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 #include <lustre_ioctl.h>
  57 /* struct ptlrpc_request */
  58 #include <lustre_net.h>
  59 /* struct obd_export */
  60 #include <lustre_export.h>
  61 /* struct obd_device */
  62 #include <obd.h>
  63 /* lu2dt_dev() */
  64 #include <dt_object.h>
  65 #include <lustre_mds.h>
  66 #include <lustre_log.h>
  67 #include "mdt_internal.h"
  68 #include <lustre_acl.h>
  69 #include <lustre_param.h>
  70 #include <lustre_quota.h>
  71 #include <lustre_lfsck.h>
  72
  73 mdl_mode_t mdt_mdl_lock_modes[] = {
  74         [LCK_MINMODE] = MDL_MINMODE,
  75         [LCK_EX]      = MDL_EX,
  76         [LCK_PW]      = MDL_PW,
  77         [LCK_PR]      = MDL_PR,
  78         [LCK_CW]      = MDL_CW,
  79         [LCK_CR]      = MDL_CR,
  80         [LCK_NL]      = MDL_NL,
  81         [LCK_GROUP]   = MDL_GROUP
  82 };
  83
  84 ldlm_mode_t mdt_dlm_lock_modes[] = {
  85         [MDL_MINMODE] = LCK_MINMODE,
  86         [MDL_EX]      = LCK_EX,
  87         [MDL_PW]      = LCK_PW,
  88         [MDL_PR]      = LCK_PR,
  89         [MDL_CW]      = LCK_CW,
  90         [MDL_CR]      = LCK_CR,
  91         [MDL_NL]      = LCK_NL,
  92         [MDL_GROUP]   = LCK_GROUP
  93 };
  94
  95 static struct mdt_device *mdt_dev(struct lu_device *d);
  96 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
  97
  98 static const struct lu_object_operations mdt_obj_ops;
  99
 100 /* Slab for MDT object allocation */
 101 static struct kmem_cache *mdt_object_kmem;
 102
 103 /* For HSM restore handles */
 104 struct kmem_cache *mdt_hsm_cdt_kmem;
 105
 106 /* For HSM request handles */
 107 struct kmem_cache *mdt_hsm_car_kmem;
 108
 109 static struct lu_kmem_descr mdt_caches[] = {
 110         {
 111                 .ckd_cache = &mdt_object_kmem,
 112                 .ckd_name  = "mdt_obj",
 113                 .ckd_size  = sizeof(struct mdt_object)
 114         },
 115         {
 116                 .ckd_cache      = &mdt_hsm_cdt_kmem,
 117                 .ckd_name       = "mdt_cdt_restore_handle",
 118                 .ckd_size       = sizeof(struct cdt_restore_handle)
 119         },
 120         {
 121                 .ckd_cache      = &mdt_hsm_car_kmem,
 122                 .ckd_name       = "mdt_cdt_agent_req",
 123                 .ckd_size       = sizeof(struct cdt_agent_req)
 124         },
 125         {
 126                 .ckd_cache = NULL
 127         }
 128 };
 129
 130 int mdt_get_disposition(struct ldlm_reply *rep, int flag)
 131 {
 132         if (!rep)
 133                 return 0;
 134         return (rep->lock_policy_res1 & flag);
 135 }
 136
 137 void mdt_clear_disposition(struct mdt_thread_info *info,
 138                            struct ldlm_reply *rep, int flag)
 139 {
 140         if (info) {
 141                 info->mti_opdata &= ~flag;
 142                 tgt_opdata_clear(info->mti_env, flag);
 143         }
 144         if (rep)
 145                 rep->lock_policy_res1 &= ~flag;
 146 }
 147
 148 void mdt_set_disposition(struct mdt_thread_info *info,
 149                          struct ldlm_reply *rep, int flag)
 150 {
 151         if (info) {
 152                 info->mti_opdata |= flag;
 153                 tgt_opdata_set(info->mti_env, flag);
 154         }
 155         if (rep)
 156                 rep->lock_policy_res1 |= flag;
 157 }
 158
 159 void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
 160 {
 161         lh->mlh_pdo_hash = 0;
 162         lh->mlh_reg_mode = lm;
 163         lh->mlh_rreg_mode = lm;
