size = 0;
 796                 } else
 797                         GOTO(out, rc = rc2);
 798         }
 799 #endif
 800 out:
 801         ma->ma_need = need;
 802         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
 803                rc, ma->ma_valid, ma->ma_lmm);
 804         RETURN(rc);
 805 }
 806
 807 static int mdt_getattr_internal(struct mdt_thread_info *info,
 808                                 struct mdt_object *o, int ma_need)
 809 {
 810         struct md_object        *next = mdt_object_child(o);
 811         const struct mdt_body   *reqbody = info->mti_body;
 812         struct ptlrpc_request   *req = mdt_info_req(info);
 813         struct md_attr          *ma = &info->mti_attr;
 814         struct lu_attr          *la = &ma->ma_attr;
 815         struct req_capsule      *pill = info->mti_pill;
 816         const struct lu_env     *env = info->mti_env;
 817         struct mdt_body         *repbody;
 818         struct lu_buf           *buffer = &info->mti_buf;
 819         struct obd_export       *exp = info->mti_exp;
 820         struct lu_nodemap       *nodemap = exp->exp_target_data.ted_nodemap;
 821         int                      rc;
 822         int                      is_root;
 823         ENTRY;
 824
 825         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
 826                 RETURN(err_serious(-ENOMEM));
 827
 828         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 829
 830         ma->ma_valid = 0;
 831
 832         if (mdt_object_remote(o)) {
 833                 /* This object is located on remote node.*/
 834                 /* Return -ENOTSUPP for old client */
 835                 if (!mdt_is_dne_client(req->rq_export))
 836                         GOTO(out, rc = -ENOTSUPP);
 837
 838                 repbody->mbo_fid1 = *mdt_object_fid(o);
 839                 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
 840                 GOTO(out, rc = 0);
 841         }
 842
 843         if (reqbody->mbo_eadatasize > 0) {
 844                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
 845                 if (buffer->lb_buf == NULL)
 846                         GOTO(out, rc = -EPROTO);
 847                 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
 848                                                       RCL_SERVER);
 849         } else {
 850                 buffer->lb_buf = NULL;
 851                 buffer->lb_len = 0;
 852                 ma_need &= ~(MA_LOV | MA_LMV);
 853                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
 854                        mdt_obd_name(info->mti_mdt),
 855                        req->rq_export->exp_client_uuid.uuid);
 856         }
 857
 858         /* If it is dir object and client require MEA, then we got MEA */
 859         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 860             (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
 861                 /* Assumption: MDT_MD size is enough for lmv size. */
 862                 ma->ma_lmv = buffer->lb_buf;
 863                 ma->ma_lmv_size = buffer->lb_len;
 864                 ma->ma_need = MA_INODE;
 865                 if (ma->ma_lmv_size > 0) {
 866                         if (reqbody->mbo_valid & OBD_MD_MEA)
 867                                 ma->ma_need |= MA_LMV;
 868                         else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA)
 869                                 ma->ma_need |= MA_LMV_DEF;
 870                 }
 871         } else {
 872                 ma->ma_lmm = buffer->lb_buf;
 873                 ma->ma_lmm_size = buffer->lb_len;
 874                 ma->ma_need = MA_INODE | MA_HSM;
 875                 if (ma->ma_lmm_size > 0)
 876                         ma->ma_need |= MA_LOV;
 877         }
 878
 879         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 880             reqbody->mbo_valid & OBD_MD_FLDIREA  &&
 881             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
 882                 /* get default stripe info for this dir. */
 883                 ma->ma_need |= MA_LOV_DEF;
 884         }
 885         ma->ma_need |= ma_need;
 886
 887         rc = mdt_attr_get_complex(info, o, ma);
 888         if (unlikely(rc)) {
 889                 CERROR("%s: getattr error for "DFID": rc = %d\n",
 890                        mdt_obd_name(info->mti_mdt),
 891                        PFID(mdt_object_fid(o)), rc);
 892                 RETURN(rc);
 893         }
 894
 895         /* if file is released, check if a restore is running */
 896         if ((ma->ma_valid & MA_HSM) && (ma->ma_hsm.mh_flags & HS_RELEASED) &&
 897             mdt_hsm_restore_is_running(info, mdt_object_fid(o))) {
 898                 repbody->mbo_t_state = MS_RESTORE;
 899                 repbody->mbo_valid |= OBD_MD_TSTATE;
 900         }
 901
 902         is_root = lu_fid_eq(mdt_object_fid(o), &info->mti_mdt->mdt_md_root_fid);
 903
 904         /* the Lustre protocol supposes to return default striping
 905          * on the user-visible root if explicitly requested */
 906         if ((ma->ma_valid & MA_LOV) == 0 && S_ISDIR(la->la_mode) &&
 907             (ma->ma_need & MA_LOV_DEF && is_root) && ma->ma_need & MA_LOV) {
 908                 struct lu_fid      rootfid;
 909                 struct mdt_object *root;
 910                 struct mdt_device *mdt = info->mti_mdt;
 911
 912                 rc = dt_root_get(env, mdt->mdt_bottom, &rootfid);
 913                 if (rc)
 914                         RETURN(rc);
 915                 root = mdt_object_find(env, mdt, &rootfid);
 916                 if (IS_ERR(root))
 917                         RETURN(PTR_ERR(root));
 918                 rc = mdt_stripe_get(info, root, ma, XATTR_NAME_LOV);
 919                 mdt_object_put(info->mti_env, root);
 920                 if (unlikely(rc)) {
 921                         CERROR("%s: getattr error for "DFID": rc = %d\n",
 922                                mdt_obd_name(info->mti_mdt),
 923                                PFID(mdt_object_fid(o)), rc);
 924                         RETURN(rc);
 925                 }
 926         }
 927
 928         if (likely(ma->ma_valid & MA_INODE))
 929                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
 930         else
 931                 RETURN(-EFAULT);
 932
 933         if (mdt_body_has_lov(la, reqbody)) {
 934                 if (ma->ma_valid & MA_LOV) {
 935                         LASSERT(ma->ma_lmm_size);
 936                         repbody->mbo_eadatasize = ma->ma_lmm_size;
 937                         if (S_ISDIR(la->la_mode))
 938                                 repbody->mbo_valid |= OBD_MD_FLDIREA;
 939                         else
 940                                 repbody->mbo_valid |= OBD_MD_FLEASIZE;
 941                         mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
 942                 }
 943                 if (ma->ma_valid & MA_LMV) {
 944                         /* Return -ENOTSUPP for old client */
 945                         if (!mdt_is_striped_client(req->rq_export))
 946                                 RETURN(-ENOTSUPP);
 947
 948                         LASSERT(S_ISDIR(la->la_mode));
 949                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
 950                         repbody->mbo_eadatasize = ma->ma_lmv_size;
 951                         repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
 952                 }
 953                 if (ma->ma_valid & MA_LMV_DEF) {
 954                         /* Return -ENOTSUPP for old client */
 955                         if (!mdt_is_striped_client(req->rq_export))
 956                                 RETURN(-ENOTSUPP);
 957                         LASSERT(S_ISDIR(la->la_mode));
 958                         repbody->mbo_eadatasize = ma->ma_lmv_size;
 959                         repbody->mbo_valid |= (OBD_MD_FLDIREA |
 960                                                OBD_MD_DEFAULT_MEA);
 961                 }
 962         } else if (S_ISLNK(la->la_mode) &&
 963                    reqbody->mbo_valid & OBD_MD_LINKNAME) {
 964                 buffer->lb_buf = ma->ma_lmm;
 965                 /* eadatasize from client includes NULL-terminator, so
 966                  * there is no need to read it */
 967                 buffer->lb_len = reqbody->mbo_eadatasize - 1;
 968                 rc = mo_readlink(env, next, buffer);
 969                 if (unlikely(rc <= 0)) {
 970                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
 971                                mdt_obd_name(info->mti_mdt),
 972                                PFID(mdt_object_fid(o)), rc);
 973                         rc = -EFAULT;
 974                 } else {
 975                         int print_limit = min_t(int, PAGE_CACHE_SIZE - 128, rc);
 976
 977                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
 978                                 rc -= 2;
 979                         repbody->mbo_valid |= OBD_MD_LINKNAME;
 980                         /* we need to report back size with NULL-terminator
 981                          * because client expects that */
 982                         repbody->mbo_eadatasize = rc + 1;
 983                         if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
 984                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
 985                                        "on "DFID ", expected %d\n",
 986                                        mdt_obd_name(info->mti_mdt),
 987                                        rc, PFID(mdt_object_fid(o)),
 988                                        reqbody->mbo_eadatasize - 1);
 989                         /* NULL terminate */
 990                         ((char *)ma->ma_lmm)[rc] = 0;
 991
 992                         /* If the total CDEBUG() size is larger than a page, it
 993                          * will print a warning to the console, avoid this by
 994                          * printing just the last part of the symlink. */
 995                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
 996                                print_limit < rc ? "..." : "", print_limit,
 997                                (char *)ma->ma_lmm + rc - print_limit, rc);
 998                         rc = 0;
 999                 }
1000         }
1001
1002         if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
1003                 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1004                 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1005                 CDEBUG(D_INODE, "changing the max MD size to %u\n",
1006                        repbody->mbo_max_mdsize);
1007         }
1008
1009         if (exp_connect_rmtclient(info->mti_exp) &&
1010             reqbody->mbo_valid & OBD_MD_FLRMTPERM) {
1011                 void *buf = req_capsule_server_get(pill, &RMF_ACL);
1012
1013                 /* mdt_getattr_lock only */
1014                 rc = mdt_pack_remote_perm(info, o, buf);
1015                 if (rc) {
1016                         repbody->mbo_valid &= ~OBD_MD_FLRMTPERM;
1017                         repbody->mbo_aclsize = 0;
1018                         RETURN(rc);
1019                 } else {
1020                         repbody->mbo_valid |= OBD_MD_FLRMTPERM;
1021                         repbody->mbo_aclsize = sizeof(struct mdt_remote_perm);
1022                 }
1023         }
1024 #ifdef CONFIG_FS_POSIX_ACL
1025         else if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1026                  (reqbody->mbo_valid & OBD_MD_FLACL))
1027                 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
1028 #endif
1029
1030 out:
1031         if (rc == 0)
1032                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
1033
1034         RETURN(rc);
1035 }
1036
1037 static int mdt_getattr(struct tgt_session_info *tsi)
1038 {
1039         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1040         struct mdt_object       *obj = info->mti_object;
1041         struct req_capsule      *pill = info->mti_pill;
1042         struct mdt_body         *reqbody;
1043         struct mdt_body         *repbody;
1044         int rc, rc2;
1045         ENTRY;
1046
1047         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1048         LASSERT(reqbody);
1049         LASSERT(obj != NULL);
1050         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1051
1052         /* Unlike intent case where we need to pre-fill out buffers early on
1053          * in intent policy for ldlm reasons, here we can have a much better
1054          * guess at EA size by just reading it from disk.
1055          * Exceptions are readdir and (missing) directory striping */
1056         /* Readlink */
1057         if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1058                 /* No easy way to know how long is the symlink, but it cannot
1059                  * be more than PATH_MAX, so we allocate +1 */
1060                 rc = PATH_MAX + 1;
1061
1062         /* A special case for fs ROOT: getattr there might fetch
1063          * default EA for entire fs, not just for this dir!
1064          */
1065         } else if (lu_fid_eq(mdt_object_fid(obj),
1066                              &info->mti_mdt->mdt_md_root_fid) &&
1067                    (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1068                    (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1069                                                                  MDS_GETATTR)) {
1070                 /* Should the default strping be bigger, mdt_fix_reply
1071                  * will reallocate */
1072                 rc = DEF_REP_MD_SIZE;
1073         } else {
1074                 /* Read the actual EA size from disk */
1075                 rc = mdt_attr_get_eabuf_size(info, obj);
1076         }
1077
1078         if (rc < 0)
1079                 GOTO(out, rc = err_serious(rc));
1080
1081         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1082
1083         rc = req_capsule_server_pack(pill);
1084         if (unlikely(rc != 0))
1085                 GOTO(out, rc = err_serious(rc));
1086
1087         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1088         LASSERT(repbody != NULL);
1089         repbody->mbo_eadatasize = 0;
1090         repbody->mbo_aclsize = 0;
1091
1092         if (reqbody->mbo_valid & OBD_MD_FLRMTPERM)
1093                 rc = mdt_init_ucred(info, reqbody);
1094         else
1095                 rc = mdt_check_ucred(info);
1096         if (unlikely(rc))
1097                 GOTO(out_shrink, rc);
1098
1099         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1100
1101         rc = mdt_getattr_internal(info, obj, 0);
1102         if (reqbody->mbo_valid & OBD_MD_FLRMTPERM)
1103                 mdt_exit_ucred(info);
1104         EXIT;
1105 out_shrink:
1106         mdt_client_compatibility(info);
1107         rc2 = mdt_fix_reply(info);
1108         if (rc == 0)
1109                 rc = rc2;
1110 out:
1111         mdt_thread_info_fini(info);
1112         return rc;
1113 }
1114
1115 /**
1116  * Exchange MOF_LOV_CREATED flags between two objects after a
1117  * layout swap. No assumption is made on whether o1 or o2 have
1118  * created objects or not.
1119  *
1120  * \param[in,out] o1    First swap layout object
1121  * \param[in,out] o2    Second swap layout object
1122  */
1123 static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2)
1124 {
1125         __u64   o1_flags;
1126
1127         mutex_lock(&o1->mot_lov_mutex);
1128         mutex_lock(&o2->mot_lov_mutex);
1129
1130         o1_flags = o1->mot_flags;
1131         o1->mot_flags = (o1->mot_flags & ~MOF_LOV_CREATED) |
1132                         (o2->mot_flags & MOF_LOV_CREATED);
1133
1134         o2->mot_flags = (o2->mot_flags & ~MOF_LOV_CREATED) |
1135                         (o1_flags & MOF_LOV_CREATED);
1136
1137         mutex_unlock(&o2->mot_lov_mutex);
1138         mutex_unlock(&o1->mot_lov_mutex);
1139 }
1140
1141 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1142 {
1143         struct mdt_thread_info  *info;
1144         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1145         struct obd_export       *exp = req->rq_export;
1146         struct mdt_object       *o1, *o2, *o;
1147         struct mdt_lock_handle  *lh1, *lh2;
1148         struct mdc_swap_layouts *msl;
1149         int                      rc;
1150         ENTRY;
1151
1152         /* client does not support layout lock, so layout swaping
1153          * is disabled.
1154          * FIXME: there is a problem for old clients which don't support
1155          * layout lock yet. If those clients have already opened the file
1156          * they won't be notified at all so that old layout may still be
1157          * used to do IO. This can be fixed after file release is landed by
1158          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1159         if (!exp_connect_layout(exp))
1160                 RETURN(-EOPNOTSUPP);
1161
1162         info = tsi2mdt_info(tsi);
1163
1164         if (info->mti_dlm_req != NULL)
1165                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1166
1167         o1 = info->mti_object;
1168         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1169                                 &info->mti_body->mbo_fid2);
1170         if (IS_ERR(o))
1171                 GOTO(out, rc = PTR_ERR(o));
1172
1173         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1174                 GOTO(put, rc = -ENOENT);
1175
1176         rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1177         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1178                 GOTO(put, rc);
1179
1180         if (rc < 0)
1181                 swap(o1, o2);
1182
1183         /* permission check. Make sure the calling process having permission
1184          * to write both files. */
1185         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1186                                 MAY_WRITE);
1187         if (rc < 0)
1188                 GOTO(put, rc);
1189
1190         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1191                                 MAY_WRITE);
1192         if (rc < 0)
1193                 GOTO(put, rc);
1194
1195         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1196         if (msl == NULL)
1197                 GOTO(put, rc = -EPROTO);
1198
1199         lh1 = &info->mti_lh[MDT_LH_NEW];
1200         mdt_lock_reg_init(lh1, LCK_EX);
1201         lh2 = &info->mti_lh[MDT_LH_OLD];
1202         mdt_lock_reg_init(lh2, LCK_EX);
1203
1204         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1205                              MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK);
1206         if (rc < 0)
1207                 GOTO(put, rc);
1208
1209         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1210                              MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK);
1211         if (rc < 0)
1212                 GOTO(unlock1, rc);
1213
1214         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1215                              mdt_object_child(o2), msl->msl_flags);
1216         if (rc < 0)
1217                 GOTO(unlock2, rc);
1218
1219         mdt_swap_lov_flag(o1, o2);
1220
1221 unlock2:
1222         mdt_object_unlock(info, o2, lh2, rc);
1223 unlock1:
1224         mdt_object_unlock(info, o1, lh1, rc);
1225 put:
1226         mdt_object_put(info->mti_env, o);
1227 out:
1228         mdt_thread_info_fini(info);
1229         RETURN(rc);
1230 }
1231
1232 static int mdt_raw_lookup(struct mdt_thread_info *info,
1233                           struct mdt_object *parent,
1234                           const struct lu_name *lname,
1235                           struct ldlm_reply *ldlm_rep)
1236 {
1237         struct lu_fid   *child_fid = &info->mti_tmp_fid1;
1238         int              rc;
1239         ENTRY;
1240
1241         LASSERT(!info->mti_cross_ref);
1242
1243         /* Only got the fid of this obj by name */
1244         fid_zero(child_fid);
1245         rc = mdo_lookup(info->mti_env, mdt_object_child(info->mti_object),
1246                         lname, child_fid, &info->mti_spec);
1247         if (rc == 0) {
1248                 struct mdt_body *repbody;
1249
1250                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1251                 repbody->mbo_fid1 = *child_fid;
1252                 repbody->mbo_valid = OBD_MD_FLID;
1253                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1254         } else if (rc == -ENOENT) {
1255                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1256         }
1257
1258         RETURN(rc);
1259 }
1260
1261 /*
1262  * UPDATE lock should be taken against parent, and be released before exit;
1263  * child_bits lock should be taken against child, and be returned back:
1264  *            (1)normal request should release the child lock;
1265  *            (2)intent request will grant the lock to client.
1266  */
1267 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1268                                  struct mdt_lock_handle *lhc,
1269                                  __u64 child_bits,
1270                                  struct ldlm_reply *ldlm_rep)
1271 {
1272         struct ptlrpc_request  *req       = mdt_info_req(info);
1273         struct mdt_body        *reqbody   = NULL;
1274         struct mdt_object      *parent    = info->mti_object;
1275         struct mdt_object      *child;
1276         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1277         struct lu_name         *lname     = NULL;
1278         struct mdt_lock_handle *lhp       = NULL;
1279         struct ldlm_lock       *lock;
1280         bool                    is_resent;
1281         bool                    try_layout;
1282         int                     ma_need = 0;
1283         int                     rc;
1284         ENTRY;
1285
1286         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1287         LASSERT(ergo(is_resent,
1288                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1289
1290         if (parent == NULL)
1291                 RETURN(-ENOENT);
1292
1293         if (info->mti_cross_ref) {
1294                 /* Only getattr on the child. Parent is on another node. */
1295                 mdt_set_disposition(info, ldlm_rep,
1296                                     DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1297                 child = parent;
1298                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1299                        "ldlm_rep = %p\n",
1300                        PFID(mdt_object_fid(child)), ldlm_rep);
1301
1302                 rc = mdt_check_resent_lock(info, child, lhc);
1303                 if (rc < 0) {
1304                         RETURN(rc);
1305                 } else if (rc > 0) {
1306                         mdt_lock_handle_init(lhc);
1307                         mdt_lock_reg_init(lhc, LCK_PR);
1308
1309                         /*
1310                          * Object's name is on another MDS, no lookup or layout
1311                          * lock is needed here but update lock is.
1312                          */
1313                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
1314                                         MDS_INODELOCK_LAYOUT);
1315                         child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1316
1317                         rc = mdt_object_lock(info, child, lhc, child_bits,
1318                                              MDT_LOCAL_LOCK);
1319                         if (rc < 0)
1320                                 RETURN(rc);
1321                 }
1322
1323                 /* Finally, we can get attr for child. */
1324                 if (!mdt_object_exists(child)) {
1325                         LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1326                                         &child->mot_obj,
1327                                         "remote object doesn't exist.\n");
1328                         mdt_object_unlock(info, child, lhc, 1);
1329                         RETURN(-ENOENT);
1330                 }
1331
1332                 rc = mdt_getattr_internal(info, child, 0);
1333                 if (unlikely(rc != 0))
1334                         mdt_object_unlock(info, child, lhc, 1);
1335
1336                 RETURN(rc);
1337         }
1338
1339         lname = &info->mti_name;
1340         mdt_name_unpack(info->mti_pill, &RMF_NAME, lname, MNF_FIX_ANON);
1341
1342         if (lu_name_is_valid(lname)) {
1343                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1344                        "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1345                        PNAME(lname), ldlm_rep);
1346         } else {
1347                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1348                 if (unlikely(reqbody == NULL))
1349                         RETURN(err_serious(-EPROTO));
1350
1351                 *child_fid = reqbody->mbo_fid2;
1352
1353                 if (unlikely(!fid_is_sane(child_fid)))
1354                         RETURN(err_serious(-EINVAL));
1355
1356                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1357                        "ldlm_rep = %p\n",
1358                        PFID(mdt_object_fid(parent)),
1359                        PFID(&reqbody->mbo_fid2), ldlm_rep);
1360         }
1361
1362         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1363
1364         if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1365                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1366                                 &parent->mot_obj,
1367                                 "Parent doesn't exist!\n");
1368                 RETURN(-ESTALE);
1369         }
1370
1371         if (mdt_object_remote(parent)) {
1372                 CERROR("%s: parent "DFID" is on remote target\n",
1373                        mdt_obd_name(info->mti_mdt),
1374                        PFID(mdt_object_fid(parent)));
1375                 RETURN(-EIO);
1376         }
1377
1378         if (lu_name_is_valid(lname)) {
1379                 /* Always allow to lookup ".." */
1380                 if (unlikely(lname->ln_namelen == 2 &&
1381                              lname->ln_name[0] == '.' &&
1382                              lname->ln_name[1] == '.'))
