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, 2013, 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 /* struct ptlrpc_request */
  57 #include <lustre_net.h>
  58 /* struct obd_export */
  59 #include <lustre_export.h>
  60 /* struct obd_device */
  61 #include <obd.h>
  62 /* lu2dt_dev() */
  63 #include <dt_object.h>
  64 #include <lustre_mds.h>
  65 #include <lustre_log.h>
  66 #include "mdt_internal.h"
  67 #include <lustre_acl.h>
  68 #include <lustre_param.h>
  69 #include <lustre_quota.h>
  70 #include <lustre_lfsck.h>
  71
  72 mdl_mode_t mdt_mdl_lock_modes[] = {
  73         [LCK_MINMODE] = MDL_MINMODE,
  74         [LCK_EX]      = MDL_EX,
  75         [LCK_PW]      = MDL_PW,
  76         [LCK_PR]      = MDL_PR,
  77         [LCK_CW]      = MDL_CW,
  78         [LCK_CR]      = MDL_CR,
  79         [LCK_NL]      = MDL_NL,
  80         [LCK_GROUP]   = MDL_GROUP
  81 };
  82
  83 ldlm_mode_t mdt_dlm_lock_modes[] = {
  84         [MDL_MINMODE] = LCK_MINMODE,
  85         [MDL_EX]      = LCK_EX,
  86         [MDL_PW]      = LCK_PW,
  87         [MDL_PR]      = LCK_PR,
  88         [MDL_CW]      = LCK_CW,
  89         [MDL_CR]      = LCK_CR,
  90         [MDL_NL]      = LCK_NL,
  91         [MDL_GROUP]   = LCK_GROUP
  92 };
  93
  94 static struct mdt_device *mdt_dev(struct lu_device *d);
  95 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
  96
  97 static const struct lu_object_operations mdt_obj_ops;
  98
  99 /* Slab for MDT object allocation */
 100 static struct kmem_cache *mdt_object_kmem;
 101
 102 /* For HSM restore handles */
 103 struct kmem_cache *mdt_hsm_cdt_kmem;
 104
 105 /* For HSM request handles */
 106 struct kmem_cache *mdt_hsm_car_kmem;
 107
 108 static struct lu_kmem_descr mdt_caches[] = {
 109         {
 110                 .ckd_cache = &mdt_object_kmem,
 111                 .ckd_name  = "mdt_obj",
 112                 .ckd_size  = sizeof(struct mdt_object)
 113         },
 114         {
 115                 .ckd_cache      = &mdt_hsm_cdt_kmem,
 116                 .ckd_name       = "mdt_cdt_restore_handle",
 117                 .ckd_size       = sizeof(struct cdt_restore_handle)
 118         },
 119         {
 120                 .ckd_cache      = &mdt_hsm_car_kmem,
 121                 .ckd_name       = "mdt_cdt_agent_req",
 122                 .ckd_size       = sizeof(struct cdt_agent_req)
 123         },
 124         {
 125                 .ckd_cache = NULL
 126         }
 127 };
 128
 129 int mdt_get_disposition(struct ldlm_reply *rep, int flag)
 130 {
 131         if (!rep)
 132                 return 0;
 133         return (rep->lock_policy_res1 & flag);
 134 }
 135
 136 void mdt_clear_disposition(struct mdt_thread_info *info,
 137                            struct ldlm_reply *rep, int flag)
 138 {
 139         if (info) {
 140                 info->mti_opdata &= ~flag;
 141                 tgt_opdata_clear(info->mti_env, flag);
 142         }
 143         if (rep)
 144                 rep->lock_policy_res1 &= ~flag;
 145 }
 146
 147 void mdt_set_disposition(struct mdt_thread_info *info,
 148                          struct ldlm_reply *rep, int flag)
 149 {
 150         if (info) {
 151                 info->mti_opdata |= flag;
 152                 tgt_opdata_set(info->mti_env, flag);
 153         }
 154         if (rep)
 155                 rep->lock_policy_res1 |= flag;
 156 }
 157
 158 void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
 159 {
 160         lh->mlh_pdo_hash = 0;
 161         lh->mlh_reg_mode = lm;
 162         lh->mlh_rreg_mode = lm;
 163         lh->mlh_type = MDT_REG_LOCK;
 164 }
 165
 166 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lock_mode,
 167                        const struct lu_name *lname)
 168 {
 169         lh->mlh_reg_mode = lock_mode;
 170         lh->mlh_rreg_mode = lock_mode;
 171         lh->mlh_type = MDT_PDO_LOCK;
 172
 173         if (lu_name_is_valid(lname)) {
 174                 lh->mlh_pdo_hash = full_name_hash(lname->ln_name,
 175                                                   lname->ln_namelen);
 176                 /* XXX Workaround for LU-2856
 177                  *
 178                  * Zero is a valid return value of full_name_hash, but
 179                  * several users of mlh_pdo_hash assume a non-zero
 180                  * hash value. We therefore map zero onto an
 181                  * arbitrary, but consistent value (1) to avoid
 182                  * problems further down the road. */
 183                 if (unlikely(lh->mlh_pdo_hash == 0))
 184                         lh->mlh_pdo_hash = 1;
 185         } else {
 186                 lh->mlh_pdo_hash = 0;
 187         }
 188 }
 189
 190 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 191                               struct mdt_lock_handle *lh)
 192 {
 193         mdl_mode_t mode;
 194         ENTRY;
 195
 196         /*
 197          * Any dir access needs couple of locks:
 198          *
 199          * 1) on part of dir we gonna take lookup/modify;
 200          *
 201          * 2) on whole dir to protect it from concurrent splitting and/or to
 202          * flush client's cache for readdir().
 203          *
 204          * so, for a given mode and object this routine decides what lock mode
 205          * to use for lock #2:
 206          *
 207          * 1) if caller's gonna lookup in dir then we need to protect dir from
 208          * being splitted only - LCK_CR
 209          *
 210          * 2) if caller's gonna modify dir then we need to protect dir from
 211          * being splitted and to flush cache - LCK_CW
 212          *
 213          * 3) if caller's gonna modify dir and that dir seems ready for
 214          * splitting then we need to protect it from any type of access
 215          * (lookup/modify/split) - LCK_EX --bzzz
 216          */
 217
 218         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 219         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 220
 221         /*
 222          * Ask underlaying level its opinion about preferable PDO lock mode
 223          * having access type passed as regular lock mode:
 224          *
 225          * - MDL_MINMODE means that lower layer does not want to specify lock
 226          * mode;
 227          *
 228          * - MDL_NL means that no PDO lock should be taken. This is used in some
 229          * cases. Say, for non-splittable directories no need to use PDO locks
 230          * at all.
 231          */
 232         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 233                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 234
 235         if (mode != MDL_MINMODE) {
 236                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 237         } else {
 238                 /*
 239                  * Lower layer does not want to specify locking mode. We do it
 240                  * our selves. No special protection is needed, just flush
 241                  * client's cache on modification and allow concurrent
 242                  * mondification.
 243                  */
 244                 switch (lh->mlh_reg_mode) {
 245                 case LCK_EX:
 246                         lh->mlh_pdo_mode = LCK_EX;
 247                         break;
 248                 case LCK_PR:
 249                         lh->mlh_pdo_mode = LCK_CR;
 250                         break;
 251                 case LCK_PW:
 252                         lh->mlh_pdo_mode = LCK_CW;
 253                         break;
 254                 default:
 255                         CERROR("Not expected lock type (0x%x)\n",
 256                                (int)lh->mlh_reg_mode);
 257                         LBUG();
 258                 }
 259         }
 260
 261         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 262         EXIT;
 263 }
 264
 265 int mdt_getstatus(struct tgt_session_info *tsi)
 266 {
 267         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
 268         struct mdt_device       *mdt = info->mti_mdt;
 269         struct mdt_body         *repbody;
 270         int                      rc;
 271
 272         ENTRY;
 273
 274         rc = mdt_check_ucred(info);
 275         if (rc)
 276                 GOTO(out, rc = err_serious(rc));
 277
 278         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
 279                 GOTO(out, rc = err_serious(-ENOMEM));
 280
 281         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 282         repbody->fid1 = mdt->mdt_md_root_fid;
 283         repbody->valid |= OBD_MD_FLID;
 284
 285         if (tsi->tsi_tgt->lut_mds_capa &&
 286             exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) {
 287                 struct mdt_object       *root;
 288                 struct lustre_capa      *capa;
 289
 290                 root = mdt_object_find(info->mti_env, mdt, &repbody->fid1);
 291                 if (IS_ERR(root))
 292                         GOTO(out, rc = PTR_ERR(root));
 293
 294                 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
 295                 LASSERT(capa);
 296                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 297                 rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa,
 298                                  0);
 299                 mdt_object_put(info->mti_env, root);
 300                 if (rc == 0)
 301                         repbody->valid |= OBD_MD_FLMDSCAPA;
 302         }
 303         EXIT;
 304 out:
 305         mdt_thread_info_fini(info);
 306         return rc;
 307 }
 308
 309 int mdt_statfs(struct tgt_session_info *tsi)
 310 {
 311         struct ptlrpc_request           *req = tgt_ses_req(tsi);
 312         struct mdt_thread_info          *info = tsi2mdt_info(tsi);
 313         struct md_device                *next = info->mti_mdt->mdt_child;
 314         struct ptlrpc_service_part      *svcpt;
 315         struct obd_statfs               *osfs;
 316         int                             rc;
 317
 318         ENTRY;
 319
 320         svcpt = req->rq_rqbd->rqbd_svcpt;
 321
 322         /* This will trigger a watchdog timeout */
 323         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 324                          (MDT_SERVICE_WATCHDOG_FACTOR *
 325                           at_get(&svcpt->scp_at_estimate)) + 1);
 326
 327         rc = mdt_check_ucred(info);
 328         if (rc)
 329                 GOTO(out, rc = err_serious(rc));
 330
 331         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
 332                 GOTO(out, rc = err_serious(-ENOMEM));
 333
 334         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 335         if (!osfs)
 336                 GOTO(out, rc = -EPROTO);
 337
 338         /** statfs information are cached in the mdt_device */
 339         if (cfs_time_before_64(info->mti_mdt->mdt_osfs_age,
 340                                cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS))) {
 341                 /** statfs data is too old, get up-to-date one */
 342                 rc = next->md_ops->mdo_statfs(info->mti_env, next, osfs);
 343                 if (rc)
 344                         GOTO(out, rc);
 345                 spin_lock(&info->mti_mdt->mdt_osfs_lock);
 346                 info->mti_mdt->mdt_osfs = *osfs;
 347                 info->mti_mdt->mdt_osfs_age = cfs_time_current_64();
 348                 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
 349         } else {
 350                 /** use cached statfs data */
 351                 spin_lock(&info->mti_mdt->mdt_osfs_lock);
 352                 *osfs = info->mti_mdt->mdt_osfs;
 353                 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
 354         }
 355
 356         if (rc == 0)
 357                 mdt_counter_incr(req, LPROC_MDT_STATFS);
 358 out:
 359         mdt_thread_info_fini(info);
 360         RETURN(rc);
 361 }
 362
 363 /**
 364  * Pack SOM attributes into the reply.
 365  * Call under a DLM UPDATE lock.
 366  */
 367 static void mdt_pack_size2body(struct mdt_thread_info *info,
 368                                struct mdt_object *mo)
 369 {
 370         struct mdt_body *b;
 371         struct md_attr *ma = &info->mti_attr;
 372
 373         LASSERT(ma->ma_attr.la_valid & LA_MODE);
 374         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 375
 376         /* Check if Size-on-MDS is supported, if this is a regular file,
 377          * if SOM is enabled on the object and if SOM cache exists and valid.
 378          * Otherwise do not pack Size-on-MDS attributes to the reply. */
 379         if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) ||
 380             !S_ISREG(ma->ma_attr.la_mode) ||
 381             !mdt_object_is_som_enabled(mo) ||
 382             !(ma->ma_valid & MA_SOM))
 383                 return;
 384
 385         b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 386         b->size = ma->ma_som->msd_size;
 387         b->blocks = ma->ma_som->msd_blocks;
 388 }
 389
 390 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 391                         const struct lu_attr *attr, const struct lu_fid *fid)
 392 {
 393         struct md_attr *ma = &info->mti_attr;
 394
 395         LASSERT(ma->ma_valid & MA_INODE);
 396
 397         b->atime      = attr->la_atime;
 398         b->mtime      = attr->la_mtime;
 399         b->ctime      = attr->la_ctime;
 400         b->mode       = attr->la_mode;
 401         b->size       = attr->la_size;
 402         b->blocks     = attr->la_blocks;
 403         b->uid        = attr->la_uid;
 404         b->gid        = attr->la_gid;
 405         b->flags      = attr->la_flags;
 406         b->nlink      = attr->la_nlink;
 407         b->rdev       = attr->la_rdev;
 408
 409         /*XXX should pack the reply body according to lu_valid*/
 410         b->valid |= OBD_MD_FLCTIME | OBD_MD_FLUID   |
 411                     OBD_MD_FLGID   | OBD_MD_FLTYPE  |
 412                     OBD_MD_FLMODE  | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
 413                     OBD_MD_FLATIME | OBD_MD_FLMTIME ;
 414
 415         if (!S_ISREG(attr->la_mode)) {
 416                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 417         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
 418                 /* means no objects are allocated on osts. */
 419                 LASSERT(!(ma->ma_valid & MA_LOV));
 420                 /* just ignore blocks occupied by extend attributes on MDS */
 421                 b->blocks = 0;
 422                 /* if no object is allocated on osts, the size on mds is valid. b=22272 */
 423                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 424         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL &&
 425                    ma->ma_lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) {
 426                 /* A released file stores its size on MDS. */
 427                 /* But return 1 block for released file, unless tools like tar
 428                  * will consider it fully sparse. (LU-3864)
 429                  */
 430                 if (unlikely(b->size == 0))
 431                         b->blocks = 0;
 432                 else
 433                         b->blocks = 1;
 434                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 435         }
 436
 437         if (fid) {
 438                 b->fid1 = *fid;
 439                 b->valid |= OBD_MD_FLID;
 440                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n",
 441                                 PFID(fid), b->nlink, b->mode, b->size);
 442         }
 443
 444         if (info)
 445                 mdt_body_reverse_idmap(info, b);
 446
 447         if (fid != NULL && (b->valid & OBD_MD_FLSIZE))
 448                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 449                        PFID(fid), (unsigned long long)b->size);
 450 }
 451
 452 static inline int mdt_body_has_lov(const struct lu_attr *la,
 453                                    const struct mdt_body *body)
 454 {
 455         return ((S_ISREG(la->la_mode) && (body->valid & OBD_MD_FLEASIZE)) ||
 456                 (S_ISDIR(la->la_mode) && (body->valid & OBD_MD_FLDIREA )) );
 457 }
 458
 459 void mdt_client_compatibility(struct mdt_thread_info *info)
 460 {
 461         struct mdt_body       *body;
 462         struct ptlrpc_request *req = mdt_info_req(info);
 463         struct obd_export     *exp = req->rq_export;
 464         struct md_attr        *ma = &info->mti_attr;
 465         struct lu_attr        *la = &ma->ma_attr;
 466         ENTRY;
 467
 468         if (exp_connect_layout(exp))
 469                 /* the client can deal with 16-bit lmm_stripe_count */
 470                 RETURN_EXIT;
 471
 472         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 473
 474         if (!mdt_body_has_lov(la, body))
 475                 RETURN_EXIT;
 476
 477         /* now we have a reply with a lov for a client not compatible with the
 478          * layout lock so we have to clean the layout generation number */
 479         if (S_ISREG(la->la_mode))
 480                 ma->ma_lmm->lmm_layout_gen = 0;
 481         EXIT;
 482 }
 483
 484 static int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
 485                              const char *name)
 486 {
 487         const struct lu_env *env = info->mti_env;
 488         int rc;
 489         ENTRY;
 490
 491         LASSERT(info->mti_big_lmm_used == 0);
 492         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
 493         if (rc < 0)
 494                 RETURN(rc);
 495
 496         /* big_lmm may need to be grown */
 497         if (info->mti_big_lmmsize < rc) {
 498                 int size = size_roundup_power2(rc);
 499
 500                 if (info->mti_big_lmmsize > 0) {
 501                         /* free old buffer */
 502                         LASSERT(info->mti_big_lmm);
 503                         OBD_FREE_LARGE(info->mti_big_lmm,
 504                                        info->mti_big_lmmsize);
 505                         info->mti_big_lmm = NULL;
 506                         info->mti_big_lmmsize = 0;
 507                 }
 508
 509                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
 510                 if (info->mti_big_lmm == NULL)
 511                         RETURN(-ENOMEM);
 512                 info->mti_big_lmmsize = size;
 513         }
 514         LASSERT(info->mti_big_lmmsize >= rc);
 515
 516         info->mti_buf.lb_buf = info->mti_big_lmm;
 517         info->mti_buf.lb_len = info->mti_big_lmmsize;
 518         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
 519
 520         RETURN(rc);
 521 }
 522
 523 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
 524                    struct md_attr *ma, const char *name)
 525 {
 526         struct md_object *next = mdt_object_child(o);
 527         struct lu_buf    *buf = &info->mti_buf;
 528         int rc;
 529
 530         if (strcmp(name, XATTR_NAME_LOV) == 0) {
 531                 buf->lb_buf = ma->ma_lmm;
 532                 buf->lb_len = ma->ma_lmm_size;
 533                 LASSERT(!(ma->ma_valid & MA_LOV));
 534         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 535                 buf->lb_buf = ma->ma_lmv;
 536                 buf->lb_len = ma->ma_lmv_size;
 537                 LASSERT(!(ma->ma_valid & MA_LMV));
 538         } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 539                 buf->lb_buf = ma->ma_lmv;
 540                 buf->lb_len = ma->ma_lmv_size;
 541                 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
 542         } else {
 543                 return -EINVAL;
 544         }
 545
 546         rc = mo_xattr_get(info->mti_env, next, buf, name);
 547         if (rc > 0) {
 548                 if (strcmp(name, XATTR_NAME_LOV) == 0) {
 549                         ma->ma_lmm_size = rc;
 550                         ma->ma_valid |= MA_LOV;
 551                 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 552                         ma->ma_lmv_size = rc;
 553                         ma->ma_valid |= MA_LMV;
 554                 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 555                         ma->ma_lmv_size = rc;
 556                         ma->ma_valid |= MA_LMV_DEF;
 557                 }
 558
 559                 rc = 0;
 560         } else if (rc == -ENODATA) {
 561                 /* no LOV EA */
 562                 rc = 0;
 563         } else if (rc == -ERANGE) {
 564                 /* Default LMV has fixed size, so it must be able to fit
 565                  * in the original buffer */
 566                 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
 567                         return rc;
 568                 rc = mdt_big_xattr_get(info, o, name);
 569                 if (rc > 0) {
 570                         info->mti_big_lmm_used = 1;
 571                         if (!strcmp(name, XATTR_NAME_LOV)) {
 572                                 ma->ma_valid |= MA_LOV;
 573                                 ma->ma_lmm = info->mti_big_lmm;
 574                                 ma->ma_lmm_size = rc;
 575                         } else if (!strcmp(name, XATTR_NAME_LMV)) {
 576                                 ma->ma_valid |= MA_LMV;
 577                                 ma->ma_lmv = info->mti_big_lmm;
 578                                 ma->ma_lmv_size = rc;
 579                         } else {
 580                                 return -EINVAL;
 581                         }
 582
 583                         /* update mdt_max_mdsize so all clients
 584                          * will be aware about that */
 585                         if (info->mti_mdt->mdt_max_mdsize < rc)
 586                                 info->mti_mdt->mdt_max_mdsize = rc;
 587                         rc = 0;
 588                 }
 589         }
 590
 591         return rc;
 592 }
 593
 594 int mdt_attr_get_pfid(struct mdt_thread_info *info,
 595                       struct mdt_object *o, struct lu_fid *pfid)
 596 {
 597         struct lu_buf           *buf = &info->mti_buf;
 598         struct link_ea_header   *leh;
 599         struct link_ea_entry    *lee;
 600         int                      rc;
 601         ENTRY;
 602
 603         buf->lb_buf = info->mti_big_lmm;
 604         buf->lb_len = info->mti_big_lmmsize;
 605         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
 606                           buf, XATTR_NAME_LINK);
 607         /* ignore errors, MA_PFID won't be set and it is
 608          * up to the caller to treat this as an error */
 609         if (rc == -ERANGE || buf->lb_len == 0) {
 610                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
 611                 buf->lb_buf = info->mti_big_lmm;
 612                 buf->lb_len = info->mti_big_lmmsize;
 613         }
 614
 615         if (rc < 0)
 616                 RETURN(rc);
 617         if (rc < sizeof(*leh)) {
 618                 CERROR("short LinkEA on "DFID": rc = %d\n",
 619                        PFID(mdt_object_fid(o)), rc);
 620                 RETURN(-ENODATA);
 621         }
 622
 623         leh = (struct link_ea_header *) buf->lb_buf;
 624         lee = (struct link_ea_entry *)(leh + 1);
 625         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
 626                 leh->leh_magic = LINK_EA_MAGIC;
 627                 leh->leh_reccount = __swab32(leh->leh_reccount);
 628                 leh->leh_len = __swab64(leh->leh_len);
 629         }
 630         if (leh->leh_magic != LINK_EA_MAGIC)
 631                 RETURN(-EINVAL);
 632         if (leh->leh_reccount == 0)
 633                 RETURN(-ENODATA);
 634
 635         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
 636         fid_be_to_cpu(pfid, pfid);
 637
 638         RETURN(0);
 639 }
 640
 641 int mdt_attr_get_complex(struct mdt_thread_info *info,
 642                          struct mdt_object *o, struct md_attr *ma)
 643 {
 644         const struct lu_env *env = info->mti_env;
 645         struct md_object    *next = mdt_object_child(o);
 646         struct lu_buf       *buf = &info->mti_buf;
 647         u32                  mode = lu_object_attr(&next->mo_lu);
 648         int                  need = ma->ma_need;
 649         int                  rc = 0, rc2;
 650         ENTRY;
 651
 652         ma->ma_valid = 0;
 653
 654         if (need & MA_INODE) {
 655                 ma->ma_need = MA_INODE;
 656                 rc = mo_attr_get(env, next, ma);
 657                 if (rc)
 658                         GOTO(out, rc);
 659                 ma->ma_valid |= MA_INODE;
 660         }
 661
 662         if (need & MA_PFID) {
 663                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
 664                 if (rc == 0)
 665                         ma->ma_valid |= MA_PFID;
 666                 /* ignore this error, parent fid is not mandatory */
 667                 rc = 0;
 668         }
 669
 670         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
 671                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
 672                 if (rc)
 673                         GOTO(out, rc);
 674         }
 675
 676         if (need & MA_LMV && S_ISDIR(mode)) {
 677                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
 678                 if (rc != 0)
 679                         GOTO(out, rc);
 680         }
 681
 682         if (need & MA_LMV_DEF && S_ISDIR(mode)) {
 683                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
 684                 if (rc != 0)
 685                         GOTO(out, rc);
 686         }
 687
 688         if (need & MA_SOM && S_ISREG(mode)) {
 689                 buf->lb_buf = info->mti_xattr_buf;
 690                 buf->lb_len = sizeof(info->mti_xattr_buf);
 691                 CLASSERT(sizeof(struct som_attrs) <=
 692                          sizeof(info->mti_xattr_buf));
 693                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_SOM);
 694                 rc2 = lustre_buf2som(info->mti_xattr_buf, rc2, ma->ma_som);
 695                 if (rc2 == 0)
 696                         ma->ma_valid |= MA_SOM;
 697                 else if (rc2 < 0 && rc2 != -ENODATA)
 698                         GOTO(out, rc = rc2);
 699         }
 700
 701         if (need & MA_HSM && S_ISREG(mode)) {
 702                 buf->lb_buf = info->mti_xattr_buf;
 703                 buf->lb_len = sizeof(info->mti_xattr_buf);
 704                 CLASSERT(sizeof(struct hsm_attrs) <=
 705                          sizeof(info->mti_xattr_buf));
 706                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
 707                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
 708                 if (rc2 == 0)
 709                         ma->ma_valid |= MA_HSM;
 710                 else if (rc2 < 0 && rc2 != -ENODATA)
 711                         GOTO(out, rc = rc2);
 712         }
 713
 714 #ifdef CONFIG_FS_POSIX_ACL
 715         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
 716                 buf->lb_buf = ma->ma_acl;
 717                 buf->lb_len = ma->ma_acl_size;
 718                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
 719                 if (rc2 > 0) {
 720                         ma->ma_acl_size = rc2;
 721                         ma->ma_valid |= MA_ACL_DEF;
 722                 } else if (rc2 == -ENODATA) {
 723                         /* no ACLs */
 724                         ma->ma_acl_size = 0;
 725                 } else
 726                         GOTO(out, rc = rc2);
 727         }
 728 #endif
 729 out:
 730         ma->ma_need = need;
 731         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
 732                rc, ma->ma_valid, ma->ma_lmm);
 733         RETURN(rc);
 734 }
 735
 736 static int mdt_getattr_internal(struct mdt_thread_info *info,
 737                                 struct mdt_object *o, int ma_need)
 738 {
 739         struct md_object        *next = mdt_object_child(o);
 740         const struct mdt_body   *reqbody = info->mti_body;
 741         struct ptlrpc_request   *req = mdt_info_req(info);
 742         struct md_attr          *ma = &info->mti_attr;
 743         struct lu_attr          *la = &ma->ma_attr;
 744         struct req_capsule      *pill = info->mti_pill;
 745         const struct lu_env     *env = info->mti_env;
 746         struct mdt_body         *repbody;
 747         struct lu_buf           *buffer = &info->mti_buf;
 748         int                     rc;
 749         int                     is_root;
 750         ENTRY;
 751
 752         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
 753                 RETURN(err_serious(-ENOMEM));
 754
 755         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 756
 757         ma->ma_valid = 0;
 758
 759         if (mdt_object_remote(o)) {
 760                 /* This object is located on remote node.*/
 761                 /* Return -EIO for old client */
 762                 if (!mdt_is_dne_client(req->rq_export))
 763                         GOTO(out, rc = -EIO);
 764
 765                 repbody->fid1 = *mdt_object_fid(o);
 766                 repbody->valid = OBD_MD_FLID | OBD_MD_MDS;
 767                 GOTO(out, rc = 0);
 768         }
 769
 770         buffer->lb_len = reqbody->eadatasize;
 771         if (buffer->lb_len > 0) {
 772                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
 773                 if (buffer->lb_buf == NULL)
 774                         GOTO(out, rc = -EPROTO);
 775         } else {
 776                 buffer->lb_buf = NULL;
 777                 ma_need &= ~(MA_LOV | MA_LMV);
 778                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