 164         lh->mlh_type = MDT_REG_LOCK;
 165 }
 166
 167 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lock_mode,
 168                        const struct lu_name *lname)
 169 {
 170         lh->mlh_reg_mode = lock_mode;
 171         lh->mlh_rreg_mode = lock_mode;
 172         lh->mlh_type = MDT_PDO_LOCK;
 173
 174         if (lu_name_is_valid(lname)) {
 175                 lh->mlh_pdo_hash = full_name_hash(lname->ln_name,
 176                                                   lname->ln_namelen);
 177                 /* XXX Workaround for LU-2856
 178                  *
 179                  * Zero is a valid return value of full_name_hash, but
 180                  * several users of mlh_pdo_hash assume a non-zero
 181                  * hash value. We therefore map zero onto an
 182                  * arbitrary, but consistent value (1) to avoid
 183                  * problems further down the road. */
 184                 if (unlikely(lh->mlh_pdo_hash == 0))
 185                         lh->mlh_pdo_hash = 1;
 186         } else {
 187                 lh->mlh_pdo_hash = 0;
 188         }
 189 }
 190
 191 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 192                               struct mdt_lock_handle *lh)
 193 {
 194         mdl_mode_t mode;
 195         ENTRY;
 196
 197         /*
 198          * Any dir access needs couple of locks:
 199          *
 200          * 1) on part of dir we gonna take lookup/modify;
 201          *
 202          * 2) on whole dir to protect it from concurrent splitting and/or to
 203          * flush client's cache for readdir().
 204          *
 205          * so, for a given mode and object this routine decides what lock mode
 206          * to use for lock #2:
 207          *
 208          * 1) if caller's gonna lookup in dir then we need to protect dir from
 209          * being splitted only - LCK_CR
 210          *
 211          * 2) if caller's gonna modify dir then we need to protect dir from
 212          * being splitted and to flush cache - LCK_CW
 213          *
 214          * 3) if caller's gonna modify dir and that dir seems ready for
 215          * splitting then we need to protect it from any type of access
 216          * (lookup/modify/split) - LCK_EX --bzzz
 217          */
 218
 219         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 220         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 221
 222         /*
 223          * Ask underlaying level its opinion about preferable PDO lock mode
 224          * having access type passed as regular lock mode:
 225          *
 226          * - MDL_MINMODE means that lower layer does not want to specify lock
 227          * mode;
 228          *
 229          * - MDL_NL means that no PDO lock should be taken. This is used in some
 230          * cases. Say, for non-splittable directories no need to use PDO locks
 231          * at all.
 232          */
 233         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 234                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 235
 236         if (mode != MDL_MINMODE) {
 237                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 238         } else {
 239                 /*
 240                  * Lower layer does not want to specify locking mode. We do it
 241                  * our selves. No special protection is needed, just flush
 242                  * client's cache on modification and allow concurrent
 243                  * mondification.