1383                         info->mti_spec.sp_permitted = 1;
1384
1385                 if (info->mti_body->mbo_valid == OBD_MD_FLID) {
1386                         rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1387
1388                         RETURN(rc);
1389                 }
1390
1391                 /* step 1: lock parent only if parent is a directory */
1392                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1393                         lhp = &info->mti_lh[MDT_LH_PARENT];
1394                         mdt_lock_pdo_init(lhp, LCK_PR, lname);
1395                         rc = mdt_object_lock(info, parent, lhp,
1396                                              MDS_INODELOCK_UPDATE,
1397                                              MDT_LOCAL_LOCK);
1398                         if (unlikely(rc != 0))
1399                                 RETURN(rc);
1400                 }
1401
1402                 /* step 2: lookup child's fid by name */
1403                 fid_zero(child_fid);
1404                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
1405                                 child_fid, &info->mti_spec);
1406                 if (rc == -ENOENT)
1407                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1408
1409                 if (rc != 0)
1410                         GOTO(out_parent, rc);
1411         }
1412
1413         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1414
1415         /*
1416          *step 3: find the child object by fid & lock it.
1417          *        regardless if it is local or remote.
1418          *
1419          *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1420          *      set parent dir fid the same as child fid in getattr by fid case
1421          *      we should not lu_object_find() the object again, could lead
1422          *      to hung if there is a concurrent unlink destroyed the object.
1423          */
1424         if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1425                 mdt_object_get(info->mti_env, parent);
1426                 child = parent;
1427         } else {
1428                 child = mdt_object_find(info->mti_env, info->mti_mdt,
1429                                         child_fid);
1430         }
1431
1432         if (unlikely(IS_ERR(child)))
1433                 GOTO(out_parent, rc = PTR_ERR(child));
1434
1435         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
1436         rc = mdt_check_resent_lock(info, child, lhc);
1437         if (rc < 0) {
1438                 GOTO(out_child, rc);
1439         } else if (rc > 0) {
1440                 mdt_lock_handle_init(lhc);
1441                 mdt_lock_reg_init(lhc, LCK_PR);
1442                 try_layout = false;
1443
1444                 if (!mdt_object_exists(child)) {
1445                         LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1446                                         &child->mot_obj,
1447                                         "Object doesn't exist!\n");
1448                         GOTO(out_child, rc = -ENOENT);
1449                 }
1450
1451                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1452                       mdt_object_exists(child) && !mdt_object_remote(child)) {
1453                         struct md_attr *ma = &info->mti_attr;
1454
1455                         ma->ma_valid = 0;
1456                         ma->ma_need = MA_INODE;
1457                         rc = mdt_attr_get_complex(info, child, ma);
1458                         if (unlikely(rc != 0))
1459                                 GOTO(out_child, rc);
1460
1461                         /* If the file has not been changed for some time, we
1462                          * return not only a LOOKUP lock, but also an UPDATE
1463                          * lock and this might save us RPC on later STAT. For
1464                          * directories, it also let negative dentry cache start
1465                          * working for this dir. */
1466                         if (ma->ma_valid & MA_INODE &&
1467                             ma->ma_attr.la_valid & LA_CTIME &&
1468                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1469                                 ma->ma_attr.la_ctime < cfs_time_current_sec())
1470                                 child_bits |= MDS_INODELOCK_UPDATE;
1471                 }
1472
1473                 /* layout lock must be granted in a best-effort way
1474                  * for IT operations */
1475                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1476                 if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1477                     exp_connect_layout(info->mti_exp) &&
1478                     S_ISREG(lu_object_attr(&child->mot_obj)) &&
1479                     !mdt_object_remote(child) && ldlm_rep != NULL) {
1480                         /* try to grant layout lock for regular file. */
1481                         try_layout = true;
1482                 }
1483
1484                 rc = 0;
1485                 if (try_layout) {
1486                         child_bits |= MDS_INODELOCK_LAYOUT;
1487                         /* try layout lock, it may fail to be granted due to
1488                          * contention at LOOKUP or UPDATE */
1489                         if (!mdt_object_lock_try(info, child, lhc, child_bits,
1490                                                  MDT_CROSS_LOCK)) {
1491                                 child_bits &= ~MDS_INODELOCK_LAYOUT;
1492                                 LASSERT(child_bits != 0);
1493                                 rc = mdt_object_lock(info, child, lhc,
1494                                                 child_bits, MDT_CROSS_LOCK);
1495                         } else {
1496                                 ma_need |= MA_LOV;
1497                         }
1498                 } else {
1499                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1500                          * client will enqueue the lock to the remote MDT */
1501                         if (mdt_object_remote(child))
1502                                 child_bits &= ~MDS_INODELOCK_UPDATE;
1503                         rc = mdt_object_lock(info, child, lhc, child_bits,
1504                                                 MDT_CROSS_LOCK);
1505                 }
1506                 if (unlikely(rc != 0))
1507                         GOTO(out_child, rc);
1508         }
1509
1510         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1511
1512         /* finally, we can get attr for child. */
1513         rc = mdt_getattr_internal(info, child, ma_need);
1514         if (unlikely(rc != 0)) {
1515                 mdt_object_unlock(info, child, lhc, 1);
1516         } else if (lock) {
1517                 /* Debugging code. */
1518                 LDLM_DEBUG(lock, "Returning lock to client");
1519                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1520                                          &lock->l_resource->lr_name),
1521                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1522                          PLDLMRES(lock->l_resource),
1523                          PFID(mdt_object_fid(child)));
1524         }
1525         if (lock)
1526                 LDLM_LOCK_PUT(lock);
1527
1528         EXIT;
1529 out_child:
1530         mdt_object_put(info->mti_env, child);
1531 out_parent:
1532         if (lhp)
1533                 mdt_object_unlock(info, parent, lhp, 1);
1534         return rc;
1535 }
1536
1537 /* normal handler: should release the child lock */
1538 static int mdt_getattr_name(struct tgt_session_info *tsi)
1539 {
1540         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1541         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1542         struct mdt_body        *reqbody;
1543         struct mdt_body        *repbody;
1544         int rc, rc2;
1545         ENTRY;
1546
1547         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1548         LASSERT(reqbody != NULL);
1549         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1550         LASSERT(repbody != NULL);
1551
1552         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1553         repbody->mbo_eadatasize = 0;
1554         repbody->mbo_aclsize = 0;
1555
1556         rc = mdt_init_ucred(info, reqbody);
1557         if (unlikely(rc))
1558                 GOTO(out_shrink, rc);
1559
1560         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1561         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1562                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1563                 lhc->mlh_reg_lh.cookie = 0;
1564         }
1565         mdt_exit_ucred(info);
1566         EXIT;
1567 out_shrink:
1568         mdt_client_compatibility(info);
1569         rc2 = mdt_fix_reply(info);
1570         if (rc == 0)
1571                 rc = rc2;
1572         mdt_thread_info_fini(info);
1573         return rc;
1574 }
1575
1576 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1577                          void *karg, void *uarg);
1578
1579 static int mdt_set_info(struct tgt_session_info *tsi)
1580 {
1581         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1582         char                    *key;
1583         void                    *val;
1584         int                      keylen, vallen, rc = 0;
1585
1586         ENTRY;
1587
1588         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
1589         if (key == NULL) {
1590                 DEBUG_REQ(D_HA, req, "no set_info key");
1591                 RETURN(err_serious(-EFAULT));
1592         }
1593
1594         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
1595                                       RCL_CLIENT);
1596
1597         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
1598         if (val == NULL) {
1599                 DEBUG_REQ(D_HA, req, "no set_info val");
1600                 RETURN(err_serious(-EFAULT));
1601         }
1602
1603         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
1604                                       RCL_CLIENT);
1605
1606         /* Swab any part of val you need to here */
1607         if (KEY_IS(KEY_READ_ONLY)) {
1608                 spin_lock(&req->rq_export->exp_lock);
1609                 if (*(__u32 *)val)
1610                         *exp_connect_flags_ptr(req->rq_export) |=
1611                                 OBD_CONNECT_RDONLY;
1612                 else
1613                         *exp_connect_flags_ptr(req->rq_export) &=
1614                                 ~OBD_CONNECT_RDONLY;
1615                 spin_unlock(&req->rq_export->exp_lock);
1616         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1617                 struct changelog_setinfo *cs = val;
1618
1619                 if (vallen != sizeof(*cs)) {
1620                         CERROR("%s: bad changelog_clear setinfo size %d\n",
1621                                tgt_name(tsi->tsi_tgt), vallen);
1622                         RETURN(-EINVAL);
1623                 }
1624                 if (ptlrpc_req_need_swab(req)) {
1625                         __swab64s(&cs->cs_recno);
1626                         __swab32s(&cs->cs_id);
1627                 }
1628
1629                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
1630                                    vallen, val, NULL);
1631         } else {
1632                 RETURN(-EINVAL);
1633         }
1634         RETURN(rc);
1635 }
1636
1637 static int mdt_readpage(struct tgt_session_info *tsi)
1638 {
1639         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
1640         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
1641         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
1642         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
1643         struct mdt_body         *repbody;
1644         int                      rc;
1645         int                      i;
1646
1647         ENTRY;
1648
1649         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1650                 RETURN(err_serious(-ENOMEM));
1651
1652         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
1653         if (repbody == NULL || reqbody == NULL)
1654                 RETURN(err_serious(-EFAULT));
1655
1656         /*
1657          * prepare @rdpg before calling lower layers and transfer itself. Here
1658          * reqbody->size contains offset of where to start to read and
1659          * reqbody->nlink contains number bytes to read.
1660          */
1661         rdpg->rp_hash = reqbody->mbo_size;
1662         if (rdpg->rp_hash != reqbody->mbo_size) {
1663                 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1664                        rdpg->rp_hash, reqbody->mbo_size);
1665                 RETURN(-EFAULT);
1666         }
1667
1668         rdpg->rp_attrs = reqbody->mbo_mode;
1669         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
1670                 rdpg->rp_attrs |= LUDA_64BITHASH;
1671         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
1672                                 exp_max_brw_size(tsi->tsi_exp));
1673         rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE - 1) >>
1674                           PAGE_CACHE_SHIFT;
1675         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1676         if (rdpg->rp_pages == NULL)
1677                 RETURN(-ENOMEM);
1678
1679         for (i = 0; i < rdpg->rp_npages; ++i) {
1680                 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1681                 if (rdpg->rp_pages[i] == NULL)
1682                         GOTO(free_rdpg, rc = -ENOMEM);
1683         }
1684
1685         /* call lower layers to fill allocated pages with directory data */
1686         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
1687         if (rc < 0)
1688                 GOTO(free_rdpg, rc);
1689
1690         /* send pages to client */
1691         rc = tgt_sendpage(tsi, rdpg, rc);
1692
1693         EXIT;
1694 free_rdpg:
1695
1696         for (i = 0; i < rdpg->rp_npages; i++)
1697                 if (rdpg->rp_pages[i] != NULL)
1698                         __free_page(rdpg->rp_pages[i]);
1699         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1700
1701         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1702                 RETURN(0);
1703
1704         return rc;
1705 }
1706
1707 static int mdt_reint_internal(struct mdt_thread_info *info,
1708                               struct mdt_lock_handle *lhc,
1709                               __u32 op)
1710 {
1711         struct req_capsule      *pill = info->mti_pill;
1712         struct mdt_body         *repbody;
1713         int                      rc = 0, rc2;
1714
1715         ENTRY;
1716
1717         rc = mdt_reint_unpack(info, op);
1718         if (rc != 0) {
1719                 CERROR("Can't unpack reint, rc %d\n", rc);
1720                 RETURN(err_serious(rc));
1721         }
1722
1723         /* for replay (no_create) lmm is not needed, client has it already */
1724         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1725                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1726                                      DEF_REP_MD_SIZE);
1727
1728         /* llog cookies are always 0, the field is kept for compatibility */
1729         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1730                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
1731
1732         rc = req_capsule_server_pack(pill);
1733         if (rc != 0) {
1734                 CERROR("Can't pack response, rc %d\n", rc);
1735                 RETURN(err_serious(rc));
1736         }
1737
1738         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1739                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1740                 LASSERT(repbody);
1741                 repbody->mbo_eadatasize = 0;
1742                 repbody->mbo_aclsize = 0;
1743         }
1744
1745         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1746
1747         /* for replay no cookkie / lmm need, because client have this already */
1748         if (info->mti_spec.no_create)
1749                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1750                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1751
1752         rc = mdt_init_ucred_reint(info);
1753         if (rc)
1754                 GOTO(out_shrink, rc);
1755
1756         rc = mdt_fix_attr_ucred(info, op);
1757         if (rc != 0)
1758                 GOTO(out_ucred, rc = err_serious(rc));
1759
1760         if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1761                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
1762                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1763                 GOTO(out_ucred, rc);
1764         }
1765         rc = mdt_reint_rec(info, lhc);
1766         EXIT;
1767 out_ucred:
1768         mdt_exit_ucred(info);
1769 out_shrink:
1770         mdt_client_compatibility(info);
1771         rc2 = mdt_fix_reply(info);
1772         if (rc == 0)
1773                 rc = rc2;
1774         return rc;
1775 }
1776
1777 static long mdt_reint_opcode(struct ptlrpc_request *req,
1778                              const struct req_format **fmt)
1779 {
1780         struct mdt_device       *mdt;
1781         struct mdt_rec_reint    *rec;
1782         long                     opc;
1783
1784         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
1785         if (rec != NULL) {
1786                 opc = rec->rr_opcode;
1787                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
1788                 if (opc < REINT_MAX && fmt[opc] != NULL)
1789                         req_capsule_extend(&req->rq_pill, fmt[opc]);
1790                 else {
1791                         mdt = mdt_exp2dev(req->rq_export);
1792                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
1793                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
1794                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
1795                         opc = err_serious(-EFAULT);
1796                 }
1797         } else {
1798                 opc = err_serious(-EFAULT);
1799         }
1800         return opc;
1801 }
1802
1803 static int mdt_reint(struct tgt_session_info *tsi)
1804 {
1805         long opc;
1806         int  rc;
1807         static const struct req_format *reint_fmts[REINT_MAX] = {
1808                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
1809                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
1810                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
1811                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
1812                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
1813                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
1814                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
1815                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
1816                 [REINT_MIGRATE]  = &RQF_MDS_REINT_RENAME
1817         };
1818
1819         ENTRY;
1820
1821         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
1822         if (opc >= 0) {
1823                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1824                 /*
1825                  * No lock possible here from client to pass it to reint code
1826                  * path.
1827                  */
1828                 rc = mdt_reint_internal(info, NULL, opc);
1829                 mdt_thread_info_fini(info);
1830         } else {
1831                 rc = opc;
1832         }
1833
1834         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1835         RETURN(rc);
1836 }
1837
1838 /* this should sync the whole device */
1839 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1840 {
1841         struct dt_device *dt = mdt->mdt_bottom;
1842         int rc;
1843         ENTRY;
1844
1845         rc = dt->dd_ops->dt_sync(env, dt);
1846         RETURN(rc);
1847 }
1848
1849 /* this should sync this object */
1850 static int mdt_object_sync(struct mdt_thread_info *info)
1851 {
1852         struct md_object *next;
1853         int rc;
1854         ENTRY;
1855
1856         if (!mdt_object_exists(info->mti_object)) {
1857                 CWARN("%s: non existing object "DFID": rc = %d\n",
1858                       mdt_obd_name(info->mti_mdt),
1859                       PFID(mdt_object_fid(info->mti_object)), -ESTALE);
1860                 RETURN(-ESTALE);
1861         }
1862         next = mdt_object_child(info->mti_object);
1863         rc = mo_object_sync(info->mti_env, next);
1864
1865         RETURN(rc);
1866 }
1867
1868 static int mdt_sync(struct tgt_session_info *tsi)
1869 {
1870         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1871         struct req_capsule      *pill = tsi->tsi_pill;
1872         struct mdt_body         *body;
1873         int                      rc;
1874
1875         ENTRY;
1876
1877         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1878                 RETURN(err_serious(-ENOMEM));
1879
1880         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
1881                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
1882         } else {
1883                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1884
1885                 /* sync an object */
1886                 rc = mdt_object_sync(info);
1887                 if (rc == 0) {
1888                         const struct lu_fid *fid;
1889                         struct lu_attr *la = &info->mti_attr.ma_attr;
1890
1891                         info->mti_attr.ma_need = MA_INODE;
1892                         info->mti_attr.ma_valid = 0;
1893                         rc = mdt_attr_get_complex(info, info->mti_object,
1894                                                   &info->mti_attr);
1895                         if (rc == 0) {
1896                                 body = req_capsule_server_get(pill,
1897                                                               &RMF_MDT_BODY);
1898                                 fid = mdt_object_fid(info->mti_object);
1899                                 mdt_pack_attr2body(info, body, la, fid);
1900                         }
1901                 }
1902                 mdt_thread_info_fini(info);
1903         }
1904         if (rc == 0)
1905                 mdt_counter_incr(req, LPROC_MDT_SYNC);
1906
1907         RETURN(rc);
1908 }
1909
1910 /*
1911  * Handle quota control requests to consult current usage/limit, but also
1912  * to configure quota enforcement
1913  */
1914 static int mdt_quotactl(struct tgt_session_info *tsi)
1915 {
1916         struct obd_export       *exp  = tsi->tsi_exp;
1917         struct req_capsule      *pill = tsi->tsi_pill;
1918         struct obd_quotactl     *oqctl, *repoqc;
1919         int                      id, rc;
1920         struct mdt_device       *mdt = mdt_exp2dev(exp);
1921         struct lu_device        *qmt = mdt->mdt_qmt_dev;
1922         struct lu_nodemap       *nodemap = exp->exp_target_data.ted_nodemap;
1923         ENTRY;
1924
1925         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1926         if (oqctl == NULL)
1927                 RETURN(err_serious(-EPROTO));
1928
1929         rc = req_capsule_server_pack(pill);
1930         if (rc)
1931                 RETURN(err_serious(rc));
1932
1933         switch (oqctl->qc_cmd) {
1934                 /* master quotactl */
1935         case Q_SETINFO:
1936         case Q_SETQUOTA:
1937                 if (!nodemap_can_setquota(nodemap))
1938                         RETURN(-EPERM);
1939         case Q_GETINFO:
1940         case Q_GETQUOTA:
1941                 if (qmt == NULL)
1942                         RETURN(-EOPNOTSUPP);
1943                 /* slave quotactl */
1944         case Q_GETOINFO:
1945         case Q_GETOQUOTA:
1946                 break;
1947         default:
1948                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
1949                 RETURN(-EFAULT);
1950         }
1951
1952         /* map uid/gid for remote client */
1953         id = oqctl->qc_id;
1954         if (exp_connect_rmtclient(exp)) {
1955                 struct lustre_idmap_table *idmap;
1956
1957                 idmap = exp->exp_mdt_data.med_idmap;
1958
1959                 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
1960                              oqctl->qc_cmd != Q_GETINFO))
1961                         RETURN(-EPERM);
1962
1963                 if (oqctl->qc_type == USRQUOTA)
1964                         id = lustre_idmap_lookup_uid(NULL, idmap, 0,
1965                                                      oqctl->qc_id);
1966                 else if (oqctl->qc_type == GRPQUOTA)
1967                         id = lustre_idmap_lookup_gid(NULL, idmap, 0,
1968                                                      oqctl->qc_id);
1969                 else
1970                         RETURN(-EINVAL);
1971
1972                 if (id == CFS_IDMAP_NOTFOUND) {
1973                         CDEBUG(D_QUOTA, "no mapping for id %u\n", oqctl->qc_id);
1974                         RETURN(-EACCES);
1975                 }
1976         }
1977
1978         if (oqctl->qc_type == USRQUOTA)
1979                 id = nodemap_map_id(nodemap, NODEMAP_UID,
1980                                     NODEMAP_CLIENT_TO_FS, id);
1981         else if (oqctl->qc_type == GRPQUOTA)
1982                 id = nodemap_map_id(nodemap, NODEMAP_UID,
1983                                     NODEMAP_CLIENT_TO_FS, id);
1984
1985         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
1986         if (repoqc == NULL)
1987                 RETURN(err_serious(-EFAULT));
1988
1989         if (oqctl->qc_id != id)
1990                 swap(oqctl->qc_id, id);
1991
1992         switch (oqctl->qc_cmd) {
1993
1994         case Q_GETINFO:
1995         case Q_SETINFO:
1996         case Q_SETQUOTA:
1997         case Q_GETQUOTA:
1998                 /* forward quotactl request to QMT */
1999                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2000                 break;
2001
2002         case Q_GETOINFO:
2003         case Q_GETOQUOTA:
2004                 /* slave quotactl */
2005                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2006                                    oqctl);
2007                 break;
2008
2009         default:
2010                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2011                 RETURN(-EFAULT);
2012         }
2013
2014         if (oqctl->qc_id != id)
2015                 swap(oqctl->qc_id, id);
2016
2017         *repoqc = *oqctl;
2018         RETURN(rc);
2019 }
2020
2021 /** clone llog ctxt from child (mdd)
2022  * This allows remote llog (replicator) access.