 779                        mdt_obd_name(info->mti_mdt),
 780                        req->rq_export->exp_client_uuid.uuid);
 781         }
 782
 783         /* If it is dir object and client require MEA, then we got MEA */
 784         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 785             (reqbody->valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
 786                 /* Assumption: MDT_MD size is enough for lmv size. */
 787                 ma->ma_lmv = buffer->lb_buf;
 788                 ma->ma_lmv_size = buffer->lb_len;
 789                 ma->ma_need = MA_INODE;
 790                 if (ma->ma_lmv_size > 0) {
 791                         if (reqbody->valid & OBD_MD_MEA)
 792                                 ma->ma_need |= MA_LMV;
 793                         else if (reqbody->valid & OBD_MD_DEFAULT_MEA)
 794                                 ma->ma_need |= MA_LMV_DEF;
 795                 }
 796         } else {
 797                 ma->ma_lmm = buffer->lb_buf;
 798                 ma->ma_lmm_size = buffer->lb_len;
 799                 ma->ma_need = MA_INODE | MA_HSM;
 800                 if (ma->ma_lmm_size > 0)
 801                         ma->ma_need |= MA_LOV;
 802         }
 803
 804         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 805             reqbody->valid & OBD_MD_FLDIREA  &&
 806             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
 807                 /* get default stripe info for this dir. */
 808                 ma->ma_need |= MA_LOV_DEF;
 809         }
 810         ma->ma_need |= ma_need;
 811         if (ma->ma_need & MA_SOM)
 812                 ma->ma_som = &info->mti_u.som.data;
 813
 814         rc = mdt_attr_get_complex(info, o, ma);
 815         if (unlikely(rc)) {
 816                 CERROR("%s: getattr error for "DFID": rc = %d\n",
 817                        mdt_obd_name(info->mti_mdt),
 818                        PFID(mdt_object_fid(o)), rc);
 819                 RETURN(rc);
 820         }
 821
 822         /* if file is released, check if a restore is running */
 823         if ((ma->ma_valid & MA_HSM) && (ma->ma_hsm.mh_flags & HS_RELEASED) &&
 824             mdt_hsm_restore_is_running(info, mdt_object_fid(o))) {
 825                 repbody->t_state = MS_RESTORE;
 826                 repbody->valid |= OBD_MD_TSTATE;
 827         }
 828
 829         is_root = lu_fid_eq(mdt_object_fid(o), &info->mti_mdt->mdt_md_root_fid);
 830
 831         /* the Lustre protocol supposes to return default striping
 832          * on the user-visible root if explicitly requested */
 833         if ((ma->ma_valid & MA_LOV) == 0 && S_ISDIR(la->la_mode) &&
 834             (ma->ma_need & MA_LOV_DEF && is_root) && ma->ma_need & MA_LOV) {
 835                 struct lu_fid      rootfid;
 836                 struct mdt_object *root;
 837                 struct mdt_device *mdt = info->mti_mdt;
 838
 839                 rc = dt_root_get(env, mdt->mdt_bottom, &rootfid);
 840                 if (rc)
 841                         RETURN(rc);
 842                 root = mdt_object_find(env, mdt, &rootfid);
 843                 if (IS_ERR(root))
 844                         RETURN(PTR_ERR(root));
 845                 rc = mdt_stripe_get(info, root, ma, XATTR_NAME_LOV);
 846                 mdt_object_put(info->mti_env, root);
 847                 if (unlikely(rc)) {
 848                         CERROR("%s: getattr error for "DFID": rc = %d\n",
 849                                mdt_obd_name(info->mti_mdt),
 850                                PFID(mdt_object_fid(o)), rc);
 851                         RETURN(rc);
 852                 }
 853         }
 854
 855         if (likely(ma->ma_valid & MA_INODE))
 856                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
 857         else
 858                 RETURN(-EFAULT);
 859
 860         if (mdt_body_has_lov(la, reqbody)) {
 861                 if (ma->ma_valid & MA_LOV) {
 862                         LASSERT(ma->ma_lmm_size);
 863                         mdt_dump_lmm(D_INFO, ma->ma_lmm);
 864                         repbody->eadatasize = ma->ma_lmm_size;
 865                         if (S_ISDIR(la->la_mode))
 866                                 repbody->valid |= OBD_MD_FLDIREA;
 867                         else
 868                                 repbody->valid |= OBD_MD_FLEASIZE;
 869                 }
 870                 if (ma->ma_valid & MA_LMV) {
 871                         LASSERT(S_ISDIR(la->la_mode));
 872                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
 873                         repbody->eadatasize = ma->ma_lmv_size;
 874                         repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
 875                 }
 876                 if (ma->ma_valid & MA_LMV_DEF) {
 877                         LASSERT(S_ISDIR(la->la_mode));
 878                         repbody->eadatasize = ma->ma_lmv_size;
 879                         repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_DEFAULT_MEA);
 880                 }
 881         } else if (S_ISLNK(la->la_mode) &&
 882                    reqbody->valid & OBD_MD_LINKNAME) {
 883                 buffer->lb_buf = ma->ma_lmm;
 884                 /* eadatasize from client includes NULL-terminator, so
 885                  * there is no need to read it */
 886                 buffer->lb_len = reqbody->eadatasize - 1;
 887                 rc = mo_readlink(env, next, buffer);
 888                 if (unlikely(rc <= 0)) {
 889                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
 890                                mdt_obd_name(info->mti_mdt),
 891                                PFID(mdt_object_fid(o)), rc);
 892                         rc = -EFAULT;
 893                 } else {
 894                         int print_limit = min_t(int, PAGE_CACHE_SIZE - 128, rc);
 895
 896                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
 897                                 rc -= 2;
 898                         repbody->valid |= OBD_MD_LINKNAME;
 899                         /* we need to report back size with NULL-terminator
 900                          * because client expects that */
 901                         repbody->eadatasize = rc + 1;
 902                         if (repbody->eadatasize != reqbody->eadatasize)
 903                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
 904                                        "on "DFID ", expected %d\n",
 905                                        mdt_obd_name(info->mti_mdt),
 906                                        rc, PFID(mdt_object_fid(o)),
 907                                        reqbody->eadatasize - 1);
 908                         /* NULL terminate */
 909                         ((char *)ma->ma_lmm)[rc] = 0;
 910
 911                         /* If the total CDEBUG() size is larger than a page, it
 912                          * will print a warning to the console, avoid this by
 913                          * printing just the last part of the symlink. */
 914                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
 915                                print_limit < rc ? "..." : "", print_limit,
 916                                (char *)ma->ma_lmm + rc - print_limit, rc);
 917                         rc = 0;
 918                 }
 919         }
 920
 921         if (reqbody->valid & OBD_MD_FLMODEASIZE) {
 922                 repbody->max_cookiesize = 0;
 923                 repbody->max_mdsize = info->mti_mdt->mdt_max_mdsize;
 924                 repbody->valid |= OBD_MD_FLMODEASIZE;
 925                 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
 926                        "MAX_COOKIE to : %d:%d\n", repbody->max_mdsize,
 927                        repbody->max_cookiesize);
 928         }
 929
 930         if (exp_connect_rmtclient(info->mti_exp) &&
 931             reqbody->valid & OBD_MD_FLRMTPERM) {
 932                 void *buf = req_capsule_server_get(pill, &RMF_ACL);
 933
 934                 /* mdt_getattr_lock only */
 935                 rc = mdt_pack_remote_perm(info, o, buf);
 936                 if (rc) {
 937                         repbody->valid &= ~OBD_MD_FLRMTPERM;
 938                         repbody->aclsize = 0;
 939                         RETURN(rc);
 940                 } else {
 941                         repbody->valid |= OBD_MD_FLRMTPERM;
 942                         repbody->aclsize = sizeof(struct mdt_remote_perm);
 943                 }
 944         }
 945 #ifdef CONFIG_FS_POSIX_ACL
 946         else if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
 947                  (reqbody->valid & OBD_MD_FLACL)) {
 948                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
 949                 buffer->lb_len = req_capsule_get_size(pill,
 950                                                       &RMF_ACL, RCL_SERVER);
 951                 if (buffer->lb_len > 0) {
 952                         rc = mo_xattr_get(env, next, buffer,
 953                                           XATTR_NAME_ACL_ACCESS);
 954                         if (rc < 0) {
 955                                 if (rc == -ENODATA) {
 956                                         repbody->aclsize = 0;
 957                                         repbody->valid |= OBD_MD_FLACL;
 958                                         rc = 0;
 959                                 } else if (rc == -EOPNOTSUPP) {
 960                                         rc = 0;
 961                                 } else {
 962                                         CERROR("%s: unable to read "DFID
 963                                                " ACL: rc = %d\n",
 964                                                mdt_obd_name(info->mti_mdt),
 965                                                PFID(mdt_object_fid(o)), rc);
 966                                 }
 967                         } else {
 968                                 repbody->aclsize = rc;
 969                                 repbody->valid |= OBD_MD_FLACL;
 970                                 rc = 0;
 971                         }
 972                 }
 973         }
 974 #endif
 975
 976         if (reqbody->valid & OBD_MD_FLMDSCAPA &&
 977             info->mti_mdt->mdt_lut.lut_mds_capa &&
 978             exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) {
 979                 struct lustre_capa *capa;
 980
 981                 capa = req_capsule_server_get(pill, &RMF_CAPA1);
 982                 LASSERT(capa);
 983                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 984                 rc = mo_capa_get(env, next, capa, 0);
 985                 if (rc)
 986                         RETURN(rc);
 987                 repbody->valid |= OBD_MD_FLMDSCAPA;
 988         }
 989
 990 out:
 991         if (rc == 0)
 992                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
 993
 994         RETURN(rc);
 995 }
 996
 997 static int mdt_renew_capa(struct mdt_thread_info *info)
 998 {
 999         struct mdt_object  *obj = info->mti_object;
1000         struct mdt_body    *body;
1001         struct lustre_capa *capa, *c;
1002         int rc;
1003         ENTRY;
1004
1005         /* if object doesn't exist, or server has disabled capability,
1006          * return directly, client will find body->valid OBD_MD_FLOSSCAPA
1007          * flag not set.
1008          */
1009         if (!obj || !info->mti_mdt->mdt_lut.lut_oss_capa ||
1010             !(exp_connect_flags(info->mti_exp) & OBD_CONNECT_OSS_CAPA))
1011                 RETURN(0);
1012
1013         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1014         LASSERT(body != NULL);
1015
1016         c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
1017         LASSERT(c);
1018
1019         capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
1020         LASSERT(capa);
1021
1022         *capa = *c;
1023         rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
1024         if (rc == 0)
1025                 body->valid |= OBD_MD_FLOSSCAPA;
1026         RETURN(rc);
1027 }
1028
1029 int mdt_getattr(struct tgt_session_info *tsi)
1030 {
1031         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1032         struct mdt_object       *obj = info->mti_object;
1033         struct req_capsule      *pill = info->mti_pill;
1034         struct mdt_body         *reqbody;
1035         struct mdt_body         *repbody;
1036         mode_t                   mode;
1037         int rc, rc2;
1038         ENTRY;
1039
1040         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1041         LASSERT(reqbody);
1042
1043         if (reqbody->valid & OBD_MD_FLOSSCAPA) {
1044                 rc = req_capsule_server_pack(pill);
1045                 if (unlikely(rc))
1046                         RETURN(err_serious(rc));
1047                 rc = mdt_renew_capa(info);
1048                 GOTO(out_shrink, rc);
1049         }
1050
1051         LASSERT(obj != NULL);
1052         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1053
1054         mode = lu_object_attr(&obj->mot_obj);
1055
1056         /* old clients may not report needed easize, use max value then */
1057         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1058                              reqbody->eadatasize == 0 ?
1059                              info->mti_mdt->mdt_max_mdsize :
1060                              reqbody->eadatasize);
1061
1062         rc = req_capsule_server_pack(pill);
1063         if (unlikely(rc != 0))
1064                 GOTO(out, rc = err_serious(rc));
1065
1066         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1067         LASSERT(repbody != NULL);
1068         repbody->eadatasize = 0;
1069         repbody->aclsize = 0;
1070
1071         if (reqbody->valid & OBD_MD_FLRMTPERM)
1072                 rc = mdt_init_ucred(info, reqbody);
1073         else
1074                 rc = mdt_check_ucred(info);
1075         if (unlikely(rc))
1076                 GOTO(out_shrink, rc);
1077
1078         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1079
1080         /*
1081          * Don't check capability at all, because rename might getattr for
1082          * remote obj, and at that time no capability is available.
1083          */
1084         mdt_set_capainfo(info, 1, &reqbody->fid1, BYPASS_CAPA);
1085         rc = mdt_getattr_internal(info, obj, 0);
1086         if (reqbody->valid & OBD_MD_FLRMTPERM)
1087                 mdt_exit_ucred(info);
1088         EXIT;
1089 out_shrink:
1090         mdt_client_compatibility(info);
1091         rc2 = mdt_fix_reply(info);
1092         if (rc == 0)
1093                 rc = rc2;
1094 out:
1095         mdt_thread_info_fini(info);
1096         return rc;
1097 }
1098
1099 int mdt_is_subdir(struct tgt_session_info *tsi)
1100 {
1101         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1102         struct mdt_object     *o = info->mti_object;
1103         struct req_capsule    *pill = info->mti_pill;
1104         const struct mdt_body *body = info->mti_body;
1105         struct mdt_body       *repbody;
1106         int                    rc;
1107         ENTRY;
1108
1109         LASSERT(o != NULL);
1110
1111         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1112
1113         /*
1114          * We save last checked parent fid to @repbody->fid1 for remote
1115          * directory case.
1116          */
1117         LASSERT(fid_is_sane(&body->fid2));
1118         LASSERT(mdt_object_exists(o) && !mdt_object_remote(o));
1119         rc = mdo_is_subdir(info->mti_env, mdt_object_child(o),
1120                            &body->fid2, &repbody->fid1);
1121         if (rc == 0 || rc == -EREMOTE)
1122                 repbody->valid |= OBD_MD_FLID;
1123
1124         mdt_thread_info_fini(info);
1125         RETURN(rc);
1126 }
1127
1128 int mdt_swap_layouts(struct tgt_session_info *tsi)
1129 {
1130         struct mdt_thread_info  *info;
1131         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1132         struct obd_export       *exp = req->rq_export;
1133         struct mdt_object       *o1, *o2, *o;
1134         struct mdt_lock_handle  *lh1, *lh2;
1135         struct mdc_swap_layouts *msl;
1136         int                      rc;
1137         ENTRY;
1138
1139         /* client does not support layout lock, so layout swaping
1140          * is disabled.
1141          * FIXME: there is a problem for old clients which don't support
1142          * layout lock yet. If those clients have already opened the file
1143          * they won't be notified at all so that old layout may still be
1144          * used to do IO. This can be fixed after file release is landed by
1145          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1146         if (!exp_connect_layout(exp))
1147                 RETURN(-EOPNOTSUPP);
1148
1149         info = tsi2mdt_info(tsi);
1150
1151         if (info->mti_dlm_req != NULL)
1152                 ldlm_request_cancel(req, info->mti_dlm_req, 0);
1153
1154         if (req_capsule_get_size(info->mti_pill, &RMF_CAPA1, RCL_CLIENT))
1155                 mdt_set_capainfo(info, 0, &info->mti_body->fid1,
1156                                  req_capsule_client_get(info->mti_pill,
1157                                                         &RMF_CAPA1));
1158
1159         if (req_capsule_get_size(info->mti_pill, &RMF_CAPA2, RCL_CLIENT))
1160                 mdt_set_capainfo(info, 1, &info->mti_body->fid2,
1161                                  req_capsule_client_get(info->mti_pill,
1162                                                         &RMF_CAPA2));
1163
1164         o1 = info->mti_object;
1165         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1166                                 &info->mti_body->fid2);
1167         if (IS_ERR(o))
1168                 GOTO(out, rc = PTR_ERR(o));
1169
1170         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1171                 GOTO(put, rc = -ENOENT);
1172
1173         rc = lu_fid_cmp(&info->mti_body->fid1, &info->mti_body->fid2);
1174         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1175                 GOTO(put, rc);
1176
1177         if (rc < 0)
1178                 swap(o1, o2);
1179
1180         /* permission check. Make sure the calling process having permission
1181          * to write both files. */
1182         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1183                                 MAY_WRITE);
1184         if (rc < 0)
1185                 GOTO(put, rc);
1186
1187         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1188                                 MAY_WRITE);
1189         if (rc < 0)
1190                 GOTO(put, rc);
1191
1192         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1193         if (msl == NULL)
1194                 GOTO(put, rc = -EPROTO);
1195
1196         lh1 = &info->mti_lh[MDT_LH_NEW];
1197         mdt_lock_reg_init(lh1, LCK_EX);
1198         lh2 = &info->mti_lh[MDT_LH_OLD];
1199         mdt_lock_reg_init(lh2, LCK_EX);
1200
1201         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1202                              MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK);
1203         if (rc < 0)
1204                 GOTO(put, rc);
1205
1206         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1207                              MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK);
1208         if (rc < 0)
1209                 GOTO(unlock1, rc);
1210
1211         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1212                              mdt_object_child(o2), msl->msl_flags);
1213         GOTO(unlock2, rc);
1214 unlock2:
1215         mdt_object_unlock(info, o2, lh2, rc);
1216 unlock1:
1217         mdt_object_unlock(info, o1, lh1, rc);
1218 put:
1219         mdt_object_put(info->mti_env, o);
1220 out:
1221         mdt_thread_info_fini(info);
1222         RETURN(rc);
1223 }
1224
1225 static int mdt_raw_lookup(struct mdt_thread_info *info,
1226                           struct mdt_object *parent,
1227                           const struct lu_name *lname,
1228                           struct ldlm_reply *ldlm_rep)
1229 {
1230         struct md_object *next = mdt_object_child(info->mti_object);
1231         const struct mdt_body *reqbody = info->mti_body;
1232         struct lu_fid *child_fid = &info->mti_tmp_fid1;
1233         struct mdt_body *repbody;
1234         int rc;
1235         ENTRY;
1236
1237         if (reqbody->valid != OBD_MD_FLID)
1238                 RETURN(0);
1239
1240         LASSERT(!info->mti_cross_ref);
1241
1242         /* Only got the fid of this obj by name */
1243         fid_zero(child_fid);
1244         rc = mdo_lookup(info->mti_env, next, lname, child_fid,
1245                         &info->mti_spec);
1246 #if 0
1247         /* XXX is raw_lookup possible as intent operation? */
1248         if (rc != 0) {
1249                 if (rc == -ENOENT)
1250                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1251                 RETURN(rc);
1252         } else
1253                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1254
1255         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1256 #endif
1257         if (rc == 0) {
1258                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1259                 repbody->fid1 = *child_fid;
1260                 repbody->valid = OBD_MD_FLID;
1261         }
1262         RETURN(1);
1263 }
1264
1265 /*
1266  * UPDATE lock should be taken against parent, and be release before exit;
1267  * child_bits lock should be taken against child, and be returned back:
1268  *            (1)normal request should release the child lock;
1269  *            (2)intent request will grant the lock to client.
1270  */
1271 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1272                                  struct mdt_lock_handle *lhc,
1273                                  __u64 child_bits,
1274                                  struct ldlm_reply *ldlm_rep)
1275 {
1276         struct ptlrpc_request  *req       = mdt_info_req(info);
1277         struct mdt_body        *reqbody   = NULL;
1278         struct mdt_object      *parent    = info->mti_object;
1279         struct mdt_object      *child;
1280         struct md_object       *next      = mdt_object_child(parent);
1281         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1282         struct lu_name         *lname     = NULL;
1283         struct mdt_lock_handle *lhp       = NULL;
1284         struct ldlm_lock       *lock;
1285         struct ldlm_res_id     *res_id;
1286         int                     is_resent;
1287         int                     ma_need = 0;
1288         int                     rc;
1289
1290         ENTRY;
1291
1292         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1293         LASSERT(ergo(is_resent,
1294                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1295
1296         LASSERT(parent != NULL);
1297
1298         lname = &info->mti_name;
1299         mdt_name_unpack(info->mti_pill, &RMF_NAME, lname, MNF_FIX_ANON);
1300
1301         if (!info->mti_cross_ref) {
1302                 /*
1303                  * XXX: Check for anonymous name is for getattr by fid
1304                  * (OBD_CONNECT_ATTRFID), otherwise do not allow empty name,
1305                  * that is the name must contain at least one character and
1306                  * the terminating '\0'.
1307                  */
1308                 if (!lu_name_is_valid(lname)) {
1309                         reqbody = req_capsule_client_get(info->mti_pill,
1310                                                          &RMF_MDT_BODY);
1311                         if (unlikely(reqbody == NULL))
1312                                 RETURN(err_serious(-EFAULT));
1313
1314                         if (unlikely(!fid_is_sane(&reqbody->fid2)))
1315                                 RETURN(err_serious(-EINVAL));
1316
1317                         CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1318                                "ldlm_rep = %p\n",
1319                                PFID(mdt_object_fid(parent)),
1320                                PFID(&reqbody->fid2), ldlm_rep);
1321                 } else {
1322                         CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1323                                "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1324                                PNAME(lname), ldlm_rep);
1325                 }
1326         }
1327
1328         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1329
1330         if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1331                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1332                                 &parent->mot_obj,
1333                                 "Parent doesn't exist!\n");
1334                 RETURN(-ESTALE);
1335         } else if (!info->mti_cross_ref) {
1336                 LASSERTF(!mdt_object_remote(parent),
1337                          "Parent "DFID" is on remote server\n",
1338                          PFID(mdt_object_fid(parent)));
1339         }
1340
1341         if (lu_name_is_valid(lname)) {
1342                 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1343                 if (rc != 0) {
1344                         if (rc > 0)
1345                                 rc = 0;
1346                         RETURN(rc);
1347                 }
1348         }
1349
1350         if (info->mti_cross_ref) {
1351                 /* Only getattr on the child. Parent is on another node. */
1352                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1353                 child = parent;
1354                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1355                        "ldlm_rep=%p\n", PFID(mdt_object_fid(child)), ldlm_rep);
1356
1357                 if (is_resent) {
1358                         /* Do not take lock for resent case. */
1359                         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1360                         LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
1361                                  lhc->mlh_reg_lh.cookie);
1362                         LASSERT(fid_res_name_eq(mdt_object_fid(child),
1363                                                 &lock->l_resource->lr_name));
1364                         LDLM_LOCK_PUT(lock);
1365                         rc = 0;
1366                 } else {
1367                         mdt_lock_handle_init(lhc);
1368                         mdt_lock_reg_init(lhc, LCK_PR);
1369
1370                         /*
1371                          * Object's name is on another MDS, no lookup or layout
1372                          * lock is needed here but update lock is.
1373                          */
1374                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
1375                                         MDS_INODELOCK_LAYOUT);
1376                         child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1377
1378                         rc = mdt_object_lock(info, child, lhc, child_bits,
1379                                              MDT_LOCAL_LOCK);
1380                 }
1381                 if (rc == 0) {
1382                         /* Finally, we can get attr for child. */
1383                         if (!mdt_object_exists(child)) {
1384                                 LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1385                                                 &child->mot_obj,
1386                                              "remote object doesn't exist.\n");
1387                                 mdt_object_unlock(info, child, lhc, 1);
1388                                 RETURN(-ENOENT);
1389                         }
1390
1391                         mdt_set_capainfo(info, 0, mdt_object_fid(child),
1392                                          BYPASS_CAPA);
1393                         rc = mdt_getattr_internal(info, child, 0);
1394                         if (unlikely(rc != 0))
1395                                 mdt_object_unlock(info, child, lhc, 1);
1396                 }
1397                 RETURN(rc);
1398         }
1399
1400         if (lu_name_is_valid(lname)) {
1401                 /* step 1: lock parent only if parent is a directory */
1402                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1403                         lhp = &info->mti_lh[MDT_LH_PARENT];
1404                         mdt_lock_pdo_init(lhp, LCK_PR, lname);
1405                         rc = mdt_object_lock(info, parent, lhp,
1406                                              MDS_INODELOCK_UPDATE,
1407                                              MDT_LOCAL_LOCK);
1408                         if (unlikely(rc != 0))
1409                                 RETURN(rc);
1410                 }
1411
1412                 /* step 2: lookup child's fid by name */
1413                 fid_zero(child_fid);
1414                 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
1415                                 &info->mti_spec);
1416
1417                 if (rc != 0) {
1418                         if (rc == -ENOENT)
1419                                 mdt_set_disposition(info, ldlm_rep,
1420                                                     DISP_LOOKUP_NEG);
1421                         GOTO(out_parent, rc);
1422                 } else
1423                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1424         } else {
1425                 *child_fid = reqbody->fid2;
1426                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1427         }
1428
1429         /*
1430          *step 3: find the child object by fid & lock it.