 244                  */
 245                 switch (lh->mlh_reg_mode) {
 246                 case LCK_EX:
 247                         lh->mlh_pdo_mode = LCK_EX;
 248                         break;
 249                 case LCK_PR:
 250                         lh->mlh_pdo_mode = LCK_CR;
 251                         break;
 252                 case LCK_PW:
 253                         lh->mlh_pdo_mode = LCK_CW;
 254                         break;
 255                 default:
 256                         CERROR("Not expected lock type (0x%x)\n",
 257                                (int)lh->mlh_reg_mode);
 258                         LBUG();
 259                 }
 260         }
 261
 262         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 263         EXIT;
 264 }
 265
 266 static int mdt_getstatus(struct tgt_session_info *tsi)
 267 {
 268         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
 269         struct mdt_device       *mdt = info->mti_mdt;
 270         struct mdt_body         *repbody;
 271         int                      rc;
 272
 273         ENTRY;
 274
 275         rc = mdt_check_ucred(info);
 276         if (rc)
 277                 GOTO(out, rc = err_serious(rc));
 278
 279         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
 280                 GOTO(out, rc = err_serious(-ENOMEM));
 281
 282         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 283         repbody->fid1 = mdt->mdt_md_root_fid;
 284         repbody->valid |= OBD_MD_FLID;
 285
 286         if (tsi->tsi_tgt->lut_mds_capa &&
 287             exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) {
 288                 struct mdt_object       *root;
 289                 struct lustre_capa      *capa;
 290
 291                 root = mdt_object_find(info->mti_env, mdt, &repbody->fid1);
 292                 if (IS_ERR(root))
 293                         GOTO(out, rc = PTR_ERR(root));
 294
 295                 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
 296                 LASSERT(capa);
 297                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 298                 rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa,
 299                                  0);
 300                 mdt_object_put(info->mti_env, root);
 301                 if (rc == 0)
 302                         repbody->valid |= OBD_MD_FLMDSCAPA;
 303         }
 304         EXIT;
 305 out:
 306         mdt_thread_info_fini(info);
 307         return rc;
 308 }
 309
 310 static int mdt_statfs(struct tgt_session_info *tsi)
 311 {
 312         struct ptlrpc_request           *req = tgt_ses_req(tsi);
 313         struct mdt_thread_info          *info = tsi2mdt_info(tsi);
 314         struct md_device                *next = info->mti_mdt->mdt_child;
 315         struct ptlrpc_service_part      *svcpt;
 316         struct obd_statfs               *osfs;
 317         int                             rc;
 318
 319         ENTRY;
 320
 321         svcpt = req->rq_rqbd->rqbd_svcpt;
 322
 323         /* This will trigger a watchdog timeout */
 324         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 325                          (MDT_SERVICE_WATCHDOG_FACTOR *
 326                           at_get(&svcpt->scp_at_estimate)) + 1);
 327
 328         rc = mdt_check_ucred(info);
 329         if (rc)
 330                 GOTO(out, rc = err_serious(rc));
 331
 332         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
 333                 GOTO(out, rc = err_serious(-ENOMEM));
 334
 335         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 336         if (!osfs)
 337                 GOTO(out, rc = -EPROTO);
 338
 339         /** statfs information are cached in the mdt_device */
 340         if (cfs_time_before_64(info->mti_mdt->mdt_osfs_age,
 341                                cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS))) {
 342                 /** statfs data is too old, get up-to-date one */
 343                 rc = next->md_ops->mdo_statfs(info->mti_env, next, osfs);
 344                 if (rc)
 345                         GOTO(out, rc);
 346                 spin_lock(&info->mti_mdt->mdt_osfs_lock);
 347                 info->mti_mdt->mdt_osfs = *osfs;
 348                 info->mti_mdt->mdt_osfs_age = cfs_time_current_64();
 349                 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
 350         } else {
 351                 /** use cached statfs data */
 352                 spin_lock(&info->mti_mdt->mdt_osfs_lock);
 353                 *osfs = info->mti_mdt->mdt_osfs;
 354                 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
 355         }
 356
 357         if (rc == 0)
 358                 mdt_counter_incr(req, LPROC_MDT_STATFS);
 359 out:
 360         mdt_thread_info_fini(info);
 361         RETURN(rc);
 362 }
 363
 364 /**
 365  * Pack SOM attributes into the reply.
 366  * Call under a DLM UPDATE lock.
 367  */
 368 static void mdt_pack_size2body(struct mdt_thread_info *info,
 369                                struct mdt_object *mo)
 370 {
 371         struct mdt_body *b;
 372         struct md_attr *ma = &info->mti_attr;
 373
 374         LASSERT(ma->ma_attr.la_valid & LA_MODE);
 375         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 376
 377         /* Check if Size-on-MDS is supported, if this is a regular file,
 378          * if SOM is enabled on the object and if SOM cache exists and valid.