2023  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2024  * context was originally set up, or we can handle them directly.
2025  * I choose the latter, but that means I need any llog
2026  * contexts set up by child to be accessable by the mdt.  So we clone the
2027  * context into our context list here.
2028  */
2029 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2030                                int idx)
2031 {
2032         struct md_device  *next = mdt->mdt_child;
2033         struct llog_ctxt *ctxt;
2034         int rc;
2035
2036         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2037                 return 0;
2038
2039         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2040         if (rc || ctxt == NULL) {
2041                 return 0;
2042         }
2043
2044         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2045         if (rc)
2046                 CERROR("Can't set mdt ctxt %d\n", rc);
2047
2048         return rc;
2049 }
2050
2051 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2052                                  struct mdt_device *mdt, int idx)
2053 {
2054         struct llog_ctxt *ctxt;
2055
2056         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2057         if (ctxt == NULL)
2058                 return 0;
2059         /* Put once for the get we just did, and once for the clone */
2060         llog_ctxt_put(ctxt);
2061         llog_ctxt_put(ctxt);
2062         return 0;
2063 }
2064
2065 /*
2066  * sec context handlers
2067  */
2068 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2069 {
2070         int rc;
2071
2072         rc = mdt_handle_idmap(tsi);
2073         if (unlikely(rc)) {
2074                 struct ptlrpc_request   *req = tgt_ses_req(tsi);
2075                 __u32                    opc;
2076
2077                 opc = lustre_msg_get_opc(req->rq_reqmsg);
2078                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2079                         sptlrpc_svc_ctx_invalidate(req);
2080         }
2081
2082         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2083
2084         return rc;
2085 }
2086
2087 /*
2088  * quota request handlers
2089  */
2090 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2091 {
2092         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
2093         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2094         int                      rc;
2095         ENTRY;
2096
2097         if (qmt == NULL)
2098                 RETURN(err_serious(-EOPNOTSUPP));
2099
2100         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2101         RETURN(rc);
2102 }
2103
2104 struct mdt_object *mdt_object_new(const struct lu_env *env,
2105                                   struct mdt_device *d,
2106                                   const struct lu_fid *f)
2107 {
2108         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2109         struct lu_object *o;
2110         struct mdt_object *m;
2111         ENTRY;
2112
2113         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2114         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2115         if (unlikely(IS_ERR(o)))
2116                 m = (struct mdt_object *)o;
2117         else
2118                 m = mdt_obj(o);
2119         RETURN(m);
2120 }
2121
2122 struct mdt_object *mdt_object_find(const struct lu_env *env,
2123                                    struct mdt_device *d,
2124                                    const struct lu_fid *f)
2125 {
2126         struct lu_object *o;
2127         struct mdt_object *m;
2128         ENTRY;
2129
2130         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2131         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2132         if (unlikely(IS_ERR(o)))
2133                 m = (struct mdt_object *)o;
2134         else
2135                 m = mdt_obj(o);
2136
2137         RETURN(m);
2138 }
2139
2140 /**
2141  * Asyncronous commit for mdt device.
2142  *
2143  * Pass asynchonous commit call down the MDS stack.
2144  *
2145  * \param env environment
2146  * \param mdt the mdt device
2147  */
2148 static void mdt_device_commit_async(const struct lu_env *env,
2149                                     struct mdt_device *mdt)
2150 {
2151         struct dt_device *dt = mdt->mdt_bottom;
2152         int rc;
2153
2154         rc = dt->dd_ops->dt_commit_async(env, dt);
2155         if (unlikely(rc != 0))
2156                 CWARN("%s: async commit start failed: rc = %d\n",
2157                       mdt_obd_name(mdt), rc);
2158 }
2159
2160 /**
2161  * Mark the lock as "synchonous".
2162  *
2163  * Mark the lock to deffer transaction commit to the unlock time.
2164  *
2165  * \param lock the lock to mark as "synchonous"
2166  *
2167  * \see mdt_is_lock_sync
2168  * \see mdt_save_lock
2169  */
2170 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2171 {
2172         lock->l_ast_data = (void*)1;
2173 }
2174
2175 /**
2176  * Check whehter the lock "synchonous" or not.
2177  *
2178  * \param lock the lock to check
2179  * \retval 1 the lock is "synchonous"
2180  * \retval 0 the lock isn't "synchronous"
2181  *
2182  * \see mdt_set_lock_sync
2183  * \see mdt_save_lock
2184  */
2185 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2186 {
2187         return lock->l_ast_data != NULL;
2188 }
2189
2190 /**
2191  * Blocking AST for mdt locks.
2192  *
2193  * Starts transaction commit if in case of COS lock conflict or
2194  * deffers such a commit to the mdt_save_lock.
2195  *
2196  * \param lock the lock which blocks a request or cancelling lock
2197  * \param desc unused
2198  * \param data unused
2199  * \param flag indicates whether this cancelling or blocking callback
2200  * \retval 0
2201  * \see ldlm_blocking_ast_nocheck
2202  */
2203 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2204                      void *data, int flag)
2205 {
2206         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2207         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2208         int rc;
2209         ENTRY;
2210
2211         if (flag == LDLM_CB_CANCELING)
2212                 RETURN(0);
2213         lock_res_and_lock(lock);
2214         if (lock->l_blocking_ast != mdt_blocking_ast) {
2215                 unlock_res_and_lock(lock);
2216                 RETURN(0);
2217         }
2218         if (mdt_cos_is_enabled(mdt) &&
2219             lock->l_req_mode & (LCK_PW | LCK_EX) &&
2220             lock->l_blocking_lock != NULL &&
2221             lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
2222                 mdt_set_lock_sync(lock);
2223         }
2224         rc = ldlm_blocking_ast_nocheck(lock);
2225
2226         /* There is no lock conflict if l_blocking_lock == NULL,
2227          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2228          * when the last reference to a local lock was released */
2229         if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2230                 struct lu_env env;
2231
2232                 rc = lu_env_init(&env, LCT_LOCAL);
2233                 if (unlikely(rc != 0))
2234                         CWARN("%s: lu_env initialization failed, cannot "
2235                               "start asynchronous commit: rc = %d\n",
2236                               obd->obd_name, rc);
2237                 else
2238                         mdt_device_commit_async(&env, mdt);
2239                 lu_env_fini(&env);
2240         }
2241         RETURN(rc);
2242 }
2243
2244 /* Used for cross-MDT lock */
2245 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2246                             void *data, int flag)
2247 {
2248         struct lustre_handle lockh;
2249         int               rc;
2250
2251         switch (flag) {
2252         case LDLM_CB_BLOCKING:
2253                 ldlm_lock2handle(lock, &lockh);
2254                 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
2255                 if (rc < 0) {
2256                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2257                         RETURN(rc);
2258                 }
2259                 break;
2260         case LDLM_CB_CANCELING:
2261                 LDLM_DEBUG(lock, "Revoke remote lock\n");
2262                 break;
2263         default:
2264                 LBUG();
2265         }
2266         RETURN(0);
2267 }
2268
2269 int mdt_check_resent_lock(struct mdt_thread_info *info,
2270                           struct mdt_object *mo,
2271                           struct mdt_lock_handle *lhc)
2272 {
2273         /* the lock might already be gotten in ldlm_handle_enqueue() */
2274         if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
2275                 struct ptlrpc_request *req = mdt_info_req(info);
2276                 struct ldlm_lock      *lock;
2277
2278                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2279                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
2280                 if (lock == NULL) {
2281                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
2282                          * a resend case, however, it could be already destroyed
2283                          * due to client eviction or a raced cancel RPC. */
2284                         LDLM_DEBUG_NOLOCK("Invalid lock handle "LPX64"\n",
2285                                           lhc->mlh_reg_lh.cookie);
2286                         RETURN(-ESTALE);
2287                 }
2288
2289                 if (!fid_res_name_eq(mdt_object_fid(mo),
2290                                      &lock->l_resource->lr_name)) {
2291                         CWARN("%s: Although resent, but still not "
2292                               "get child lock:"DFID"\n",
2293                               info->mti_exp->exp_obd->obd_name,
2294                               PFID(mdt_object_fid(mo)));
2295                         LDLM_LOCK_PUT(lock);
2296                         RETURN(-EPROTO);
2297                 }
2298                 LDLM_LOCK_PUT(lock);
2299                 return 0;
2300         }
2301         return 1;
2302 }
2303
2304 int mdt_remote_object_lock(struct mdt_thread_info *mti,
2305                            struct mdt_object *o, const struct lu_fid *fid,
2306                            struct lustre_handle *lh, ldlm_mode_t mode,
2307                            __u64 ibits)
2308 {
2309         struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
2310         ldlm_policy_data_t *policy = &mti->mti_policy;
2311         struct ldlm_res_id *res_id = &mti->mti_res_id;
2312         int rc = 0;
2313         ENTRY;
2314
2315         LASSERT(mdt_object_remote(o));
2316
2317         LASSERT(ibits == MDS_INODELOCK_UPDATE);
2318
2319         fid_build_reg_res_name(fid, res_id);
2320
2321         memset(einfo, 0, sizeof(*einfo));
2322         einfo->ei_type = LDLM_IBITS;
2323         einfo->ei_mode = mode;
2324         einfo->ei_cb_bl = mdt_remote_blocking_ast;
2325         einfo->ei_cb_cp = ldlm_completion_ast;
2326         einfo->ei_enq_slave = 0;
2327         einfo->ei_res_id = res_id;
2328
2329         memset(policy, 0, sizeof(*policy));
2330         policy->l_inodebits.bits = ibits;
2331
2332         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2333                             policy);
2334         RETURN(rc);
2335 }
2336
2337 static int mdt_object_local_lock(struct mdt_thread_info *info,
2338                                  struct mdt_object *o,
2339                                  struct mdt_lock_handle *lh, __u64 ibits,
2340                                  bool nonblock, int locality)
2341 {
2342         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2343         ldlm_policy_data_t *policy = &info->mti_policy;
2344         struct ldlm_res_id *res_id = &info->mti_res_id;
2345         __u64 dlmflags;
2346         int rc;
2347         ENTRY;
2348
2349         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2350         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2351         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2352         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2353
2354         /* Only enqueue LOOKUP lock for remote object */
2355         if (mdt_object_remote(o))
2356                 LASSERT(ibits == MDS_INODELOCK_LOOKUP);
2357
2358         if (lh->mlh_type == MDT_PDO_LOCK) {
2359                 /* check for exists after object is locked */
2360                 if (mdt_object_exists(o) == 0) {
2361                         /* Non-existent object shouldn't have PDO lock */
2362                         RETURN(-ESTALE);
2363                 } else {
2364                         /* Non-dir object shouldn't have PDO lock */
2365                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2366                                 RETURN(-ENOTDIR);
2367                 }
2368         }
2369
2370         memset(policy, 0, sizeof(*policy));
2371         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2372
2373         dlmflags = LDLM_FL_ATOMIC_CB;
2374         if (nonblock)
2375                 dlmflags |= LDLM_FL_BLOCK_NOWAIT;
2376
2377         /*
2378          * Take PDO lock on whole directory and build correct @res_id for lock
2379          * on part of directory.
2380          */
2381         if (lh->mlh_pdo_hash != 0) {
2382                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2383                 mdt_lock_pdo_mode(info, o, lh);
2384                 if (lh->mlh_pdo_mode != LCK_NL) {
2385                         /*
2386                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2387                          * is never going to be sent to client and we do not
2388                          * want it slowed down due to possible cancels.
2389                          */
2390                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2391                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2392                                           policy, res_id, dlmflags,
2393                                           info->mti_exp == NULL ? NULL :
2394                                           &info->mti_exp->exp_handle.h_cookie);
2395                         if (unlikely(rc != 0))
2396                                 GOTO(out_unlock, rc);
2397                 }
2398
2399                 /*
2400                  * Finish res_id initializing by name hash marking part of
2401                  * directory which is taking modification.
2402                  */
2403                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2404         }
2405
2406         policy->l_inodebits.bits = ibits;
2407
2408         /*
2409          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2410          * going to be sent to client. If it is - mdt_intent_policy() path will
2411          * fix it up and turn FL_LOCAL flag off.
2412          */
2413         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2414                           res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2415                           info->mti_exp == NULL ? NULL :
2416                           &info->mti_exp->exp_handle.h_cookie);
2417 out_unlock:
2418         if (rc != 0)
2419                 mdt_object_unlock(info, o, lh, 1);
2420         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2421                    lh->mlh_pdo_hash != 0 &&
2422                    (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
2423                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2424
2425         RETURN(rc);
2426 }
2427
2428 static int
2429 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
2430                          struct mdt_lock_handle *lh, __u64 ibits,
2431                          bool nonblock, int locality)
2432 {
2433         int rc;
2434         ENTRY;
2435
2436         if (!mdt_object_remote(o))
2437                 return mdt_object_local_lock(info, o, lh, ibits, nonblock,
2438                                              locality);
2439
2440         if (locality == MDT_LOCAL_LOCK) {
2441                 CERROR("%s: try to get local lock for remote object"
2442                        DFID".\n", mdt_obd_name(info->mti_mdt),
2443                        PFID(mdt_object_fid(o)));
2444                 RETURN(-EPROTO);
2445         }
2446
2447         /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
2448         ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
2449                    MDS_INODELOCK_XATTR);
2450         if (ibits & MDS_INODELOCK_UPDATE) {
2451                 /* Sigh, PDO needs to enqueue 2 locks right now, but
2452                  * enqueue RPC can only request 1 lock, to avoid extra
2453                  * RPC, so it will instead enqueue EX lock for remote
2454                  * object anyway XXX*/
2455                 if (lh->mlh_type == MDT_PDO_LOCK &&
2456                     lh->mlh_pdo_hash != 0) {
2457                         CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
2458                                "EX lock.\n", mdt_obd_name(info->mti_mdt),
2459                                PFID(mdt_object_fid(o)));
2460                         lh->mlh_pdo_hash = 0;
2461                         lh->mlh_rreg_mode = LCK_EX;
2462                         lh->mlh_type = MDT_REG_LOCK;
2463                 }
2464                 rc = mdt_remote_object_lock(info, o, mdt_object_fid(o),
2465                                             &lh->mlh_rreg_lh,
2466                                             lh->mlh_rreg_mode,
2467                                             MDS_INODELOCK_UPDATE);
2468                 if (rc != ELDLM_OK)
2469                         RETURN(rc);
2470         }
2471
2472         /* Only enqueue LOOKUP lock for remote object */
2473         if (ibits & MDS_INODELOCK_LOOKUP) {
2474                 rc = mdt_object_local_lock(info, o, lh,
2475                                            MDS_INODELOCK_LOOKUP,
2476                                            nonblock, locality);
2477                 if (rc != ELDLM_OK)
2478                         RETURN(rc);
2479         }
2480
2481         RETURN(0);
2482 }
2483
2484 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2485                     struct mdt_lock_handle *lh, __u64 ibits, int locality)
2486 {
2487         return mdt_object_lock_internal(info, o, lh, ibits, false, locality);
2488 }
2489
2490 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
2491                         struct mdt_lock_handle *lh, __u64 ibits, int locality)
2492 {
2493         struct mdt_lock_handle tmp = *lh;
2494         int rc;
2495
2496         rc = mdt_object_lock_internal(info, o, &tmp, ibits, true, locality);
2497         if (rc == 0)
2498                 *lh = tmp;
2499
2500         return rc == 0;
2501 }
2502
2503 /**
2504  * Save a lock within request object.
2505  *
2506  * Keep the lock referenced until whether client ACK or transaction
2507  * commit happens or release the lock immediately depending on input
2508  * parameters. If COS is ON, a write lock is converted to COS lock
2509  * before saving.
2510  *
2511  * \param info thead info object
2512  * \param h lock handle
2513  * \param mode lock mode
2514  * \param decref force immediate lock releasing
2515  */
2516 static
2517 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2518                    ldlm_mode_t mode, int decref)
2519 {
2520         ENTRY;
2521
2522         if (lustre_handle_is_used(h)) {
2523                 if (decref || !info->mti_has_trans ||
2524                     !(mode & (LCK_PW | LCK_EX))){
2525                         mdt_fid_unlock(h, mode);
2526                 } else {
2527                         struct mdt_device *mdt = info->mti_mdt;
2528                         struct ldlm_lock *lock = ldlm_handle2lock(h);
2529                         struct ptlrpc_request *req = mdt_info_req(info);
2530                         int no_ack = 0;
2531
2532                         LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2533                                  h->cookie);
2534                         /* there is no request if mdt_object_unlock() is called
2535                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
2536                         if (likely(req != NULL)) {
2537                                 CDEBUG(D_HA, "request = %p reply state = %p"
2538                                        " transno = "LPD64"\n", req,
2539                                        req->rq_reply_state, req->rq_transno);
2540                                 if (mdt_cos_is_enabled(mdt)) {
2541                                         no_ack = 1;
2542                                         ldlm_lock_downgrade(lock, LCK_COS);
2543                                         mode = LCK_COS;
2544                                 }
2545                                 ptlrpc_save_lock(req, h, mode, no_ack);
2546                         } else {
2547                                 ldlm_lock_decref(h, mode);
2548                         }
2549                         if (mdt_is_lock_sync(lock)) {
2550                                 CDEBUG(D_HA, "found sync-lock,"
2551                                        " async commit started\n");
2552                                 mdt_device_commit_async(info->mti_env,
2553                                                         mdt);
2554                         }
2555                         LDLM_LOCK_PUT(lock);
2556                 }
2557                 h->cookie = 0ull;
2558         }
2559
2560         EXIT;
2561 }
2562
2563 /**
2564  * Unlock mdt object.
2565  *
2566  * Immeditely release the regular lock and the PDO lock or save the
2567  * lock in request and keep them referenced until client ACK or
2568  * transaction commit.
2569  *
2570  * \param info thread info object
2571  * \param o mdt object
2572  * \param lh mdt lock handle referencing regular and PDO locks
2573  * \param decref force immediate lock releasing
2574  */
2575 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2576                        struct mdt_lock_handle *lh, int decref)
2577 {
2578         ENTRY;
2579
2580         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2581         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2582
2583         if (lustre_handle_is_used(&lh->mlh_rreg_lh))
2584                 ldlm_lock_decref(&lh->mlh_rreg_lh, lh->mlh_rreg_mode);
2585
2586         EXIT;
2587 }
2588
2589 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2590                                         const struct lu_fid *f,
2591                                         struct mdt_lock_handle *lh,
2592                                         __u64 ibits)
2593 {
2594         struct mdt_object *o;
2595
2596         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2597         if (!IS_ERR(o)) {
2598                 int rc;
2599
2600                 rc = mdt_object_lock(info, o, lh, ibits,
2601                                      MDT_LOCAL_LOCK);
2602                 if (rc != 0) {
2603                         mdt_object_put(info->mti_env, o);
2604                         o = ERR_PTR(rc);
2605                 }
2606         }
2607         return o;
2608 }
2609
2610 void mdt_object_unlock_put(struct mdt_thread_info * info,
2611                            struct mdt_object * o,
2612                            struct mdt_lock_handle *lh,
2613                            int decref)
2614 {
2615         mdt_object_unlock(info, o, lh, decref);
2616         mdt_object_put(info->mti_env, o);
2617 }
2618
2619 /*
2620  * Generic code handling requests that have struct mdt_body passed in:
2621  *
2622  *  - extract mdt_body from request and save it in @info, if present;
2623  *
2624  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
2625  *  @info;
2626  *
2627  *  - if HABEO_CORPUS flag is set for this request type check whether object
2628  *  actually exists on storage (lu_object_exists()).
2629  *
2630  */
2631 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2632 {
2633         const struct mdt_body    *body;
2634         struct mdt_object        *obj;
2635         const struct lu_env      *env;
2636         struct req_capsule       *pill;
2637         int                       rc;
2638         ENTRY;
2639
2640         env = info->mti_env;
2641         pill = info->mti_pill;
2642
2643         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2644         if (body == NULL)
2645                 RETURN(-EFAULT);
2646
2647         if (!(body->mbo_valid & OBD_MD_FLID))
2648                 RETURN(0);
2649
2650         if (!fid_is_sane(&body->mbo_fid1)) {
2651                 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
2652                 RETURN(-EINVAL);
2653         }
2654
2655         obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
2656         if (!IS_ERR(obj)) {
2657                 if ((flags & HABEO_CORPUS) && !mdt_object_exists(obj)) {
2658                         mdt_object_put(env, obj);
2659                         rc = -ENOENT;
2660                 } else {
2661                         info->mti_object = obj;
2662                         rc = 0;
2663                 }
2664         } else
2665                 rc = PTR_ERR(obj);
2666
2667         RETURN(rc);
2668 }
2669
2670 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2671 {
2672         struct req_capsule *pill = info->mti_pill;
2673         int rc;
2674         ENTRY;
2675
2676         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2677                 rc = mdt_body_unpack(info, flags);
2678         else
2679                 rc = 0;
2680
2681         if (rc == 0 && (flags & HABEO_REFERO)) {
2682                 /* Pack reply. */
2683                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2684                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2685                                              DEF_REP_MD_SIZE);
2686                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2687                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
2688                                              RCL_SERVER, 0);
2689
2690                 rc = req_capsule_server_pack(pill);
2691         }
2692         RETURN(rc);
2693 }
2694
2695 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2696 {
2697         lh->mlh_type = MDT_NUL_LOCK;
2698         lh->mlh_reg_lh.cookie = 0ull;
2699         lh->mlh_reg_mode = LCK_MINMODE;
2700         lh->mlh_pdo_lh.cookie = 0ull;
2701         lh->mlh_pdo_mode = LCK_MINMODE;
2702         lh->mlh_rreg_lh.cookie = 0ull;
2703         lh->mlh_rreg_mode = LCK_MINMODE;
2704 }
2705
2706 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2707 {
2708         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2709         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2710 }
2711
2712 /*
2713  * Initialize fields of struct mdt_thread_info. Other fields are left in
2714  * uninitialized state, because it's too expensive to zero out whole
2715  * mdt_thread_info (> 1K) on each request arrival.