1431          *        regardless if it is local or remote.
1432          *
1433          *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1434          *      set parent dir fid the same as child fid in getattr by fid case
1435          *      we should not lu_object_find() the object again, could lead
1436          *      to hung if there is a concurrent unlink destroyed the object.
1437          */
1438         if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1439                 mdt_object_get(info->mti_env, parent);
1440                 child = parent;
1441         } else {
1442                 child = mdt_object_find(info->mti_env, info->mti_mdt,
1443                                         child_fid);
1444         }
1445
1446         if (unlikely(IS_ERR(child)))
1447                 GOTO(out_parent, rc = PTR_ERR(child));
1448         if (is_resent) {
1449                 /* Do not take lock for resent case. */
1450                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1451                 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
1452                          lhc->mlh_reg_lh.cookie);
1453
1454                 res_id = &lock->l_resource->lr_name;
1455                 if (!fid_res_name_eq(mdt_object_fid(child),
1456                                      &lock->l_resource->lr_name)) {
1457                         LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
1458                                                  &lock->l_resource->lr_name),
1459                                  "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1460                                  PLDLMRES(lock->l_resource),
1461                                  PFID(mdt_object_fid(parent)));
1462                         CWARN("Although resent, but still not get child lock"
1463                               "parent:"DFID" child:"DFID"\n",
1464                               PFID(mdt_object_fid(parent)),
1465                               PFID(mdt_object_fid(child)));
1466                         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
1467                         LDLM_LOCK_PUT(lock);
1468                         GOTO(relock, 0);
1469                 }
1470                 LDLM_LOCK_PUT(lock);
1471                 rc = 0;
1472         } else {
1473                 bool try_layout = false;
1474
1475 relock:
1476                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout*2);
1477                 mdt_lock_handle_init(lhc);
1478                 mdt_lock_reg_init(lhc, LCK_PR);
1479
1480                 if (!mdt_object_exists(child)) {
1481                         LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1482                                         &child->mot_obj,
1483                                         "Object doesn't exist!\n");
1484                         GOTO(out_child, rc = -ENOENT);
1485                 }
1486
1487                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1488                       mdt_object_exists(child) && !mdt_object_remote(child)) {
1489                         struct md_attr *ma = &info->mti_attr;
1490
1491                         ma->ma_valid = 0;
1492                         ma->ma_need = MA_INODE;
1493                         rc = mdt_attr_get_complex(info, child, ma);
1494                         if (unlikely(rc != 0))
1495                                 GOTO(out_child, rc);
1496
1497                         /* If the file has not been changed for some time, we
1498                          * return not only a LOOKUP lock, but also an UPDATE
1499                          * lock and this might save us RPC on later STAT. For
1500                          * directories, it also let negative dentry cache start
1501                          * working for this dir. */
1502                         if (ma->ma_valid & MA_INODE &&
1503                             ma->ma_attr.la_valid & LA_CTIME &&
1504                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1505                                 ma->ma_attr.la_ctime < cfs_time_current_sec())
1506                                 child_bits |= MDS_INODELOCK_UPDATE;
1507                 }
1508
1509                 /* layout lock must be granted in a best-effort way
1510                  * for IT operations */
1511                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1512                 if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1513                     exp_connect_layout(info->mti_exp) &&
1514                     S_ISREG(lu_object_attr(&child->mot_obj)) &&
1515                     !mdt_object_remote(child) && ldlm_rep != NULL) {
1516                         /* try to grant layout lock for regular file. */
1517                         try_layout = true;
1518                 }
1519
1520                 rc = 0;
1521                 if (try_layout) {
1522                         child_bits |= MDS_INODELOCK_LAYOUT;
1523                         /* try layout lock, it may fail to be granted due to
1524                          * contention at LOOKUP or UPDATE */
1525                         if (!mdt_object_lock_try(info, child, lhc, child_bits,
1526                                                  MDT_CROSS_LOCK)) {
1527                                 child_bits &= ~MDS_INODELOCK_LAYOUT;
1528                                 LASSERT(child_bits != 0);
1529                                 rc = mdt_object_lock(info, child, lhc,
1530                                                 child_bits, MDT_CROSS_LOCK);
1531                         } else {
1532                                 ma_need |= MA_LOV;
1533                         }
1534                 } else {
1535                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1536                          * client will enqueue the lock to the remote MDT */
1537                         if (mdt_object_remote(child))
1538                                 child_bits &= ~MDS_INODELOCK_UPDATE;
1539                         rc = mdt_object_lock(info, child, lhc, child_bits,
1540                                                 MDT_CROSS_LOCK);
1541                 }
1542                 if (unlikely(rc != 0))
1543                         GOTO(out_child, rc);
1544         }
1545
1546         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1547         /* Get MA_SOM attributes if update lock is given. */
1548         if (lock &&
1549             lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_UPDATE &&
1550             S_ISREG(lu_object_attr(&mdt_object_child(child)->mo_lu)))
1551                 ma_need |= MA_SOM;
1552
1553         /* finally, we can get attr for child. */
1554         mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
1555         rc = mdt_getattr_internal(info, child, ma_need);
1556         if (unlikely(rc != 0)) {
1557                 mdt_object_unlock(info, child, lhc, 1);
1558         } else if (lock) {
1559                 /* Debugging code. */
1560                 res_id = &lock->l_resource->lr_name;
1561                 LDLM_DEBUG(lock, "Returning lock to client");
1562                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1563                                          &lock->l_resource->lr_name),
1564                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1565                          PLDLMRES(lock->l_resource),
1566                          PFID(mdt_object_fid(child)));
1567                 if (mdt_object_exists(child) && !mdt_object_remote(child))
1568                         mdt_pack_size2body(info, child);
1569         }
1570         if (lock)
1571                 LDLM_LOCK_PUT(lock);
1572
1573         EXIT;
1574 out_child:
1575         mdt_object_put(info->mti_env, child);
1576 out_parent:
1577         if (lhp)
1578                 mdt_object_unlock(info, parent, lhp, 1);
1579         return rc;
1580 }
1581
1582 /* normal handler: should release the child lock */
1583 int mdt_getattr_name(struct tgt_session_info *tsi)
1584 {
1585         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1586         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1587         struct mdt_body        *reqbody;
1588         struct mdt_body        *repbody;
1589         int rc, rc2;
1590         ENTRY;
1591
1592         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1593         LASSERT(reqbody != NULL);
1594         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1595         LASSERT(repbody != NULL);
1596
1597         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1598         repbody->eadatasize = 0;
1599         repbody->aclsize = 0;
1600
1601         rc = mdt_init_ucred(info, reqbody);
1602         if (unlikely(rc))
1603                 GOTO(out_shrink, rc);
1604
1605         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1606         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1607                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1608                 lhc->mlh_reg_lh.cookie = 0;
1609         }
1610         mdt_exit_ucred(info);
1611         EXIT;
1612 out_shrink:
1613         mdt_client_compatibility(info);
1614         rc2 = mdt_fix_reply(info);
1615         if (rc == 0)
1616                 rc = rc2;
1617         mdt_thread_info_fini(info);
1618         return rc;
1619 }
1620
1621 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1622                          void *karg, void *uarg);
1623
1624 int mdt_set_info(struct tgt_session_info *tsi)
1625 {
1626         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1627         char                    *key;
1628         void                    *val;
1629         int                      keylen, vallen, rc = 0;
1630
1631         ENTRY;
1632
1633         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
1634         if (key == NULL) {
1635                 DEBUG_REQ(D_HA, req, "no set_info key");
1636                 RETURN(err_serious(-EFAULT));
1637         }
1638
1639         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
1640                                       RCL_CLIENT);
1641
1642         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
1643         if (val == NULL) {
1644                 DEBUG_REQ(D_HA, req, "no set_info val");
1645                 RETURN(err_serious(-EFAULT));
1646         }
1647
1648         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
1649                                       RCL_CLIENT);
1650
1651         /* Swab any part of val you need to here */
1652         if (KEY_IS(KEY_READ_ONLY)) {
1653                 spin_lock(&req->rq_export->exp_lock);
1654                 if (*(__u32 *)val)
1655                         *exp_connect_flags_ptr(req->rq_export) |=
1656                                 OBD_CONNECT_RDONLY;
1657                 else
1658                         *exp_connect_flags_ptr(req->rq_export) &=
1659                                 ~OBD_CONNECT_RDONLY;
1660                 spin_unlock(&req->rq_export->exp_lock);
1661         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1662                 struct changelog_setinfo *cs = val;
1663
1664                 if (vallen != sizeof(*cs)) {
1665                         CERROR("%s: bad changelog_clear setinfo size %d\n",
1666                                tgt_name(tsi->tsi_tgt), vallen);
1667                         RETURN(-EINVAL);
1668                 }
1669                 if (ptlrpc_req_need_swab(req)) {
1670                         __swab64s(&cs->cs_recno);
1671                         __swab32s(&cs->cs_id);
1672                 }
1673
1674                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
1675                                    vallen, val, NULL);
1676         } else {
1677                 RETURN(-EINVAL);
1678         }
1679         RETURN(rc);
1680 }
1681
1682 int mdt_readpage(struct tgt_session_info *tsi)
1683 {
1684         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
1685         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
1686         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
1687         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
1688         struct mdt_body         *repbody;
1689         int                      rc;
1690         int                      i;
1691
1692         ENTRY;
1693
1694         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1695                 RETURN(err_serious(-ENOMEM));
1696
1697         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
1698         if (repbody == NULL || reqbody == NULL)
1699                 RETURN(err_serious(-EFAULT));
1700
1701         /*
1702          * prepare @rdpg before calling lower layers and transfer itself. Here
1703          * reqbody->size contains offset of where to start to read and
1704          * reqbody->nlink contains number bytes to read.
1705          */
1706         rdpg->rp_hash = reqbody->size;
1707         if (rdpg->rp_hash != reqbody->size) {
1708                 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1709                        rdpg->rp_hash, reqbody->size);
1710                 RETURN(-EFAULT);
1711         }
1712
1713         rdpg->rp_attrs = reqbody->mode;
1714         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
1715                 rdpg->rp_attrs |= LUDA_64BITHASH;
1716         rdpg->rp_count  = min_t(unsigned int, reqbody->nlink,
1717                                 exp_max_brw_size(tsi->tsi_exp));
1718         rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE - 1) >>
1719                           PAGE_CACHE_SHIFT;
1720         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1721         if (rdpg->rp_pages == NULL)
1722                 RETURN(-ENOMEM);
1723
1724         for (i = 0; i < rdpg->rp_npages; ++i) {
1725                 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1726                 if (rdpg->rp_pages[i] == NULL)
1727                         GOTO(free_rdpg, rc = -ENOMEM);
1728         }
1729
1730         /* call lower layers to fill allocated pages with directory data */
1731         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
1732         if (rc < 0)
1733                 GOTO(free_rdpg, rc);
1734
1735         /* send pages to client */
1736         rc = tgt_sendpage(tsi, rdpg, rc);
1737
1738         EXIT;
1739 free_rdpg:
1740
1741         for (i = 0; i < rdpg->rp_npages; i++)
1742                 if (rdpg->rp_pages[i] != NULL)
1743                         __free_page(rdpg->rp_pages[i]);
1744         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1745
1746         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1747                 RETURN(0);
1748
1749         return rc;
1750 }
1751
1752 static int mdt_reint_internal(struct mdt_thread_info *info,
1753                               struct mdt_lock_handle *lhc,
1754                               __u32 op)
1755 {
1756         struct req_capsule      *pill = info->mti_pill;
1757         struct mdt_body         *repbody;
1758         int                      rc = 0, rc2;
1759
1760         ENTRY;
1761
1762         rc = mdt_reint_unpack(info, op);
1763         if (rc != 0) {
1764                 CERROR("Can't unpack reint, rc %d\n", rc);
1765                 RETURN(err_serious(rc));
1766         }
1767
1768         /* for replay (no_create) lmm is not needed, client has it already */
1769         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1770                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1771                                      info->mti_rr.rr_eadatalen);
1772
1773         /* llog cookies are always 0, the field is kept for compatibility */
1774         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1775                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
1776
1777         rc = req_capsule_server_pack(pill);
1778         if (rc != 0) {
1779                 CERROR("Can't pack response, rc %d\n", rc);
1780                 RETURN(err_serious(rc));
1781         }
1782
1783         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1784                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1785                 LASSERT(repbody);
1786                 repbody->eadatasize = 0;
1787                 repbody->aclsize = 0;
1788         }
1789
1790         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1791
1792         /* for replay no cookkie / lmm need, because client have this already */
1793         if (info->mti_spec.no_create)
1794                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1795                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1796
1797         rc = mdt_init_ucred_reint(info);
1798         if (rc)
1799                 GOTO(out_shrink, rc);
1800
1801         rc = mdt_fix_attr_ucred(info, op);
1802         if (rc != 0)
1803                 GOTO(out_ucred, rc = err_serious(rc));
1804
1805         if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1806                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
1807                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1808                 GOTO(out_ucred, rc);
1809         }
1810         rc = mdt_reint_rec(info, lhc);
1811         EXIT;
1812 out_ucred:
1813         mdt_exit_ucred(info);
1814 out_shrink:
1815         mdt_client_compatibility(info);
1816         rc2 = mdt_fix_reply(info);
1817         if (rc == 0)
1818                 rc = rc2;
1819         return rc;
1820 }
1821
1822 static long mdt_reint_opcode(struct ptlrpc_request *req,
1823                              const struct req_format **fmt)
1824 {
1825         struct mdt_device       *mdt;
1826         struct mdt_rec_reint    *rec;
1827         long                     opc;
1828
1829         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
1830         if (rec != NULL) {
1831                 opc = rec->rr_opcode;
1832                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
1833                 if (opc < REINT_MAX && fmt[opc] != NULL)
1834                         req_capsule_extend(&req->rq_pill, fmt[opc]);
1835                 else {
1836                         mdt = mdt_exp2dev(req->rq_export);
1837                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
1838                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
1839                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
1840                         opc = err_serious(-EFAULT);
1841                 }
1842         } else {
1843                 opc = err_serious(-EFAULT);
1844         }
1845         return opc;
1846 }
1847
1848 int mdt_reint(struct tgt_session_info *tsi)
1849 {
1850         long opc;
1851         int  rc;
1852         static const struct req_format *reint_fmts[REINT_MAX] = {
1853                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
1854                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
1855                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
1856                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
1857                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
1858                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
1859                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
1860                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
1861                 [REINT_MIGRATE]  = &RQF_MDS_REINT_RENAME
1862         };
1863
1864         ENTRY;
1865
1866         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
1867         if (opc >= 0) {
1868                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1869                 /*
1870                  * No lock possible here from client to pass it to reint code
1871                  * path.
1872                  */
1873                 rc = mdt_reint_internal(info, NULL, opc);
1874                 mdt_thread_info_fini(info);
1875         } else {
1876                 rc = opc;
1877         }
1878
1879         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1880         RETURN(rc);
1881 }
1882
1883 /* this should sync the whole device */
1884 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1885 {
1886         struct dt_device *dt = mdt->mdt_bottom;
1887         int rc;
1888         ENTRY;
1889
1890         rc = dt->dd_ops->dt_sync(env, dt);
1891         RETURN(rc);
1892 }
1893
1894 /* this should sync this object */
1895 static int mdt_object_sync(struct mdt_thread_info *info)
1896 {
1897         struct md_object *next;
1898         int rc;
1899         ENTRY;
1900
1901         if (!mdt_object_exists(info->mti_object)) {
1902                 CWARN("Non existing object  "DFID"!\n",
1903                       PFID(mdt_object_fid(info->mti_object)));
1904                 RETURN(-ESTALE);
1905         }
1906         next = mdt_object_child(info->mti_object);
1907         rc = mo_object_sync(info->mti_env, next);
1908
1909         RETURN(rc);
1910 }
1911
1912 int mdt_sync(struct tgt_session_info *tsi)
1913 {
1914         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1915         struct req_capsule      *pill = tsi->tsi_pill;
1916         struct mdt_body         *body;
1917         int                      rc;
1918
1919         ENTRY;
1920
1921         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1922                 RETURN(err_serious(-ENOMEM));
1923
1924         if (fid_seq(&tsi->tsi_mdt_body->fid1) == 0) {
1925                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
1926         } else {
1927                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1928
1929                 /* sync an object */
1930                 rc = mdt_object_sync(info);
1931                 if (rc == 0) {
1932                         const struct lu_fid *fid;
1933                         struct lu_attr *la = &info->mti_attr.ma_attr;
1934
1935                         info->mti_attr.ma_need = MA_INODE;
1936                         info->mti_attr.ma_valid = 0;
1937                         rc = mdt_attr_get_complex(info, info->mti_object,
1938                                                   &info->mti_attr);
1939                         if (rc == 0) {
1940                                 body = req_capsule_server_get(pill,
1941                                                               &RMF_MDT_BODY);
1942                                 fid = mdt_object_fid(info->mti_object);
1943                                 mdt_pack_attr2body(info, body, la, fid);
1944                         }
1945                 }
1946                 mdt_thread_info_fini(info);
1947         }
1948         if (rc == 0)
1949                 mdt_counter_incr(req, LPROC_MDT_SYNC);
1950
1951         RETURN(rc);
1952 }
1953
1954 /*
1955  * Handle quota control requests to consult current usage/limit, but also
1956  * to configure quota enforcement
1957  */
1958 int mdt_quotactl(struct tgt_session_info *tsi)
1959 {
1960         struct obd_export       *exp  = tsi->tsi_exp;
1961         struct req_capsule      *pill = tsi->tsi_pill;
1962         struct obd_quotactl     *oqctl, *repoqc;
1963         int                      id, rc;
1964         struct mdt_device       *mdt = mdt_exp2dev(exp);
1965         struct lu_device        *qmt = mdt->mdt_qmt_dev;
1966         ENTRY;
1967
1968         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1969         if (oqctl == NULL)
1970                 RETURN(err_serious(-EPROTO));
1971
1972         rc = req_capsule_server_pack(pill);
1973         if (rc)
1974                 RETURN(err_serious(rc));
1975
1976         switch (oqctl->qc_cmd) {
1977         case Q_QUOTACHECK:
1978         case LUSTRE_Q_INVALIDATE:
1979         case LUSTRE_Q_FINVALIDATE:
1980         case Q_QUOTAON:
1981         case Q_QUOTAOFF:
1982         case Q_INITQUOTA:
1983                 /* deprecated, not used any more */
1984                 RETURN(-EOPNOTSUPP);
1985                 /* master quotactl */
1986         case Q_GETINFO:
1987         case Q_SETINFO:
1988         case Q_SETQUOTA:
1989         case Q_GETQUOTA:
1990                 if (qmt == NULL)
1991                         RETURN(-EOPNOTSUPP);
1992                 /* slave quotactl */
1993         case Q_GETOINFO:
1994         case Q_GETOQUOTA:
1995                 break;
1996         default:
1997                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
1998                 RETURN(-EFAULT);
1999         }
2000
2001         /* map uid/gid for remote client */
2002         id = oqctl->qc_id;
2003         if (exp_connect_rmtclient(exp)) {
2004                 struct lustre_idmap_table *idmap;
2005
2006                 idmap = exp->exp_mdt_data.med_idmap;
2007
2008                 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
2009                              oqctl->qc_cmd != Q_GETINFO))
2010                         RETURN(-EPERM);
2011
2012                 if (oqctl->qc_type == USRQUOTA)
2013                         id = lustre_idmap_lookup_uid(NULL, idmap, 0,
2014                                                      oqctl->qc_id);
2015                 else if (oqctl->qc_type == GRPQUOTA)
2016                         id = lustre_idmap_lookup_gid(NULL, idmap, 0,
2017                                                      oqctl->qc_id);
2018                 else
2019                         RETURN(-EINVAL);
2020
2021                 if (id == CFS_IDMAP_NOTFOUND) {
2022                         CDEBUG(D_QUOTA, "no mapping for id %u\n", oqctl->qc_id);
2023                         RETURN(-EACCES);
2024                 }
2025         }
2026
2027         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2028         if (repoqc == NULL)
2029                 RETURN(err_serious(-EFAULT));
2030
2031         if (oqctl->qc_id != id)
2032                 swap(oqctl->qc_id, id);
2033
2034         switch (oqctl->qc_cmd) {
2035
2036         case Q_GETINFO:
2037         case Q_SETINFO:
2038         case Q_SETQUOTA:
2039         case Q_GETQUOTA:
2040                 /* forward quotactl request to QMT */
2041                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2042                 break;
2043
2044         case Q_GETOINFO:
2045         case Q_GETOQUOTA:
2046                 /* slave quotactl */
2047                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2048                                    oqctl);
2049                 break;
2050
2051         default:
2052                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2053                 RETURN(-EFAULT);
2054         }
2055
2056         if (oqctl->qc_id != id)
2057                 swap(oqctl->qc_id, id);
2058
2059         *repoqc = *oqctl;
2060         RETURN(rc);
2061 }
2062
2063 /** clone llog ctxt from child (mdd)
2064  * This allows remote llog (replicator) access.
2065  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2066  * context was originally set up, or we can handle them directly.
2067  * I choose the latter, but that means I need any llog
2068  * contexts set up by child to be accessable by the mdt.  So we clone the
2069  * context into our context list here.
2070  */
2071 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2072                                int idx)
2073 {
2074         struct md_device  *next = mdt->mdt_child;
2075         struct llog_ctxt *ctxt;
2076         int rc;
2077
2078         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2079                 return 0;
2080
2081         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2082         if (rc || ctxt == NULL) {
2083                 return 0;
2084         }
2085
2086         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2087         if (rc)
2088                 CERROR("Can't set mdt ctxt %d\n", rc);
2089
2090         return rc;
2091 }
2092
2093 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2094                                  struct mdt_device *mdt, int idx)
2095 {
2096         struct llog_ctxt *ctxt;
2097
2098         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2099         if (ctxt == NULL)
2100                 return 0;
2101         /* Put once for the get we just did, and once for the clone */
2102         llog_ctxt_put(ctxt);
2103         llog_ctxt_put(ctxt);
2104         return 0;
2105 }
2106
2107 /*
2108  * sec context handlers
2109  */
2110 int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2111 {
2112         int rc;
2113
2114         rc = mdt_handle_idmap(tsi);
2115         if (unlikely(rc)) {
2116                 struct ptlrpc_request   *req = tgt_ses_req(tsi);
2117                 __u32                    opc;
2118
2119                 opc = lustre_msg_get_opc(req->rq_reqmsg);
2120                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2121                         sptlrpc_svc_ctx_invalidate(req);
2122         }
2123
2124         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2125
2126         return rc;
2127 }
2128
2129 /*
2130  * quota request handlers
2131  */
2132 int mdt_quota_dqacq(struct tgt_session_info *tsi)
2133 {
2134         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
2135         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2136         int                      rc;
2137         ENTRY;
2138
2139         if (qmt == NULL)
2140                 RETURN(err_serious(-EOPNOTSUPP));
2141
2142         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2143         RETURN(rc);
2144 }
2145
2146 struct mdt_object *mdt_object_new(const struct lu_env *env,
2147                                   struct mdt_device *d,
2148                                   const struct lu_fid *f)
2149 {
2150         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2151         struct lu_object *o;
2152         struct mdt_object *m;
2153         ENTRY;
2154
2155         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2156         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2157         if (unlikely(IS_ERR(o)))
2158                 m = (struct mdt_object *)o;
2159         else
2160                 m = mdt_obj(o);
2161         RETURN(m);
2162 }
2163
2164 struct mdt_object *mdt_object_find(const struct lu_env *env,
2165                                    struct mdt_device *d,
2166                                    const struct lu_fid *f)
2167 {
2168         struct lu_object *o;
2169         struct mdt_object *m;
2170         ENTRY;
2171
2172         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2173         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2174         if (unlikely(IS_ERR(o)))
2175                 m = (struct mdt_object *)o;
2176         else
2177                 m = mdt_obj(o);
2178
2179         RETURN(m);
2180 }
2181
2182 /**
2183  * Asyncronous commit for mdt device.
2184  *
2185  * Pass asynchonous commit call down the MDS stack.
2186  *
2187  * \param env environment
2188  * \param mdt the mdt device
2189  */
2190 static void mdt_device_commit_async(const struct lu_env *env,
2191                                     struct mdt_device *mdt)
2192 {
2193         struct dt_device *dt = mdt->mdt_bottom;
2194         int rc;
2195
2196         rc = dt->dd_ops->dt_commit_async(env, dt);
2197         if (unlikely(rc != 0))
2198                 CWARN("async commit start failed with rc = %d", rc);
2199 }
2200
2201 /**
2202  * Mark the lock as "synchonous".