 379          * Otherwise do not pack Size-on-MDS attributes to the reply. */
 380         if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) ||
 381             !S_ISREG(ma->ma_attr.la_mode) ||
 382             !mdt_object_is_som_enabled(mo) ||
 383             !(ma->ma_valid & MA_SOM))
 384                 return;
 385
 386         b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 387         b->size = ma->ma_som->msd_size;
 388         b->blocks = ma->ma_som->msd_blocks;
 389 }
 390
 391 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 392                         const struct lu_attr *attr, const struct lu_fid *fid)
 393 {
 394         struct md_attr *ma = &info->mti_attr;
 395
 396         LASSERT(ma->ma_valid & MA_INODE);
 397
 398         b->atime      = attr->la_atime;
 399         b->mtime      = attr->la_mtime;
 400         b->ctime      = attr->la_ctime;
 401         b->mode       = attr->la_mode;
 402         b->size       = attr->la_size;
 403         b->blocks     = attr->la_blocks;
 404         b->uid        = attr->la_uid;
 405         b->gid        = attr->la_gid;
 406         b->flags      = attr->la_flags;
 407         b->nlink      = attr->la_nlink;
 408         b->rdev       = attr->la_rdev;
 409
 410         /*XXX should pack the reply body according to lu_valid*/
 411         b->valid |= OBD_MD_FLCTIME | OBD_MD_FLUID   |
 412                     OBD_MD_FLGID   | OBD_MD_FLTYPE  |
 413                     OBD_MD_FLMODE  | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
 414                     OBD_MD_FLATIME | OBD_MD_FLMTIME ;
 415
 416         if (!S_ISREG(attr->la_mode)) {
 417                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 418         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
 419                 /* means no objects are allocated on osts. */
 420                 LASSERT(!(ma->ma_valid & MA_LOV));
 421                 /* just ignore blocks occupied by extend attributes on MDS */
 422                 b->blocks = 0;
 423                 /* if no object is allocated on osts, the size on mds is valid. b=22272 */
 424                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 425         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL &&
 426                    ma->ma_lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) {
 427                 /* A released file stores its size on MDS. */
 428                 /* But return 1 block for released file, unless tools like tar
 429                  * will consider it fully sparse. (LU-3864)
 430                  */
 431                 if (unlikely(b->size == 0))
 432                         b->blocks = 0;
 433                 else
 434                         b->blocks = 1;
 435                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 436         }
 437
 438         if (fid) {
 439                 b->fid1 = *fid;
 440                 b->valid |= OBD_MD_FLID;
 441                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n",
 442                                 PFID(fid), b->nlink, b->mode, b->size);
 443         }
 444
 445         if (info)
 446                 mdt_body_reverse_idmap(info, b);
 447
 448         if (fid != NULL && (b->valid & OBD_MD_FLSIZE))
 449                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 450                        PFID(fid), (unsigned long long)b->size);
 451 }
 452
 453 static inline int mdt_body_has_lov(const struct lu_attr *la,
 454                                    const struct mdt_body *body)
 455 {
 456         return ((S_ISREG(la->la_mode) && (body->valid & OBD_MD_FLEASIZE)) ||
 457                 (S_ISDIR(la->la_mode) && (body->valid & OBD_MD_FLDIREA )) );
 458 }
 459
 460 void mdt_client_compatibility(struct mdt_thread_info *info)
 461 {
 462         struct mdt_body       *body;
 463         struct ptlrpc_request *req = mdt_info_req(info);
 464         struct obd_export     *exp = req->rq_export;
 465         struct md_attr        *ma = &info->mti_attr;
 466         struct lu_attr        *la = &ma->ma_attr;
 467         ENTRY;
 468
 469         if (exp_connect_layout(exp))
 470                 /* the client can deal with 16-bit lmm_stripe_count */
 471                 RETURN_EXIT;
 472