2716  */
2717 void mdt_thread_info_init(struct ptlrpc_request *req,
2718                           struct mdt_thread_info *info)
2719 {
2720         int i;
2721
2722         info->mti_pill = &req->rq_pill;
2723
2724         /* lock handle */
2725         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2726                 mdt_lock_handle_init(&info->mti_lh[i]);
2727
2728         /* mdt device: it can be NULL while CONNECT */
2729         if (req->rq_export) {
2730                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2731                 info->mti_exp = req->rq_export;
2732         } else
2733                 info->mti_mdt = NULL;
2734         info->mti_env = req->rq_svc_thread->t_env;
2735         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2736
2737         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2738         info->mti_big_buf = LU_BUF_NULL;
2739         info->mti_body = NULL;
2740         info->mti_object = NULL;
2741         info->mti_dlm_req = NULL;
2742         info->mti_has_trans = 0;
2743         info->mti_cross_ref = 0;
2744         info->mti_opdata = 0;
2745         info->mti_big_lmm_used = 0;
2746
2747         info->mti_spec.no_create = 0;
2748         info->mti_spec.sp_rm_entry = 0;
2749         info->mti_spec.sp_permitted = 0;
2750
2751         info->mti_spec.u.sp_ea.eadata = NULL;
2752         info->mti_spec.u.sp_ea.eadatalen = 0;
2753 }
2754
2755 void mdt_thread_info_fini(struct mdt_thread_info *info)
2756 {
2757         int i;
2758
2759         if (info->mti_object != NULL) {
2760                 mdt_object_put(info->mti_env, info->mti_object);
2761                 info->mti_object = NULL;
2762         }
2763
2764         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2765                 mdt_lock_handle_fini(&info->mti_lh[i]);
2766         info->mti_env = NULL;
2767         info->mti_pill = NULL;
2768         info->mti_exp = NULL;
2769
2770         if (unlikely(info->mti_big_buf.lb_buf != NULL))
2771                 lu_buf_free(&info->mti_big_buf);
2772 }
2773
2774 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
2775 {
2776         struct mdt_thread_info  *mti;
2777
2778         mti = mdt_th_info(tsi->tsi_env);
2779         LASSERT(mti != NULL);
2780
2781         mdt_thread_info_init(tgt_ses_req(tsi), mti);
2782         if (tsi->tsi_corpus != NULL) {
2783                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
2784                 lu_object_get(tsi->tsi_corpus);
2785         }
2786         mti->mti_body = tsi->tsi_mdt_body;
2787         mti->mti_dlm_req = tsi->tsi_dlm_req;
2788
2789         return mti;
2790 }
2791
2792 static int mdt_tgt_connect(struct tgt_session_info *tsi)
2793 {
2794         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2795         int                      rc;
2796
2797         ENTRY;
2798
2799         if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
2800             cfs_fail_val ==
2801             tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id) {
2802                 set_current_state(TASK_UNINTERRUPTIBLE);
2803                 schedule_timeout(msecs_to_jiffies(3 * MSEC_PER_SEC));
2804         }
2805
2806         rc = tgt_connect(tsi);
2807         if (rc != 0)
2808                 RETURN(rc);
2809
2810         rc = mdt_init_idmap(tsi);
2811         if (rc != 0)
2812                 GOTO(err, rc);
2813         RETURN(0);
2814 err:
2815         obd_disconnect(class_export_get(req->rq_export));
2816         return rc;
2817 }
2818
2819 enum mdt_it_code {
2820         MDT_IT_OPEN,
2821         MDT_IT_OCREAT,
2822         MDT_IT_CREATE,
2823         MDT_IT_GETATTR,
2824         MDT_IT_READDIR,
2825         MDT_IT_LOOKUP,
2826         MDT_IT_UNLINK,
2827         MDT_IT_TRUNC,
2828         MDT_IT_GETXATTR,
2829         MDT_IT_LAYOUT,
2830         MDT_IT_QUOTA,
2831         MDT_IT_NR
2832 };
2833
2834 static int mdt_intent_getattr(enum mdt_it_code opcode,
2835                               struct mdt_thread_info *info,
2836                               struct ldlm_lock **,
2837                               __u64);
2838
2839 static int mdt_intent_getxattr(enum mdt_it_code opcode,
2840                                 struct mdt_thread_info *info,
2841                                 struct ldlm_lock **lockp,
2842                                 __u64 flags);
2843
2844 static int mdt_intent_layout(enum mdt_it_code opcode,
2845                              struct mdt_thread_info *info,
2846                              struct ldlm_lock **,
2847                              __u64);
2848 static int mdt_intent_reint(enum mdt_it_code opcode,
2849                             struct mdt_thread_info *info,
2850                             struct ldlm_lock **,
2851                             __u64);
2852
2853 static struct mdt_it_flavor {
2854         const struct req_format *it_fmt;
2855         __u32                    it_flags;
2856         int                    (*it_act)(enum mdt_it_code ,
2857                                          struct mdt_thread_info *,
2858                                          struct ldlm_lock **,
2859                                          __u64);
2860         long                     it_reint;
2861 } mdt_it_flavor[] = {
2862         [MDT_IT_OPEN]     = {
2863                 .it_fmt   = &RQF_LDLM_INTENT,
2864                 /*.it_flags = HABEO_REFERO,*/
2865                 .it_flags = 0,
2866                 .it_act   = mdt_intent_reint,
2867                 .it_reint = REINT_OPEN
2868         },
2869         [MDT_IT_OCREAT]   = {
2870                 .it_fmt   = &RQF_LDLM_INTENT,
2871                 /*
2872                  * OCREAT is not a MUTABOR request as if the file
2873                  * already exists.
2874                  * We do the extra check of OBD_CONNECT_RDONLY in
2875                  * mdt_reint_open() when we really need to create
2876                  * the object.
2877                  */
2878                 .it_flags = 0,
2879                 .it_act   = mdt_intent_reint,
2880                 .it_reint = REINT_OPEN
2881         },
2882         [MDT_IT_CREATE]   = {
2883                 .it_fmt   = &RQF_LDLM_INTENT,
2884                 .it_flags = MUTABOR,
2885                 .it_act   = mdt_intent_reint,
2886                 .it_reint = REINT_CREATE
2887         },
2888         [MDT_IT_GETATTR]  = {
2889                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2890                 .it_flags = HABEO_REFERO,
2891                 .it_act   = mdt_intent_getattr
2892         },
2893         [MDT_IT_READDIR]  = {
2894                 .it_fmt   = NULL,
2895                 .it_flags = 0,
2896                 .it_act   = NULL
2897         },
2898         [MDT_IT_LOOKUP]   = {
2899                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2900                 .it_flags = HABEO_REFERO,
2901                 .it_act   = mdt_intent_getattr
2902         },
2903         [MDT_IT_UNLINK]   = {
2904                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
2905                 .it_flags = MUTABOR,
2906                 .it_act   = NULL,
2907                 .it_reint = REINT_UNLINK
2908         },
2909         [MDT_IT_TRUNC]    = {
2910                 .it_fmt   = NULL,
2911                 .it_flags = MUTABOR,
2912                 .it_act   = NULL
2913         },
2914         [MDT_IT_GETXATTR] = {
2915                 .it_fmt   = &RQF_LDLM_INTENT_GETXATTR,
2916                 .it_flags = HABEO_CORPUS,
2917                 .it_act   = mdt_intent_getxattr
2918         },
2919         [MDT_IT_LAYOUT] = {
2920                 .it_fmt   = &RQF_LDLM_INTENT_LAYOUT,
2921                 .it_flags = 0,
2922                 .it_act   = mdt_intent_layout
2923         }
2924 };
2925
2926 static int
2927 mdt_intent_lock_replace(struct mdt_thread_info *info,
2928                         struct ldlm_lock **lockp,
2929                         struct mdt_lock_handle *lh,
2930                         __u64 flags)
2931 {
2932         struct ptlrpc_request  *req = mdt_info_req(info);
2933         struct ldlm_lock       *lock = *lockp;
2934         struct ldlm_lock       *new_lock;
2935
2936         /* If possible resent found a lock, @lh is set to its handle */
2937         new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
2938
2939         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
2940                 lh->mlh_reg_lh.cookie = 0;
2941                 RETURN(0);
2942         }
2943
2944         LASSERTF(new_lock != NULL,
2945                  "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
2946
2947         /*
2948          * If we've already given this lock to a client once, then we should
2949          * have no readers or writers.  Otherwise, we should have one reader
2950          * _or_ writer ref (which will be zeroed below) before returning the
2951          * lock to a client.
2952          */
2953         if (new_lock->l_export == req->rq_export) {
2954                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
2955         } else {
2956                 LASSERT(new_lock->l_export == NULL);
2957                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
2958         }
2959
2960         *lockp = new_lock;
2961
2962         if (new_lock->l_export == req->rq_export) {
2963                 /*
2964                  * Already gave this to the client, which means that we
2965                  * reconstructed a reply.
2966                  */
2967                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
2968                         MSG_RESENT);
2969
2970                 LDLM_LOCK_RELEASE(new_lock);
2971                 lh->mlh_reg_lh.cookie = 0;
2972                 RETURN(ELDLM_LOCK_REPLACED);
2973         }
2974
2975         /*
2976          * Fixup the lock to be given to the client.
2977          */
2978         lock_res_and_lock(new_lock);
2979         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
2980          * possible blocking AST. */
2981         while (new_lock->l_readers > 0) {
2982                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
2983                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
2984                 new_lock->l_readers--;
2985         }
2986         while (new_lock->l_writers > 0) {
2987                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
2988                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
2989                 new_lock->l_writers--;
2990         }
2991
2992         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
2993         new_lock->l_blocking_ast = lock->l_blocking_ast;
2994         new_lock->l_completion_ast = lock->l_completion_ast;
2995         new_lock->l_remote_handle = lock->l_remote_handle;
2996         new_lock->l_flags &= ~LDLM_FL_LOCAL;
2997
2998         unlock_res_and_lock(new_lock);
2999
3000         cfs_hash_add(new_lock->l_export->exp_lock_hash,
3001                      &new_lock->l_remote_handle,
3002                      &new_lock->l_exp_hash);
3003
3004         LDLM_LOCK_RELEASE(new_lock);
3005         lh->mlh_reg_lh.cookie = 0;
3006
3007         RETURN(ELDLM_LOCK_REPLACED);
3008 }
3009
3010 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3011                                     struct ldlm_lock *new_lock,
3012                                     struct mdt_lock_handle *lh,
3013                                     __u64 flags)
3014 {
3015         struct ptlrpc_request  *req = mdt_info_req(info);
3016         struct ldlm_request    *dlmreq;
3017
3018         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3019                 return;
3020
3021         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3022
3023         /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
3024          * lock was found by ldlm_handle_enqueue(); if so @lh must be
3025          * initialized. */
3026         if (flags & LDLM_FL_RESENT) {
3027                 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
3028                 lh->mlh_reg_mode = new_lock->l_granted_mode;
3029
3030                 LDLM_DEBUG(new_lock, "Restoring lock cookie");
3031                 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie "LPX64,
3032                           lh->mlh_reg_lh.cookie);
3033                 return;
3034         }
3035
3036         /*
3037          * If the xid matches, then we know this is a resent request, and allow
3038          * it. (It's probably an OPEN, for which we don't send a lock.
3039          */
3040         if (req_xid_is_last(req))
3041                 return;
3042
3043         /*
3044          * This remote handle isn't enqueued, so we never received or processed
3045          * this request.  Clear MSG_RESENT, because it can be handled like any
3046          * normal request now.
3047          */
3048         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3049
3050         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
3051                   dlmreq->lock_handle[0].cookie);
3052 }
3053
3054 static int mdt_intent_getxattr(enum mdt_it_code opcode,
3055                                 struct mdt_thread_info *info,
3056                                 struct ldlm_lock **lockp,
3057                                 __u64 flags)
3058 {
3059         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3060         struct ldlm_reply      *ldlm_rep = NULL;
3061         int rc, grc;
3062
3063         /*
3064          * Initialize lhc->mlh_reg_lh either from a previously granted lock
3065          * (for the resend case) or a new lock. Below we will use it to
3066          * replace the original lock.
3067          */
3068         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3069         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3070                 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3071                 rc = mdt_object_lock(info, info->mti_object, lhc,
3072                                         MDS_INODELOCK_XATTR,
3073                                         MDT_LOCAL_LOCK);
3074                 if (rc)
3075                         return rc;
3076         }
3077
3078         grc = mdt_getxattr(info);
3079
3080         rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3081
3082         if (mdt_info_req(info)->rq_repmsg != NULL)
3083                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3084         if (ldlm_rep == NULL)
3085                 RETURN(err_serious(-EFAULT));
3086
3087         ldlm_rep->lock_policy_res2 = grc;
3088
3089         return rc;
3090 }
3091
3092 static int mdt_intent_getattr(enum mdt_it_code opcode,
3093                               struct mdt_thread_info *info,
3094                               struct ldlm_lock **lockp,
3095                               __u64 flags)
3096 {
3097         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3098         __u64                   child_bits;
3099         struct ldlm_reply      *ldlm_rep;
3100         struct mdt_body        *reqbody;
3101         struct mdt_body        *repbody;
3102         int                     rc, rc2;
3103         ENTRY;
3104
3105         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3106         LASSERT(reqbody);
3107
3108         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3109         LASSERT(repbody);
3110
3111         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
3112         repbody->mbo_eadatasize = 0;
3113         repbody->mbo_aclsize = 0;
3114
3115         switch (opcode) {
3116         case MDT_IT_LOOKUP:
3117                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3118                 break;
3119         case MDT_IT_GETATTR:
3120                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3121                              MDS_INODELOCK_PERM;
3122                 break;
3123         default:
3124                 CERROR("Unsupported intent (%d)\n", opcode);
3125                 GOTO(out_shrink, rc = -EINVAL);
3126         }
3127
3128         rc = mdt_init_ucred(info, reqbody);
3129         if (rc)
3130                 GOTO(out_shrink, rc);
3131
3132         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3133         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3134
3135         /* Get lock from request for possible resent case. */
3136         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3137
3138         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3139         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3140
3141         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3142                 ldlm_rep->lock_policy_res2 = 0;
3143         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3144             ldlm_rep->lock_policy_res2) {
3145                 lhc->mlh_reg_lh.cookie = 0ull;
3146                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3147         }
3148
3149         rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3150         EXIT;
3151 out_ucred:
3152         mdt_exit_ucred(info);
3153 out_shrink:
3154         mdt_client_compatibility(info);
3155         rc2 = mdt_fix_reply(info);
3156         if (rc == 0)
3157                 rc = rc2;
3158         return rc;
3159 }
3160
3161 static int mdt_intent_layout(enum mdt_it_code opcode,
3162                              struct mdt_thread_info *info,
3163                              struct ldlm_lock **lockp,
3164                              __u64 flags)
3165 {
3166         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_LAYOUT];
3167         struct layout_intent *layout;
3168         struct lu_fid *fid;
3169         struct mdt_object *obj = NULL;
3170         int rc = 0;
3171         ENTRY;
3172
3173         if (opcode != MDT_IT_LAYOUT) {
3174                 CERROR("%s: Unknown intent (%d)\n", mdt_obd_name(info->mti_mdt),
3175                         opcode);
3176                 RETURN(-EINVAL);
3177         }
3178
3179         fid = &info->mti_tmp_fid2;
3180         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3181
3182         /* Get lock from request for possible resent case. */
3183         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3184
3185         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3186         if (IS_ERR(obj))
3187                 RETURN(PTR_ERR(obj));
3188
3189         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3190                 /* get the length of lsm */
3191                 rc = mdt_attr_get_eabuf_size(info, obj);
3192                 if (rc < 0) {
3193                         mdt_object_put(info->mti_env, obj);
3194                         RETURN(rc);
3195                 }
3196
3197                 if (rc > info->mti_mdt->mdt_max_mdsize)
3198                         info->mti_mdt->mdt_max_mdsize = rc;
3199         }
3200
3201         mdt_object_put(info->mti_env, obj);
3202
3203         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3204         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER, rc);
3205         rc = req_capsule_server_pack(info->mti_pill);
3206         if (rc != 0)
3207                 RETURN(-EINVAL);
3208
3209         if (lustre_handle_is_used(&lhc->mlh_reg_lh))
3210                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3211
3212         layout = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3213         LASSERT(layout != NULL);
3214         if (layout->li_opc == LAYOUT_INTENT_ACCESS)
3215                 /* return to normal/resent ldlm handling */
3216                 RETURN(rc);
3217
3218         CERROR("%s: Unsupported layout intent (%d)\n",
3219                 mdt_obd_name(info->mti_mdt), layout->li_opc);
3220         RETURN(-EINVAL);
3221 }
3222
3223 static int mdt_intent_reint(enum mdt_it_code opcode,
3224                             struct mdt_thread_info *info,
3225                             struct ldlm_lock **lockp,
3226                             __u64 flags)
3227 {
3228         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3229         struct ldlm_reply      *rep = NULL;
3230         long                    opc;
3231         int                     rc;
3232
3233         static const struct req_format *intent_fmts[REINT_MAX] = {
3234                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3235                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3236         };
3237
3238         ENTRY;
3239
3240         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
3241         if (opc < 0)
3242                 RETURN(opc);
3243
3244         if (mdt_it_flavor[opcode].it_reint != opc) {
3245                 CERROR("Reint code %ld doesn't match intent: %d\n",
3246                        opc, opcode);
3247                 RETURN(err_serious(-EPROTO));
3248         }
3249
3250         /* Get lock from request for possible resent case. */
3251         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3252
3253         rc = mdt_reint_internal(info, lhc, opc);
3254
3255         /* Check whether the reply has been packed successfully. */
3256         if (mdt_info_req(info)->rq_repmsg != NULL)
3257                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3258         if (rep == NULL)
3259                 RETURN(err_serious(-EFAULT));
3260
3261         /* MDC expects this in any case */
3262         if (rc != 0)
3263                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3264
3265         /* the open lock or the lock for cross-ref object should be
3266          * returned to the client */
3267         if (lustre_handle_is_used(&lhc->mlh_reg_lh) &&
3268             (rc == 0 || rc == -MDT_EREMOTE_OPEN)) {
3269                 rep->lock_policy_res2 = 0;
3270                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3271                 RETURN(rc);
3272         }
3273
3274         rep->lock_policy_res2 = clear_serious(rc);
3275
3276         if (rep->lock_policy_res2 == -ENOENT &&
3277             mdt_get_disposition(rep, DISP_LOOKUP_NEG))
3278                 rep->lock_policy_res2 = 0;
3279
3280         lhc->mlh_reg_lh.cookie = 0ull;
3281         if (rc == -ENOTCONN || rc == -ENODEV ||
3282             rc == -EOVERFLOW) { /**< if VBR failure then return error */
3283                 /*
3284                  * If it is the disconnect error (ENODEV & ENOCONN), the error
3285                  * will be returned by rq_status, and client at ptlrpc layer
3286                  * will detect this, then disconnect, reconnect the import
3287                  * immediately, instead of impacting the following the rpc.
3288                  */
3289                 RETURN(rc);
3290         }
3291         /*
3292          * For other cases, the error will be returned by intent, and client
3293          * will retrieve the result from intent.