2203  *
2204  * Mark the lock to deffer transaction commit to the unlock time.
2205  *
2206  * \param lock the lock to mark as "synchonous"
2207  *
2208  * \see mdt_is_lock_sync
2209  * \see mdt_save_lock
2210  */
2211 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2212 {
2213         lock->l_ast_data = (void*)1;
2214 }
2215
2216 /**
2217  * Check whehter the lock "synchonous" or not.
2218  *
2219  * \param lock the lock to check
2220  * \retval 1 the lock is "synchonous"
2221  * \retval 0 the lock isn't "synchronous"
2222  *
2223  * \see mdt_set_lock_sync
2224  * \see mdt_save_lock
2225  */
2226 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2227 {
2228         return lock->l_ast_data != NULL;
2229 }
2230
2231 /**
2232  * Blocking AST for mdt locks.
2233  *
2234  * Starts transaction commit if in case of COS lock conflict or
2235  * deffers such a commit to the mdt_save_lock.
2236  *
2237  * \param lock the lock which blocks a request or cancelling lock
2238  * \param desc unused
2239  * \param data unused
2240  * \param flag indicates whether this cancelling or blocking callback
2241  * \retval 0
2242  * \see ldlm_blocking_ast_nocheck
2243  */
2244 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2245                      void *data, int flag)
2246 {
2247         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2248         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2249         int rc;
2250         ENTRY;
2251
2252         if (flag == LDLM_CB_CANCELING)
2253                 RETURN(0);
2254         lock_res_and_lock(lock);
2255         if (lock->l_blocking_ast != mdt_blocking_ast) {
2256                 unlock_res_and_lock(lock);
2257                 RETURN(0);
2258         }
2259         if (mdt_cos_is_enabled(mdt) &&
2260             lock->l_req_mode & (LCK_PW | LCK_EX) &&
2261             lock->l_blocking_lock != NULL &&
2262             lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
2263                 mdt_set_lock_sync(lock);
2264         }
2265         rc = ldlm_blocking_ast_nocheck(lock);
2266
2267         /* There is no lock conflict if l_blocking_lock == NULL,
2268          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2269          * when the last reference to a local lock was released */
2270         if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2271                 struct lu_env env;
2272
2273                 rc = lu_env_init(&env, LCT_LOCAL);
2274                 if (unlikely(rc != 0))
2275                         CWARN("lu_env initialization failed with rc = %d,"
2276                               "cannot start asynchronous commit\n", rc);
2277                 else
2278                         mdt_device_commit_async(&env, mdt);
2279                 lu_env_fini(&env);
2280         }
2281         RETURN(rc);
2282 }
2283
2284 /* Used for cross-MDT lock */
2285 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2286                             void *data, int flag)
2287 {
2288         struct lustre_handle lockh;
2289         int               rc;
2290
2291         switch (flag) {
2292         case LDLM_CB_BLOCKING:
2293                 ldlm_lock2handle(lock, &lockh);
2294                 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
2295                 if (rc < 0) {
2296                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2297                         RETURN(rc);
2298                 }
2299                 break;
2300         case LDLM_CB_CANCELING:
2301                 LDLM_DEBUG(lock, "Revoke remote lock\n");
2302                 break;
2303         default:
2304                 LBUG();
2305         }
2306         RETURN(0);
2307 }
2308
2309 int mdt_remote_object_lock(struct mdt_thread_info *mti,
2310                            struct mdt_object *o, const struct lu_fid *fid,
2311                            struct lustre_handle *lh, ldlm_mode_t mode,
2312                            __u64 ibits)
2313 {
2314         struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
2315         ldlm_policy_data_t *policy = &mti->mti_policy;
2316         struct ldlm_res_id *res_id = &mti->mti_res_id;
2317         int rc = 0;
2318         ENTRY;
2319
2320         LASSERT(mdt_object_remote(o));
2321
2322         LASSERT(ibits == MDS_INODELOCK_UPDATE);
2323
2324         fid_build_reg_res_name(fid, res_id);
2325
2326         memset(einfo, 0, sizeof(*einfo));
2327         einfo->ei_type = LDLM_IBITS;
2328         einfo->ei_mode = mode;
2329         einfo->ei_cb_bl = mdt_remote_blocking_ast;
2330         einfo->ei_cb_cp = ldlm_completion_ast;
2331         einfo->ei_enq_slave = 0;
2332         einfo->ei_res_id = res_id;
2333
2334         memset(policy, 0, sizeof(*policy));
2335         policy->l_inodebits.bits = ibits;
2336
2337         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2338                             policy);
2339         RETURN(rc);
2340 }
2341
2342 static int mdt_object_local_lock(struct mdt_thread_info *info,
2343                                  struct mdt_object *o,
2344                                  struct mdt_lock_handle *lh, __u64 ibits,
2345                                  bool nonblock, int locality)
2346 {
2347         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2348         ldlm_policy_data_t *policy = &info->mti_policy;
2349         struct ldlm_res_id *res_id = &info->mti_res_id;
2350         __u64 dlmflags;
2351         int rc;
2352         ENTRY;
2353
2354         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2355         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2356         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2357         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2358
2359         /* Only enqueue LOOKUP lock for remote object */
2360         if (mdt_object_remote(o))
2361                 LASSERT(ibits == MDS_INODELOCK_LOOKUP);
2362
2363         if (lh->mlh_type == MDT_PDO_LOCK) {
2364                 /* check for exists after object is locked */
2365                 if (mdt_object_exists(o) == 0) {
2366                         /* Non-existent object shouldn't have PDO lock */
2367                         RETURN(-ESTALE);
2368                 } else {
2369                         /* Non-dir object shouldn't have PDO lock */
2370                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2371                                 RETURN(-ENOTDIR);
2372                 }
2373         }
2374
2375         memset(policy, 0, sizeof(*policy));
2376         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2377
2378         dlmflags = LDLM_FL_ATOMIC_CB;
2379         if (nonblock)
2380                 dlmflags |= LDLM_FL_BLOCK_NOWAIT;
2381
2382         /*
2383          * Take PDO lock on whole directory and build correct @res_id for lock
2384          * on part of directory.
2385          */
2386         if (lh->mlh_pdo_hash != 0) {
2387                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2388                 mdt_lock_pdo_mode(info, o, lh);
2389                 if (lh->mlh_pdo_mode != LCK_NL) {
2390                         /*
2391                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2392                          * is never going to be sent to client and we do not
2393                          * want it slowed down due to possible cancels.
2394                          */
2395                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2396                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2397                                           policy, res_id, dlmflags,
2398                                           &info->mti_exp->exp_handle.h_cookie);
2399                         if (unlikely(rc))
2400                                 RETURN(rc);
2401                 }
2402
2403                 /*
2404                  * Finish res_id initializing by name hash marking part of
2405                  * directory which is taking modification.
2406                  */
2407                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2408         }
2409
2410         policy->l_inodebits.bits = ibits;
2411
2412         /*
2413          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2414          * going to be sent to client. If it is - mdt_intent_policy() path will
2415          * fix it up and turn FL_LOCAL flag off.
2416          */
2417         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2418                           res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2419                           &info->mti_exp->exp_handle.h_cookie);
2420         if (rc)
2421                 mdt_object_unlock(info, o, lh, 1);
2422         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2423                  lh->mlh_pdo_hash != 0 &&
2424                  (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX)) {
2425                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2426         }
2427
2428         RETURN(rc);
2429 }
2430
2431 int mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
2432                              struct mdt_lock_handle *lh, __u64 ibits,
2433                              bool nonblock, int locality)
2434 {
2435         int rc;
2436         ENTRY;
2437
2438         if (!mdt_object_remote(o))
2439                 return mdt_object_local_lock(info, o, lh, ibits, nonblock,
2440                                              locality);
2441
2442         if (locality == MDT_LOCAL_LOCK) {
2443                 CERROR("%s: try to get local lock for remote object"
2444                        DFID".\n", mdt_obd_name(info->mti_mdt),
2445                        PFID(mdt_object_fid(o)));
2446                 RETURN(-EPROTO);
2447         }
2448
2449         /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
2450         ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
2451                    MDS_INODELOCK_XATTR);
2452         if (ibits & MDS_INODELOCK_UPDATE) {
2453                 /* Sigh, PDO needs to enqueue 2 locks right now, but
2454                  * enqueue RPC can only request 1 lock, to avoid extra
2455                  * RPC, so it will instead enqueue EX lock for remote
2456                  * object anyway XXX*/
2457                 if (lh->mlh_type == MDT_PDO_LOCK &&
2458                     lh->mlh_pdo_hash != 0) {
2459                         CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
2460                                "EX lock.\n", mdt_obd_name(info->mti_mdt),
2461                                PFID(mdt_object_fid(o)));
2462                         lh->mlh_pdo_hash = 0;
2463                         lh->mlh_rreg_mode = LCK_EX;
2464                         lh->mlh_type = MDT_REG_LOCK;
2465                 }
2466                 rc = mdt_remote_object_lock(info, o, mdt_object_fid(o),
2467                                             &lh->mlh_rreg_lh,
2468                                             lh->mlh_rreg_mode,
2469                                             MDS_INODELOCK_UPDATE);
2470                 if (rc != ELDLM_OK)
2471                         RETURN(rc);
2472         }
2473
2474         /* Only enqueue LOOKUP lock for remote object */
2475         if (ibits & MDS_INODELOCK_LOOKUP) {
2476                 rc = mdt_object_local_lock(info, o, lh,
2477                                            MDS_INODELOCK_LOOKUP,
2478                                            nonblock, locality);
2479                 if (rc != ELDLM_OK)
2480                         RETURN(rc);
2481         }
2482
2483         RETURN(0);
2484 }
2485
2486 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2487                     struct mdt_lock_handle *lh, __u64 ibits, int locality)
2488 {
2489         return mdt_object_lock_internal(info, o, lh, ibits, false, locality);
2490 }
2491
2492 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
2493                         struct mdt_lock_handle *lh, __u64 ibits, int locality)
2494 {
2495         struct mdt_lock_handle tmp = *lh;
2496         int rc;
2497
2498         rc = mdt_object_lock_internal(info, o, &tmp, ibits, true, locality);
2499         if (rc == 0)
2500                 *lh = tmp;
2501
2502         return rc == 0;
2503 }
2504
2505 /**
2506  * Save a lock within request object.
2507  *
2508  * Keep the lock referenced until whether client ACK or transaction
2509  * commit happens or release the lock immediately depending on input
2510  * parameters. If COS is ON, a write lock is converted to COS lock
2511  * before saving.
2512  *
2513  * \param info thead info object
2514  * \param h lock handle
2515  * \param mode lock mode
2516  * \param decref force immediate lock releasing
2517  */
2518 static
2519 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2520                    ldlm_mode_t mode, int decref)
2521 {
2522         ENTRY;
2523
2524         if (lustre_handle_is_used(h)) {
2525                 if (decref || !info->mti_has_trans ||
2526                     !(mode & (LCK_PW | LCK_EX))){
2527                         mdt_fid_unlock(h, mode);
2528                 } else {
2529                         struct mdt_device *mdt = info->mti_mdt;
2530                         struct ldlm_lock *lock = ldlm_handle2lock(h);
2531                         struct ptlrpc_request *req = mdt_info_req(info);
2532                         int no_ack = 0;
2533
2534                         LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2535                                  h->cookie);
2536                         /* there is no request if mdt_object_unlock() is called
2537                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
2538                         if (likely(req != NULL)) {
2539                                 CDEBUG(D_HA, "request = %p reply state = %p"
2540                                        " transno = "LPD64"\n", req,
2541                                        req->rq_reply_state, req->rq_transno);
2542                                 if (mdt_cos_is_enabled(mdt)) {
2543                                         no_ack = 1;
2544                                         ldlm_lock_downgrade(lock, LCK_COS);
2545                                         mode = LCK_COS;
2546                                 }
2547                                 ptlrpc_save_lock(req, h, mode, no_ack);
2548                         } else {
2549                                 ldlm_lock_decref(h, mode);
2550                         }
2551                         if (mdt_is_lock_sync(lock)) {
2552                                 CDEBUG(D_HA, "found sync-lock,"
2553                                        " async commit started\n");
2554                                 mdt_device_commit_async(info->mti_env,
2555                                                         mdt);
2556                         }
2557                         LDLM_LOCK_PUT(lock);
2558                 }
2559                 h->cookie = 0ull;
2560         }
2561
2562         EXIT;
2563 }
2564
2565 /**
2566  * Unlock mdt object.
2567  *
2568  * Immeditely release the regular lock and the PDO lock or save the
2569  * lock in reqeuest and keep them referenced until client ACK or
2570  * transaction commit.
2571  *
2572  * \param info thread info object
2573  * \param o mdt object
2574  * \param lh mdt lock handle referencing regular and PDO locks
2575  * \param decref force immediate lock releasing
2576  */
2577 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2578                        struct mdt_lock_handle *lh, int decref)
2579 {
2580         ENTRY;
2581
2582         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2583         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2584
2585         if (lustre_handle_is_used(&lh->mlh_rreg_lh))
2586                 ldlm_lock_decref(&lh->mlh_rreg_lh, lh->mlh_rreg_mode);
2587
2588         EXIT;
2589 }
2590
2591 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2592                                         const struct lu_fid *f,
2593                                         struct mdt_lock_handle *lh,
2594                                         __u64 ibits)
2595 {
2596         struct mdt_object *o;
2597
2598         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2599         if (!IS_ERR(o)) {
2600                 int rc;
2601
2602                 rc = mdt_object_lock(info, o, lh, ibits,
2603                                      MDT_LOCAL_LOCK);
2604                 if (rc != 0) {
2605                         mdt_object_put(info->mti_env, o);
2606                         o = ERR_PTR(rc);
2607                 }
2608         }
2609         return o;
2610 }
2611
2612 void mdt_object_unlock_put(struct mdt_thread_info * info,
2613                            struct mdt_object * o,
2614                            struct mdt_lock_handle *lh,
2615                            int decref)
2616 {
2617         mdt_object_unlock(info, o, lh, decref);
2618         mdt_object_put(info->mti_env, o);
2619 }
2620
2621 /*
2622  * Generic code handling requests that have struct mdt_body passed in:
2623  *
2624  *  - extract mdt_body from request and save it in @info, if present;
2625  *
2626  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
2627  *  @info;
2628  *
2629  *  - if HABEO_CORPUS flag is set for this request type check whether object
2630  *  actually exists on storage (lu_object_exists()).
2631  *
2632  */
2633 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2634 {
2635         const struct mdt_body    *body;
2636         struct mdt_object        *obj;
2637         const struct lu_env      *env;
2638         struct req_capsule       *pill;
2639         int                       rc;
2640         ENTRY;
2641
2642         env = info->mti_env;
2643         pill = info->mti_pill;
2644
2645         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2646         if (body == NULL)
2647                 RETURN(-EFAULT);
2648
2649         if (!(body->valid & OBD_MD_FLID))
2650                 RETURN(0);
2651
2652         if (!fid_is_sane(&body->fid1)) {
2653                 CERROR("Invalid fid: "DFID"\n", PFID(&body->fid1));
2654                 RETURN(-EINVAL);
2655         }
2656
2657         /*
2658          * Do not get size or any capa fields before we check that request
2659          * contains capa actually. There are some requests which do not, for
2660          * instance MDS_IS_SUBDIR.
2661          */
2662         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2663             req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
2664                 mdt_set_capainfo(info, 0, &body->fid1,
2665                                  req_capsule_client_get(pill, &RMF_CAPA1));
2666
2667         obj = mdt_object_find(env, info->mti_mdt, &body->fid1);
2668         if (!IS_ERR(obj)) {
2669                 if ((flags & HABEO_CORPUS) &&
2670                     !mdt_object_exists(obj)) {
2671                         mdt_object_put(env, obj);
2672                         /* for capability renew ENOENT will be handled in
2673                          * mdt_renew_capa */
2674                         if (body->valid & OBD_MD_FLOSSCAPA)
2675                                 rc = 0;
2676                         else
2677                                 rc = -ENOENT;
2678                 } else {
2679                         info->mti_object = obj;
2680                         rc = 0;
2681                 }
2682         } else
2683                 rc = PTR_ERR(obj);
2684
2685         RETURN(rc);
2686 }
2687
2688 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2689 {
2690         struct req_capsule *pill = info->mti_pill;
2691         int rc;
2692         ENTRY;
2693
2694         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2695                 rc = mdt_body_unpack(info, flags);
2696         else
2697                 rc = 0;
2698
2699         if (rc == 0 && (flags & HABEO_REFERO)) {
2700                 /* Pack reply. */
2701                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2702                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2703                                              info->mti_body->eadatasize);
2704                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2705                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
2706                                              RCL_SERVER, 0);
2707
2708                 rc = req_capsule_server_pack(pill);
2709         }
2710         RETURN(rc);
2711 }
2712
2713 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2714 {
2715         struct md_device *next = m->mdt_child;
2716
2717         return next->md_ops->mdo_init_capa_ctxt(env, next,
2718                                                 m->mdt_lut.lut_mds_capa,
2719                                                 m->mdt_capa_timeout,
2720                                                 m->mdt_capa_alg,
2721                                                 m->mdt_capa_keys);
2722 }
2723
2724 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2725 {
2726         lh->mlh_type = MDT_NUL_LOCK;
2727         lh->mlh_reg_lh.cookie = 0ull;
2728         lh->mlh_reg_mode = LCK_MINMODE;
2729         lh->mlh_pdo_lh.cookie = 0ull;
2730         lh->mlh_pdo_mode = LCK_MINMODE;
2731         lh->mlh_rreg_lh.cookie = 0ull;
2732         lh->mlh_rreg_mode = LCK_MINMODE;
2733 }
2734
2735 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2736 {
2737         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2738         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2739 }
2740
2741 /*
2742  * Initialize fields of struct mdt_thread_info. Other fields are left in
2743  * uninitialized state, because it's too expensive to zero out whole
2744  * mdt_thread_info (> 1K) on each request arrival.
2745  */
2746 void mdt_thread_info_init(struct ptlrpc_request *req,
2747                           struct mdt_thread_info *info)
2748 {
2749         int i;
2750
2751         info->mti_pill = &req->rq_pill;
2752
2753         /* lock handle */
2754         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2755                 mdt_lock_handle_init(&info->mti_lh[i]);
2756
2757         /* mdt device: it can be NULL while CONNECT */
2758         if (req->rq_export) {
2759                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2760                 info->mti_exp = req->rq_export;
2761         } else
2762                 info->mti_mdt = NULL;
2763         info->mti_env = req->rq_svc_thread->t_env;
2764         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2765
2766         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2767         info->mti_big_buf = LU_BUF_NULL;
2768         info->mti_body = NULL;
2769         info->mti_object = NULL;
2770         info->mti_dlm_req = NULL;
2771         info->mti_has_trans = 0;
2772         info->mti_cross_ref = 0;
2773         info->mti_opdata = 0;
2774         info->mti_big_lmm_used = 0;
2775
2776         info->mti_spec.no_create = 0;
2777         info->mti_spec.sp_rm_entry = 0;
2778
2779         info->mti_spec.u.sp_ea.eadata = NULL;
2780         info->mti_spec.u.sp_ea.eadatalen = 0;
2781 }
2782
2783 void mdt_thread_info_fini(struct mdt_thread_info *info)
2784 {
2785         int i;
2786
2787         if (info->mti_object != NULL) {
2788                 mdt_object_put(info->mti_env, info->mti_object);
2789                 info->mti_object = NULL;
2790         }
2791
2792         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2793                 mdt_lock_handle_fini(&info->mti_lh[i]);
2794         info->mti_env = NULL;
2795         info->mti_pill = NULL;
2796         info->mti_exp = NULL;
2797
2798         if (unlikely(info->mti_big_buf.lb_buf != NULL))
2799                 lu_buf_free(&info->mti_big_buf);
2800 }
2801
2802 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
2803 {
2804         struct mdt_thread_info  *mti;
2805         struct lustre_capa      *lc;
2806
2807         mti = mdt_th_info(tsi->tsi_env);
2808         LASSERT(mti != NULL);
2809
2810         mdt_thread_info_init(tgt_ses_req(tsi), mti);
2811         if (tsi->tsi_corpus != NULL) {
2812                 struct req_capsule *pill = tsi->tsi_pill;
2813
2814                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
2815                 lu_object_get(tsi->tsi_corpus);
2816
2817                 /*
2818                  * XXX: must be part of tgt_mdt_body_unpack but moved here
2819                  * due to mdt_set_capainfo().
2820                  */
2821                 if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2822                     req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT) > 0) {
2823                         lc = req_capsule_client_get(pill, &RMF_CAPA1);
2824                         mdt_set_capainfo(mti, 0, &tsi->tsi_mdt_body->fid1, lc);
2825                 }
2826         }
2827         mti->mti_body = tsi->tsi_mdt_body;
2828         mti->mti_dlm_req = tsi->tsi_dlm_req;
2829
2830         return mti;
2831 }
2832
2833 int mdt_tgt_connect(struct tgt_session_info *tsi)
2834 {
2835         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2836         int                      rc;
2837
2838         ENTRY;
2839
2840         rc = tgt_connect(tsi);
2841         if (rc != 0)
2842                 RETURN(rc);
2843
2844         rc = mdt_init_idmap(tsi);
2845         if (rc != 0)
2846                 GOTO(err, rc);
2847         RETURN(0);
2848 err:
2849         obd_disconnect(class_export_get(req->rq_export));
2850         return rc;
2851 }
2852
2853 enum mdt_it_code {
2854         MDT_IT_OPEN,
2855         MDT_IT_OCREAT,
2856         MDT_IT_CREATE,
2857         MDT_IT_GETATTR,
2858         MDT_IT_READDIR,
2859         MDT_IT_LOOKUP,
2860         MDT_IT_UNLINK,
2861         MDT_IT_TRUNC,
2862         MDT_IT_GETXATTR,
2863         MDT_IT_LAYOUT,
2864         MDT_IT_QUOTA,
2865         MDT_IT_NR
2866 };
2867
2868 static int mdt_intent_getattr(enum mdt_it_code opcode,
2869                               struct mdt_thread_info *info,
2870                               struct ldlm_lock **,
2871                               __u64);
2872
2873 static int mdt_intent_getxattr(enum mdt_it_code opcode,
2874                                 struct mdt_thread_info *info,
2875                                 struct ldlm_lock **lockp,
2876                                 __u64 flags);
2877
2878 static int mdt_intent_layout(enum mdt_it_code opcode,
2879                              struct mdt_thread_info *info,
2880                              struct ldlm_lock **,
2881                              __u64);
2882 static int mdt_intent_reint(enum mdt_it_code opcode,
2883                             struct mdt_thread_info *info,
2884                             struct ldlm_lock **,
2885                             __u64);
2886
2887 static struct mdt_it_flavor {
2888         const struct req_format *it_fmt;
2889         __u32                    it_flags;
2890         int                    (*it_act)(enum mdt_it_code ,
2891                                          struct mdt_thread_info *,
2892                                          struct ldlm_lock **,
2893                                          __u64);
2894         long                     it_reint;
2895 } mdt_it_flavor[] = {
2896         [MDT_IT_OPEN]     = {
2897                 .it_fmt   = &RQF_LDLM_INTENT,
2898                 /*.it_flags = HABEO_REFERO,*/
2899                 .it_flags = 0,
2900                 .it_act   = mdt_intent_reint,
2901                 .it_reint = REINT_OPEN
2902         },
2903         [MDT_IT_OCREAT]   = {
2904                 .it_fmt   = &RQF_LDLM_INTENT,
2905                 /*
2906                  * OCREAT is not a MUTABOR request as if the file
2907                  * already exists.
2908                  * We do the extra check of OBD_CONNECT_RDONLY in
2909                  * mdt_reint_open() when we really need to create
2910                  * the object.
2911                  */
2912                 .it_flags = 0,
2913                 .it_act   = mdt_intent_reint,
2914                 .it_reint = REINT_OPEN
2915         },
2916         [MDT_IT_CREATE]   = {
2917                 .it_fmt   = &RQF_LDLM_INTENT,
2918                 .it_flags = MUTABOR,
2919                 .it_act   = mdt_intent_reint,
2920                 .it_reint = REINT_CREATE
2921         },
2922         [MDT_IT_GETATTR]  = {
2923                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2924                 .it_flags = HABEO_REFERO,
2925                 .it_act   = mdt_intent_getattr
2926         },
2927         [MDT_IT_READDIR]  = {
2928                 .it_fmt   = NULL,
2929                 .it_flags = 0,
2930                 .it_act   = NULL
2931         },
2932         [MDT_IT_LOOKUP]   = {
2933                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2934                 .it_flags = HABEO_REFERO,
2935                 .it_act   = mdt_intent_getattr
2936         },
2937         [MDT_IT_UNLINK]   = {
2938                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
2939                 .it_flags = MUTABOR,
2940                 .it_act   = NULL,
2941                 .it_reint = REINT_UNLINK
2942         },
2943         [MDT_IT_TRUNC]    = {
2944                 .it_fmt   = NULL,
2945                 .it_flags = MUTABOR,
2946                 .it_act   = NULL
2947         },
2948         [MDT_IT_GETXATTR] = {
2949                 .it_fmt   = &RQF_LDLM_INTENT_GETXATTR,
2950                 .it_flags = HABEO_CORPUS,
2951                 .it_act   = mdt_intent_getxattr
2952         },
2953         [MDT_IT_LAYOUT] = {
2954                 .it_fmt   = &RQF_LDLM_INTENT_LAYOUT,
2955                 .it_flags = 0,
2956                 .it_act   = mdt_intent_layout
2957         }
2958 };
2959
2960 int mdt_intent_lock_replace(struct mdt_thread_info *info,
2961                             struct ldlm_lock **lockp,
2962                             struct ldlm_lock *new_lock,
2963                             struct mdt_lock_handle *lh,
2964                             __u64 flags)
2965 {
2966         struct ptlrpc_request  *req = mdt_info_req(info);
2967         struct ldlm_lock       *lock = *lockp;
2968
2969         /*
2970          * Get new lock only for cases when possible resent did not find any
2971          * lock.