 473         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 474
 475         if (!mdt_body_has_lov(la, body))
 476                 RETURN_EXIT;
 477
 478         /* now we have a reply with a lov for a client not compatible with the
 479          * layout lock so we have to clean the layout generation number */
 480         if (S_ISREG(la->la_mode))
 481                 ma->ma_lmm->lmm_layout_gen = 0;
 482         EXIT;
 483 }
 484
 485 int mdt_attr_get_eabuf_size(struct mdt_thread_info *info, struct mdt_object *o)
 486 {
 487         const struct lu_env *env = info->mti_env;
 488         int rc, rc2;
 489
 490         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 491                           XATTR_NAME_LOV);
 492
 493         if (rc == -ENODATA)
 494                 rc = 0;
 495
 496         if (rc < 0)
 497                 goto out;
 498
 499         /* Is it a directory? Let's check for the LMV as well */
 500         if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
 501                 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 502                                    XATTR_NAME_LMV);
 503
 504                 if (rc2 == -ENODATA)
 505                         rc2 = mo_xattr_get(env, mdt_object_child(o),
 506                                            &LU_BUF_NULL,
 507                                            XATTR_NAME_DEFAULT_LMV);
 508
 509                 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
 510                         rc = rc2;
 511         }
 512
 513 out:
 514         return rc;
 515 }
 516
 517 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
 518                       const char *name)
 519 {
 520         const struct lu_env *env = info->mti_env;
 521         int rc;
 522         ENTRY;
 523
 524         LASSERT(info->mti_big_lmm_used == 0);
 525         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
 526         if (rc < 0)
 527                 RETURN(rc);
 528
 529         /* big_lmm may need to be grown */
 530         if (info->mti_big_lmmsize < rc) {
 531                 int size = size_roundup_power2(rc);
 532
 533                 if (info->mti_big_lmmsize > 0) {
 534                         /* free old buffer */
 535                         LASSERT(info->mti_big_lmm);
 536                         OBD_FREE_LARGE(info->mti_big_lmm,
 537                                        info->mti_big_lmmsize);
 538                         info->mti_big_lmm = NULL;
 539                         info->mti_big_lmmsize = 0;
 540                 }
 541
 542                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
 543                 if (info->mti_big_lmm == NULL)
 544                         RETURN(-ENOMEM);
 545                 info->mti_big_lmmsize = size;
 546         }
 547         LASSERT(info->mti_big_lmmsize >= rc);
 548
 549         info->mti_buf.lb_buf = info->mti_big_lmm;
 550         info->mti_buf.lb_len = info->mti_big_lmmsize;
 551         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
 552
 553         RETURN(rc);
 554 }
 555
 556 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
 557                    struct md_attr *ma, const char *name)
 558 {
 559         struct md_object *next = mdt_object_child(o);
 560         struct lu_buf    *buf = &info->mti_buf;
 561         int rc;
 562
 563         if (strcmp(name, XATTR_NAME_LOV) == 0) {
 564                 buf->lb_buf = ma->ma_lmm;
 565                 buf->lb_len = ma->ma_lmm_size;
 566                 LASSERT(!(ma->ma_valid & MA_LOV));
 567         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 568                 buf->lb_buf = ma->ma_lmv;
 569                 buf->lb_len = ma->ma_lmv_size;
 570                 LASSERT(!(ma->ma_valid & MA_LMV));
 571         } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 572                 buf->lb_buf = ma->ma_lmv;
 573                 buf->lb_len = ma->ma_lmv_size;
 574                 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
 575         } else {
 576                 return -EINVAL;
 577         }
 578
 579         rc = mo_xattr_get(info->mti_env, next, buf, name);
 580         if (rc > 0) {
 581
 582 got:
 583                 if (strcmp(name, XATTR_NAME_LOV) == 0) {