3294          */
3295         RETURN(ELDLM_LOCK_ABORTED);
3296 }
3297
3298 static int mdt_intent_code(long itcode)
3299 {
3300         int rc;
3301
3302         switch(itcode) {
3303         case IT_OPEN:
3304                 rc = MDT_IT_OPEN;
3305                 break;
3306         case IT_OPEN|IT_CREAT:
3307                 rc = MDT_IT_OCREAT;
3308                 break;
3309         case IT_CREAT:
3310                 rc = MDT_IT_CREATE;
3311                 break;
3312         case IT_READDIR:
3313                 rc = MDT_IT_READDIR;
3314                 break;
3315         case IT_GETATTR:
3316                 rc = MDT_IT_GETATTR;
3317                 break;
3318         case IT_LOOKUP:
3319                 rc = MDT_IT_LOOKUP;
3320                 break;
3321         case IT_UNLINK:
3322                 rc = MDT_IT_UNLINK;
3323                 break;
3324         case IT_TRUNC:
3325                 rc = MDT_IT_TRUNC;
3326                 break;
3327         case IT_GETXATTR:
3328                 rc = MDT_IT_GETXATTR;
3329                 break;
3330         case IT_LAYOUT:
3331                 rc = MDT_IT_LAYOUT;
3332                 break;
3333         case IT_QUOTA_DQACQ:
3334         case IT_QUOTA_CONN:
3335                 rc = MDT_IT_QUOTA;
3336                 break;
3337         default:
3338                 CERROR("Unknown intent opcode: %ld\n", itcode);
3339                 rc = -EINVAL;
3340                 break;
3341         }
3342         return rc;
3343 }
3344
3345 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
3346                           struct ldlm_lock **lockp, __u64 flags)
3347 {
3348         struct req_capsule      *pill = info->mti_pill;
3349         struct ptlrpc_request   *req = mdt_info_req(info);
3350         struct mdt_it_flavor    *flv;
3351         int opc;
3352         int rc;
3353         ENTRY;
3354
3355         opc = mdt_intent_code(itopc);
3356         if (opc < 0)
3357                 RETURN(-EINVAL);
3358
3359         if (opc == MDT_IT_QUOTA) {
3360                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
3361
3362                 if (qmt == NULL)
3363                         RETURN(-EOPNOTSUPP);
3364
3365                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
3366                 /* pass the request to quota master */
3367                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
3368                                                  mdt_info_req(info), lockp,
3369                                                  flags);
3370                 RETURN(rc);
3371         }
3372
3373         flv = &mdt_it_flavor[opc];
3374         if (flv->it_fmt != NULL)
3375                 req_capsule_extend(pill, flv->it_fmt);
3376
3377         rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3378         if (rc < 0)
3379                 RETURN(rc);
3380
3381         if (flv->it_flags & MUTABOR &&
3382             exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
3383                 RETURN(-EROFS);
3384
3385         if (flv->it_act != NULL) {
3386                 struct ldlm_reply *rep;
3387
3388                 /* execute policy */
3389                 rc = flv->it_act(opc, info, lockp, flags);
3390
3391                 /* Check whether the reply has been packed successfully. */
3392                 if (req->rq_repmsg != NULL) {
3393                         rep = req_capsule_server_get(info->mti_pill,
3394                                                      &RMF_DLM_REP);
3395                         rep->lock_policy_res2 =
3396                                 ptlrpc_status_hton(rep->lock_policy_res2);
3397                 }
3398         }
3399
3400         RETURN(rc);
3401 }
3402
3403 static int mdt_intent_policy(struct ldlm_namespace *ns,
3404                              struct ldlm_lock **lockp, void *req_cookie,
3405                              ldlm_mode_t mode, __u64 flags, void *data)
3406 {
3407         struct tgt_session_info *tsi;
3408         struct mdt_thread_info  *info;
3409         struct ptlrpc_request   *req  =  req_cookie;
3410         struct ldlm_intent      *it;
3411         struct req_capsule      *pill;
3412         int rc;
3413
3414         ENTRY;
3415
3416         LASSERT(req != NULL);
3417
3418         tsi = tgt_ses_info(req->rq_svc_thread->t_env);
3419
3420         info = tsi2mdt_info(tsi);
3421         LASSERT(info != NULL);
3422         pill = info->mti_pill;
3423         LASSERT(pill->rc_req == req);
3424
3425         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3426                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
3427                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3428                 if (it != NULL) {
3429                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
3430                         if (rc == 0)
3431                                 rc = ELDLM_OK;
3432
3433                         /* Lock without inodebits makes no sense and will oops
3434                          * later in ldlm. Let's check it now to see if we have
3435                          * ibits corrupted somewhere in mdt_intent_opc().
3436                          * The case for client miss to set ibits has been
3437                          * processed by others. */
3438                         LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
3439                                         lr_type == LDLM_IBITS,
3440                                      info->mti_dlm_req->lock_desc.\
3441                                         l_policy_data.l_inodebits.bits != 0));
3442                 } else
3443                         rc = err_serious(-EFAULT);
3444         } else {
3445                 /* No intent was provided */
3446                 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3447                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
3448                 rc = req_capsule_server_pack(pill);
3449                 if (rc)
3450                         rc = err_serious(rc);
3451         }
3452         mdt_thread_info_fini(info);
3453         RETURN(rc);
3454 }
3455
3456 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
3457 {
3458         struct seq_server_site  *ss = mdt_seq_site(mdt);
3459
3460         if (ss->ss_node_id == 0)
3461                 return;
3462
3463         if (ss->ss_client_seq != NULL) {
3464                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3465                 ss->ss_client_seq->lcs_exp = NULL;
3466         }
3467
3468         if (ss->ss_server_fld != NULL) {
3469                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
3470                 ss->ss_server_fld->lsf_control_exp = NULL;
3471         }
3472 }
3473
3474 static void mdt_seq_fini_cli(struct mdt_device *mdt)
3475 {
3476         struct seq_server_site *ss = mdt_seq_site(mdt);
3477
3478         if (ss == NULL)
3479                 return;
3480
3481         if (ss->ss_server_seq != NULL)
3482                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
3483 }
3484
3485 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
3486 {
3487         mdt_seq_fini_cli(mdt);
3488         mdt_deregister_seq_exp(mdt);
3489
3490         return seq_site_fini(env, mdt_seq_site(mdt));
3491 }
3492
3493 /**
3494  * It will retrieve its FLDB entries from MDT0, and it only happens
3495  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
3496  * and it needs to refresh FLDB from the MDT0.
3497  **/
3498 static int mdt_register_lwp_callback(void *data)
3499 {
3500         struct lu_env           env;
3501         struct mdt_device       *mdt = data;
3502         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
3503         int                     rc;
3504         ENTRY;
3505
3506         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
3507
3508         rc = lu_env_init(&env, LCT_MD_THREAD);
3509         if (rc < 0) {
3510                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
3511                 RETURN(rc);
3512         }
3513
3514         /* Allocate new sequence now to avoid creating local transaction
3515          * in the normal transaction process */
3516         rc = seq_server_check_and_alloc_super(&env,
3517                                               mdt_seq_site(mdt)->ss_server_seq);
3518         if (rc < 0)
3519                 GOTO(out, rc);
3520
3521         if (fld->lsf_new) {
3522                 rc = fld_update_from_controller(&env, fld);
3523                 if (rc != 0) {
3524                         CERROR("%s: cannot update controller: rc = %d\n",
3525                                mdt_obd_name(mdt), rc);
3526                         GOTO(out, rc);
3527                 }
3528         }
3529 out:
3530         lu_env_fini(&env);
3531         RETURN(rc);
3532 }
3533
3534 static int mdt_register_seq_exp(struct mdt_device *mdt)
3535 {
3536         struct seq_server_site  *ss = mdt_seq_site(mdt);
3537         char                    *lwp_name = NULL;
3538         int                     rc;
3539
3540         if (ss->ss_node_id == 0)
3541                 return 0;
3542
3543         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
3544         if (lwp_name == NULL)
3545                 GOTO(out_free, rc = -ENOMEM);
3546
3547         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
3548         if (rc != 0)
3549                 GOTO(out_free, rc);
3550
3551         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
3552                                       NULL, NULL);
3553         if (rc != 0)
3554                 GOTO(out_free, rc);
3555
3556         rc = lustre_register_lwp_item(lwp_name,
3557                                       &ss->ss_server_fld->lsf_control_exp,
3558                                       mdt_register_lwp_callback, mdt);
3559         if (rc != 0) {
3560                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3561                 ss->ss_client_seq->lcs_exp = NULL;
3562                 GOTO(out_free, rc);
3563         }
3564 out_free:
3565         if (lwp_name != NULL)
3566                 OBD_FREE(lwp_name, MAX_OBD_NAME);
3567
3568         return rc;
3569 }
3570
3571 /*
3572  * Init client sequence manager which is used by local MDS to talk to sequence
3573  * controller on remote node.
3574  */
3575 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
3576 {
3577         struct seq_server_site  *ss = mdt_seq_site(mdt);
3578         int                     rc;
3579         char                    *prefix;
3580         ENTRY;
3581
3582         /* check if this is adding the first MDC and controller is not yet
3583          * initialized. */
3584         OBD_ALLOC_PTR(ss->ss_client_seq);
3585         if (ss->ss_client_seq == NULL)
3586                 RETURN(-ENOMEM);
3587
3588         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3589         if (prefix == NULL) {
3590                 OBD_FREE_PTR(ss->ss_client_seq);
3591                 ss->ss_client_seq = NULL;
3592                 RETURN(-ENOMEM);
3593         }
3594
3595         /* Note: seq_client_fini will be called in seq_site_fini */
3596         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
3597         rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
3598                              prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
3599                                                             NULL);
3600         OBD_FREE(prefix, MAX_OBD_NAME + 5);
3601         if (rc != 0) {
3602                 OBD_FREE_PTR(ss->ss_client_seq);
3603                 ss->ss_client_seq = NULL;
3604                 RETURN(rc);
3605         }
3606
3607         rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
3608
3609         RETURN(rc);
3610 }
3611
3612 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
3613 {
3614         struct seq_server_site  *ss;
3615         int                     rc;
3616         ENTRY;
3617
3618         ss = mdt_seq_site(mdt);
3619         /* init sequence controller server(MDT0) */
3620         if (ss->ss_node_id == 0) {
3621                 OBD_ALLOC_PTR(ss->ss_control_seq);
3622                 if (ss->ss_control_seq == NULL)
3623                         RETURN(-ENOMEM);
3624
3625                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
3626                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
3627                                      ss);
3628                 if (rc)
3629                         GOTO(out_seq_fini, rc);
3630         }
3631
3632         /* Init normal sequence server */
3633         OBD_ALLOC_PTR(ss->ss_server_seq);
3634         if (ss->ss_server_seq == NULL)
3635                 GOTO(out_seq_fini, rc = -ENOMEM);
3636
3637         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
3638                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
3639         if (rc)
3640                 GOTO(out_seq_fini, rc);
3641
3642         /* init seq client for seq server to talk to seq controller(MDT0) */
3643         rc = mdt_seq_init_cli(env, mdt);
3644         if (rc != 0)
3645                 GOTO(out_seq_fini, rc);
3646
3647         if (ss->ss_node_id != 0)
3648                 /* register controller export through lwp */
3649                 rc = mdt_register_seq_exp(mdt);
3650
3651         EXIT;
3652 out_seq_fini:
3653         if (rc)
3654                 mdt_seq_fini(env, mdt);
3655
3656         return rc;
3657 }
3658
3659 /*
3660  * FLD wrappers
3661  */
3662 static int mdt_fld_fini(const struct lu_env *env,
3663                         struct mdt_device *m)
3664 {
3665         struct seq_server_site *ss = mdt_seq_site(m);
3666         ENTRY;
3667
3668         if (ss && ss->ss_server_fld) {
3669                 fld_server_fini(env, ss->ss_server_fld);
3670                 OBD_FREE_PTR(ss->ss_server_fld);
3671                 ss->ss_server_fld = NULL;
3672         }
3673
3674         RETURN(0);
3675 }
3676
3677 static int mdt_fld_init(const struct lu_env *env,
3678                         const char *uuid,
3679                         struct mdt_device *m)
3680 {
3681         struct seq_server_site *ss;
3682         int rc;
3683         ENTRY;
3684
3685         ss = mdt_seq_site(m);
3686
3687         OBD_ALLOC_PTR(ss->ss_server_fld);
3688         if (ss->ss_server_fld == NULL)
3689                 RETURN(rc = -ENOMEM);
3690
3691         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
3692                              LU_SEQ_RANGE_MDT);
3693         if (rc) {
3694                 OBD_FREE_PTR(ss->ss_server_fld);
3695                 ss->ss_server_fld = NULL;
3696                 RETURN(rc);
3697         }
3698
3699         RETURN(0);
3700 }
3701
3702 static void mdt_stack_pre_fini(const struct lu_env *env,
3703                            struct mdt_device *m, struct lu_device *top)
3704 {
3705         struct lustre_cfg_bufs  *bufs;
3706         struct lustre_cfg       *lcfg;
3707         struct mdt_thread_info  *info;
3708         ENTRY;
3709
3710         LASSERT(top);
3711
3712         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3713         LASSERT(info != NULL);
3714
3715         bufs = &info->mti_u.bufs;
3716
3717         LASSERT(m->mdt_child_exp);
3718         LASSERT(m->mdt_child_exp->exp_obd);
3719
3720         /* process cleanup, pass mdt obd name to get obd umount flags */
3721         /* XXX: this is needed because all layers are referenced by
3722          * objects (some of them are pinned by osd, for example *
3723          * the proper solution should be a model where object used
3724          * by osd only doesn't have mdt/mdd slices -bzzz */
3725         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
3726         lustre_cfg_bufs_set_string(bufs, 1, NULL);
3727         lcfg = lustre_cfg_new(LCFG_PRE_CLEANUP, bufs);
3728         if (lcfg == NULL)
3729                 RETURN_EXIT;
3730
3731         top->ld_ops->ldo_process_config(env, top, lcfg);
3732         lustre_cfg_free(lcfg);
3733         EXIT;
3734 }
3735
3736 static void mdt_stack_fini(const struct lu_env *env,
3737                            struct mdt_device *m, struct lu_device *top)
3738 {
3739         struct obd_device       *obd = mdt2obd_dev(m);
3740         struct lustre_cfg_bufs  *bufs;
3741         struct lustre_cfg       *lcfg;
3742         struct mdt_thread_info  *info;
3743         char                     flags[3] = "";
3744         ENTRY;
3745
3746         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3747         LASSERT(info != NULL);
3748
3749         lu_dev_del_linkage(top->ld_site, top);
3750
3751         lu_site_purge(env, top->ld_site, -1);
3752
3753         bufs = &info->mti_u.bufs;
3754         /* process cleanup, pass mdt obd name to get obd umount flags */
3755         /* another purpose is to let all layers to release their objects */
3756         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
3757         if (obd->obd_force)
3758                 strcat(flags, "F");
3759         if (obd->obd_fail)
3760                 strcat(flags, "A");
3761         lustre_cfg_bufs_set_string(bufs, 1, flags);
3762         lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
3763         if (lcfg == NULL)
3764                 RETURN_EXIT;
3765
3766         LASSERT(top);
3767         top->ld_ops->ldo_process_config(env, top, lcfg);
3768         lustre_cfg_free(lcfg);
3769
3770         lu_site_purge(env, top->ld_site, -1);
3771
3772         m->mdt_child = NULL;
3773         m->mdt_bottom = NULL;
3774
3775         obd_disconnect(m->mdt_child_exp);
3776         m->mdt_child_exp = NULL;
3777
3778         obd_disconnect(m->mdt_bottom_exp);
3779         m->mdt_child_exp = NULL;
3780 }
3781
3782 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
3783                                const char *next, struct obd_export **exp)
3784 {
3785         struct obd_connect_data *data = NULL;
3786         struct obd_device       *obd;
3787         int                      rc;
3788         ENTRY;
3789
3790         OBD_ALLOC_PTR(data);
3791         if (data == NULL)
3792                 GOTO(out, rc = -ENOMEM);
3793
3794         obd = class_name2obd(next);
3795         if (obd == NULL) {
3796                 CERROR("%s: can't locate next device: %s\n",
3797                        mdt_obd_name(m), next);
3798                 GOTO(out, rc = -ENOTCONN);
3799         }
3800
3801         data->ocd_connect_flags = OBD_CONNECT_VERSION;
3802         data->ocd_version = LUSTRE_VERSION_CODE;
3803
3804         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
3805         if (rc) {
3806                 CERROR("%s: cannot connect to next dev %s (%d)\n",
3807                        mdt_obd_name(m), next, rc);
3808                 GOTO(out, rc);
3809         }
3810
3811 out:
3812         if (data)
3813                 OBD_FREE_PTR(data);
3814         RETURN(rc);
3815 }
3816
3817 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
3818                           struct lustre_cfg *cfg)
3819 {
3820         char                   *dev = lustre_cfg_string(cfg, 0);
3821         int                     rc, name_size, uuid_size;
3822         char                   *name, *uuid, *p;
3823         struct lustre_cfg_bufs *bufs;
3824         struct lustre_cfg      *lcfg;
3825         struct obd_device      *obd;
3826         struct lustre_profile  *lprof;
3827         struct lu_site         *site;
3828         ENTRY;
3829
3830         /* in 1.8 we had the only device in the stack - MDS.