2972          */
2973         if (new_lock == NULL)
2974                 new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
2975
2976         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
2977                 lh->mlh_reg_lh.cookie = 0;
2978                 RETURN(0);
2979         }
2980
2981         LASSERTF(new_lock != NULL,
2982                  "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
2983
2984         /*
2985          * If we've already given this lock to a client once, then we should
2986          * have no readers or writers.  Otherwise, we should have one reader
2987          * _or_ writer ref (which will be zeroed below) before returning the
2988          * lock to a client.
2989          */
2990         if (new_lock->l_export == req->rq_export) {
2991                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
2992         } else {
2993                 LASSERT(new_lock->l_export == NULL);
2994                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
2995         }
2996
2997         *lockp = new_lock;
2998
2999         if (new_lock->l_export == req->rq_export) {
3000                 /*
3001                  * Already gave this to the client, which means that we
3002                  * reconstructed a reply.
3003                  */
3004                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3005                         MSG_RESENT);
3006                 lh->mlh_reg_lh.cookie = 0;
3007                 RETURN(ELDLM_LOCK_REPLACED);
3008         }
3009
3010         /*
3011          * Fixup the lock to be given to the client.
3012          */
3013         lock_res_and_lock(new_lock);
3014         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3015          * possible blocking AST. */
3016         while (new_lock->l_readers > 0) {
3017                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3018                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3019                 new_lock->l_readers--;
3020         }
3021         while (new_lock->l_writers > 0) {
3022                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3023                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3024                 new_lock->l_writers--;
3025         }
3026
3027         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3028         new_lock->l_blocking_ast = lock->l_blocking_ast;
3029         new_lock->l_completion_ast = lock->l_completion_ast;
3030         new_lock->l_remote_handle = lock->l_remote_handle;
3031         new_lock->l_flags &= ~LDLM_FL_LOCAL;
3032
3033         unlock_res_and_lock(new_lock);
3034
3035         cfs_hash_add(new_lock->l_export->exp_lock_hash,
3036                      &new_lock->l_remote_handle,
3037                      &new_lock->l_exp_hash);
3038
3039         LDLM_LOCK_RELEASE(new_lock);
3040         lh->mlh_reg_lh.cookie = 0;
3041
3042         RETURN(ELDLM_LOCK_REPLACED);
3043 }
3044
3045 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3046                                     struct ldlm_lock *new_lock,
3047                                     struct ldlm_lock **old_lock,
3048                                     struct mdt_lock_handle *lh,
3049                                     enum mdt_it_code opcode)
3050 {
3051         struct ptlrpc_request  *req = mdt_info_req(info);
3052         struct obd_export      *exp = req->rq_export;
3053         struct lustre_handle    remote_hdl;
3054         struct ldlm_request    *dlmreq;
3055         struct ldlm_lock       *lock;
3056
3057         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3058                 return;
3059
3060         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3061         remote_hdl = dlmreq->lock_handle[0];
3062         /* If the client does not require open lock, it does not need to
3063          * search lock in exp_lock_hash, since the server thread will
3064          * make sure the lock will be released, and the resend request
3065          * can always re-enqueue the lock */
3066         if ((opcode != MDT_IT_OPEN) || (opcode == MDT_IT_OPEN &&
3067             info->mti_spec.sp_cr_flags & MDS_OPEN_LOCK)) {
3068                 /* In the function below, .hs_keycmp resolves to
3069                  * ldlm_export_lock_keycmp() */
3070                 /* coverity[overrun-buffer-val] */
3071                 lock = cfs_hash_lookup(exp->exp_lock_hash, &remote_hdl);
3072                 if (lock) {
3073                         lock_res_and_lock(lock);
3074                         if (lock != new_lock) {
3075                                 lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie;
3076                                 lh->mlh_reg_mode = lock->l_granted_mode;
3077
3078                                 LDLM_DEBUG(lock, "Restoring lock cookie");
3079                                 DEBUG_REQ(D_DLMTRACE, req,
3080                                           "restoring lock cookie "LPX64,
3081                                           lh->mlh_reg_lh.cookie);
3082                                 if (old_lock)
3083                                         *old_lock = LDLM_LOCK_GET(lock);
3084                                 cfs_hash_put(exp->exp_lock_hash,
3085                                              &lock->l_exp_hash);
3086                                 unlock_res_and_lock(lock);
3087                                 return;
3088                         }
3089                         cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
3090                         unlock_res_and_lock(lock);
3091                 }
3092         }
3093         /*
3094          * If the xid matches, then we know this is a resent request, and allow
3095          * it. (It's probably an OPEN, for which we don't send a lock.
3096          */
3097         if (req_xid_is_last(req))
3098                 return;
3099
3100         /*
3101          * This remote handle isn't enqueued, so we never received or processed
3102          * this request.  Clear MSG_RESENT, because it can be handled like any
3103          * normal request now.
3104          */
3105         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3106
3107         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
3108                   remote_hdl.cookie);
3109 }
3110
3111 static int mdt_intent_getxattr(enum mdt_it_code opcode,
3112                                 struct mdt_thread_info *info,
3113                                 struct ldlm_lock **lockp,
3114                                 __u64 flags)
3115 {
3116         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3117         struct ldlm_reply      *ldlm_rep = NULL;
3118         int rc, grc;
3119
3120         /*
3121          * Initialize lhc->mlh_reg_lh either from a previously granted lock
3122          * (for the resend case) or a new lock. Below we will use it to
3123          * replace the original lock.
3124          */
3125         mdt_intent_fixup_resent(info, *lockp, NULL, lhc, opcode);
3126         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3127                 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3128                 rc = mdt_object_lock(info, info->mti_object, lhc,
3129                                         MDS_INODELOCK_XATTR,
3130                                         MDT_LOCAL_LOCK);
3131                 if (rc)
3132                         return rc;
3133         }
3134
3135         grc = mdt_getxattr(info);
3136
3137         rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3138
3139         if (mdt_info_req(info)->rq_repmsg != NULL)
3140                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3141         if (ldlm_rep == NULL)
3142                 RETURN(err_serious(-EFAULT));
3143
3144         ldlm_rep->lock_policy_res2 = grc;
3145
3146         return rc;
3147 }
3148
3149 static int mdt_intent_getattr(enum mdt_it_code opcode,
3150                               struct mdt_thread_info *info,
3151                               struct ldlm_lock **lockp,
3152                               __u64 flags)
3153 {
3154         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3155         struct ldlm_lock       *new_lock = NULL;
3156         __u64                   child_bits;
3157         struct ldlm_reply      *ldlm_rep;
3158         struct ptlrpc_request  *req;
3159         struct mdt_body        *reqbody;
3160         struct mdt_body        *repbody;
3161         int                     rc, rc2;
3162         ENTRY;
3163
3164         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3165         LASSERT(reqbody);
3166
3167         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3168         LASSERT(repbody);
3169
3170         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
3171         repbody->eadatasize = 0;
3172         repbody->aclsize = 0;
3173
3174         switch (opcode) {
3175         case MDT_IT_LOOKUP:
3176                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3177                 break;
3178         case MDT_IT_GETATTR:
3179                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3180                              MDS_INODELOCK_PERM;
3181                 break;
3182         default:
3183                 CERROR("Unsupported intent (%d)\n", opcode);
3184                 GOTO(out_shrink, rc = -EINVAL);
3185         }
3186
3187         rc = mdt_init_ucred(info, reqbody);
3188         if (rc)
3189                 GOTO(out_shrink, rc);
3190
3191         req = info->mti_pill->rc_req;
3192         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3193         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3194
3195         /* Get lock from request for possible resent case. */
3196         mdt_intent_fixup_resent(info, *lockp, &new_lock, lhc, opcode);
3197
3198         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3199         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3200
3201         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3202                 ldlm_rep->lock_policy_res2 = 0;
3203         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3204             ldlm_rep->lock_policy_res2) {
3205                 lhc->mlh_reg_lh.cookie = 0ull;
3206                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3207         }
3208
3209         rc = mdt_intent_lock_replace(info, lockp, new_lock, lhc, flags);
3210         EXIT;
3211 out_ucred:
3212         mdt_exit_ucred(info);
3213 out_shrink:
3214         mdt_client_compatibility(info);
3215         rc2 = mdt_fix_reply(info);
3216         if (rc == 0)
3217                 rc = rc2;
3218         return rc;
3219 }
3220
3221 static int mdt_intent_layout(enum mdt_it_code opcode,
3222                              struct mdt_thread_info *info,
3223                              struct ldlm_lock **lockp,
3224                              __u64 flags)
3225 {
3226         struct layout_intent *layout;
3227         struct lu_fid *fid;
3228         struct mdt_object *obj = NULL;
3229         struct md_object *child = NULL;
3230         int rc;
3231         ENTRY;
3232
3233         if (opcode != MDT_IT_LAYOUT) {
3234                 CERROR("%s: Unknown intent (%d)\n", mdt_obd_name(info->mti_mdt),
3235                         opcode);
3236                 RETURN(-EINVAL);
3237         }
3238
3239         fid = &info->mti_tmp_fid2;
3240         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3241
3242         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3243         if (IS_ERR(obj))
3244                 RETURN(PTR_ERR(obj));
3245
3246         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3247                 child = mdt_object_child(obj);
3248
3249                 /* get the length of lsm */
3250                 rc = mo_xattr_get(info->mti_env, child, &LU_BUF_NULL,
3251                                   XATTR_NAME_LOV);
3252
3253                 if (rc > info->mti_mdt->mdt_max_mdsize)
3254                         info->mti_mdt->mdt_max_mdsize = rc;
3255         }
3256
3257         mdt_object_put(info->mti_env, obj);
3258
3259         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3260         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
3261                         ldlm_lvbo_size(*lockp));
3262         rc = req_capsule_server_pack(info->mti_pill);
3263         if (rc != 0)
3264                 RETURN(-EINVAL);
3265
3266         layout = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3267         LASSERT(layout != NULL);
3268         if (layout->li_opc == LAYOUT_INTENT_ACCESS)
3269                 /* return to normal ldlm handling */
3270                 RETURN(0);
3271
3272         CERROR("%s: Unsupported layout intent (%d)\n",
3273                 mdt_obd_name(info->mti_mdt), layout->li_opc);
3274         RETURN(-EINVAL);
3275 }
3276
3277 static int mdt_intent_reint(enum mdt_it_code opcode,
3278                             struct mdt_thread_info *info,
3279                             struct ldlm_lock **lockp,
3280                             __u64 flags)
3281 {
3282         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3283         struct ldlm_reply      *rep = NULL;
3284         long                    opc;
3285         int                     rc;
3286
3287         static const struct req_format *intent_fmts[REINT_MAX] = {
3288                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3289                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3290         };
3291
3292         ENTRY;
3293
3294         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
3295         if (opc < 0)
3296                 RETURN(opc);
3297
3298         if (mdt_it_flavor[opcode].it_reint != opc) {
3299                 CERROR("Reint code %ld doesn't match intent: %d\n",
3300                        opc, opcode);
3301                 RETURN(err_serious(-EPROTO));
3302         }
3303
3304         /* Get lock from request for possible resent case. */
3305         mdt_intent_fixup_resent(info, *lockp, NULL, lhc, opcode);
3306
3307         rc = mdt_reint_internal(info, lhc, opc);
3308
3309         /* Check whether the reply has been packed successfully. */
3310         if (mdt_info_req(info)->rq_repmsg != NULL)
3311                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3312         if (rep == NULL)
3313                 RETURN(err_serious(-EFAULT));
3314
3315         /* MDC expects this in any case */
3316         if (rc != 0)
3317                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3318
3319         /* the open lock or the lock for cross-ref object should be
3320          * returned to the client */
3321         if (rc == -EREMOTE || mdt_get_disposition(rep, DISP_OPEN_LOCK)) {
3322                 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
3323                 rep->lock_policy_res2 = 0;
3324                 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3325                 RETURN(rc);
3326         }
3327
3328         rep->lock_policy_res2 = clear_serious(rc);
3329
3330         if (rep->lock_policy_res2 == -ENOENT &&
3331             mdt_get_disposition(rep, DISP_LOOKUP_NEG))
3332                 rep->lock_policy_res2 = 0;
3333
3334         if (rc == -ENOTCONN || rc == -ENODEV ||
3335             rc == -EOVERFLOW) { /**< if VBR failure then return error */
3336                 /*
3337                  * If it is the disconnect error (ENODEV & ENOCONN), the error
3338                  * will be returned by rq_status, and client at ptlrpc layer
3339                  * will detect this, then disconnect, reconnect the import
3340                  * immediately, instead of impacting the following the rpc.
3341                  */
3342                 lhc->mlh_reg_lh.cookie = 0ull;
3343                 RETURN(rc);
3344         } else {
3345                 /*
3346                  * For other cases, the error will be returned by intent.
3347                  * and client will retrieve the result from intent.
3348                  */
3349                  /*
3350                   * FIXME: when open lock is finished, that should be
3351                   * checked here.
3352                   */
3353                 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3354                         LASSERTF(rc == 0, "Error occurred but lock handle "
3355                                  "is still in use, rc = %d\n", rc);
3356                         rep->lock_policy_res2 = 0;
3357                         rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3358                         RETURN(rc);
3359                 } else {
3360                         lhc->mlh_reg_lh.cookie = 0ull;
3361                         RETURN(ELDLM_LOCK_ABORTED);
3362                 }
3363         }
3364 }
3365
3366 static int mdt_intent_code(long itcode)
3367 {
3368         int rc;
3369
3370         switch(itcode) {
3371         case IT_OPEN:
3372                 rc = MDT_IT_OPEN;
3373                 break;
3374         case IT_OPEN|IT_CREAT:
3375                 rc = MDT_IT_OCREAT;
3376                 break;
3377         case IT_CREAT:
3378                 rc = MDT_IT_CREATE;
3379                 break;
3380         case IT_READDIR:
3381                 rc = MDT_IT_READDIR;
3382                 break;
3383         case IT_GETATTR:
3384                 rc = MDT_IT_GETATTR;
3385                 break;
3386         case IT_LOOKUP:
3387                 rc = MDT_IT_LOOKUP;
3388                 break;
3389         case IT_UNLINK:
3390                 rc = MDT_IT_UNLINK;
3391                 break;
3392         case IT_TRUNC:
3393                 rc = MDT_IT_TRUNC;
3394                 break;
3395         case IT_GETXATTR:
3396                 rc = MDT_IT_GETXATTR;
3397                 break;
3398         case IT_LAYOUT:
3399                 rc = MDT_IT_LAYOUT;
3400                 break;
3401         case IT_QUOTA_DQACQ:
3402         case IT_QUOTA_CONN:
3403                 rc = MDT_IT_QUOTA;
3404                 break;
3405         default:
3406                 CERROR("Unknown intent opcode: %ld\n", itcode);
3407                 rc = -EINVAL;
3408                 break;
3409         }
3410         return rc;
3411 }
3412
3413 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
3414                           struct ldlm_lock **lockp, __u64 flags)
3415 {
3416         struct req_capsule   *pill;
3417         struct mdt_it_flavor *flv;
3418         int opc;
3419         int rc;
3420         ENTRY;
3421
3422         opc = mdt_intent_code(itopc);
3423         if (opc < 0)
3424                 RETURN(-EINVAL);
3425
3426         pill = info->mti_pill;
3427
3428         if (opc == MDT_IT_QUOTA) {
3429                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
3430
3431                 if (qmt == NULL)
3432                         RETURN(-EOPNOTSUPP);
3433
3434                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
3435                 /* pass the request to quota master */
3436                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
3437                                                  mdt_info_req(info), lockp,
3438                                                  flags);
3439                 RETURN(rc);
3440         }
3441
3442         flv  = &mdt_it_flavor[opc];
3443         if (flv->it_fmt != NULL)
3444                 req_capsule_extend(pill, flv->it_fmt);
3445
3446         rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3447         if (rc == 0) {
3448                 struct ptlrpc_request *req = mdt_info_req(info);
3449                 if (flv->it_flags & MUTABOR &&
3450                     exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
3451                         RETURN(-EROFS);
3452         }
3453         if (rc == 0 && flv->it_act != NULL) {
3454                 struct ldlm_reply *rep;
3455
3456                 /* execute policy */
3457                 rc = flv->it_act(opc, info, lockp, flags);
3458
3459                 /* Check whether the reply has been packed successfully. */
3460                 if (mdt_info_req(info)->rq_repmsg != NULL) {
3461                         rep = req_capsule_server_get(info->mti_pill,
3462                                                      &RMF_DLM_REP);
3463                         rep->lock_policy_res2 =
3464                                 ptlrpc_status_hton(rep->lock_policy_res2);
3465                 }
3466         } else {
3467                 rc = -EPROTO;
3468         }
3469         RETURN(rc);
3470 }
3471
3472 static int mdt_intent_policy(struct ldlm_namespace *ns,
3473                              struct ldlm_lock **lockp, void *req_cookie,
3474                              ldlm_mode_t mode, __u64 flags, void *data)
3475 {
3476         struct tgt_session_info *tsi;
3477         struct mdt_thread_info  *info;
3478         struct ptlrpc_request   *req  =  req_cookie;
3479         struct ldlm_intent      *it;
3480         struct req_capsule      *pill;
3481         int rc;
3482
3483         ENTRY;
3484
3485         LASSERT(req != NULL);
3486
3487         tsi = tgt_ses_info(req->rq_svc_thread->t_env);
3488
3489         info = tsi2mdt_info(tsi);
3490         LASSERT(info != NULL);
3491         pill = info->mti_pill;
3492         LASSERT(pill->rc_req == req);
3493
3494         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3495                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
3496                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3497                 if (it != NULL) {
3498                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
3499                         if (rc == 0)
3500                                 rc = ELDLM_OK;
3501
3502                         /* Lock without inodebits makes no sense and will oops
3503                          * later in ldlm. Let's check it now to see if we have
3504                          * ibits corrupted somewhere in mdt_intent_opc().
3505                          * The case for client miss to set ibits has been
3506                          * processed by others. */
3507                         LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
3508                                         lr_type == LDLM_IBITS,
3509                                      info->mti_dlm_req->lock_desc.\
3510                                         l_policy_data.l_inodebits.bits != 0));
3511                 } else
3512                         rc = err_serious(-EFAULT);
3513         } else {
3514                 /* No intent was provided */
3515                 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3516                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
3517                 rc = req_capsule_server_pack(pill);
3518                 if (rc)
3519                         rc = err_serious(rc);
3520         }
3521         mdt_thread_info_fini(info);
3522         RETURN(rc);
3523 }
3524
3525 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
3526 {
3527         struct seq_server_site  *ss = mdt_seq_site(mdt);
3528
3529         if (ss->ss_node_id == 0)
3530                 return;
3531
3532         if (ss->ss_client_seq != NULL) {
3533                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3534                 ss->ss_client_seq->lcs_exp = NULL;
3535         }
3536
3537         if (ss->ss_server_fld != NULL) {
3538                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
3539                 ss->ss_server_fld->lsf_control_exp = NULL;
3540         }
3541 }
3542
3543 static void mdt_seq_fini_cli(struct mdt_device *mdt)
3544 {
3545         struct seq_server_site *ss = mdt_seq_site(mdt);
3546
3547         if (ss == NULL)
3548                 return;
3549
3550         if (ss->ss_server_seq == NULL)
3551                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
3552
3553         return;
3554 }
3555
3556 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
3557 {
3558         mdt_seq_fini_cli(mdt);
3559         mdt_deregister_seq_exp(mdt);
3560
3561         return seq_site_fini(env, mdt_seq_site(mdt));
3562 }
3563
3564 /**
3565  * It will retrieve its FLDB entries from MDT0, and it only happens
3566  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
3567  * and it needs to refresh FLDB from the MDT0.
3568  **/
3569 static int mdt_register_lwp_callback(void *data)
3570 {
3571         struct lu_env           env;
3572         struct mdt_device       *mdt = data;
3573         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
3574         int                     rc;
3575         ENTRY;
3576
3577         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
3578
3579         if (!likely(fld->lsf_new))
3580                 RETURN(0);
3581
3582         rc = lu_env_init(&env, LCT_MD_THREAD);
3583         if (rc) {
3584                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
3585                 RETURN(rc);
3586         }
3587
3588         rc = fld_update_from_controller(&env, fld);
3589         if (rc != 0) {
3590                 CERROR("%s: cannot update controller: rc = %d\n",
3591                        mdt_obd_name(mdt), rc);
3592                 GOTO(out, rc);
3593         }
3594 out:
3595         lu_env_fini(&env);
3596         RETURN(rc);
3597 }
3598
3599 static int mdt_register_seq_exp(struct mdt_device *mdt)
3600 {
3601         struct seq_server_site  *ss = mdt_seq_site(mdt);
3602         char                    *lwp_name = NULL;
3603         int                     rc;
3604
3605         if (ss->ss_node_id == 0)
3606                 return 0;
3607
3608         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
3609         if (lwp_name == NULL)
3610                 GOTO(out_free, rc = -ENOMEM);
3611
3612         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
3613         if (rc != 0)
3614                 GOTO(out_free, rc);
3615
3616         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
3617                                       NULL, NULL);
3618         if (rc != 0)
3619                 GOTO(out_free, rc);
3620
3621         rc = lustre_register_lwp_item(lwp_name,
3622                                       &ss->ss_server_fld->lsf_control_exp,
3623                                       mdt_register_lwp_callback, mdt);
3624         if (rc != 0) {
3625                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3626                 ss->ss_client_seq->lcs_exp = NULL;
3627                 GOTO(out_free, rc);
3628         }
3629 out_free:
3630         if (lwp_name != NULL)
3631                 OBD_FREE(lwp_name, MAX_OBD_NAME);
3632
3633         return rc;
3634 }
3635
3636 /*
3637  * Init client sequence manager which is used by local MDS to talk to sequence
3638  * controller on remote node.