 584                         if (info->mti_big_lmm_used)
 585                                 ma->ma_lmm = info->mti_big_lmm;
 586
 587                         /* NOT return LOV EA with hole to old client. */
 588                         if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
 589                                      LOV_PATTERN_F_HOLE) &&
 590                             !(exp_connect_flags(info->mti_exp) &
 591                               OBD_CONNECT_LFSCK)) {
 592                                 return -EIO;
 593                         } else {
 594                                 ma->ma_lmm_size = rc;
 595                                 ma->ma_valid |= MA_LOV;
 596                         }
 597                 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 598                         if (info->mti_big_lmm_used)
 599                                 ma->ma_lmv = info->mti_big_lmm;
 600
 601                         ma->ma_lmv_size = rc;
 602                         ma->ma_valid |= MA_LMV;
 603                 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 604                         ma->ma_lmv_size = rc;
 605                         ma->ma_valid |= MA_LMV_DEF;
 606                 }
 607
 608                 /* Update mdt_max_mdsize so all clients will be aware that */
 609                 if (info->mti_mdt->mdt_max_mdsize < rc)
 610                         info->mti_mdt->mdt_max_mdsize = rc;
 611
 612                 rc = 0;
 613         } else if (rc == -ENODATA) {
 614                 /* no LOV EA */
 615                 rc = 0;
 616         } else if (rc == -ERANGE) {
 617                 /* Default LMV has fixed size, so it must be able to fit
 618                  * in the original buffer */
 619                 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
 620                         return rc;
 621                 rc = mdt_big_xattr_get(info, o, name);
 622                 if (rc > 0) {
 623                         info->mti_big_lmm_used = 1;
 624                         goto got;
 625                 }
 626         }
 627
 628         return rc;
 629 }
 630
 631 static int mdt_attr_get_pfid(struct mdt_thread_info *info,
 632                              struct mdt_object *o, struct lu_fid *pfid)
 633 {
 634         struct lu_buf           *buf = &info->mti_buf;
 635         struct link_ea_header   *leh;
 636         struct link_ea_entry    *lee;
 637         int                      rc;
 638         ENTRY;
 639
 640         buf->lb_buf = info->mti_big_lmm;
 641         buf->lb_len = info->mti_big_lmmsize;
 642         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
 643                           buf, XATTR_NAME_LINK);
 644         /* ignore errors, MA_PFID won't be set and it is
 645          * up to the caller to treat this as an error */
 646         if (rc == -ERANGE || buf->lb_len == 0) {
 647                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
 648                 buf->lb_buf = info->mti_big_lmm;
 649                 buf->lb_len = info->mti_big_lmmsize;
 650         }
 651
 652         if (rc < 0)
 653                 RETURN(rc);
 654         if (rc < sizeof(*leh)) {
 655                 CERROR("short LinkEA on "DFID": rc = %d\n",
 656                        PFID(mdt_object_fid(o)), rc);
 657                 RETURN(-ENODATA);
 658         }
 659
 660         leh = (struct link_ea_header *) buf->lb_buf;
 661         lee = (struct link_ea_entry *)(leh + 1);
 662         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
 663                 leh->leh_magic = LINK_EA_MAGIC;
 664                 leh->leh_reccount = __swab32(leh->leh_reccount);
 665                 leh->leh_len = __swab64(leh->leh_len);
 666         }
 667         if (leh->leh_magic != LINK_EA_MAGIC)
 668                 RETURN(-EINVAL);
 669         if (leh->leh_reccount == 0)
 670                 RETURN(-ENODATA);
 671
 672         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
 673         fid_be_to_cpu(pfid, pfid);
 674
 675         RETURN(0);
 676 }
 677
 678 int mdt_attr_get_complex(struct mdt_thread_info *info,
 679                          struct mdt_object *o, struct md_attr *ma)
 680 {
 681         const struct lu_env *env = info->mti_env;