3831          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
3832          * in 2.3 OSD is instantiated by obd_mount.c, so we need
3833          * to generate names and setup MDT, MDD. MDT will be using
3834          * generated name to connect to MDD. for MDD the next device
3835          * will be LOD with name taken from so called "profile" which
3836          * is generated by mount_option line
3837          *
3838          * 1.8 MGS generates config. commands like this:
3839          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
3840          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
3841          * 2.0 MGS generates config. commands like this:
3842          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
3843          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
3844          *                    3:lustre-MDT0000-mdtlov  4:f
3845          *
3846          * we generate MDD name from MDT one, just replacing T with D
3847          *
3848          * after all the preparations, the logical equivalent will be
3849          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
3850          *                    3:lustre-MDT0000-mdtlov  4:f
3851          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
3852          *                    3:lustre-MDD0000  4:f
3853          *
3854          *  notice we build the stack from down to top: MDD first, then MDT */
3855
3856         name_size = MAX_OBD_NAME;
3857         uuid_size = MAX_OBD_NAME;
3858
3859         OBD_ALLOC(name, name_size);
3860         OBD_ALLOC(uuid, uuid_size);
3861         if (name == NULL || uuid == NULL)
3862                 GOTO(cleanup_mem, rc = -ENOMEM);
3863
3864         OBD_ALLOC_PTR(bufs);
3865         if (!bufs)
3866                 GOTO(cleanup_mem, rc = -ENOMEM);
3867
3868         strcpy(name, dev);
3869         p = strstr(name, "-MDT");
3870         if (p == NULL)
3871                 GOTO(free_bufs, rc = -ENOMEM);
3872         p[3] = 'D';
3873
3874         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
3875
3876         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
3877         if (lprof == NULL || lprof->lp_dt == NULL) {
3878                 CERROR("can't find the profile: %s\n",
3879                        lustre_cfg_string(cfg, 0));
3880                 GOTO(free_bufs, rc = -EINVAL);
3881         }
3882
3883         lustre_cfg_bufs_reset(bufs, name);
3884         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
3885         lustre_cfg_bufs_set_string(bufs, 2, uuid);
3886         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
3887
3888         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
3889         if (lcfg == NULL)
3890                 GOTO(free_bufs, rc = -ENOMEM);
3891
3892         rc = class_attach(lcfg);
3893         if (rc)
3894                 GOTO(lcfg_cleanup, rc);
3895
3896         obd = class_name2obd(name);
3897         if (!obd) {
3898                 CERROR("Can not find obd %s (%s in config)\n",
3899                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
3900                 GOTO(lcfg_cleanup, rc = -EINVAL);
3901         }
3902
3903         lustre_cfg_free(lcfg);
3904
3905         lustre_cfg_bufs_reset(bufs, name);
3906         lustre_cfg_bufs_set_string(bufs, 1, uuid);
3907         lustre_cfg_bufs_set_string(bufs, 2, dev);
3908         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
3909
3910         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
3911         if (lcfg == NULL)
3912                 GOTO(class_detach, rc = -ENOMEM);
3913
3914         rc = class_setup(obd, lcfg);
3915         if (rc)
3916                 GOTO(class_detach, rc);
3917
3918         /* connect to MDD we just setup */
3919         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
3920         if (rc)
3921                 GOTO(class_detach, rc);
3922
3923         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
3924         LASSERT(site);
3925         LASSERT(mdt_lu_site(mdt) == NULL);
3926         mdt->mdt_lu_dev.ld_site = site;
3927         site->ls_top_dev = &mdt->mdt_lu_dev;
3928         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
3929
3930         /* now connect to bottom OSD */
3931         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
3932         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
3933         if (rc)
3934                 GOTO(class_detach, rc);
3935         mdt->mdt_bottom =
3936                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
3937
3938         rc = lu_env_refill((struct lu_env *)env);
3939         if (rc != 0)
3940                 CERROR("Failure to refill session: '%d'\n", rc);
3941
3942         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
3943
3944         EXIT;
3945 class_detach:
3946         if (rc)
3947                 class_detach(obd, lcfg);
3948 lcfg_cleanup:
3949         lustre_cfg_free(lcfg);
3950 free_bufs:
3951         OBD_FREE_PTR(bufs);
3952 cleanup_mem:
3953         if (name)
3954                 OBD_FREE(name, name_size);
3955         if (uuid)
3956                 OBD_FREE(uuid, uuid_size);
3957         RETURN(rc);
3958 }
3959
3960 /* setup quota master target on MDT0 */
3961 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
3962                           struct lustre_cfg *cfg)
3963 {
3964         struct obd_device       *obd;
3965         char                    *dev = lustre_cfg_string(cfg, 0);
3966         char                    *qmtname, *uuid, *p;
3967         struct lustre_cfg_bufs  *bufs;
3968         struct lustre_cfg       *lcfg;
3969         struct lustre_profile   *lprof;
3970         struct obd_connect_data *data;
3971         int                      rc;
3972         ENTRY;
3973
3974         LASSERT(mdt->mdt_qmt_exp == NULL);
3975         LASSERT(mdt->mdt_qmt_dev == NULL);
3976
3977         /* quota master is on MDT0 only for now */
3978         if (mdt->mdt_seq_site.ss_node_id != 0)
3979                 RETURN(0);
3980
3981         /* MGS generates config commands which look as follows:
3982          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
3983          *                    3:lustre-MDT0000-mdtlov  4:f
3984          *
3985          * We generate the QMT name from the MDT one, just replacing MD with QM
3986          * after all the preparations, the logical equivalent will be:
3987          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
3988          *                    3:lustre-MDT0000-osd  4:f */
3989         OBD_ALLOC(qmtname, MAX_OBD_NAME);
3990         OBD_ALLOC(uuid, UUID_MAX);
3991         OBD_ALLOC_PTR(bufs);
3992         OBD_ALLOC_PTR(data);
3993         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
3994                 GOTO(cleanup_mem, rc = -ENOMEM);
3995
3996         strcpy(qmtname, dev);
3997         p = strstr(qmtname, "-MDT");
3998         if (p == NULL)
3999                 GOTO(cleanup_mem, rc = -ENOMEM);
4000         /* replace MD with QM */
4001         p[1] = 'Q';
4002         p[2] = 'M';
4003
4004         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
4005
4006         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4007         if (lprof == NULL || lprof->lp_dt == NULL) {
4008                 CERROR("can't find profile for %s\n",
4009                        lustre_cfg_string(cfg, 0));
4010                 GOTO(cleanup_mem, rc = -EINVAL);
4011         }
4012
4013         lustre_cfg_bufs_reset(bufs, qmtname);
4014         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
4015         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4016         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4017
4018         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4019         if (lcfg == NULL)
4020                 GOTO(cleanup_mem, rc = -ENOMEM);
4021
4022         rc = class_attach(lcfg);
4023         if (rc)
4024                 GOTO(lcfg_cleanup, rc);
4025
4026         obd = class_name2obd(qmtname);
4027         if (!obd) {
4028                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
4029                        lustre_cfg_string(cfg, 0));
4030                 GOTO(lcfg_cleanup, rc = -EINVAL);
4031         }
4032
4033         lustre_cfg_free(lcfg);
4034
4035         lustre_cfg_bufs_reset(bufs, qmtname);
4036         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4037         lustre_cfg_bufs_set_string(bufs, 2, dev);
4038
4039         /* for quota, the next device should be the OSD device */
4040         lustre_cfg_bufs_set_string(bufs, 3,
4041                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
4042
4043         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4044         if (lcfg == NULL)
4045                 GOTO(class_detach, rc = -ENOMEM);
4046
4047         rc = class_setup(obd, lcfg);
4048         if (rc)
4049                 GOTO(class_detach, rc);
4050
4051         mdt->mdt_qmt_dev = obd->obd_lu_dev;
4052
4053         /* configure local quota objects */
4054         rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
4055                                                    &mdt->mdt_lu_dev,
4056                                                    mdt->mdt_qmt_dev);
4057         if (rc)
4058                 GOTO(class_cleanup, rc);
4059
4060         /* connect to quota master target */
4061         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4062         data->ocd_version = LUSTRE_VERSION_CODE;
4063         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
4064                          data, NULL);
4065         if (rc) {
4066                 CERROR("cannot connect to quota master device %s (%d)\n",
4067                        qmtname, rc);
4068                 GOTO(class_cleanup, rc);
4069         }
4070
4071         EXIT;
4072 class_cleanup:
4073         if (rc) {
4074                 class_manual_cleanup(obd);
4075                 mdt->mdt_qmt_dev = NULL;
4076         }
4077 class_detach:
4078         if (rc)
4079                 class_detach(obd, lcfg);
4080 lcfg_cleanup:
4081         lustre_cfg_free(lcfg);
4082 cleanup_mem:
4083         if (bufs)
4084                 OBD_FREE_PTR(bufs);
4085         if (qmtname)
4086                 OBD_FREE(qmtname, MAX_OBD_NAME);
4087         if (uuid)
4088                 OBD_FREE(uuid, UUID_MAX);
4089         if (data)
4090                 OBD_FREE_PTR(data);
4091         return rc;
4092 }
4093
4094 /* Shutdown quota master target associated with mdt */
4095 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
4096 {
4097         ENTRY;
4098
4099         if (mdt->mdt_qmt_exp == NULL)
4100                 RETURN_EXIT;
4101         LASSERT(mdt->mdt_qmt_dev != NULL);
4102
4103         /* the qmt automatically shuts down when the mdt disconnects */
4104         obd_disconnect(mdt->mdt_qmt_exp);
4105         mdt->mdt_qmt_exp = NULL;
4106         mdt->mdt_qmt_dev = NULL;
4107         EXIT;
4108 }
4109
4110 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
4111  * for now. This will be removed along with converting rest of MDT code
4112  * to use tgt_session_info */
4113 static int mdt_tgt_getxattr(struct tgt_session_info *tsi)
4114 {
4115         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
4116         int                      rc;
4117
4118         rc = mdt_getxattr(info);
4119
4120         mdt_thread_info_fini(info);
4121         return rc;
4122 }
4123
4124 static struct tgt_handler mdt_tgt_handlers[] = {
4125 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4126                 0,                      MDS_CONNECT,    mdt_tgt_connect,
4127                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
4128 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4129                 0,                      MDS_DISCONNECT, tgt_disconnect,
4130                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
4131 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4132                 HABEO_REFERO,           MDS_SET_INFO,   mdt_set_info,
4133                 &RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION),
4134 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
4135 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_GETSTATUS,  mdt_getstatus),
4136 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETATTR,    mdt_getattr),
4137 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_GETATTR_NAME,
4138                                                         mdt_getattr_name),
4139 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETXATTR,   mdt_tgt_getxattr),
4140 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_STATFS,     mdt_statfs),
4141 TGT_MDT_HDL(0           | MUTABOR,      MDS_REINT,      mdt_reint),
4142 TGT_MDT_HDL(HABEO_CORPUS,               MDS_CLOSE,      mdt_close),
4143 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_READPAGE,   mdt_readpage),
4144 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_SYNC,       mdt_sync),
4145 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
4146 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_PROGRESS,
4147                                                         mdt_hsm_progress),
4148 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_REGISTER,
4149                                                         mdt_hsm_ct_register),
4150 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_UNREGISTER,
4151                                                         mdt_hsm_ct_unregister),
4152 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_STATE_GET,
4153                                                         mdt_hsm_state_get),
4154 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_STATE_SET,
4155                                                         mdt_hsm_state_set),
4156 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_ACTION, mdt_hsm_action),
4157 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_REQUEST,
4158                                                         mdt_hsm_request),
4159 TGT_MDT_HDL(HABEO_CLAVIS | HABEO_CORPUS | HABEO_REFERO | MUTABOR,
4160             MDS_SWAP_LAYOUTS,
4161             mdt_swap_layouts),
4162 };
4163
4164 static struct tgt_handler mdt_sec_ctx_ops[] = {
4165 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
4166 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
4167 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
4168 };
4169
4170 static struct tgt_handler mdt_quota_ops[] = {
4171 TGT_QUOTA_HDL(HABEO_REFERO,             QUOTA_DQACQ,      mdt_quota_dqacq),
4172 };
4173
4174 static struct tgt_opc_slice mdt_common_slice[] = {
4175         {
4176                 .tos_opc_start  = MDS_FIRST_OPC,
4177                 .tos_opc_end    = MDS_LAST_OPC,
4178                 .tos_hs         = mdt_tgt_handlers
4179         },
4180         {
4181                 .tos_opc_start  = OBD_FIRST_OPC,
4182                 .tos_opc_end    = OBD_LAST_OPC,
4183                 .tos_hs         = tgt_obd_handlers
4184         },
4185         {
4186                 .tos_opc_start  = LDLM_FIRST_OPC,
4187                 .tos_opc_end    = LDLM_LAST_OPC,
4188                 .tos_hs         = tgt_dlm_handlers
4189         },
4190         {
4191                 .tos_opc_start  = SEC_FIRST_OPC,
4192                 .tos_opc_end    = SEC_LAST_OPC,
4193                 .tos_hs         = mdt_sec_ctx_ops
4194         },
4195         {
4196                 .tos_opc_start  = OUT_UPDATE_FIRST_OPC,
4197                 .tos_opc_end    = OUT_UPDATE_LAST_OPC,
4198                 .tos_hs         = tgt_out_handlers
4199         },
4200         {
4201                 .tos_opc_start  = FLD_FIRST_OPC,
4202                 .tos_opc_end    = FLD_LAST_OPC,
4203                 .tos_hs         = fld_handlers
4204         },
4205         {
4206                 .tos_opc_start  = SEQ_FIRST_OPC,
4207                 .tos_opc_end    = SEQ_LAST_OPC,
4208                 .tos_hs         = seq_handlers
4209         },
4210         {
4211                 .tos_opc_start  = QUOTA_DQACQ,
4212                 .tos_opc_end    = QUOTA_LAST_OPC,
4213                 .tos_hs         = mdt_quota_ops
4214         },
4215         {
4216                 .tos_opc_start  = LLOG_FIRST_OPC,
4217                 .tos_opc_end    = LLOG_LAST_OPC,
4218                 .tos_hs         = tgt_llog_handlers
4219         },
4220         {
4221                 .tos_opc_start  = LFSCK_FIRST_OPC,
4222                 .tos_opc_end    = LFSCK_LAST_OPC,
4223                 .tos_hs         = tgt_lfsck_handlers
4224         },
4225
4226         {
4227                 .tos_hs         = NULL
4228         }
4229 };
4230
4231 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4232 {
4233         struct md_device        *next = m->mdt_child;
4234         struct lu_device        *d    = &m->mdt_lu_dev;
4235         struct obd_device       *obd  = mdt2obd_dev(m);
4236         struct lfsck_stop        stop;
4237         ENTRY;
4238
4239         stop.ls_status = LS_PAUSED;
4240         stop.ls_flags = 0;
4241         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
4242
4243         target_recovery_fini(obd);
4244         ping_evictor_stop();
4245         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
4246
4247         if (m->mdt_opts.mo_coordinator)
4248                 mdt_hsm_cdt_stop(m);
4249
4250         mdt_hsm_cdt_fini(m);
4251
4252         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
4253         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4254         obd_exports_barrier(obd);
4255         obd_zombie_barrier();
4256
4257         mdt_procfs_fini(m);
4258
4259         tgt_fini(env, &m->mdt_lut);
4260         mdt_fs_cleanup(env, m);
4261         upcall_cache_cleanup(m->mdt_identity_cache);
4262         m->mdt_identity_cache = NULL;
4263
4264         if (m->mdt_namespace != NULL) {
4265                 ldlm_namespace_free(m->mdt_namespace, NULL,
4266                                     d->ld_obd->obd_force);
4267                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4268         }
4269
4270         mdt_quota_fini(env, m);
4271
4272         cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
4273
4274         mdt_seq_fini(env, m);
4275         mdt_fld_fini(env, m);
4276
4277         /*
4278          * Finish the stack
4279          */
4280         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4281
4282         LASSERT(atomic_read(&d->ld_ref) == 0);
4283
4284         server_put_mount(mdt_obd_name(m), true);
4285
4286         EXIT;
4287 }
4288
4289 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4290
4291 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4292                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
4293 {
4294         struct mdt_thread_info    *info;
4295         struct obd_device         *obd;
4296         const char                *dev = lustre_cfg_string(cfg, 0);
4297         const char                *num = lustre_cfg_string(cfg, 2);
4298         struct lustre_mount_info  *lmi = NULL;
4299         struct lustre_sb_info     *lsi;
4300         struct lu_site            *s;
4301         struct seq_server_site    *ss_site;
4302         const char                *identity_upcall = "NONE";
4303         struct md_device          *next;
4304         int                        rc;
4305         long                       node_id;
4306         mntopt_t                   mntopts;
4307         ENTRY;
4308
4309         lu_device_init(&m->mdt_lu_dev, ldt);
4310         /*
4311          * Environment (env) might be missing mdt_thread_key values at that
4312          * point, if device is allocated when mdt_thread_key is in QUIESCENT
4313          * mode.
4314          *
4315          * Usually device allocation path doesn't use module key values, but
4316          * mdt has to do a lot of work here, so allocate key value.
4317          */
4318         rc = lu_env_refill((struct lu_env *)env);
4319         if (rc != 0)
4320                 RETURN(rc);
4321
4322         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4323         LASSERT(info != NULL);
4324
4325         obd = class_name2obd(dev);
4326         LASSERT(obd != NULL);
4327
4328         m->mdt_max_mdsize = MAX_MD_SIZE; /* 4 stripes */
4329
4330         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4331
4332         /* default is coordinator off, it is started through conf_param
4333          * or /proc */
4334         m->mdt_opts.mo_coordinator = 0;
4335
4336         lmi = server_get_mount(dev);
4337         if (lmi == NULL) {
4338                 CERROR("Cannot get mount info for %s!\n", dev);
4339                 RETURN(-EFAULT);
4340         } else {
4341                 lsi = s2lsi(lmi->lmi_sb);
4342                 /* CMD is supported only in IAM mode */
4343                 LASSERT(num);
4344                 node_id = simple_strtol(num, NULL, 10);
4345                 obd->u.obt.obt_magic = OBT_MAGIC;
4346         }
4347
4348         m->mdt_squash.rsi_uid = 0;
4349         m->mdt_squash.rsi_gid = 0;
4350         INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
4351         init_rwsem(&m->mdt_squash.rsi_sem);
4352         spin_lock_init(&m->mdt_osfs_lock);
4353         m->mdt_osfs_age = cfs_time_shift_64(-1000);
4354         m->mdt_enable_remote_dir = 0;
4355         m->mdt_enable_remote_dir_gid = 0;
4356
4357         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
4358         m->mdt_lu_dev.ld_obd = obd;
4359         /* Set this lu_device to obd for error handling purposes. */
4360         obd->obd_lu_dev = &m->mdt_lu_dev;
4361
4362         /* init the stack */
4363         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
4364         if (rc) {
4365                 CERROR("%s: Can't init device stack, rc %d\n",
4366                        mdt_obd_name(m), rc);
4367                 GOTO(err_lmi, rc);
4368         }
4369
4370         s = mdt_lu_site(m);
4371         ss_site = mdt_seq_site(m);
4372         s->ld_seq_site = ss_site;
4373         ss_site->ss_lu = s;
4374
4375         /* set server index */
4376         ss_site->ss_node_id = node_id;
4377
4378         /* failover is the default
4379          * FIXME: we do not failout mds0/mgs, which may cause some problems.
4380          * assumed whose ss_node_id == 0 XXX
4381          * */
4382         obd->obd_replayable = 1;
4383         /* No connection accepted until configurations will finish */
4384         obd->obd_no_conn = 1;
4385
4386         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4387                 char *str = lustre_cfg_string(cfg, 4);
4388                 if (strchr(str, 'n')) {
4389                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
4390                         obd->obd_replayable = 0;
4391                 }
4392         }
4393
4394         rc = mdt_fld_init(env, mdt_obd_name(m), m);
4395         if (rc)
4396                 GOTO(err_fini_stack, rc);
4397
4398         rc = mdt_seq_init(env, m);
4399         if (rc)
4400                 GOTO(err_fini_fld, rc);
4401
4402         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
4403                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
4404         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4405                                               LDLM_NAMESPACE_SERVER,
4406                                               LDLM_NAMESPACE_GREEDY,
4407                                               LDLM_NS_TYPE_MDT);
4408         if (m->mdt_namespace == NULL)
4409                 GOTO(err_fini_seq, rc = -ENOMEM);
4410
4411         m->mdt_namespace->ns_lvbp = m;
4412         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
4413
4414         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4415         /* set obd_namespace for compatibility with old code */
4416         obd->obd_namespace = m->mdt_namespace;
4417
4418         rc = mdt_hsm_cdt_init(m);
4419         if (rc != 0) {
4420                 CERROR("%s: error initializing coordinator, rc %d\n",
4421                        mdt_obd_name(m), rc);
4422                 GOTO(err_free_ns, rc);
4423         }
4424
4425         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
4426                       OBD_FAIL_MDS_ALL_REQUEST_NET,
4427                       OBD_FAIL_MDS_ALL_REPLY_NET);
4428         if (rc)
4429                 GOTO(err_free_hsm, rc);
4430
4431         rc = mdt_fs_setup(env, m, obd, lsi);
4432         if (rc)
4433                 GOTO(err_tgt, rc);
4434
4435         tgt_adapt_sptlrpc_conf(&m->mdt_lut, 1);
4436
4437         next = m->mdt_child;
4438         rc = next->md_ops->mdo_iocontrol(env, next, OBD_IOC_GET_MNTOPT, 0,
4439                                          &mntopts);
4440         if (rc)
4441                 GOTO(err_fs_cleanup, rc);
4442
4443         if (mntopts & MNTOPT_USERXATTR)
4444                 m->mdt_opts.mo_user_xattr = 1;
4445         else
4446                 m->mdt_opts.mo_user_xattr = 0;
4447
4448         rc = next->md_ops->mdo_maxeasize_get(env, next, &m->mdt_max_ea_size);
4449         if (rc)
4450                 GOTO(err_fs_cleanup, rc);
4451
4452         if (mntopts & MNTOPT_ACL)
4453                 m->mdt_opts.mo_acl = 1;
4454         else
4455                 m->mdt_opts.mo_acl = 0;
4456
4457         /* XXX: to support suppgid for ACL, we enable identity_upcall
4458          * by default, otherwise, maybe got unexpected -EACCESS. */
4459         if (m->mdt_opts.mo_acl)
4460                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4461
4462         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
4463                                                 identity_upcall,
4464                                                 &mdt_identity_upcall_cache_ops);
4465         if (IS_ERR(m->mdt_identity_cache)) {
4466                 rc = PTR_ERR(m->mdt_identity_cache);
4467                 m->mdt_identity_cache = NULL;
4468                 GOTO(err_fs_cleanup, rc);
4469         }
4470
4471         rc = mdt_procfs_init(m, dev);
4472         if (rc) {
4473                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4474                 GOTO(err_recovery, rc);
4475         }
4476
4477         rc = mdt_quota_init(env, m, cfg);
4478         if (rc)
4479                 GOTO(err_procfs, rc);
4480
4481         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
4482         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
4483                            "mdt_ldlm_client", m->mdt_ldlm_client);
4484
4485         ping_evictor_start();
4486
4487         /* recovery will be started upon mdt_prepare()
4488          * when the whole stack is complete and ready
4489          * to serve the requests */
4490
4491         /* Reduce the initial timeout on an MDS because it doesn't need such
4492          * a long timeout as an OST does. Adaptive timeouts will adjust this
4493          * value appropriately. */
4494         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4495                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4496
4497         RETURN(0);
4498 err_procfs:
4499         mdt_procfs_fini(m);
4500 err_recovery:
4501         target_recovery_fini(obd);
4502         upcall_cache_cleanup(m->mdt_identity_cache);
4503         m->mdt_identity_cache = NULL;
4504 err_fs_cleanup:
4505         mdt_fs_cleanup(env, m);
4506 err_tgt:
4507         tgt_fini(env, &m->mdt_lut);
4508 err_free_hsm:
4509         mdt_hsm_cdt_fini(m);
4510 err_free_ns:
4511         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4512         obd->obd_namespace = m->mdt_namespace = NULL;
4513 err_fini_seq:
4514         mdt_seq_fini(env, m);
4515 err_fini_fld:
4516         mdt_fld_fini(env, m);
4517 err_fini_stack:
4518         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4519 err_lmi:
4520         if (lmi)
4521                 server_put_mount(dev, true);
4522         return(rc);
4523 }
4524
4525 /* For interoperability, the left element is old parameter, the right one
4526  * is the new version of the parameter, if some parameter is deprecated,
4527  * the new version should be set as NULL. */
4528 static struct cfg_interop_param mdt_interop_param[] = {
4529         { "mdt.group_upcall",   NULL },
4530         { "mdt.quota_type",     NULL },
4531         { "mdd.quota_type",     NULL },
4532         { "mdt.rootsquash",     "mdt.root_squash" },
4533         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
4534         { NULL }
4535 };
4536
4537 /* used by MGS to process specific configurations */
4538 static int mdt_process_config(const struct lu_env *env,
4539                               struct lu_device *d, struct lustre_cfg *cfg)
4540 {
4541         struct mdt_device *m = mdt_dev(d);
4542         struct md_device *md_next = m->mdt_child;
4543         struct lu_device *next = md2lu_dev(md_next);
4544         int rc;
4545         ENTRY;
4546
4547         switch (cfg->lcfg_command) {
4548         case LCFG_PARAM: {
4549                 struct obd_device          *obd = d->ld_obd;
4550
4551                 /* For interoperability */
4552                 struct cfg_interop_param   *ptr = NULL;
4553                 struct lustre_cfg          *old_cfg = NULL;
4554                 char                       *param = NULL;
4555
4556                 param = lustre_cfg_string(cfg, 1);
4557                 if (param == NULL) {
4558                         CERROR("param is empty\n");
4559                         rc = -EINVAL;
4560                         break;
4561                 }
4562
4563                 ptr = class_find_old_param(param, mdt_interop_param);
4564                 if (ptr != NULL) {
4565                         if (ptr->new_param == NULL) {
4566                                 rc = 0;