3639  */
3640 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
3641 {
3642         struct seq_server_site  *ss = mdt_seq_site(mdt);
3643         int                     rc;
3644         char                    *prefix;
3645         ENTRY;
3646
3647         /* check if this is adding the first MDC and controller is not yet
3648          * initialized. */
3649         OBD_ALLOC_PTR(ss->ss_client_seq);
3650         if (ss->ss_client_seq == NULL)
3651                 RETURN(-ENOMEM);
3652
3653         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3654         if (prefix == NULL) {
3655                 OBD_FREE_PTR(ss->ss_client_seq);
3656                 ss->ss_client_seq = NULL;
3657                 RETURN(-ENOMEM);
3658         }
3659
3660         /* Note: seq_client_fini will be called in seq_site_fini */
3661         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
3662         rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
3663                              prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
3664                                                             NULL);
3665         OBD_FREE(prefix, MAX_OBD_NAME + 5);
3666         if (rc != 0) {
3667                 OBD_FREE_PTR(ss->ss_client_seq);
3668                 ss->ss_client_seq = NULL;
3669                 RETURN(rc);
3670         }
3671
3672         rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
3673
3674         RETURN(rc);
3675 }
3676
3677 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
3678 {
3679         struct seq_server_site  *ss;
3680         int                     rc;
3681         ENTRY;
3682
3683         ss = mdt_seq_site(mdt);
3684         /* init sequence controller server(MDT0) */
3685         if (ss->ss_node_id == 0) {
3686                 OBD_ALLOC_PTR(ss->ss_control_seq);
3687                 if (ss->ss_control_seq == NULL)
3688                         RETURN(-ENOMEM);
3689
3690                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
3691                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
3692                                      ss);
3693                 if (rc)
3694                         GOTO(out_seq_fini, rc);
3695         }
3696
3697         /* Init normal sequence server */
3698         OBD_ALLOC_PTR(ss->ss_server_seq);
3699         if (ss->ss_server_seq == NULL)
3700                 GOTO(out_seq_fini, rc = -ENOMEM);
3701
3702         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
3703                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
3704         if (rc)
3705                 GOTO(out_seq_fini, rc);
3706
3707         /* init seq client for seq server to talk to seq controller(MDT0) */
3708         rc = mdt_seq_init_cli(env, mdt);
3709         if (rc != 0)
3710                 GOTO(out_seq_fini, rc);
3711
3712         if (ss->ss_node_id != 0)
3713                 /* register controler export through lwp */
3714                 rc = mdt_register_seq_exp(mdt);
3715
3716         EXIT;
3717 out_seq_fini:
3718         if (rc)
3719                 mdt_seq_fini(env, mdt);
3720
3721         return rc;
3722 }
3723
3724 /*
3725  * FLD wrappers
3726  */
3727 static int mdt_fld_fini(const struct lu_env *env,
3728                         struct mdt_device *m)
3729 {
3730         struct seq_server_site *ss = mdt_seq_site(m);
3731         ENTRY;
3732
3733         if (ss && ss->ss_server_fld) {
3734                 fld_server_fini(env, ss->ss_server_fld);
3735                 OBD_FREE_PTR(ss->ss_server_fld);
3736                 ss->ss_server_fld = NULL;
3737         }
3738
3739         RETURN(0);
3740 }
3741
3742 static int mdt_fld_init(const struct lu_env *env,
3743                         const char *uuid,
3744                         struct mdt_device *m)
3745 {
3746         struct seq_server_site *ss;
3747         int rc;
3748         ENTRY;
3749
3750         ss = mdt_seq_site(m);
3751
3752         OBD_ALLOC_PTR(ss->ss_server_fld);
3753         if (ss->ss_server_fld == NULL)
3754                 RETURN(rc = -ENOMEM);
3755
3756         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
3757                              LU_SEQ_RANGE_MDT);
3758         if (rc) {
3759                 OBD_FREE_PTR(ss->ss_server_fld);
3760                 ss->ss_server_fld = NULL;
3761                 RETURN(rc);
3762         }
3763
3764         RETURN(0);
3765 }
3766
3767 static void mdt_stack_pre_fini(const struct lu_env *env,
3768                            struct mdt_device *m, struct lu_device *top)
3769 {
3770         struct obd_device       *obd;
3771         struct lustre_cfg_bufs  *bufs;
3772         struct lustre_cfg       *lcfg;
3773         struct mdt_thread_info  *info;
3774         ENTRY;
3775
3776         LASSERT(top);
3777
3778         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3779         LASSERT(info != NULL);
3780
3781         bufs = &info->mti_u.bufs;
3782
3783         LASSERT(m->mdt_child_exp);
3784         LASSERT(m->mdt_child_exp->exp_obd);
3785         obd = m->mdt_child_exp->exp_obd;
3786
3787         /* process cleanup, pass mdt obd name to get obd umount flags */
3788         /* XXX: this is needed because all layers are referenced by
3789          * objects (some of them are pinned by osd, for example *
3790          * the proper solution should be a model where object used
3791          * by osd only doesn't have mdt/mdd slices -bzzz */
3792         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
3793         lustre_cfg_bufs_set_string(bufs, 1, NULL);
3794         lcfg = lustre_cfg_new(LCFG_PRE_CLEANUP, bufs);
3795         if (!lcfg) {
3796                 CERROR("%s:Cannot alloc lcfg!\n", mdt_obd_name(m));
3797                 return;
3798         }
3799         top->ld_ops->ldo_process_config(env, top, lcfg);
3800         lustre_cfg_free(lcfg);
3801         EXIT;
3802 }
3803
3804 static void mdt_stack_fini(const struct lu_env *env,
3805                            struct mdt_device *m, struct lu_device *top)
3806 {
3807         struct obd_device       *obd = mdt2obd_dev(m);
3808         struct lustre_cfg_bufs  *bufs;
3809         struct lustre_cfg       *lcfg;
3810         struct mdt_thread_info  *info;
3811         char                     flags[3] = "";
3812         ENTRY;
3813
3814         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3815         LASSERT(info != NULL);
3816
3817         lu_dev_del_linkage(top->ld_site, top);
3818
3819         lu_site_purge(env, top->ld_site, -1);
3820
3821         bufs = &info->mti_u.bufs;
3822         /* process cleanup, pass mdt obd name to get obd umount flags */
3823         /* another purpose is to let all layers to release their objects */
3824         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
3825         if (obd->obd_force)
3826                 strcat(flags, "F");
3827         if (obd->obd_fail)
3828                 strcat(flags, "A");
3829         lustre_cfg_bufs_set_string(bufs, 1, flags);
3830         lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
3831         if (!lcfg) {
3832                 CERROR("Cannot alloc lcfg!\n");
3833                 return;
3834         }
3835         LASSERT(top);
3836         top->ld_ops->ldo_process_config(env, top, lcfg);
3837         lustre_cfg_free(lcfg);
3838
3839         lu_site_purge(env, top->ld_site, -1);
3840
3841         m->mdt_child = NULL;
3842         m->mdt_bottom = NULL;
3843
3844         obd_disconnect(m->mdt_child_exp);
3845         m->mdt_child_exp = NULL;
3846
3847         obd_disconnect(m->mdt_bottom_exp);
3848         m->mdt_child_exp = NULL;
3849 }
3850
3851 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
3852                                const char *next, struct obd_export **exp)
3853 {
3854         struct obd_connect_data *data = NULL;
3855         struct obd_device       *obd;
3856         int                      rc;
3857         ENTRY;
3858
3859         OBD_ALLOC_PTR(data);
3860         if (data == NULL)
3861                 GOTO(out, rc = -ENOMEM);
3862
3863         obd = class_name2obd(next);
3864         if (obd == NULL) {
3865                 CERROR("%s: can't locate next device: %s\n",
3866                        mdt_obd_name(m), next);
3867                 GOTO(out, rc = -ENOTCONN);
3868         }
3869
3870         data->ocd_connect_flags = OBD_CONNECT_VERSION;
3871         data->ocd_version = LUSTRE_VERSION_CODE;
3872
3873         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
3874         if (rc) {
3875                 CERROR("%s: cannot connect to next dev %s (%d)\n",
3876                        mdt_obd_name(m), next, rc);
3877                 GOTO(out, rc);
3878         }
3879
3880 out:
3881         if (data)
3882                 OBD_FREE_PTR(data);
3883         RETURN(rc);
3884 }
3885
3886 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
3887                           struct lustre_cfg *cfg)
3888 {
3889         char                   *dev = lustre_cfg_string(cfg, 0);
3890         int                     rc, name_size, uuid_size;
3891         char                   *name, *uuid, *p;
3892         struct lustre_cfg_bufs *bufs;
3893         struct lustre_cfg      *lcfg;
3894         struct obd_device      *obd;
3895         struct lustre_profile  *lprof;
3896         struct lu_site         *site;
3897         ENTRY;
3898
3899         /* in 1.8 we had the only device in the stack - MDS.
3900          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
3901          * in 2.3 OSD is instantiated by obd_mount.c, so we need
3902          * to generate names and setup MDT, MDD. MDT will be using
3903          * generated name to connect to MDD. for MDD the next device
3904          * will be LOD with name taken from so called "profile" which
3905          * is generated by mount_option line
3906          *
3907          * 1.8 MGS generates config. commands like this:
3908          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
3909          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
3910          * 2.0 MGS generates config. commands like this:
3911          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
3912          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
3913          *                    3:lustre-MDT0000-mdtlov  4:f
3914          *
3915          * we generate MDD name from MDT one, just replacing T with D
3916          *
3917          * after all the preparations, the logical equivalent will be
3918          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
3919          *                    3:lustre-MDT0000-mdtlov  4:f
3920          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
3921          *                    3:lustre-MDD0000  4:f
3922          *
3923          *  notice we build the stack from down to top: MDD first, then MDT */
3924
3925         name_size = MAX_OBD_NAME;
3926         uuid_size = MAX_OBD_NAME;
3927
3928         OBD_ALLOC(name, name_size);
3929         OBD_ALLOC(uuid, uuid_size);
3930         if (name == NULL || uuid == NULL)
3931                 GOTO(cleanup_mem, rc = -ENOMEM);
3932
3933         OBD_ALLOC_PTR(bufs);
3934         if (!bufs)
3935                 GOTO(cleanup_mem, rc = -ENOMEM);
3936
3937         strcpy(name, dev);
3938         p = strstr(name, "-MDT");
3939         if (p == NULL)
3940                 GOTO(free_bufs, rc = -ENOMEM);
3941         p[3] = 'D';
3942
3943         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
3944
3945         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
3946         if (lprof == NULL || lprof->lp_dt == NULL) {
3947                 CERROR("can't find the profile: %s\n",
3948                        lustre_cfg_string(cfg, 0));
3949                 GOTO(free_bufs, rc = -EINVAL);
3950         }
3951
3952         lustre_cfg_bufs_reset(bufs, name);
3953         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
3954         lustre_cfg_bufs_set_string(bufs, 2, uuid);
3955         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
3956
3957         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
3958         if (!lcfg)
3959                 GOTO(free_bufs, rc = -ENOMEM);
3960
3961         rc = class_attach(lcfg);
3962         if (rc)
3963                 GOTO(lcfg_cleanup, rc);
3964
3965         obd = class_name2obd(name);
3966         if (!obd) {
3967                 CERROR("Can not find obd %s (%s in config)\n",
3968                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
3969                 GOTO(class_detach, rc = -EINVAL);
3970         }
3971
3972         lustre_cfg_free(lcfg);
3973
3974         lustre_cfg_bufs_reset(bufs, name);
3975         lustre_cfg_bufs_set_string(bufs, 1, uuid);
3976         lustre_cfg_bufs_set_string(bufs, 2, dev);
3977         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
3978
3979         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
3980
3981         rc = class_setup(obd, lcfg);
3982         if (rc)
3983                 GOTO(class_detach, rc);
3984
3985         /* connect to MDD we just setup */
3986         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
3987         if (rc)
3988                 GOTO(class_detach, rc);
3989
3990         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
3991         LASSERT(site);
3992         LASSERT(mdt_lu_site(mdt) == NULL);
3993         mdt->mdt_lu_dev.ld_site = site;
3994         site->ls_top_dev = &mdt->mdt_lu_dev;
3995         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
3996
3997         /* now connect to bottom OSD */
3998         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
3999         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
4000         if (rc)
4001                 GOTO(class_detach, rc);
4002         mdt->mdt_bottom =
4003                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
4004
4005         rc = lu_env_refill((struct lu_env *)env);
4006         if (rc != 0)
4007                 CERROR("Failure to refill session: '%d'\n", rc);
4008
4009         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
4010
4011         EXIT;
4012 class_detach:
4013         if (rc)
4014                 class_detach(obd, lcfg);
4015 lcfg_cleanup:
4016         lustre_cfg_free(lcfg);
4017 free_bufs:
4018         OBD_FREE_PTR(bufs);
4019 cleanup_mem:
4020         if (name)
4021                 OBD_FREE(name, name_size);
4022         if (uuid)
4023                 OBD_FREE(uuid, uuid_size);
4024         RETURN(rc);
4025 }
4026
4027 /* setup quota master target on MDT0 */
4028 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
4029                           struct lustre_cfg *cfg)
4030 {
4031         struct obd_device       *obd;
4032         char                    *dev = lustre_cfg_string(cfg, 0);
4033         char                    *qmtname, *uuid, *p;
4034         struct lustre_cfg_bufs  *bufs;
4035         struct lustre_cfg       *lcfg;
4036         struct lustre_profile   *lprof;
4037         struct obd_connect_data *data;
4038         int                      rc;
4039         ENTRY;
4040
4041         LASSERT(mdt->mdt_qmt_exp == NULL);
4042         LASSERT(mdt->mdt_qmt_dev == NULL);
4043
4044         /* quota master is on MDT0 only for now */
4045         if (mdt->mdt_seq_site.ss_node_id != 0)
4046                 RETURN(0);
4047
4048         /* MGS generates config commands which look as follows:
4049          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4050          *                    3:lustre-MDT0000-mdtlov  4:f
4051          *
4052          * We generate the QMT name from the MDT one, just replacing MD with QM
4053          * after all the preparations, the logical equivalent will be:
4054          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
4055          *                    3:lustre-MDT0000-osd  4:f */
4056         OBD_ALLOC(qmtname, MAX_OBD_NAME);
4057         OBD_ALLOC(uuid, UUID_MAX);
4058         OBD_ALLOC_PTR(bufs);
4059         OBD_ALLOC_PTR(data);
4060         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
4061                 GOTO(cleanup_mem, rc = -ENOMEM);
4062
4063         strcpy(qmtname, dev);
4064         p = strstr(qmtname, "-MDT");
4065         if (p == NULL)
4066                 GOTO(cleanup_mem, rc = -ENOMEM);
4067         /* replace MD with QM */
4068         p[1] = 'Q';
4069         p[2] = 'M';
4070
4071         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
4072
4073         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4074         if (lprof == NULL || lprof->lp_dt == NULL) {
4075                 CERROR("can't find profile for %s\n",
4076                        lustre_cfg_string(cfg, 0));
4077                 GOTO(cleanup_mem, rc = -EINVAL);
4078         }
4079
4080         lustre_cfg_bufs_reset(bufs, qmtname);
4081         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
4082         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4083         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4084
4085         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4086         if (!lcfg)
4087                 GOTO(cleanup_mem, rc = -ENOMEM);
4088
4089         rc = class_attach(lcfg);
4090         if (rc)
4091                 GOTO(lcfg_cleanup, rc);
4092
4093         obd = class_name2obd(qmtname);
4094         if (!obd) {
4095                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
4096                        lustre_cfg_string(cfg, 0));
4097                 GOTO(class_detach, rc = -EINVAL);
4098         }
4099
4100         lustre_cfg_free(lcfg);
4101
4102         lustre_cfg_bufs_reset(bufs, qmtname);
4103         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4104         lustre_cfg_bufs_set_string(bufs, 2, dev);
4105
4106         /* for quota, the next device should be the OSD device */
4107         lustre_cfg_bufs_set_string(bufs, 3,
4108                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
4109
4110         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4111
4112         rc = class_setup(obd, lcfg);
4113         if (rc)
4114                 GOTO(class_detach, rc);
4115
4116         mdt->mdt_qmt_dev = obd->obd_lu_dev;
4117
4118         /* configure local quota objects */
4119         rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
4120                                                    &mdt->mdt_lu_dev,
4121                                                    mdt->mdt_qmt_dev);
4122         if (rc)
4123                 GOTO(class_cleanup, rc);
4124
4125         /* connect to quota master target */
4126         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4127         data->ocd_version = LUSTRE_VERSION_CODE;
4128         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
4129                          data, NULL);
4130         if (rc) {
4131                 CERROR("cannot connect to quota master device %s (%d)\n",
4132                        qmtname, rc);
4133                 GOTO(class_cleanup, rc);
4134         }
4135
4136         EXIT;
4137 class_cleanup:
4138         if (rc) {
4139                 class_manual_cleanup(obd);
4140                 mdt->mdt_qmt_dev = NULL;
4141         }
4142 class_detach:
4143         if (rc)
4144                 class_detach(obd, lcfg);
4145 lcfg_cleanup:
4146         lustre_cfg_free(lcfg);
4147 cleanup_mem:
4148         if (bufs)
4149                 OBD_FREE_PTR(bufs);
4150         if (qmtname)
4151                 OBD_FREE(qmtname, MAX_OBD_NAME);
4152         if (uuid)
4153                 OBD_FREE(uuid, UUID_MAX);
4154         if (data)
4155                 OBD_FREE_PTR(data);
4156         return rc;
4157 }
4158
4159 /* Shutdown quota master target associated with mdt */
4160 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
4161 {
4162         ENTRY;
4163
4164         if (mdt->mdt_qmt_exp == NULL)
4165                 RETURN_EXIT;
4166         LASSERT(mdt->mdt_qmt_dev != NULL);
4167
4168         /* the qmt automatically shuts down when the mdt disconnects */
4169         obd_disconnect(mdt->mdt_qmt_exp);
4170         mdt->mdt_qmt_exp = NULL;
4171         mdt->mdt_qmt_dev = NULL;
4172         EXIT;
4173 }
4174
4175 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
4176  * for now. This will be removed along with converting rest of MDT code
4177  * to use tgt_session_info */
4178 int mdt_tgt_getxattr(struct tgt_session_info *tsi)
4179 {
4180         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
4181         int                      rc;
4182
4183         rc = mdt_getxattr(info);
4184
4185         mdt_thread_info_fini(info);
4186         return rc;
4187 }
4188
4189 static struct tgt_handler mdt_tgt_handlers[] = {
4190 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4191                 0,                      MDS_CONNECT,    mdt_tgt_connect,
4192                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
4193 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4194                 0,                      MDS_DISCONNECT, tgt_disconnect,
4195                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
4196 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4197                 HABEO_REFERO,           MDS_SET_INFO,   mdt_set_info,
4198                 &RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION),
4199 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
4200 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_GETSTATUS,  mdt_getstatus),
4201 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETATTR,    mdt_getattr),
4202 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_GETATTR_NAME,
4203                                                         mdt_getattr_name),
4204 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETXATTR,   mdt_tgt_getxattr),
4205 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_STATFS,     mdt_statfs),
4206 TGT_MDT_HDL(0           | MUTABOR,      MDS_REINT,      mdt_reint),
4207 TGT_MDT_HDL(HABEO_CORPUS,               MDS_CLOSE,      mdt_close),
4208 TGT_MDT_HDL(HABEO_CORPUS,               MDS_DONE_WRITING,
4209                                                         mdt_done_writing),
4210 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_READPAGE,   mdt_readpage),
4211 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_SYNC,       mdt_sync),
4212 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_IS_SUBDIR,  mdt_is_subdir),
4213 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
4214 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_PROGRESS,
4215                                                         mdt_hsm_progress),
4216 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_REGISTER,
4217                                                         mdt_hsm_ct_register),
4218 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_UNREGISTER,
4219                                                         mdt_hsm_ct_unregister),
4220 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_STATE_GET,
4221                                                         mdt_hsm_state_get),
4222 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_STATE_SET,
4223                                                         mdt_hsm_state_set),
4224 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_ACTION, mdt_hsm_action),
4225 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_REQUEST,
4226                                                         mdt_hsm_request),
4227 TGT_MDT_HDL(HABEO_CLAVIS | HABEO_CORPUS | HABEO_REFERO | MUTABOR,
4228             MDS_SWAP_LAYOUTS,
4229             mdt_swap_layouts),
4230 };
4231
4232 static struct tgt_handler mdt_sec_ctx_ops[] = {
4233 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
4234 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
4235 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
4236 };
4237
4238 static struct tgt_handler mdt_quota_ops[] = {
4239 TGT_QUOTA_HDL(HABEO_REFERO,             QUOTA_DQACQ,      mdt_quota_dqacq),
4240 };
4241
4242 static struct tgt_opc_slice mdt_common_slice[] = {
4243         {
4244                 .tos_opc_start  = MDS_FIRST_OPC,
4245                 .tos_opc_end    = MDS_LAST_OPC,
4246                 .tos_hs         = mdt_tgt_handlers
4247         },
4248         {
4249                 .tos_opc_start  = OBD_FIRST_OPC,
4250                 .tos_opc_end    = OBD_LAST_OPC,
4251                 .tos_hs         = tgt_obd_handlers
4252         },
4253         {
4254                 .tos_opc_start  = LDLM_FIRST_OPC,
4255                 .tos_opc_end    = LDLM_LAST_OPC,
4256                 .tos_hs         = tgt_dlm_handlers
4257         },
4258         {
4259                 .tos_opc_start  = SEC_FIRST_OPC,
4260                 .tos_opc_end    = SEC_LAST_OPC,
4261                 .tos_hs         = mdt_sec_ctx_ops
4262         },
4263         {
4264                 .tos_opc_start  = OUT_UPDATE_FIRST_OPC,
4265                 .tos_opc_end    = OUT_UPDATE_LAST_OPC,
4266                 .tos_hs         = tgt_out_handlers
4267         },
4268         {
4269                 .tos_opc_start  = FLD_FIRST_OPC,
4270                 .tos_opc_end    = FLD_LAST_OPC,
4271                 .tos_hs         = fld_handlers
4272         },
4273         {
4274                 .tos_opc_start  = SEQ_FIRST_OPC,
4275                 .tos_opc_end    = SEQ_LAST_OPC,
4276                 .tos_hs         = seq_handlers
4277         },
4278         {
4279                 .tos_opc_start  = QUOTA_DQACQ,
4280                 .tos_opc_end    = QUOTA_LAST_OPC,
4281                 .tos_hs         = mdt_quota_ops
4282         },
4283         {
4284                 .tos_opc_start  = LLOG_FIRST_OPC,
4285                 .tos_opc_end    = LLOG_LAST_OPC,
4286                 .tos_hs         = tgt_llog_handlers
4287         },
4288         {
4289                 .tos_opc_start  = LFSCK_FIRST_OPC,
4290                 .tos_opc_end    = LFSCK_LAST_OPC,
4291                 .tos_hs         = tgt_lfsck_handlers
4292         },
4293
4294         {
4295                 .tos_hs         = NULL
4296         }
4297 };
4298
4299 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4300 {
4301         struct md_device        *next = m->mdt_child;
4302         struct lu_device        *d    = &m->mdt_lu_dev;
4303         struct obd_device       *obd  = mdt2obd_dev(m);
4304         struct lfsck_stop        stop;
4305         ENTRY;
4306
4307         target_recovery_fini(obd);
4308         ping_evictor_stop();
4309         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
4310
4311         if (m->mdt_opts.mo_coordinator)
4312                 mdt_hsm_cdt_stop(m);
4313
4314         mdt_hsm_cdt_fini(m);
4315
4316         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
4317         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4318         obd_exports_barrier(obd);
4319         obd_zombie_barrier();
4320
4321         mdt_procfs_fini(m);
4322
4323         tgt_fini(env, &m->mdt_lut);
4324         mdt_fs_cleanup(env, m);
4325         upcall_cache_cleanup(m->mdt_identity_cache);
4326         m->mdt_identity_cache = NULL;
4327
4328         if (m->mdt_namespace != NULL) {
4329                 ldlm_namespace_free(m->mdt_namespace, NULL,
4330                                     d->ld_obd->obd_force);
4331                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4332         }
4333
4334         mdt_quota_fini(env, m);
4335
4336         cfs_free_nidlist(&m->mdt_nosquash_nids);
4337         if (m->mdt_nosquash_str) {
4338                 OBD_FREE(m->mdt_nosquash_str, m->mdt_nosquash_strlen);
4339                 m->mdt_nosquash_str = NULL;
4340                 m->mdt_nosquash_strlen = 0;
4341         }
4342
4343         stop.ls_status = LS_PAUSED;
4344         stop.ls_flags = 0;
4345         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
4346
4347         mdt_seq_fini(env, m);
4348         mdt_fld_fini(env, m);
4349
4350         next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL);
4351         cfs_timer_disarm(&m->mdt_ck_timer);
4352         mdt_ck_thread_stop(m);
4353
4354         /*
4355          * Finish the stack
4356          */
4357         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4358
4359         LASSERT(atomic_read(&d->ld_ref) == 0);
4360
4361         server_put_mount(mdt_obd_name(m));
4362
4363         EXIT;
4364 }
4365
4366 int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4367
4368 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4369                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
4370 {
4371         struct mdt_thread_info    *info;
4372         struct obd_device         *obd;
4373         const char                *dev = lustre_cfg_string(cfg, 0);
4374         const char                *num = lustre_cfg_string(cfg, 2);
4375         struct lustre_mount_info  *lmi = NULL;
4376         struct lustre_sb_info     *lsi;
4377         struct lu_site            *s;
4378         struct seq_server_site    *ss_site;
4379         const char                *identity_upcall = "NONE";
4380         struct md_device          *next;
4381         int                        rc;
4382         long                       node_id;
4383         mntopt_t                   mntopts;
4384         ENTRY;
4385
4386         lu_device_init(&m->mdt_lu_dev, ldt);
4387         /*
4388          * Environment (env) might be missing mdt_thread_key values at that
4389          * point, if device is allocated when mdt_thread_key is in QUIESCENT
4390          * mode.
4391          *
4392          * Usually device allocation path doesn't use module key values, but
4393          * mdt has to do a lot of work here, so allocate key value.
4394          */
4395         rc = lu_env_refill((struct lu_env *)env);
4396         if (rc != 0)
4397                 RETURN(rc);
4398
4399         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4400         LASSERT(info != NULL);
4401
4402         obd = class_name2obd(dev);
4403         LASSERT(obd != NULL);
4404
4405         m->mdt_max_mdsize = MAX_MD_SIZE; /* 4 stripes */
4406
4407         m->mdt_som_conf = 0;
4408
4409         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4410
4411         /* default is coordinator off, it is started through conf_param
4412          * or /proc */
4413         m->mdt_opts.mo_coordinator = 0;
4414
4415         lmi = server_get_mount(dev);
4416         if (lmi == NULL) {
4417                 CERROR("Cannot get mount info for %s!\n", dev);
4418                 RETURN(-EFAULT);
4419         } else {
4420                 lsi = s2lsi(lmi->lmi_sb);
4421                 /* CMD is supported only in IAM mode */
4422                 LASSERT(num);
4423                 node_id = simple_strtol(num, NULL, 10);
4424                 obd->u.obt.obt_magic = OBT_MAGIC;
4425         }
4426
4427         spin_lock_init(&m->mdt_ioepoch_lock);
4428         m->mdt_capa_timeout = CAPA_TIMEOUT;
4429         m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1;
4430         m->mdt_ck_timeout = CAPA_KEY_TIMEOUT;
4431         m->mdt_squash_uid = 0;
4432         m->mdt_squash_gid = 0;
4433         CFS_INIT_LIST_HEAD(&m->mdt_nosquash_nids);
4434         m->mdt_nosquash_str = NULL;
4435         m->mdt_nosquash_strlen = 0;
4436         init_rwsem(&m->mdt_squash_sem);
4437         spin_lock_init(&m->mdt_osfs_lock);
4438         m->mdt_osfs_age = cfs_time_shift_64(-1000);
4439         m->mdt_enable_remote_dir = 0;
4440         m->mdt_enable_remote_dir_gid = 0;
4441
4442         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
4443         m->mdt_lu_dev.ld_obd = obd;
4444         /* Set this lu_device to obd for error handling purposes. */
4445         obd->obd_lu_dev = &m->mdt_lu_dev;
4446
4447         /* init the stack */
4448         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
4449         if (rc) {
4450                 CERROR("%s: Can't init device stack, rc %d\n",
4451                        mdt_obd_name(m), rc);
4452                 GOTO(err_lmi, rc);
4453         }
4454
4455         s = mdt_lu_site(m);
4456         ss_site = mdt_seq_site(m);
4457         s->ld_seq_site = ss_site;
4458         ss_site->ss_lu = s;
4459
4460         /* set server index */
4461         ss_site->ss_node_id = node_id;
4462
4463         /* failover is the default
4464          * FIXME: we do not failout mds0/mgs, which may cause some problems.