 682         struct md_object    *next = mdt_object_child(o);
 683         struct lu_buf       *buf = &info->mti_buf;
 684         u32                  mode = lu_object_attr(&next->mo_lu);
 685         int                  need = ma->ma_need;
 686         int                  rc = 0, rc2;
 687         ENTRY;
 688
 689         ma->ma_valid = 0;
 690
 691         if (need & MA_INODE) {
 692                 ma->ma_need = MA_INODE;
 693                 rc = mo_attr_get(env, next, ma);
 694                 if (rc)
 695                         GOTO(out, rc);
 696                 ma->ma_valid |= MA_INODE;
 697         }
 698
 699         if (need & MA_PFID) {
 700                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
 701                 if (rc == 0)
 702                         ma->ma_valid |= MA_PFID;
 703                 /* ignore this error, parent fid is not mandatory */
 704                 rc = 0;
 705         }
 706
 707         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
 708                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
 709                 if (rc)
 710                         GOTO(out, rc);
 711         }
 712
 713         if (need & MA_LMV && S_ISDIR(mode)) {
 714                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
 715                 if (rc != 0)
 716                         GOTO(out, rc);
 717         }
 718
 719         if (need & MA_LMV_DEF && S_ISDIR(mode)) {
 720                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
 721                 if (rc != 0)
 722                         GOTO(out, rc);
 723         }
 724
 725         if (need & MA_SOM && S_ISREG(mode)) {
 726                 buf->lb_buf = info->mti_xattr_buf;
 727                 buf->lb_len = sizeof(info->mti_xattr_buf);
 728                 CLASSERT(sizeof(struct som_attrs) <=
 729                          sizeof(info->mti_xattr_buf));
 730                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_SOM);
 731                 rc2 = lustre_buf2som(info->mti_xattr_buf, rc2, ma->ma_som);
 732                 if (rc2 == 0)
 733                         ma->ma_valid |= MA_SOM;
 734                 else if (rc2 < 0 && rc2 != -ENODATA)
 735                         GOTO(out, rc = rc2);
 736         }
 737
 738         if (need & MA_HSM && S_ISREG(mode)) {
 739                 buf->lb_buf = info->mti_xattr_buf;
 740                 buf->lb_len = sizeof(info->mti_xattr_buf);
 741                 CLASSERT(sizeof(struct hsm_attrs) <=
 742                          sizeof(info->mti_xattr_buf));
 743                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
 744                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
 745                 if (rc2 == 0)
 746                         ma->ma_valid |= MA_HSM;
 747                 else if (rc2 < 0 && rc2 != -ENODATA)
 748                         GOTO(out, rc = rc2);
 749         }
 750
 751 #ifdef CONFIG_FS_POSIX_ACL
 752         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
 753                 buf->lb_buf = ma->ma_acl;
 754                 buf->lb_len = ma->ma_acl_size;
 755                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
 756                 if (rc2 > 0) {
 757                         ma->ma_acl_size = rc2;
 758                         ma->ma_valid |= MA_ACL_DEF;
 759                 } else if (rc2 == -ENODATA) {
 760                         /* no ACLs */
 761                         ma->ma_acl_size = 0;
 762                 } else
 763                         GOTO(out, rc = rc2);
 764         }
 765 #endif
 766 out:
 767         ma->ma_need = need;
 768         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
 769                rc, ma->ma_valid, ma->ma_lmm);
 770         RETURN(rc);
 771 }
 772
 773 static int mdt_getattr_internal(struct mdt_thread_info *info,
 774                                 struct mdt_object *o, int ma_need)
 775 {
 776         struct md_object        *next = mdt_object_child(o);
 777         const struct mdt_body   *reqbody = info->mti_body;
 778         struct ptlrpc_request   *req = mdt_info_req(info);
 779         struct md_attr          *ma = &info->mti_attr;
 780         struct lu_attr          *la = &ma->ma_attr;