4567                                 CWARN("For interoperability, skip this %s."
4568                                       " It is obsolete.\n", ptr->old_param);
4569                                 break;
4570                         }
4571
4572                         CWARN("Found old param %s, changed it to %s.\n",
4573                               ptr->old_param, ptr->new_param);
4574
4575                         old_cfg = cfg;
4576                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
4577                         if (IS_ERR(cfg)) {
4578                                 rc = PTR_ERR(cfg);
4579                                 break;
4580                         }
4581                 }
4582
4583                 rc = class_process_proc_param(PARAM_MDT, obd->obd_vars,
4584                                               cfg, obd);
4585                 if (rc > 0 || rc == -ENOSYS) {
4586                         /* is it an HSM var ? */
4587                         rc = class_process_proc_param(PARAM_HSM,
4588                                                       hsm_cdt_get_proc_vars(),
4589                                                       cfg, obd);
4590                         if (rc > 0 || rc == -ENOSYS)
4591                                 /* we don't understand; pass it on */
4592                                 rc = next->ld_ops->ldo_process_config(env, next,
4593                                                                       cfg);
4594                 }
4595
4596                 if (old_cfg != NULL)
4597                         lustre_cfg_free(cfg);
4598
4599                 break;
4600         }
4601         default:
4602                 /* others are passed further */
4603                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4604                 break;
4605         }
4606         RETURN(rc);
4607 }
4608
4609 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4610                                           const struct lu_object_header *hdr,
4611                                           struct lu_device *d)
4612 {
4613         struct mdt_object *mo;
4614
4615         ENTRY;
4616
4617         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
4618         if (mo != NULL) {
4619                 struct lu_object *o;
4620                 struct lu_object_header *h;
4621
4622                 o = &mo->mot_obj;
4623                 h = &mo->mot_header;
4624                 lu_object_header_init(h);
4625                 lu_object_init(o, h, d);
4626                 lu_object_add_top(h, o);
4627                 o->lo_ops = &mdt_obj_ops;
4628                 spin_lock_init(&mo->mot_write_lock);
4629                 mutex_init(&mo->mot_lov_mutex);
4630                 init_rwsem(&mo->mot_open_sem);
4631                 RETURN(o);
4632         }
4633         RETURN(NULL);
4634 }
4635
4636 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
4637                            const struct lu_object_conf *unused)
4638 {
4639         struct mdt_device *d = mdt_dev(o->lo_dev);
4640         struct lu_device  *under;
4641         struct lu_object  *below;
4642         int                rc = 0;
4643         ENTRY;
4644
4645         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4646                PFID(lu_object_fid(o)));
4647
4648         under = &d->mdt_child->md_lu_dev;
4649         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4650         if (below != NULL) {
4651                 lu_object_add(o, below);
4652         } else
4653                 rc = -ENOMEM;
4654
4655         RETURN(rc);
4656 }
4657
4658 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4659 {
4660         struct mdt_object *mo = mdt_obj(o);
4661         struct lu_object_header *h;
4662         ENTRY;
4663
4664         h = o->lo_header;
4665         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4666                PFID(lu_object_fid(o)));
4667
4668         LASSERT(atomic_read(&mo->mot_open_count) == 0);
4669         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
4670
4671         lu_object_fini(o);
4672         lu_object_header_fini(h);
4673         OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
4674
4675         EXIT;
4676 }
4677
4678 static int mdt_object_print(const struct lu_env *env, void *cookie,
4679                             lu_printer_t p, const struct lu_object *o)
4680 {
4681         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
4682
4683         return (*p)(env, cookie,
4684                     LUSTRE_MDT_NAME"-object@%p(flags=%d, writecount=%d)",
4685                     mdto, mdto->mot_flags, mdto->mot_write_count);
4686 }
4687
4688 static int mdt_prepare(const struct lu_env *env,
4689                 struct lu_device *pdev,
4690                 struct lu_device *cdev)
4691 {
4692         struct mdt_device *mdt = mdt_dev(cdev);
4693         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
4694         struct obd_device *obd = cdev->ld_obd;
4695         int rc;
4696
4697         ENTRY;
4698
4699         LASSERT(obd);
4700
4701         rc = next->ld_ops->ldo_prepare(env, cdev, next);
4702         if (rc)
4703                 RETURN(rc);
4704
4705         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
4706         if (rc)
4707                 RETURN(rc);
4708
4709         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
4710         if (rc)
4711                 RETURN(rc);
4712
4713         rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
4714         /* The LFSCK instance is registered just now, so it must be there when
4715          * register the namespace to such instance. */
4716         LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
4717
4718         if (mdt->mdt_seq_site.ss_node_id == 0) {
4719                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
4720                                                          &mdt->mdt_md_root_fid);
4721                 if (rc)
4722                         RETURN(rc);
4723         }
4724
4725         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
4726         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
4727         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
4728         LASSERT(obd->obd_no_conn);
4729         spin_lock(&obd->obd_dev_lock);
4730         obd->obd_no_conn = 0;
4731         spin_unlock(&obd->obd_dev_lock);
4732
4733         if (obd->obd_recovering == 0)
4734                 mdt_postrecov(env, mdt);
4735
4736         RETURN(rc);
4737 }
4738
4739 const struct lu_device_operations mdt_lu_ops = {
4740         .ldo_object_alloc   = mdt_object_alloc,
4741         .ldo_process_config = mdt_process_config,
4742         .ldo_prepare        = mdt_prepare,
4743 };
4744
4745 static const struct lu_object_operations mdt_obj_ops = {
4746         .loo_object_init    = mdt_object_init,
4747         .loo_object_free    = mdt_object_free,
4748         .loo_object_print   = mdt_object_print
4749 };
4750
4751 static int mdt_obd_set_info_async(const struct lu_env *env,
4752                                   struct obd_export *exp,
4753                                   __u32 keylen, void *key,
4754                                   __u32 vallen, void *val,
4755                                   struct ptlrpc_request_set *set)
4756 {
4757         int rc;
4758
4759         ENTRY;
4760
4761         if (KEY_IS(KEY_SPTLRPC_CONF)) {
4762                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp), 0);
4763                 RETURN(rc);
4764         }
4765
4766         RETURN(0);
4767 }
4768
4769 /**
4770  * Match client and server connection feature flags.
4771  *
4772  * Compute the compatibility flags for a connection request based on
4773  * features mutually supported by client and server.
4774  *
4775  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
4776  * must not be updated here, otherwise a partially initialized value may
4777  * be exposed. After the connection request is successfully processed,
4778  * the top-level MDT connect request handler atomically updates the export
4779  * connect flags from the obd_connect_data::ocd_connect_flags field of the
4780  * reply. \see mdt_connect().
4781  *
4782  * \param exp   the obd_export associated with this client/target pair
4783  * \param mdt   the target device for the connection
4784  * \param data  stores data for this connect request
4785  *
4786  * \retval 0       success
4787  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
4788  * \retval -EBADE  client and server feature requirements are incompatible
4789  */
4790 static int mdt_connect_internal(struct obd_export *exp,
4791                                 struct mdt_device *mdt,
4792                                 struct obd_connect_data *data)
4793 {
4794         LASSERT(data != NULL);
4795
4796         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
4797         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
4798
4799         if (!(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
4800             !(data->ocd_connect_flags & OBD_CONNECT_IBITS)) {
4801                 CWARN("%s: client %s does not support ibits lock, either "
4802                       "very old or an invalid client: flags "LPX64"\n",
4803                       mdt_obd_name(mdt), exp->exp_client_uuid.uuid,
4804                       data->ocd_connect_flags);
4805                 return -EBADE;
4806         }
4807
4808         if (!mdt->mdt_opts.mo_acl)
4809                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
4810
4811         if (!mdt->mdt_opts.mo_user_xattr)
4812                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
4813
4814         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
4815                 data->ocd_brw_size = min(data->ocd_brw_size,
4816                                          (__u32)MD_MAX_BRW_SIZE);
4817                 if (data->ocd_brw_size == 0) {
4818                         CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
4819                                " ocd_version: %x ocd_grant: %d "
4820                                "ocd_index: %u ocd_brw_size is "
4821                                "unexpectedly zero, network data "
4822                                "corruption? Refusing connection of this"
4823                                " client\n",
4824                                mdt_obd_name(mdt),
4825                                exp->exp_client_uuid.uuid,
4826                                exp, data->ocd_connect_flags, data->ocd_version,
4827                                data->ocd_grant, data->ocd_index);
4828                         return -EPROTO;
4829                 }
4830         }
4831
4832         /* NB: Disregard the rule against updating
4833          * exp_connect_data.ocd_connect_flags in this case, since
4834          * tgt_client_new() needs to know if this is a lightweight
4835          * connection, and it is safe to expose this flag before
4836          * connection processing completes. */
4837         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
4838                 spin_lock(&exp->exp_lock);
4839                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
4840                 spin_unlock(&exp->exp_lock);
4841         }
4842
4843         data->ocd_version = LUSTRE_VERSION_CODE;
4844
4845         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
4846                 CWARN("%s: MDS requires FID support, but client not\n",
4847                       mdt_obd_name(mdt));
4848                 return -EBADE;
4849         }
4850
4851         if (OCD_HAS_FLAG(data, PINGLESS)) {
4852                 if (ptlrpc_pinger_suppress_pings()) {
4853                         spin_lock(&exp->exp_obd->obd_dev_lock);
4854                         list_del_init(&exp->exp_obd_chain_timed);
4855                         spin_unlock(&exp->exp_obd->obd_dev_lock);
4856                 } else {
4857                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
4858                 }
4859         }
4860
4861         data->ocd_max_easize = mdt->mdt_max_ea_size;
4862
4863         return 0;
4864 }
4865
4866 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
4867                                    struct mdt_thread_info *info,
4868                                    struct mdt_file_data *mfd)
4869 {
4870         struct lu_context ses;
4871         int rc;
4872         ENTRY;
4873
4874         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
4875         if (rc)
4876                 RETURN(rc);
4877
4878         env->le_ses = &ses;
4879         lu_context_enter(&ses);
4880
4881         mdt_ucred(info)->uc_valid = UCRED_OLD;
4882         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
4883
4884         lu_context_exit(&ses);
4885         lu_context_fini(&ses);
4886         env->le_ses = NULL;
4887
4888         RETURN(rc);
4889 }
4890
4891 static int mdt_export_cleanup(struct obd_export *exp)
4892 {
4893         struct list_head         closing_list;
4894         struct mdt_export_data  *med = &exp->exp_mdt_data;
4895         struct obd_device       *obd = exp->exp_obd;
4896         struct mdt_device       *mdt;
4897         struct mdt_thread_info  *info;
4898         struct lu_env            env;
4899         struct mdt_file_data    *mfd, *n;
4900         int rc = 0;
4901         ENTRY;
4902
4903         INIT_LIST_HEAD(&closing_list);
4904         spin_lock(&med->med_open_lock);
4905         while (!list_empty(&med->med_open_head)) {
4906                 struct list_head *tmp = med->med_open_head.next;
4907                 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
4908
4909                 /* Remove mfd handle so it can't be found again.
4910                  * We are consuming the mfd_list reference here. */
4911                 class_handle_unhash(&mfd->mfd_handle);
4912                 list_move_tail(&mfd->mfd_list, &closing_list);
4913         }
4914         spin_unlock(&med->med_open_lock);
4915         mdt = mdt_dev(obd->obd_lu_dev);
4916         LASSERT(mdt != NULL);
4917
4918         rc = lu_env_init(&env, LCT_MD_THREAD);
4919         if (rc)
4920                 RETURN(rc);
4921
4922         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
4923         LASSERT(info != NULL);
4924         memset(info, 0, sizeof *info);
4925         info->mti_env = &env;
4926         info->mti_mdt = mdt;
4927         info->mti_exp = exp;
4928
4929         if (!list_empty(&closing_list)) {
4930                 struct md_attr *ma = &info->mti_attr;
4931
4932                 /* Close any open files (which may also cause orphan
4933                  * unlinking). */
4934                 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
4935                         list_del_init(&mfd->mfd_list);
4936                         ma->ma_need = ma->ma_valid = 0;
4937
4938                         /* This file is being closed due to an eviction, it
4939                          * could have been modified and now dirty regarding to
4940                          * HSM archive, check this!
4941                          * The logic here is to mark a file dirty if there's a
4942                          * chance it was dirtied before the client was evicted,
4943                          * so that we don't have to wait for a release attempt
4944                          * before finding out the file was actually dirty and
4945                          * fail the release. Aggressively marking it dirty here
4946                          * will cause the policy engine to attempt to
4947                          * re-archive it; when rearchiving, we can compare the
4948                          * current version to the HSM data_version and make the
4949                          * archive request into a noop if it's not actually
4950                          * dirty.
4951                          */
4952                         if (mfd->mfd_mode & FMODE_WRITE)
4953                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
4954
4955                         /* Don't unlink orphan on failover umount, LU-184 */
4956                         if (exp->exp_flags & OBD_OPT_FAILOVER) {
4957                                 ma->ma_valid = MA_FLAGS;
4958                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
4959                         }
4960                         mdt_mfd_close(info, mfd);
4961                 }
4962         }
4963         info->mti_mdt = NULL;
4964         /* cleanup client slot early */
4965         /* Do not erase record for recoverable client. */
4966         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
4967                 tgt_client_del(&env, exp);
4968         lu_env_fini(&env);
4969
4970         RETURN(rc);
4971 }
4972
4973 static int mdt_obd_disconnect(struct obd_export *exp)
4974 {
4975         int rc;
4976         ENTRY;
4977
4978         LASSERT(exp);
4979         class_export_get(exp);
4980
4981         nodemap_del_member(exp);
4982         rc = server_disconnect_export(exp);
4983         if (rc != 0)
4984                 CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
4985
4986         rc = mdt_export_cleanup(exp);
4987         class_export_put(exp);
4988         RETURN(rc);
4989 }
4990
4991 /* mds_connect copy */
4992 static int mdt_obd_connect(const struct lu_env *env,
4993                            struct obd_export **exp, struct obd_device *obd,
4994                            struct obd_uuid *cluuid,
4995                            struct obd_connect_data *data,
4996                            void *localdata)
4997 {
4998         struct obd_export       *lexp;
4999         struct lustre_handle    conn = { 0 };
5000         struct mdt_device       *mdt;
5001         int                      rc;
5002         lnet_nid_t              *client_nid = localdata;
5003         ENTRY;
5004
5005         LASSERT(env != NULL);
5006         if (!exp || !obd || !cluuid)
5007                 RETURN(-EINVAL);
5008
5009         mdt = mdt_dev(obd->obd_lu_dev);
5010
5011         /*
5012          * first, check whether the stack is ready to handle requests
5013          * XXX: probably not very appropriate method is used now
5014          *      at some point we should find a better one
5015          */
5016         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) && data != NULL &&
5017             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) &&
5018             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
5019                 rc = obd_get_info(env, mdt->mdt_child_exp,
5020                                   sizeof(KEY_OSP_CONNECTED),
5021                                   KEY_OSP_CONNECTED, NULL, NULL);
5022                 if (rc)
5023                         RETURN(-EAGAIN);
5024                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
5025         }
5026
5027         rc = class_connect(&conn, obd, cluuid);
5028         if (rc)
5029                 RETURN(rc);
5030
5031         lexp = class_conn2export(&conn);
5032         LASSERT(lexp != NULL);
5033
5034         rc = mdt_connect_internal(lexp, mdt, data);
5035         if (rc == 0) {
5036                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5037
5038                 LASSERT(lcd);
5039                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5040                 rc = tgt_client_new(env, lexp);
5041                 if (rc == 0) {
5042                         rc = nodemap_add_member(*client_nid, lexp);
5043                         if (rc != 0 && rc != -EEXIST)
5044                                 goto out;
5045
5046                         mdt_export_stats_init(obd, lexp, localdata);
5047                 }
5048
5049                 /* For phase I, sync for cross-ref operation. */
5050                 spin_lock(&lexp->exp_lock);
5051                 lexp->exp_keep_sync = 1;
5052                 spin_unlock(&lexp->exp_lock);
5053         }
5054 out:
5055         if (rc != 0) {
5056                 class_disconnect(lexp);
5057                 *exp = NULL;
5058         } else {
5059                 *exp = lexp;
5060         }
5061
5062         RETURN(rc);
5063 }
5064
5065 static int mdt_obd_reconnect(const struct lu_env *env,
5066                              struct obd_export *exp, struct obd_device *obd,
5067                              struct obd_uuid *cluuid,
5068                              struct obd_connect_data *data,
5069                              void *localdata)
5070 {
5071         lnet_nid_t             *client_nid = localdata;
5072         int                     rc;
5073         ENTRY;
5074
5075         if (exp == NULL || obd == NULL || cluuid == NULL)
5076                 RETURN(-EINVAL);
5077
5078         rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
5079         if (rc == 0) {
5080                 rc = nodemap_add_member(*client_nid, exp);
5081                 if (rc == 0 || rc == -EEXIST)
5082                         mdt_export_stats_init(obd, exp, localdata);
5083         }
5084
5085         RETURN(rc);
5086 }
5087
5088 /* FIXME: Can we avoid using these two interfaces? */
5089 static int mdt_init_export(struct obd_export *exp)
5090 {
5091         struct mdt_export_data *med = &exp->exp_mdt_data;
5092         int                     rc;
5093         ENTRY;
5094
5095         INIT_LIST_HEAD(&med->med_open_head);
5096         spin_lock_init(&med->med_open_lock);
5097         mutex_init(&med->med_idmap_mutex);
5098         med->med_idmap = NULL;
5099         spin_lock(&exp->exp_lock);
5100         exp->exp_connecting = 1;
5101         spin_unlock(&exp->exp_lock);
5102
5103         /* self-export doesn't need client data and ldlm initialization */
5104         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5105                                      &exp->exp_client_uuid)))
5106                 RETURN(0);
5107
5108         rc = tgt_client_alloc(exp);
5109         if (rc)
5110                 GOTO(err, rc);
5111
5112         rc = ldlm_init_export(exp);
5113         if (rc)
5114                 GOTO(err_free, rc);
5115
5116         RETURN(rc);
5117
5118 err_free:
5119         tgt_client_free(exp);
5120 err:
5121         CERROR("%s: Failed to initialize export: rc = %d\n",
5122                exp->exp_obd->obd_name, rc);
5123         return rc;
5124 }
5125
5126 static int mdt_destroy_export(struct obd_export *exp)
5127 {
5128         ENTRY;
5129
5130         if (exp_connect_rmtclient(exp))
5131                 mdt_cleanup_idmap(&exp->exp_mdt_data);
5132
5133         target_destroy_export(exp);
5134         /* destroy can be called from failed obd_setup, so
5135          * checking uuid is safer than obd_self_export */
5136         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5137                                      &exp->exp_client_uuid)))
5138                 RETURN(0);
5139
5140         ldlm_destroy_export(exp);
5141         tgt_client_free(exp);
5142
5143         LASSERT(list_empty(&exp->exp_outstanding_replies));
5144         LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
5145
5146         RETURN(0);
5147 }
5148
5149 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
5150                    struct linkea_data *ldata)
5151 {
5152         int rc;
5153
5154         LASSERT(ldata->ld_buf->lb_buf != NULL);
5155
5156         if (!mdt_object_exists(mdt_obj))
5157                 return -ENODATA;
5158
5159         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5160                           ldata->ld_buf, XATTR_NAME_LINK);
5161         if (rc == -ERANGE) {
5162                 /* Buf was too small, figure out what we need. */
5163                 lu_buf_free(ldata->ld_buf);
5164                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5165                                   ldata->ld_buf, XATTR_NAME_LINK);
5166                 if (rc < 0)
5167                         return rc;
5168                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
5169                 if (ldata->ld_buf->lb_buf == NULL)
5170                         return -ENOMEM;
5171                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5172                                   ldata->ld_buf, XATTR_NAME_LINK);
5173         }
5174         if (rc < 0)
5175                 return rc;
5176
5177         return linkea_init(ldata);
5178 }
5179
5180 /**
5181  * Given an MDT object, try to look up the full path to the object.
5182  * Part of the MDT layer implementation of lfs fid2path.
5183  *
5184  * \param[in]     info  Per-thread common data shared by MDT level handlers.