4465          * assumed whose ss_node_id == 0 XXX
4466          * */
4467         obd->obd_replayable = 1;
4468         /* No connection accepted until configurations will finish */
4469         obd->obd_no_conn = 1;
4470
4471         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4472                 char *str = lustre_cfg_string(cfg, 4);
4473                 if (strchr(str, 'n')) {
4474                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
4475                         obd->obd_replayable = 0;
4476                 }
4477         }
4478
4479         rc = mdt_fld_init(env, mdt_obd_name(m), m);
4480         if (rc)
4481                 GOTO(err_fini_stack, rc);
4482
4483         rc = mdt_seq_init(env, m);
4484         if (rc)
4485                 GOTO(err_fini_fld, rc);
4486
4487         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
4488                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
4489         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4490                                               LDLM_NAMESPACE_SERVER,
4491                                               LDLM_NAMESPACE_GREEDY,
4492                                               LDLM_NS_TYPE_MDT);
4493         if (m->mdt_namespace == NULL)
4494                 GOTO(err_fini_seq, rc = -ENOMEM);
4495
4496         m->mdt_namespace->ns_lvbp = m;
4497         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
4498
4499         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4500         /* set obd_namespace for compatibility with old code */
4501         obd->obd_namespace = m->mdt_namespace;
4502
4503         cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m);
4504
4505         rc = mdt_hsm_cdt_init(m);
4506         if (rc != 0) {
4507                 CERROR("%s: error initializing coordinator, rc %d\n",
4508                        mdt_obd_name(m), rc);
4509                 GOTO(err_free_ns, rc);
4510         }
4511
4512         rc = mdt_ck_thread_start(m);
4513         if (rc)
4514                 GOTO(err_free_hsm, rc);
4515
4516         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
4517                       OBD_FAIL_MDS_ALL_REQUEST_NET,
4518                       OBD_FAIL_MDS_ALL_REPLY_NET);
4519         if (rc)
4520                 GOTO(err_capa, rc);
4521
4522         rc = mdt_fs_setup(env, m, obd, lsi);
4523         if (rc)
4524                 GOTO(err_tgt, rc);
4525
4526         tgt_adapt_sptlrpc_conf(&m->mdt_lut, 1);
4527
4528         next = m->mdt_child;
4529         rc = next->md_ops->mdo_iocontrol(env, next, OBD_IOC_GET_MNTOPT, 0,
4530                                          &mntopts);
4531         if (rc)
4532                 GOTO(err_fs_cleanup, rc);
4533
4534         if (mntopts & MNTOPT_USERXATTR)
4535                 m->mdt_opts.mo_user_xattr = 1;
4536         else
4537                 m->mdt_opts.mo_user_xattr = 0;
4538
4539         rc = next->md_ops->mdo_maxeasize_get(env, next, &m->mdt_max_ea_size);
4540         if (rc)
4541                 GOTO(err_fs_cleanup, rc);
4542
4543         if (mntopts & MNTOPT_ACL)
4544                 m->mdt_opts.mo_acl = 1;
4545         else
4546                 m->mdt_opts.mo_acl = 0;
4547
4548         /* XXX: to support suppgid for ACL, we enable identity_upcall
4549          * by default, otherwise, maybe got unexpected -EACCESS. */
4550         if (m->mdt_opts.mo_acl)
4551                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4552
4553         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
4554                                                 identity_upcall,
4555                                                 &mdt_identity_upcall_cache_ops);
4556         if (IS_ERR(m->mdt_identity_cache)) {
4557                 rc = PTR_ERR(m->mdt_identity_cache);
4558                 m->mdt_identity_cache = NULL;
4559                 GOTO(err_fs_cleanup, rc);
4560         }
4561
4562         rc = mdt_procfs_init(m, dev);
4563         if (rc) {
4564                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4565                 GOTO(err_recovery, rc);
4566         }
4567
4568         rc = mdt_quota_init(env, m, cfg);
4569         if (rc)
4570                 GOTO(err_procfs, rc);
4571
4572         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
4573         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
4574                            "mdt_ldlm_client", m->mdt_ldlm_client);
4575
4576         ping_evictor_start();
4577
4578         /* recovery will be started upon mdt_prepare()
4579          * when the whole stack is complete and ready
4580          * to serve the requests */
4581
4582         mdt_init_capa_ctxt(env, m);
4583
4584         /* Reduce the initial timeout on an MDS because it doesn't need such
4585          * a long timeout as an OST does. Adaptive timeouts will adjust this
4586          * value appropriately. */
4587         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4588                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4589
4590         RETURN(0);
4591 err_procfs:
4592         mdt_procfs_fini(m);
4593 err_recovery:
4594         target_recovery_fini(obd);
4595         upcall_cache_cleanup(m->mdt_identity_cache);
4596         m->mdt_identity_cache = NULL;
4597 err_fs_cleanup:
4598         mdt_fs_cleanup(env, m);
4599 err_tgt:
4600         tgt_fini(env, &m->mdt_lut);
4601 err_capa:
4602         cfs_timer_disarm(&m->mdt_ck_timer);
4603         mdt_ck_thread_stop(m);
4604 err_free_hsm:
4605         mdt_hsm_cdt_fini(m);
4606 err_free_ns:
4607         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4608         obd->obd_namespace = m->mdt_namespace = NULL;
4609 err_fini_seq:
4610         mdt_seq_fini(env, m);
4611 err_fini_fld:
4612         mdt_fld_fini(env, m);
4613 err_fini_stack:
4614         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4615 err_lmi:
4616         if (lmi)
4617                 server_put_mount(dev);
4618         return(rc);
4619 }
4620
4621 /* For interoperability, the left element is old parameter, the right one
4622  * is the new version of the parameter, if some parameter is deprecated,
4623  * the new version should be set as NULL. */
4624 static struct cfg_interop_param mdt_interop_param[] = {
4625         { "mdt.group_upcall",   NULL },
4626         { "mdt.quota_type",     NULL },
4627         { "mdd.quota_type",     NULL },
4628         { "mdt.rootsquash",     "mdt.root_squash" },
4629         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
4630         { NULL }
4631 };
4632
4633 /* used by MGS to process specific configurations */
4634 static int mdt_process_config(const struct lu_env *env,
4635                               struct lu_device *d, struct lustre_cfg *cfg)
4636 {
4637         struct mdt_device *m = mdt_dev(d);
4638         struct md_device *md_next = m->mdt_child;
4639         struct lu_device *next = md2lu_dev(md_next);
4640         int rc;
4641         ENTRY;
4642
4643         switch (cfg->lcfg_command) {
4644         case LCFG_PARAM: {
4645                 struct lprocfs_static_vars  lvars;
4646                 struct obd_device          *obd = d->ld_obd;
4647
4648                 /* For interoperability */
4649                 struct cfg_interop_param   *ptr = NULL;
4650                 struct lustre_cfg          *old_cfg = NULL;
4651                 char                       *param = NULL;
4652
4653                 param = lustre_cfg_string(cfg, 1);
4654                 if (param == NULL) {
4655                         CERROR("param is empty\n");
4656                         rc = -EINVAL;
4657                         break;
4658                 }
4659
4660                 ptr = class_find_old_param(param, mdt_interop_param);
4661                 if (ptr != NULL) {
4662                         if (ptr->new_param == NULL) {
4663                                 rc = 0;
4664                                 CWARN("For interoperability, skip this %s."
4665                                       " It is obsolete.\n", ptr->old_param);
4666                                 break;
4667                         }
4668
4669                         CWARN("Found old param %s, changed it to %s.\n",
4670                               ptr->old_param, ptr->new_param);
4671
4672                         old_cfg = cfg;
4673                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
4674                         if (IS_ERR(cfg)) {
4675                                 rc = PTR_ERR(cfg);
4676                                 break;
4677                         }
4678                 }
4679
4680                 lprocfs_mdt_init_vars(&lvars);
4681                 rc = class_process_proc_param(PARAM_MDT, lvars.obd_vars,
4682                                               cfg, obd);
4683                 if (rc > 0 || rc == -ENOSYS) {
4684                         /* is it an HSM var ? */
4685                         rc = class_process_proc_param(PARAM_HSM,
4686                                                       hsm_cdt_get_proc_vars(),
4687                                                       cfg, obd);
4688                         if (rc > 0 || rc == -ENOSYS)
4689                                 /* we don't understand; pass it on */
4690                                 rc = next->ld_ops->ldo_process_config(env, next,
4691                                                                       cfg);
4692                 }
4693
4694                 if (old_cfg != NULL)
4695                         lustre_cfg_free(cfg);
4696
4697                 break;
4698         }
4699         default:
4700                 /* others are passed further */
4701                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4702                 break;
4703         }
4704         RETURN(rc);
4705 }
4706
4707 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4708                                           const struct lu_object_header *hdr,
4709                                           struct lu_device *d)
4710 {
4711         struct mdt_object *mo;
4712
4713         ENTRY;
4714
4715         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
4716         if (mo != NULL) {
4717                 struct lu_object *o;
4718                 struct lu_object_header *h;
4719
4720                 o = &mo->mot_obj;
4721                 h = &mo->mot_header;
4722                 lu_object_header_init(h);
4723                 lu_object_init(o, h, d);
4724                 lu_object_add_top(h, o);
4725                 o->lo_ops = &mdt_obj_ops;
4726                 mutex_init(&mo->mot_ioepoch_mutex);
4727                 mutex_init(&mo->mot_lov_mutex);
4728                 init_rwsem(&mo->mot_open_sem);
4729                 RETURN(o);
4730         }
4731         RETURN(NULL);
4732 }
4733
4734 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
4735                            const struct lu_object_conf *unused)
4736 {
4737         struct mdt_device *d = mdt_dev(o->lo_dev);
4738         struct lu_device  *under;
4739         struct lu_object  *below;
4740         int                rc = 0;
4741         ENTRY;
4742
4743         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4744                PFID(lu_object_fid(o)));
4745
4746         under = &d->mdt_child->md_lu_dev;
4747         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4748         if (below != NULL) {
4749                 lu_object_add(o, below);
4750         } else
4751                 rc = -ENOMEM;
4752
4753         RETURN(rc);
4754 }
4755
4756 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4757 {
4758         struct mdt_object *mo = mdt_obj(o);
4759         struct lu_object_header *h;
4760         ENTRY;
4761
4762         h = o->lo_header;
4763         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4764                PFID(lu_object_fid(o)));
4765
4766         LASSERT(atomic_read(&mo->mot_open_count) == 0);
4767         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
4768
4769         lu_object_fini(o);
4770         lu_object_header_fini(h);
4771         OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
4772
4773         EXIT;
4774 }
4775
4776 static int mdt_object_print(const struct lu_env *env, void *cookie,
4777                             lu_printer_t p, const struct lu_object *o)
4778 {
4779         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
4780         return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p(ioepoch="LPU64" "
4781                     "flags="LPX64", epochcount=%d, writecount=%d)",
4782                     mdto, mdto->mot_ioepoch, mdto->mot_flags,
4783                     mdto->mot_ioepoch_count, mdto->mot_writecount);
4784 }
4785
4786 static int mdt_prepare(const struct lu_env *env,
4787                 struct lu_device *pdev,
4788                 struct lu_device *cdev)
4789 {
4790         struct mdt_device *mdt = mdt_dev(cdev);
4791         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
4792         struct obd_device *obd = cdev->ld_obd;
4793         struct lfsck_start_param lsp;
4794         int rc;
4795
4796         ENTRY;
4797
4798         LASSERT(obd);
4799
4800         rc = next->ld_ops->ldo_prepare(env, cdev, next);
4801         if (rc)
4802                 RETURN(rc);
4803
4804         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
4805         if (rc)
4806                 RETURN(rc);
4807
4808         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
4809         if (rc)
4810                 RETURN(rc);
4811
4812         rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
4813         /* The LFSCK instance is registered just now, so it must be there when
4814          * register the namespace to such instance. */
4815         LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
4816
4817         lsp.lsp_start = NULL;
4818         lsp.lsp_index_valid = 0;
4819         rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
4820                                                    OBD_IOC_START_LFSCK,
4821                                                    0, &lsp);
4822         if (rc != 0) {
4823                 CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
4824                       mdt_obd_name(mdt), rc);
4825                 rc = 0;
4826         }
4827
4828         if (mdt->mdt_seq_site.ss_node_id == 0) {
4829                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
4830                                                          &mdt->mdt_md_root_fid);
4831                 if (rc)
4832                         RETURN(rc);
4833         }
4834
4835         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
4836         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
4837         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
4838         LASSERT(obd->obd_no_conn);
4839         spin_lock(&obd->obd_dev_lock);
4840         obd->obd_no_conn = 0;
4841         spin_unlock(&obd->obd_dev_lock);
4842
4843         if (obd->obd_recovering == 0)
4844                 mdt_postrecov(env, mdt);
4845
4846         RETURN(rc);
4847 }
4848
4849 const struct lu_device_operations mdt_lu_ops = {
4850         .ldo_object_alloc   = mdt_object_alloc,
4851         .ldo_process_config = mdt_process_config,
4852         .ldo_prepare        = mdt_prepare,
4853 };
4854
4855 static const struct lu_object_operations mdt_obj_ops = {
4856         .loo_object_init    = mdt_object_init,
4857         .loo_object_free    = mdt_object_free,
4858         .loo_object_print   = mdt_object_print
4859 };
4860
4861 static int mdt_obd_set_info_async(const struct lu_env *env,
4862                                   struct obd_export *exp,
4863                                   __u32 keylen, void *key,
4864                                   __u32 vallen, void *val,
4865                                   struct ptlrpc_request_set *set)
4866 {
4867         int rc;
4868
4869         ENTRY;
4870
4871         if (KEY_IS(KEY_SPTLRPC_CONF)) {
4872                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp), 0);
4873                 RETURN(rc);
4874         }
4875
4876         RETURN(0);
4877 }
4878
4879 /**
4880  * Match client and server connection feature flags.
4881  *
4882  * Compute the compatibility flags for a connection request based on
4883  * features mutually supported by client and server.
4884  *
4885  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
4886  * must not be updated here, otherwise a partially initialized value may
4887  * be exposed. After the connection request is successfully processed,
4888  * the top-level MDT connect request handler atomically updates the export
4889  * connect flags from the obd_connect_data::ocd_connect_flags field of the
4890  * reply. \see mdt_connect().
4891  *
4892  * \param exp   the obd_export associated with this client/target pair
4893  * \param mdt   the target device for the connection
4894  * \param data  stores data for this connect request
4895  *
4896  * \retval 0       success
4897  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
4898  * \retval -EBADE  client and server feature requirements are incompatible
4899  */
4900 static int mdt_connect_internal(struct obd_export *exp,
4901                                 struct mdt_device *mdt,
4902                                 struct obd_connect_data *data)
4903 {
4904         LASSERT(data != NULL);
4905
4906         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
4907         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
4908
4909         /* If no known bits (which should not happen, probably,
4910            as everybody should support LOOKUP and UPDATE bits at least)
4911            revert to compat mode with plain locks. */
4912         if (!data->ocd_ibits_known &&
4913             data->ocd_connect_flags & OBD_CONNECT_IBITS)
4914                 data->ocd_connect_flags &= ~OBD_CONNECT_IBITS;
4915
4916         if (!mdt->mdt_opts.mo_acl)
4917                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
4918
4919         if (!mdt->mdt_opts.mo_user_xattr)
4920                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
4921
4922         if (!mdt->mdt_som_conf)
4923                 data->ocd_connect_flags &= ~OBD_CONNECT_SOM;
4924
4925         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
4926                 data->ocd_brw_size = min(data->ocd_brw_size,
4927                                          (__u32)MD_MAX_BRW_SIZE);
4928                 if (data->ocd_brw_size == 0) {
4929                         CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
4930                                " ocd_version: %x ocd_grant: %d "
4931                                "ocd_index: %u ocd_brw_size is "
4932                                "unexpectedly zero, network data "
4933                                "corruption? Refusing connection of this"
4934                                " client\n",
4935                                mdt_obd_name(mdt),
4936                                exp->exp_client_uuid.uuid,
4937                                exp, data->ocd_connect_flags, data->ocd_version,
4938                                data->ocd_grant, data->ocd_index);
4939                         return -EPROTO;
4940                 }
4941         }
4942
4943         /* NB: Disregard the rule against updating
4944          * exp_connect_data.ocd_connect_flags in this case, since
4945          * tgt_client_new() needs to know if this is a lightweight
4946          * connection, and it is safe to expose this flag before
4947          * connection processing completes. */
4948         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
4949                 spin_lock(&exp->exp_lock);
4950                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
4951                 spin_unlock(&exp->exp_lock);
4952         }
4953
4954         data->ocd_version = LUSTRE_VERSION_CODE;
4955
4956         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
4957                 CWARN("%s: MDS requires FID support, but client not\n",
4958                       mdt_obd_name(mdt));
4959                 return -EBADE;
4960         }
4961
4962         if (mdt->mdt_som_conf &&
4963             !(data->ocd_connect_flags & (OBD_CONNECT_LIGHTWEIGHT |
4964                                          OBD_CONNECT_MDS_MDS |
4965                                          OBD_CONNECT_SOM))) {
4966                 CWARN("%s: MDS has SOM enabled, but client does not support "
4967                       "it\n", mdt_obd_name(mdt));
4968                 return -EBADE;
4969         }
4970
4971         if (OCD_HAS_FLAG(data, PINGLESS)) {
4972                 if (ptlrpc_pinger_suppress_pings()) {
4973                         spin_lock(&exp->exp_obd->obd_dev_lock);
4974                         list_del_init(&exp->exp_obd_chain_timed);
4975                         spin_unlock(&exp->exp_obd->obd_dev_lock);
4976                 } else {
4977                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
4978                 }
4979         }
4980
4981         data->ocd_max_easize = mdt->mdt_max_ea_size;
4982
4983         return 0;
4984 }
4985
4986 /* mds_connect copy */
4987 static int mdt_obd_connect(const struct lu_env *env,
4988                            struct obd_export **exp, struct obd_device *obd,
4989                            struct obd_uuid *cluuid,
4990                            struct obd_connect_data *data,
4991                            void *localdata)
4992 {
4993         struct obd_export      *lexp;
4994         struct lustre_handle    conn = { 0 };
4995         struct mdt_device      *mdt;
4996         int                     rc;
4997         ENTRY;
4998
4999         LASSERT(env != NULL);
5000         if (!exp || !obd || !cluuid)
5001                 RETURN(-EINVAL);
5002
5003         mdt = mdt_dev(obd->obd_lu_dev);
5004
5005         /*
5006          * first, check whether the stack is ready to handle requests
5007          * XXX: probably not very appropriate method is used now
5008          *      at some point we should find a better one
5009          */
5010         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) && data != NULL &&
5011             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
5012                 rc = obd_get_info(env, mdt->mdt_child_exp,
5013                                   sizeof(KEY_OSP_CONNECTED),
5014                                   KEY_OSP_CONNECTED, NULL, NULL, NULL);
5015                 if (rc)
5016                         RETURN(-EAGAIN);
5017                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
5018         }
5019
5020         rc = class_connect(&conn, obd, cluuid);
5021         if (rc)
5022                 RETURN(rc);
5023
5024         lexp = class_conn2export(&conn);
5025         LASSERT(lexp != NULL);
5026
5027         rc = mdt_connect_internal(lexp, mdt, data);
5028         if (rc == 0) {
5029                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5030
5031                 LASSERT(lcd);
5032                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5033                 rc = tgt_client_new(env, lexp);
5034                 if (rc == 0)
5035                         mdt_export_stats_init(obd, lexp, localdata);
5036
5037                 /* For phase I, sync for cross-ref operation. */
5038                 lexp->exp_keep_sync = 1;
5039         }
5040
5041         if (rc != 0) {
5042                 class_disconnect(lexp);
5043                 *exp = NULL;
5044         } else {
5045                 *exp = lexp;
5046         }
5047
5048         RETURN(rc);
5049 }
5050
5051 static int mdt_obd_reconnect(const struct lu_env *env,
5052                              struct obd_export *exp, struct obd_device *obd,
5053                              struct obd_uuid *cluuid,
5054                              struct obd_connect_data *data,
5055                              void *localdata)
5056 {
5057         int                     rc;
5058         ENTRY;
5059
5060         if (exp == NULL || obd == NULL || cluuid == NULL)
5061                 RETURN(-EINVAL);
5062
5063         rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
5064         if (rc == 0)
5065                 mdt_export_stats_init(obd, exp, localdata);
5066
5067         RETURN(rc);
5068 }
5069
5070 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
5071                                    struct mdt_thread_info *info,
5072                                    struct mdt_file_data *mfd)
5073 {
5074         struct lu_context ses;
5075         int rc;
5076         ENTRY;
5077
5078         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
5079         if (rc)
5080                 RETURN(rc);
5081
5082         env->le_ses = &ses;
5083         lu_context_enter(&ses);
5084
5085         mdt_ucred(info)->uc_valid = UCRED_OLD;
5086         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
5087
5088         lu_context_exit(&ses);
5089         lu_context_fini(&ses);
5090         env->le_ses = NULL;
5091
5092         RETURN(rc);
5093 }
5094
5095 static int mdt_export_cleanup(struct obd_export *exp)
5096 {
5097         struct mdt_export_data *med = &exp->exp_mdt_data;
5098         struct obd_device      *obd = exp->exp_obd;
5099         struct mdt_device      *mdt;
5100         struct mdt_thread_info *info;
5101         struct lu_env           env;
5102         CFS_LIST_HEAD(closing_list);
5103         struct mdt_file_data *mfd, *n;
5104         int rc = 0;
5105         ENTRY;
5106
5107         spin_lock(&med->med_open_lock);
5108         while (!cfs_list_empty(&med->med_open_head)) {
5109                 cfs_list_t *tmp = med->med_open_head.next;
5110                 mfd = cfs_list_entry(tmp, struct mdt_file_data, mfd_list);
5111
5112                 /* Remove mfd handle so it can't be found again.
5113                  * We are consuming the mfd_list reference here. */
5114                 class_handle_unhash(&mfd->mfd_handle);
5115                 cfs_list_move_tail(&mfd->mfd_list, &closing_list);
5116         }
5117         spin_unlock(&med->med_open_lock);
5118         mdt = mdt_dev(obd->obd_lu_dev);
5119         LASSERT(mdt != NULL);
5120
5121         rc = lu_env_init(&env, LCT_MD_THREAD);
5122         if (rc)
5123                 RETURN(rc);
5124
5125         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5126         LASSERT(info != NULL);
5127         memset(info, 0, sizeof *info);
5128         info->mti_env = &env;
5129         info->mti_mdt = mdt;
5130         info->mti_exp = exp;
5131
5132         if (!cfs_list_empty(&closing_list)) {
5133                 struct md_attr *ma = &info->mti_attr;
5134
5135                 /* Close any open files (which may also cause orphan unlinking). */
5136                 cfs_list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5137                         cfs_list_del_init(&mfd->mfd_list);
5138                         ma->ma_need = ma->ma_valid = 0;
5139
5140                         /* This file is being closed due to an eviction, it
5141                          * could have been modified and now dirty regarding to
5142                          * HSM archive, check this!
5143                          * The logic here is to mark a file dirty if there's a
5144                          * chance it was dirtied before the client was evicted,
5145                          * so that we don't have to wait for a release attempt
5146                          * before finding out the file was actually dirty and
5147                          * fail the release. Aggressively marking it dirty here
5148                          * will cause the policy engine to attempt to
5149                          * re-archive it; when rearchiving, we can compare the
5150                          * current version to the HSM data_version and make the
5151                          * archive request into a noop if it's not actually
5152                          * dirty.
5153                          */
5154                         if (mfd->mfd_mode & (FMODE_WRITE|MDS_FMODE_TRUNC))
5155                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
5156
5157                         /* Don't unlink orphan on failover umount, LU-184 */
5158                         if (exp->exp_flags & OBD_OPT_FAILOVER) {
5159                                 ma->ma_valid = MA_FLAGS;
5160                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5161                         }
5162                         mdt_mfd_close(info, mfd);
5163                 }
5164         }
5165         info->mti_mdt = NULL;
5166         /* cleanup client slot early */
5167         /* Do not erase record for recoverable client. */
5168         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5169                 tgt_client_del(&env, exp);
5170         lu_env_fini(&env);
5171
5172         RETURN(rc);
5173 }
5174
5175 static int mdt_obd_disconnect(struct obd_export *exp)
5176 {
5177         int rc;
5178         ENTRY;
5179
5180         LASSERT(exp);
5181         class_export_get(exp);
5182
5183         rc = server_disconnect_export(exp);
5184         if (rc != 0)
5185                 CDEBUG(D_IOCTL, "server disconnect error: %d\n", rc);
5186
5187         rc = mdt_export_cleanup(exp);
5188         class_export_put(exp);
5189         RETURN(rc);
5190 }
5191
5192 /* FIXME: Can we avoid using these two interfaces? */
5193 static int mdt_init_export(struct obd_export *exp)
5194 {
5195         struct mdt_export_data *med = &exp->exp_mdt_data;
5196         int                     rc;
5197         ENTRY;
5198
5199         CFS_INIT_LIST_HEAD(&med->med_open_head);
5200         spin_lock_init(&med->med_open_lock);
5201         mutex_init(&med->med_idmap_mutex);
5202         med->med_idmap = NULL;
5203         spin_lock(&exp->exp_lock);
5204         exp->exp_connecting = 1;
5205         spin_unlock(&exp->exp_lock);
5206
5207         /* self-export doesn't need client data and ldlm initialization */
5208         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5209                                      &exp->exp_client_uuid)))
5210                 RETURN(0);
5211
5212         rc = tgt_client_alloc(exp);
5213         if (rc)
5214                 GOTO(err, rc);
5215
5216         rc = ldlm_init_export(exp);
5217         if (rc)
5218                 GOTO(err_free, rc);
5219
5220         RETURN(rc);
5221
5222 err_free:
5223         tgt_client_free(exp);
5224 err:
5225         CERROR("%s: Failed to initialize export: rc = %d\n",
5226                exp->exp_obd->obd_name, rc);
5227         return rc;
5228 }
5229
5230 static int mdt_destroy_export(struct obd_export *exp)
5231 {
5232         ENTRY;
5233
5234         if (exp_connect_rmtclient(exp))
5235                 mdt_cleanup_idmap(&exp->exp_mdt_data);
5236
5237         target_destroy_export(exp);
5238         /* destroy can be called from failed obd_setup, so
5239          * checking uuid is safer than obd_self_export */
5240         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5241                                      &exp->exp_client_uuid)))
5242                 RETURN(0);
5243
5244         ldlm_destroy_export(exp);
5245         tgt_client_free(exp);
5246
5247         LASSERT(cfs_list_empty(&exp->exp_outstanding_replies));
5248         LASSERT(cfs_list_empty(&exp->exp_mdt_data.med_open_head));
5249
5250         RETURN(0);
5251 }
5252
5253 /** The maximum depth that fid2path() will search.
5254  * This is limited only because we want to store the fids for
5255  * historical path lookup purposes.
5256  */
5257 #define MAX_PATH_DEPTH 100
5258
5259 /** mdt_path() lookup structure. */
5260 struct path_lookup_info {
5261         __u64                   pli_recno;      /**< history point */
5262         __u64                   pli_currec;     /**< current record */
5263         struct lu_fid           pli_fid;
5264         struct lu_fid           pli_fids[MAX_PATH_DEPTH]; /**< path, in fids */
5265         struct mdt_object       *pli_mdt_obj;
5266         char                    *pli_path;      /**< full path */
5267         int                     pli_pathlen;
5268         int                     pli_linkno;     /**< which hardlink to follow */
5269         int                     pli_fidcount;   /**< number of \a pli_fids */
5270 };
5271
5272 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
5273                    struct linkea_data *ldata)
5274 {
5275         int rc;
5276
5277         LASSERT(ldata->ld_buf->lb_buf != NULL);
5278
5279         if (!mdt_object_exists(mdt_obj))
5280                 return -ENODATA;
5281
5282         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5283                           ldata->ld_buf, XATTR_NAME_LINK);
5284         if (rc == -ERANGE) {
5285                 /* Buf was too small, figure out what we need. */
5286                 lu_buf_free(ldata->ld_buf);
5287                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5288                                   ldata->ld_buf, XATTR_NAME_LINK);
5289                 if (rc < 0)
5290                         return rc;
5291                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
5292                 if (ldata->ld_buf->lb_buf == NULL)
5293                         return -ENOMEM;
5294                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5295                                   ldata->ld_buf, XATTR_NAME_LINK);
5296         }
5297         if (rc < 0)
5298                 return rc;
5299
5300         return linkea_init(ldata);
5301 }
5302
5303 static int mdt_path_current(struct mdt_thread_info *info,
5304                             struct path_lookup_info *pli)
5305 {
5306         struct mdt_device       *mdt = info->mti_mdt;
5307         struct mdt_object       *mdt_obj;
5308         struct link_ea_header   *leh;
5309         struct link_ea_entry    *lee;
5310         struct lu_name          *tmpname = &info->mti_name;
5311         struct lu_fid           *tmpfid = &info->mti_tmp_fid1;
5312         struct lu_buf           *buf = &info->mti_big_buf;
5313         char                    *ptr;
5314         int                     reclen;
5315         struct linkea_data      ldata = { 0 };
5316         int                     rc = 0;
5317         ENTRY;
5318
5319         /* temp buffer for path element, the buffer will be finally freed
5320          * in mdt_thread_info_fini */
5321         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
5322         if (buf->lb_buf == NULL)
5323                 RETURN(-ENOMEM);
5324
5325         ldata.ld_buf = buf;
5326         ptr = pli->pli_path + pli->pli_pathlen - 1;
5327         *ptr = 0;
5328         --ptr;
5329         pli->pli_fidcount = 0;
5330         pli->pli_fids[0] = *(struct lu_fid *)mdt_object_fid(pli->pli_mdt_obj);
5331
5332         /* root FID only exists on MDT0, and fid2path should also ends at MDT0,
5333          * so checking root_fid can only happen on MDT0. */
5334         while (!lu_fid_eq(&mdt->mdt_md_root_fid,
5335                           &pli->pli_fids[pli->pli_fidcount])) {
5336                 mdt_obj = mdt_object_find(info->mti_env, mdt,
5337                                           &pli->pli_fids[pli->pli_fidcount]);
5338                 if (IS_ERR(mdt_obj))
5339                         GOTO(out, rc = PTR_ERR(mdt_obj));
5340                 if (mdt_object_remote(mdt_obj)) {
5341                         mdt_object_put(info->mti_env, mdt_obj);
5342                         GOTO(remote_out, rc = -EREMOTE);
5343                 }
5344                 if (!mdt_object_exists(mdt_obj)) {
5345                         mdt_object_put(info->mti_env, mdt_obj);
5346                         GOTO(out, rc = -ENOENT);
5347                 }
5348
5349                 rc = mdt_links_read(info, mdt_obj, &ldata);
5350                 mdt_object_put(info->mti_env, mdt_obj);
5351                 if (rc != 0)
5352                         GOTO(out, rc);
5353
5354                 leh = buf->lb_buf;
5355                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
5356                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
5357                 /* If set, use link #linkno for path lookup, otherwise use
5358                    link #0.  Only do this for the final path element. */
5359                 if (pli->pli_fidcount == 0 &&
5360                     pli->pli_linkno < leh->leh_reccount) {
5361                         int count;
5362                         for (count = 0; count < pli->pli_linkno; count++) {
5363                                 lee = (struct link_ea_entry *)
5364                                      ((char *)lee + reclen);
5365                                 linkea_entry_unpack(lee, &reclen, tmpname,
5366                                                     tmpfid);
5367                         }
5368                         if (pli->pli_linkno < leh->leh_reccount - 1)
5369                                 /* indicate to user there are more links */
5370                                 pli->pli_linkno++;
5371                 }
5372
5373                 /* Pack the name in the end of the buffer */
5374                 ptr -= tmpname->ln_namelen;
5375                 if (ptr - 1 <= pli->pli_path)
5376                         GOTO(out, rc = -EOVERFLOW);
5377                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
5378                 *(--ptr) = '/';
5379
5380                 /* Store the parent fid for historic lookup */
5381                 if (++pli->pli_fidcount >= MAX_PATH_DEPTH)
5382                         GOTO(out, rc = -EOVERFLOW);
5383                 pli->pli_fids[pli->pli_fidcount] = *tmpfid;
5384         }
5385
5386 remote_out:
5387         ptr++; /* skip leading / */
5388         memmove(pli->pli_path, ptr, pli->pli_path + pli->pli_pathlen - ptr);
5389
5390         EXIT;
5391 out:
5392         return rc;
5393 }
5394
5395 /* Returns the full path to this fid, as of changelog record recno. */
5396 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
5397                     char *path, int pathlen, __u64 *recno, int *linkno,
5398                     struct lu_fid *fid)
5399 {
5400         struct mdt_device       *mdt = info->mti_mdt;
5401         struct path_lookup_info *pli;
5402         int                     tries = 3;
5403         int                     rc = -EAGAIN;
5404         ENTRY;
5405
5406         if (pathlen < 3)
5407                 RETURN(-EOVERFLOW);
5408
5409         if (lu_fid_eq(&mdt->mdt_md_root_fid, mdt_object_fid(obj))) {
5410                 path[0] = '\0';
5411                 RETURN(0);
5412         }
5413
5414         OBD_ALLOC_PTR(pli);
5415         if (pli == NULL)
5416                 RETURN(-ENOMEM);
5417
5418         pli->pli_mdt_obj = obj;
5419         pli->pli_recno = *recno;
5420         pli->pli_path = path;
5421         pli->pli_pathlen = pathlen;
5422         pli->pli_linkno = *linkno;
5423
5424         /* Retry multiple times in case file is being moved */
5425         while (tries-- && rc == -EAGAIN)
5426                 rc = mdt_path_current(info, pli);
5427
5428         /* return the last resolved fids to the client, so the client will
5429          * build the left path on another MDT for remote object */
5430         *fid = pli->pli_fids[pli->pli_fidcount];
5431
5432         *recno = pli->pli_currec;
5433         /* Return next link index to caller */
5434         *linkno = pli->pli_linkno;
5435
5436         OBD_FREE_PTR(pli);
5437
5438         RETURN(rc);
5439 }
5440
5441 static int mdt_fid2path(struct mdt_thread_info *info,
5442                         struct getinfo_fid2path *fp)
5443 {
5444         struct mdt_device *mdt = info->mti_mdt;
5445         struct mdt_object *obj;
5446         int    rc;
5447         ENTRY;
5448
5449         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
5450                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
5451
5452         if (!fid_is_sane(&fp->gf_fid))
5453                 RETURN(-EINVAL);
5454
5455         if (!fid_is_namespace_visible(&fp->gf_fid)) {
5456                 CWARN("%s: "DFID" is invalid, sequence should be "
5457                       ">= "LPX64"\n", mdt_obd_name(mdt),
5458                       PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
5459                 RETURN(-EINVAL);
5460         }
5461
5462         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
5463         if (obj == NULL || IS_ERR(obj)) {
5464                 CDEBUG(D_IOCTL, "no object "DFID": %ld\n", PFID(&fp->gf_fid),
5465                        PTR_ERR(obj));
5466                 RETURN(-EINVAL);
5467         }
5468
5469         if (mdt_object_remote(obj))
5470                 rc = -EREMOTE;
5471         else if (!mdt_object_exists(obj))
5472                 rc = -ENOENT;
5473         else
5474                 rc = 0;
5475
5476         if (rc < 0) {
5477                 mdt_object_put(info->mti_env, obj);
5478                 CDEBUG(D_IOCTL, "nonlocal object "DFID": %d\n",
5479                        PFID(&fp->gf_fid), rc);
5480                 RETURN(rc);
5481         }
5482
5483         rc = mdt_path(info, obj, fp->gf_path, fp->gf_pathlen, &fp->gf_recno,
5484                       &fp->gf_linkno, &fp->gf_fid);
5485
5486         CDEBUG(D_INFO, "fid "DFID", path %s recno "LPX64" linkno %u\n",
5487                PFID(&fp->gf_fid), fp->gf_path, fp->gf_recno, fp->gf_linkno);
5488
5489         mdt_object_put(info->mti_env, obj);
5490
5491         RETURN(rc);
5492 }
5493
5494 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key,
5495                             void *val, int vallen)
5496 {
5497         struct getinfo_fid2path *fpout, *fpin;
5498         int rc = 0;
5499
5500         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
5501         fpout = val;
5502
5503         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5504                 lustre_swab_fid2path(fpin);
5505
5506         memcpy(fpout, fpin, sizeof(*fpin));
5507         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
5508                 RETURN(-EINVAL);
5509
5510         rc = mdt_fid2path(info, fpout);
5511         RETURN(rc);
5512 }
5513
5514 int mdt_get_info(struct tgt_session_info *tsi)
5515 {
5516         char    *key;
5517         int      keylen;
5518         __u32   *vallen;
5519         void    *valout;
5520         int      rc;
5521
5522         ENTRY;
5523
5524         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
5525         if (key == NULL) {
5526                 CDEBUG(D_IOCTL, "No GETINFO key");
5527                 RETURN(err_serious(-EFAULT));
5528         }
5529         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
5530                                       RCL_CLIENT);
5531
5532         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
5533         if (vallen == NULL) {
5534                 CDEBUG(D_IOCTL, "Unable to get RMF_GETINFO_VALLEN buffer");
5535                 RETURN(err_serious(-EFAULT));
5536         }
5537
5538         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
5539                              *vallen);
5540         rc = req_capsule_server_pack(tsi->tsi_pill);
5541         if (rc)
5542                 RETURN(err_serious(rc));
5543
5544         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
5545         if (valout == NULL) {
5546                 CDEBUG(D_IOCTL, "Unable to get get-info RPC out buffer");
5547                 RETURN(err_serious(-EFAULT));
5548         }
5549
5550         if (KEY_IS(KEY_FID2PATH)) {
5551                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
5552
5553                 rc = mdt_rpc_fid2path(info, key, valout, *vallen);
5554                 mdt_thread_info_fini(info);
5555         } else {
5556                 rc = -EINVAL;
5557         }
5558         RETURN(rc);
5559 }
5560
5561 /* Pass the ioc down */
5562 static int mdt_ioc_child(struct lu_env *env, struct mdt_device *mdt,
5563                          unsigned int cmd, int len, void *data)
5564 {
5565         struct lu_context ioctl_session;
5566         struct md_device *next = mdt->mdt_child;
5567         int rc;
5568         ENTRY;
5569
5570         rc = lu_context_init(&ioctl_session, LCT_SERVER_SESSION);
5571         if (rc)
5572                 RETURN(rc);
5573         ioctl_session.lc_thread = (struct ptlrpc_thread *)current;
5574         lu_context_enter(&ioctl_session);
5575         env->le_ses = &ioctl_session;
5576
5577         LASSERT(next->md_ops->mdo_iocontrol);
5578         rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data);
5579
5580         lu_context_exit(&ioctl_session);
5581         lu_context_fini(&ioctl_session);
5582         RETURN(rc);
5583 }
5584
5585 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
5586 {
5587         struct obd_ioctl_data *data = karg;
5588         struct lu_fid *fid;
5589         __u64 version;
5590         struct mdt_object *obj;
5591         struct mdt_lock_handle  *lh;
5592         int rc;
5593         ENTRY;
5594
5595         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
5596             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
5597                 RETURN(-EINVAL);
5598
5599         fid = (struct lu_fid *)data->ioc_inlbuf1;
5600
5601         if (!fid_is_sane(fid))
5602                 RETURN(-EINVAL);
5603
5604         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
5605
5606         lh = &mti->mti_lh[MDT_LH_PARENT];
5607         mdt_lock_reg_init(lh, LCK_CR);
5608
5609         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
5610         if (IS_ERR(obj))
5611                 RETURN(PTR_ERR(obj));
5612
5613         if (mdt_object_remote(obj)) {
5614                 rc = -EREMOTE;
5615                 /**
5616                  * before calling version get the correct MDS should be
5617                  * fid, this is error to find remote object here
5618                  */
5619                 CERROR("nonlocal object "DFID"\n", PFID(fid));
5620         } else if (!mdt_object_exists(obj)) {
5621                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
5622                 rc = -ENOENT;
5623         } else {
5624                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
5625                *(__u64 *)data->ioc_inlbuf2 = version;
5626                 rc = 0;
5627         }
5628         mdt_object_unlock_put(mti, obj, lh, 1);
5629         RETURN(rc);
5630 }
5631
5632 /* ioctls on obd dev */
5633 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
5634                          void *karg, void *uarg)
5635 {
5636         struct lu_env      env;
5637         struct obd_device *obd = exp->exp_obd;
5638         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
5639         struct dt_device  *dt = mdt->mdt_bottom;
5640         int rc;
5641
5642         ENTRY;
5643         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
5644         rc = lu_env_init(&env, LCT_MD_THREAD);
5645         if (rc)
5646                 RETURN(rc);
5647
5648         switch (cmd) {
5649         case OBD_IOC_SYNC:
5650                 rc = mdt_device_sync(&env, mdt);
5651                 break;
5652         case OBD_IOC_SET_READONLY:
5653                 rc = dt->dd_ops->dt_ro(&env, dt);
5654                 break;
5655         case OBD_IOC_ABORT_RECOVERY:
5656                 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
5657                 target_stop_recovery_thread(obd);
5658                 rc = 0;
5659                 break;
5660         case OBD_IOC_CHANGELOG_REG:
5661         case OBD_IOC_CHANGELOG_DEREG:
5662         case OBD_IOC_CHANGELOG_CLEAR:
5663                 rc = mdt_ioc_child(&env, mdt, cmd, len, karg);
5664                 break;
5665         case OBD_IOC_START_LFSCK: {
5666                 struct md_device *next = mdt->mdt_child;
5667                 struct obd_ioctl_data *data = karg;
5668                 struct lfsck_start_param lsp;
5669
5670                 if (unlikely(data == NULL)) {
5671                         rc = -EINVAL;
5672                         break;
5673                 }
5674
5675                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
5676                 lsp.lsp_index_valid = 0;
5677                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
5678                 break;
5679         }
5680         case OBD_IOC_STOP_LFSCK: {
5681                 struct md_device        *next = mdt->mdt_child;
5682                 struct obd_ioctl_data   *data = karg;
5683                 struct lfsck_stop        stop;
5684
5685                 stop.ls_status = LS_STOPPED;
5686                 /* Old lfsck utils may pass NULL @stop. */
5687                 if (data->ioc_inlbuf1 == NULL)
5688                         stop.ls_flags = 0;
5689                 else
5690                         stop.ls_flags =
5691                         ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
5692
5693                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
5694                 break;
5695         }
5696         case OBD_IOC_GET_OBJ_VERSION: {
5697                 struct mdt_thread_info *mti;
5698                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5699                 memset(mti, 0, sizeof *mti);
5700                 mti->mti_env = &env;
5701                 mti->mti_mdt = mdt;
5702                 mti->mti_exp = exp;
5703
5704                 rc = mdt_ioc_version_get(mti, karg);
5705                 break;
5706         }
5707         case OBD_IOC_CATLOGLIST: {
5708                 struct mdt_thread_info *mti;
5709
5710                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5711                 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
5712                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
5713                                        &mti->mti_tmp_fid1);
5714                 break;
5715          }
5716         default:
5717                 rc = -EOPNOTSUPP;
5718                 CERROR("%s: Not supported cmd = %d, rc = %d\n",
5719                         mdt_obd_name(mdt), cmd, rc);
5720         }
5721
5722         lu_env_fini(&env);
5723         RETURN(rc);
5724 }
5725
5726 int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
5727 {
5728         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
5729         int rc;
5730         ENTRY;
5731
5732         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
5733         RETURN(rc);
5734 }
5735
5736 int mdt_obd_postrecov(struct obd_device *obd)
5737 {
5738         struct lu_env env;
5739         int rc;
5740
5741         rc = lu_env_init(&env, LCT_MD_THREAD);
5742         if (rc)
5743                 RETURN(rc);
5744         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
5745         lu_env_fini(&env);
5746         return rc;
5747 }
5748
5749 static struct obd_ops mdt_obd_device_ops = {
5750         .o_owner          = THIS_MODULE,
5751         .o_set_info_async = mdt_obd_set_info_async,
5752         .o_connect        = mdt_obd_connect,
5753         .o_reconnect      = mdt_obd_reconnect,
5754         .o_disconnect     = mdt_obd_disconnect,
5755         .o_init_export    = mdt_init_export,
5756         .o_destroy_export = mdt_destroy_export,
5757         .o_iocontrol      = mdt_iocontrol,
5758         .o_postrecov      = mdt_obd_postrecov,
5759 };
5760
5761 static struct lu_device* mdt_device_fini(const struct lu_env *env,
5762                                          struct lu_device *d)
5763 {
5764         struct mdt_device *m = mdt_dev(d);
5765         ENTRY;
5766
5767         mdt_fini(env, m);
5768         RETURN(NULL);
5769 }
5770
5771 static struct lu_device *mdt_device_free(const struct lu_env *env,
5772                                          struct lu_device *d)
5773 {
5774         struct mdt_device *m = mdt_dev(d);
5775         ENTRY;
5776
5777         lu_device_fini(&m->mdt_lu_dev);
5778         OBD_FREE_PTR(m);
5779
5780         RETURN(NULL);
5781 }
5782
5783 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
5784                                           struct lu_device_type *t,
5785                                           struct lustre_cfg *cfg)
5786 {
5787         struct lu_device  *l;
5788         struct mdt_device *m;
5789
5790         OBD_ALLOC_PTR(m);
5791         if (m != NULL) {
5792                 int rc;
5793
5794                 l = &m->mdt_lu_dev;
5795                 rc = mdt_init0(env, m, t, cfg);
5796                 if (rc != 0) {
5797                         mdt_device_free(env, l);
5798                         l = ERR_PTR(rc);
5799                         return l;
5800                 }
5801         } else
5802                 l = ERR_PTR(-ENOMEM);
5803         return l;
5804 }
5805
5806 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
5807 LU_KEY_INIT(mdt, struct mdt_thread_info);
5808
5809 static void mdt_key_fini(const struct lu_context *ctx,
5810                          struct lu_context_key *key, void* data)
5811 {
5812         struct mdt_thread_info *info = data;
5813
5814         if (info->mti_big_lmm) {
5815                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
5816                 info->mti_big_lmm = NULL;
5817                 info->mti_big_lmmsize = 0;
5818         }
5819         OBD_FREE_PTR(info);
5820 }
5821
5822 /* context key: mdt_thread_key */
5823 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
5824
5825 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
5826 {
5827         return lu_ucred(info->mti_env);
5828 }
5829
5830 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
5831 {
5832         return lu_ucred_check(info->mti_env);
5833 }
5834
5835 /**
5836  * Enable/disable COS (Commit On Sharing).
5837  *
5838  * Set/Clear the COS flag in mdt options.
5839  *
5840  * \param mdt mdt device
5841  * \param val 0 disables COS, other values enable COS
5842  */
5843 void mdt_enable_cos(struct mdt_device *mdt, int val)
5844 {
5845         struct lu_env env;
5846         int rc;
5847
5848         mdt->mdt_opts.mo_cos = !!val;
5849         rc = lu_env_init(&env, LCT_LOCAL);
5850         if (unlikely(rc != 0)) {
5851                 CWARN("lu_env initialization failed with rc = %d,"
5852                       "cannot sync\n", rc);
5853                 return;
5854         }
5855         mdt_device_sync(&env, mdt);
5856         lu_env_fini(&env);
5857 }
5858
5859 /**
5860  * Check COS (Commit On Sharing) status.
5861  *
5862  * Return COS flag status.
5863  *
5864  * \param mdt mdt device
5865  */
5866 int mdt_cos_is_enabled(struct mdt_device *mdt)
5867 {
5868         return mdt->mdt_opts.mo_cos != 0;
5869 }
5870
5871 static struct lu_device_type_operations mdt_device_type_ops = {
5872         .ldto_device_alloc = mdt_device_alloc,
5873         .ldto_device_free  = mdt_device_free,
5874         .ldto_device_fini  = mdt_device_fini
5875 };
5876
5877 static struct lu_device_type mdt_device_type = {
5878         .ldt_tags     = LU_DEVICE_MD,
5879         .ldt_name     = LUSTRE_MDT_NAME,
5880         .ldt_ops      = &mdt_device_type_ops,
5881         .ldt_ctx_tags = LCT_MD_THREAD
5882 };
5883
5884 static int __init mdt_mod_init(void)
5885 {
5886         struct lprocfs_static_vars lvars;
5887         int rc;
5888
5889         CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") ==
5890                  FID_NOBRACE_LEN + 1);
5891         CLASSERT(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") ==
5892                  FID_LEN + 1);
5893         rc = lu_kmem_init(mdt_caches);
5894         if (rc)
5895                 return rc;
5896
5897         rc = mds_mod_init();
5898         if (rc)
5899                 GOTO(lu_fini, rc);
5900
5901         lprocfs_mdt_init_vars(&lvars);
5902         rc = class_register_type(&mdt_obd_device_ops, NULL, NULL,
5903 #ifndef HAVE_ONLY_PROCFS_SEQ
5904                                 lvars.module_vars,
5905 #endif
5906                                 LUSTRE_MDT_NAME, &mdt_device_type);
5907         if (rc)
5908                 GOTO(mds_fini, rc);
5909 lu_fini:
5910         if (rc)
5911                 lu_kmem_fini(mdt_caches);
5912 mds_fini:
5913         if (rc)
5914                 mds_mod_exit();
5915         return rc;
5916 }
5917
5918 static void __exit mdt_mod_exit(void)
5919 {
5920         class_unregister_type(LUSTRE_MDT_NAME);
5921         mds_mod_exit();
5922         lu_kmem_fini(mdt_caches);
5923 }
5924
5925 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
5926 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
5927 MODULE_LICENSE("GPL");
5928
5929 cfs_module(mdt, LUSTRE_VERSION_STRING, mdt_mod_init, mdt_mod_exit);