5185  * \param[in]     obj   Object to do path lookup of
5186  * \param[in,out] fp    User-provided struct to store path information
5187  *
5188  * \retval 0 Lookup successful, path information stored in fp
5189  * \retval -EAGAIN Lookup failed, usually because object is being moved
5190  * \retval negative errno if there was a problem
5191  */
5192 static int mdt_path_current(struct mdt_thread_info *info,
5193                             struct mdt_object *obj,
5194                             struct getinfo_fid2path *fp)
5195 {
5196         struct mdt_device       *mdt = info->mti_mdt;
5197         struct mdt_object       *mdt_obj;
5198         struct link_ea_header   *leh;
5199         struct link_ea_entry    *lee;
5200         struct lu_name          *tmpname = &info->mti_name;
5201         struct lu_fid           *tmpfid = &info->mti_tmp_fid1;
5202         struct lu_buf           *buf = &info->mti_big_buf;
5203         char                    *ptr;
5204         int                     reclen;
5205         struct linkea_data      ldata = { NULL };
5206         int                     rc = 0;
5207         bool                    first = true;
5208         ENTRY;
5209
5210         /* temp buffer for path element, the buffer will be finally freed
5211          * in mdt_thread_info_fini */
5212         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
5213         if (buf->lb_buf == NULL)
5214                 RETURN(-ENOMEM);
5215
5216         ldata.ld_buf = buf;
5217         ptr = fp->gf_path + fp->gf_pathlen - 1;
5218         *ptr = 0;
5219         --ptr;
5220         *tmpfid = fp->gf_fid = *mdt_object_fid(obj);
5221
5222         /* root FID only exists on MDT0, and fid2path should also ends at MDT0,
5223          * so checking root_fid can only happen on MDT0. */
5224         while (!lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid)) {
5225                 struct lu_buf           lmv_buf;
5226
5227                 mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
5228                 if (IS_ERR(mdt_obj))
5229                         GOTO(out, rc = PTR_ERR(mdt_obj));
5230
5231                 if (!mdt_object_exists(mdt_obj)) {
5232                         mdt_object_put(info->mti_env, mdt_obj);
5233                         GOTO(out, rc = -ENOENT);
5234                 }
5235
5236                 if (mdt_object_remote(mdt_obj)) {
5237                         mdt_object_put(info->mti_env, mdt_obj);
5238                         GOTO(remote_out, rc = -EREMOTE);
5239                 }
5240
5241                 rc = mdt_links_read(info, mdt_obj, &ldata);
5242                 if (rc != 0) {
5243                         mdt_object_put(info->mti_env, mdt_obj);
5244                         GOTO(out, rc);
5245                 }
5246
5247                 leh = buf->lb_buf;
5248                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
5249                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
5250                 /* If set, use link #linkno for path lookup, otherwise use
5251                    link #0.  Only do this for the final path element. */
5252                 if (first && fp->gf_linkno < leh->leh_reccount) {
5253                         int count;
5254                         for (count = 0; count < fp->gf_linkno; count++) {
5255                                 lee = (struct link_ea_entry *)
5256                                      ((char *)lee + reclen);
5257                                 linkea_entry_unpack(lee, &reclen, tmpname,
5258                                                     tmpfid);
5259                         }
5260                         if (fp->gf_linkno < leh->leh_reccount - 1)
5261                                 /* indicate to user there are more links */
5262                                 fp->gf_linkno++;
5263                 }
5264
5265                 lmv_buf.lb_buf = info->mti_xattr_buf;
5266                 lmv_buf.lb_len = sizeof(info->mti_xattr_buf);
5267                 /* Check if it is slave stripes */
5268                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5269                                   &lmv_buf, XATTR_NAME_LMV);
5270                 mdt_object_put(info->mti_env, mdt_obj);
5271                 if (rc > 0) {
5272                         union lmv_mds_md *lmm = lmv_buf.lb_buf;
5273
5274                         /* For slave stripes, get its master */
5275                         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE) {
5276                                 fp->gf_fid = *tmpfid;
5277                                 continue;
5278                         }
5279                 } else if (rc < 0 && rc != -ENODATA) {
5280                         GOTO(out, rc);
5281                 }
5282
5283                 rc = 0;
5284
5285                 /* Pack the name in the end of the buffer */
5286                 ptr -= tmpname->ln_namelen;
5287                 if (ptr - 1 <= fp->gf_path)
5288                         GOTO(out, rc = -EOVERFLOW);
5289                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
5290                 *(--ptr) = '/';
5291
5292                 /* keep the last resolved fid to the client, so the
5293                  * client will build the left path on another MDT for
5294                  * remote object */
5295                 fp->gf_fid = *tmpfid;
5296
5297                 first = false;
5298         }
5299
5300 remote_out:
5301         ptr++; /* skip leading / */
5302         memmove(fp->gf_path, ptr, fp->gf_path + fp->gf_pathlen - ptr);
5303
5304 out:
5305         RETURN(rc);
5306 }
5307
5308 /**
5309  * Given an MDT object, use mdt_path_current to get the path.
5310  * Essentially a wrapper to retry mdt_path_current a set number of times
5311  * if -EAGAIN is returned (usually because an object is being moved).
5312  *
5313  * Part of the MDT layer implementation of lfs fid2path.
5314  *
5315  * \param[in]     info  Per-thread common data shared by mdt level handlers.
5316  * \param[in]     obj   Object to do path lookup of
5317  * \param[in,out] fp    User-provided struct for arguments and to store path
5318  *                      information
5319  *
5320  * \retval 0 Lookup successful, path information stored in fp
5321  * \retval negative errno if there was a problem
5322  */
5323 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
5324                     struct getinfo_fid2path *fp)
5325 {
5326         struct mdt_device       *mdt = info->mti_mdt;
5327         int                     tries = 3;
5328         int                     rc = -EAGAIN;
5329         ENTRY;
5330
5331         if (fp->gf_pathlen < 3)
5332                 RETURN(-EOVERFLOW);
5333
5334         if (lu_fid_eq(&mdt->mdt_md_root_fid, mdt_object_fid(obj))) {
5335                 fp->gf_path[0] = '\0';
5336                 RETURN(0);
5337         }
5338
5339         /* Retry multiple times in case file is being moved */
5340         while (tries-- && rc == -EAGAIN)
5341                 rc = mdt_path_current(info, obj, fp);
5342
5343         RETURN(rc);
5344 }
5345
5346 /**
5347  * Get the full path of the provided FID, as of changelog record recno.
5348  *
5349  * This checks sanity and looks up object for user provided FID
5350  * before calling the actual path lookup code.
5351  *
5352  * Part of the MDT layer implementation of lfs fid2path.
5353  *
5354  * \param[in]     info  Per-thread common data shared by mdt level handlers.
5355  * \param[in,out] fp    User-provided struct for arguments and to store path
5356  *                      information
5357  *
5358  * \retval 0 Lookup successful, path information and recno stored in fp
5359  * \retval -ENOENT, object does not exist
5360  * \retval negative errno if there was a problem
5361  */
5362 static int mdt_fid2path(struct mdt_thread_info *info,
5363                         struct getinfo_fid2path *fp)
5364 {
5365         struct mdt_device *mdt = info->mti_mdt;
5366         struct mdt_object *obj;
5367         int    rc;
5368         ENTRY;
5369
5370         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
5371                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
5372
5373         if (!fid_is_sane(&fp->gf_fid))
5374                 RETURN(-EINVAL);
5375
5376         if (!fid_is_namespace_visible(&fp->gf_fid)) {
5377                 CWARN("%s: "DFID" is invalid, sequence should be "
5378                       ">= "LPX64"\n", mdt_obd_name(mdt),
5379                       PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
5380                 RETURN(-EINVAL);
5381         }
5382
5383         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
5384         if (IS_ERR(obj)) {
5385                 rc = PTR_ERR(obj);
5386                 CDEBUG(D_IOCTL, "cannot find "DFID": rc = %d\n",
5387                        PFID(&fp->gf_fid), rc);
5388                 RETURN(rc);
5389         }
5390
5391         if (mdt_object_remote(obj))
5392                 rc = -EREMOTE;
5393         else if (!mdt_object_exists(obj))
5394                 rc = -ENOENT;
5395         else
5396                 rc = 0;
5397
5398         if (rc < 0) {
5399                 mdt_object_put(info->mti_env, obj);
5400                 CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n",
5401                        PFID(&fp->gf_fid), rc);
5402                 RETURN(rc);
5403         }
5404
5405         rc = mdt_path(info, obj, fp);
5406
5407         CDEBUG(D_INFO, "fid "DFID", path %s recno "LPX64" linkno %u\n",
5408                PFID(&fp->gf_fid), fp->gf_path, fp->gf_recno, fp->gf_linkno);
5409
5410         mdt_object_put(info->mti_env, obj);
5411
5412         RETURN(rc);
5413 }
5414
5415 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key,
5416                             void *val, int vallen)
5417 {
5418         struct getinfo_fid2path *fpout, *fpin;
5419         int rc = 0;
5420
5421         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
5422         fpout = val;
5423
5424         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5425                 lustre_swab_fid2path(fpin);
5426
5427         memcpy(fpout, fpin, sizeof(*fpin));
5428         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
5429                 RETURN(-EINVAL);
5430
5431         rc = mdt_fid2path(info, fpout);
5432         RETURN(rc);
5433 }
5434
5435 int mdt_get_info(struct tgt_session_info *tsi)
5436 {
5437         char    *key;
5438         int      keylen;
5439         __u32   *vallen;
5440         void    *valout;
5441         int      rc;
5442
5443         ENTRY;
5444
5445         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
5446         if (key == NULL) {
5447                 CDEBUG(D_IOCTL, "No GETINFO key\n");
5448                 RETURN(err_serious(-EFAULT));
5449         }
5450         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
5451                                       RCL_CLIENT);
5452
5453         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
5454         if (vallen == NULL) {
5455                 CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n",
5456                                 tgt_name(tsi->tsi_tgt));
5457                 RETURN(err_serious(-EFAULT));
5458         }
5459
5460         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
5461                              *vallen);
5462         rc = req_capsule_server_pack(tsi->tsi_pill);
5463         if (rc)
5464                 RETURN(err_serious(rc));
5465
5466         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
5467         if (valout == NULL) {
5468                 CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n",
5469                                 tgt_name(tsi->tsi_tgt));
5470                 RETURN(err_serious(-EFAULT));
5471         }
5472
5473         if (KEY_IS(KEY_FID2PATH)) {
5474                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
5475
5476                 rc = mdt_rpc_fid2path(info, key, valout, *vallen);
5477                 mdt_thread_info_fini(info);
5478         } else {
5479                 rc = -EINVAL;
5480         }
5481         RETURN(rc);
5482 }
5483
5484 /* Pass the ioc down */
5485 static int mdt_ioc_child(struct lu_env *env, struct mdt_device *mdt,
5486                          unsigned int cmd, int len, void *data)
5487 {
5488         struct lu_context ioctl_session;
5489         struct md_device *next = mdt->mdt_child;
5490         int rc;
5491         ENTRY;
5492
5493         rc = lu_context_init(&ioctl_session, LCT_SERVER_SESSION);
5494         if (rc)
5495                 RETURN(rc);
5496         ioctl_session.lc_thread = (struct ptlrpc_thread *)current;
5497         lu_context_enter(&ioctl_session);
5498         env->le_ses = &ioctl_session;
5499
5500         LASSERT(next->md_ops->mdo_iocontrol);
5501         rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data);
5502
5503         lu_context_exit(&ioctl_session);
5504         lu_context_fini(&ioctl_session);
5505         RETURN(rc);
5506 }
5507
5508 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
5509 {
5510         struct obd_ioctl_data *data = karg;
5511         struct lu_fid *fid;
5512         __u64 version;
5513         struct mdt_object *obj;
5514         struct mdt_lock_handle  *lh;
5515         int rc;
5516         ENTRY;
5517
5518         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
5519             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
5520                 RETURN(-EINVAL);
5521
5522         fid = (struct lu_fid *)data->ioc_inlbuf1;
5523
5524         if (!fid_is_sane(fid))
5525                 RETURN(-EINVAL);
5526
5527         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
5528
5529         lh = &mti->mti_lh[MDT_LH_PARENT];
5530         mdt_lock_reg_init(lh, LCK_CR);
5531
5532         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
5533         if (IS_ERR(obj))
5534                 RETURN(PTR_ERR(obj));
5535
5536         if (mdt_object_remote(obj)) {
5537                 rc = -EREMOTE;
5538                 /**
5539                  * before calling version get the correct MDS should be
5540                  * fid, this is error to find remote object here
5541                  */
5542                 CERROR("nonlocal object "DFID"\n", PFID(fid));
5543         } else if (!mdt_object_exists(obj)) {
5544                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
5545                 rc = -ENOENT;
5546         } else {
5547                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
5548                *(__u64 *)data->ioc_inlbuf2 = version;
5549                 rc = 0;
5550         }
5551         mdt_object_unlock_put(mti, obj, lh, 1);
5552         RETURN(rc);
5553 }
5554
5555 /* ioctls on obd dev */
5556 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
5557                          void *karg, void *uarg)
5558 {
5559         struct lu_env      env;
5560         struct obd_device *obd = exp->exp_obd;
5561         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
5562         struct dt_device  *dt = mdt->mdt_bottom;
5563         int rc;
5564
5565         ENTRY;
5566         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
5567         rc = lu_env_init(&env, LCT_MD_THREAD);
5568         if (rc)
5569                 RETURN(rc);
5570
5571         switch (cmd) {
5572         case OBD_IOC_SYNC:
5573                 rc = mdt_device_sync(&env, mdt);
5574                 break;
5575         case OBD_IOC_SET_READONLY:
5576                 rc = dt->dd_ops->dt_ro(&env, dt);
5577                 break;
5578         case OBD_IOC_ABORT_RECOVERY:
5579                 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
5580                 target_stop_recovery_thread(obd);
5581                 rc = 0;
5582                 break;
5583         case OBD_IOC_CHANGELOG_REG:
5584         case OBD_IOC_CHANGELOG_DEREG:
5585         case OBD_IOC_CHANGELOG_CLEAR:
5586                 rc = mdt_ioc_child(&env, mdt, cmd, len, karg);
5587                 break;
5588         case OBD_IOC_START_LFSCK: {
5589                 struct md_device *next = mdt->mdt_child;
5590                 struct obd_ioctl_data *data = karg;
5591                 struct lfsck_start_param lsp;
5592
5593                 if (unlikely(data == NULL)) {
5594                         rc = -EINVAL;
5595                         break;
5596                 }
5597
5598                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
5599                 lsp.lsp_index_valid = 0;
5600                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
5601                 break;
5602         }
5603         case OBD_IOC_STOP_LFSCK: {
5604                 struct md_device        *next = mdt->mdt_child;
5605                 struct obd_ioctl_data   *data = karg;
5606                 struct lfsck_stop        stop;
5607
5608                 stop.ls_status = LS_STOPPED;
5609                 /* Old lfsck utils may pass NULL @stop. */
5610                 if (data->ioc_inlbuf1 == NULL)
5611                         stop.ls_flags = 0;
5612                 else
5613                         stop.ls_flags =
5614                         ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
5615
5616                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
5617                 break;
5618         }
5619         case OBD_IOC_GET_OBJ_VERSION: {
5620                 struct mdt_thread_info *mti;
5621                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5622                 memset(mti, 0, sizeof *mti);
5623                 mti->mti_env = &env;
5624                 mti->mti_mdt = mdt;
5625                 mti->mti_exp = exp;
5626
5627                 rc = mdt_ioc_version_get(mti, karg);
5628                 break;
5629         }
5630         case OBD_IOC_CATLOGLIST: {
5631                 struct mdt_thread_info *mti;
5632
5633                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5634                 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
5635                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
5636                                        &mti->mti_tmp_fid1);
5637                 break;
5638          }
5639         default:
5640                 rc = -EOPNOTSUPP;
5641                 CERROR("%s: Not supported cmd = %d, rc = %d\n",
5642                         mdt_obd_name(mdt), cmd, rc);
5643         }
5644
5645         lu_env_fini(&env);
5646         RETURN(rc);
5647 }
5648
5649 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
5650 {
5651         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
5652         struct lfsck_start_param lsp;
5653         int rc;
5654         ENTRY;
5655
5656         lsp.lsp_start = NULL;
5657         lsp.lsp_index_valid = 0;
5658         rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
5659                                                    OBD_IOC_START_LFSCK,
5660                                                    0, &lsp);
5661         if (rc != 0 && rc != -EALREADY)
5662                 CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
5663                       mdt_obd_name(mdt), rc);
5664
5665         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
5666         RETURN(rc);
5667 }
5668
5669 static int mdt_obd_postrecov(struct obd_device *obd)
5670 {
5671         struct lu_env env;
5672         int rc;
5673
5674         rc = lu_env_init(&env, LCT_MD_THREAD);
5675         if (rc)
5676                 RETURN(rc);
5677         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
5678         lu_env_fini(&env);
5679         return rc;
5680 }
5681
5682 static struct obd_ops mdt_obd_device_ops = {
5683         .o_owner          = THIS_MODULE,
5684         .o_set_info_async = mdt_obd_set_info_async,
5685         .o_connect        = mdt_obd_connect,
5686         .o_reconnect      = mdt_obd_reconnect,
5687         .o_disconnect     = mdt_obd_disconnect,
5688         .o_init_export    = mdt_init_export,
5689         .o_destroy_export = mdt_destroy_export,
5690         .o_iocontrol      = mdt_iocontrol,
5691         .o_postrecov      = mdt_obd_postrecov,
5692 };
5693
5694 static struct lu_device* mdt_device_fini(const struct lu_env *env,
5695                                          struct lu_device *d)
5696 {
5697         struct mdt_device *m = mdt_dev(d);
5698         ENTRY;
5699
5700         mdt_fini(env, m);
5701         RETURN(NULL);
5702 }
5703
5704 static struct lu_device *mdt_device_free(const struct lu_env *env,
5705                                          struct lu_device *d)
5706 {
5707         struct mdt_device *m = mdt_dev(d);
5708         ENTRY;
5709
5710         lu_device_fini(&m->mdt_lu_dev);
5711         OBD_FREE_PTR(m);
5712
5713         RETURN(NULL);
5714 }
5715
5716 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
5717                                           struct lu_device_type *t,
5718                                           struct lustre_cfg *cfg)
5719 {
5720         struct lu_device  *l;
5721         struct mdt_device *m;
5722
5723         OBD_ALLOC_PTR(m);
5724         if (m != NULL) {
5725                 int rc;
5726
5727                 l = &m->mdt_lu_dev;
5728                 rc = mdt_init0(env, m, t, cfg);
5729                 if (rc != 0) {
5730                         mdt_device_free(env, l);
5731                         l = ERR_PTR(rc);
5732                         return l;
5733                 }
5734         } else
5735                 l = ERR_PTR(-ENOMEM);
5736         return l;
5737 }
5738
5739 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
5740 LU_KEY_INIT(mdt, struct mdt_thread_info);
5741
5742 static void mdt_key_fini(const struct lu_context *ctx,
5743                          struct lu_context_key *key, void* data)
5744 {
5745         struct mdt_thread_info *info = data;
5746
5747         if (info->mti_big_lmm) {
5748                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
5749                 info->mti_big_lmm = NULL;
5750                 info->mti_big_lmmsize = 0;
5751         }
5752         OBD_FREE_PTR(info);
5753 }
5754
5755 /* context key: mdt_thread_key */
5756 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
5757
5758 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
5759 {
5760         return lu_ucred(info->mti_env);
5761 }
5762
5763 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
5764 {
5765         return lu_ucred_check(info->mti_env);
5766 }
5767
5768 /**
5769  * Enable/disable COS (Commit On Sharing).
5770  *
5771  * Set/Clear the COS flag in mdt options.
5772  *
5773  * \param mdt mdt device
5774  * \param val 0 disables COS, other values enable COS
5775  */
5776 void mdt_enable_cos(struct mdt_device *mdt, int val)
5777 {
5778         struct lu_env env;
5779         int rc;
5780
5781         mdt->mdt_opts.mo_cos = !!val;
5782         rc = lu_env_init(&env, LCT_LOCAL);
5783         if (unlikely(rc != 0)) {
5784                 CWARN("%s: lu_env initialization failed, cannot "
5785                       "sync: rc = %d\n", mdt_obd_name(mdt), rc);
5786                 return;
5787         }
5788         mdt_device_sync(&env, mdt);
5789         lu_env_fini(&env);
5790 }
5791
5792 /**
5793  * Check COS (Commit On Sharing) status.
5794  *
5795  * Return COS flag status.
5796  *
5797  * \param mdt mdt device
5798  */
5799 int mdt_cos_is_enabled(struct mdt_device *mdt)
5800 {
5801         return mdt->mdt_opts.mo_cos != 0;
5802 }
5803
5804 static struct lu_device_type_operations mdt_device_type_ops = {
5805         .ldto_device_alloc = mdt_device_alloc,
5806         .ldto_device_free  = mdt_device_free,
5807         .ldto_device_fini  = mdt_device_fini
5808 };
5809
5810 static struct lu_device_type mdt_device_type = {
5811         .ldt_tags     = LU_DEVICE_MD,
5812         .ldt_name     = LUSTRE_MDT_NAME,
5813         .ldt_ops      = &mdt_device_type_ops,
5814         .ldt_ctx_tags = LCT_MD_THREAD
5815 };
5816
5817 static int __init mdt_mod_init(void)
5818 {
5819         int rc;
5820
5821         CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") ==
5822                  FID_NOBRACE_LEN + 1);
5823         CLASSERT(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") ==
5824                  FID_LEN + 1);
5825         rc = lu_kmem_init(mdt_caches);
5826         if (rc)
5827                 return rc;
5828
5829         rc = mds_mod_init();
5830         if (rc)
5831                 GOTO(lu_fini, rc);
5832
5833         rc = class_register_type(&mdt_obd_device_ops, NULL, true, NULL,
5834                                  LUSTRE_MDT_NAME, &mdt_device_type);
5835         if (rc)
5836                 GOTO(mds_fini, rc);
5837 lu_fini:
5838         if (rc)
5839                 lu_kmem_fini(mdt_caches);
5840 mds_fini:
5841         if (rc)
5842                 mds_mod_exit();
5843         return rc;
5844 }
5845
5846 static void __exit mdt_mod_exit(void)
5847 {
5848         class_unregister_type(LUSTRE_MDT_NAME);
5849         mds_mod_exit();
5850         lu_kmem_fini(mdt_caches);
5851 }
5852
5853 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
5854 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
5855 MODULE_VERSION(LUSTRE_VERSION_STRING);
5856 MODULE_LICENSE("GPL");
5857
5858 module_init(mdt_mod_init);
5859 module_exit(mdt_mod_exit);