 781         struct req_capsule      *pill = info->mti_pill;
 782         const struct lu_env     *env = info->mti_env;
 783         struct mdt_body         *repbody;
 784         struct lu_buf           *buffer = &info->mti_buf;
 785         int                     rc;
 786         int                     is_root;
 787         ENTRY;
 788
 789         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
 790                 RETURN(err_serious(-ENOMEM));
 791
 792         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 793
 794         ma->ma_valid = 0;
 795
 796         if (mdt_object_remote(o)) {
 797                 /* This object is located on remote node.*/
 798                 /* Return -ENOTSUPP for old client */
 799                 if (!mdt_is_dne_client(req->rq_export))
 800                         GOTO(out, rc = -ENOTSUPP);
 801
 802                 repbody->fid1 = *mdt_object_fid(o);
 803                 repbody->valid = OBD_MD_FLID | OBD_MD_MDS;
 804                 GOTO(out, rc = 0);
 805         }
 806
 807         if (reqbody->eadatasize > 0) {
 808                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
 809                 if (buffer->lb_buf == NULL)
 810                         GOTO(out, rc = -EPROTO);
 811                 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
 812                                                       RCL_SERVER);
 813         } else {
 814                 buffer->lb_buf = NULL;
 815                 buffer->lb_len = 0;
 816                 ma_need &= ~(MA_LOV | MA_LMV);
 817                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
 818                        mdt_obd_name(info->mti_mdt),
 819                        req->rq_export->exp_client_uuid.uuid);
 820         }
 821
 822         /* If it is dir object and client require MEA, then we got MEA */
 823         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 824             (reqbody->valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
 825                 /* Assumption: MDT_MD size is enough for lmv size. */
 826                 ma->ma_lmv = buffer->lb_buf;
 827                 ma->ma_lmv_size = buffer->lb_len;
 828                 ma->ma_need = MA_INODE;
 829                 if (ma->ma_lmv_size > 0) {
 830                         if (reqbody->valid & OBD_MD_MEA)
 831                                 ma->ma_need |= MA_LMV;
 832                         else if (reqbody->valid & OBD_MD_DEFAULT_MEA)
 833                                 ma->ma_need |= MA_LMV_DEF;
 834                 }
 835         } else {
 836                 ma->ma_lmm = buffer->lb_buf;
 837                 ma->ma_lmm_size = buffer->lb_len;
 838                 ma->ma_need = MA_INODE | MA_HSM;
 839                 if (ma->ma_lmm_size > 0)
 840                         ma->ma_need |= MA_LOV;
 841         }
 842
 843         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 844             reqbody->valid & OBD_MD_FLDIREA  &&
 845             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
 846                 /* get default stripe info for this dir. */
 847                 ma->ma_need |= MA_LOV_DEF;
 848         }
 849         ma->ma_need |= ma_need;
 850         if (ma->ma_need & MA_SOM)
 851                 ma->ma_som = &info->mti_u.som.data;
 852
 853         rc = mdt_attr_get_complex(info, o, ma);
 854         if (unlikely(rc)) {
 855                 CERROR("%s: getattr error for "DFID": rc = %d\n",
 856                        mdt_obd_name(info->mti_mdt),
 857                        PFID(mdt_object_fid(o)), rc);
 858                 RETURN(rc);
 859         }
 860
 861         /* if file is released, check if a restore is running */
 862         if ((ma->ma_valid & MA_HSM) && (ma->ma_hsm.mh_flags & HS_RELEASED) &&
 863             mdt_hsm_restore_is_running(info, mdt_object_fid(o))) {
 864                 repbody->t_state = MS_RESTORE;
 865                 repbody->valid |= OBD_MD_TSTATE;
 866         }
 867
 868         is_root = lu_fid_eq(mdt_object_fid(o), &info->mti_mdt->mdt_md_root_fid);
 869
 870         /* the Lustre protocol supposes to return default striping
 871          * on the user-visible root if explicitly requested */
 872         if ((ma->ma_valid & MA_LOV) == 0 && S_ISDIR(la->la_mode) &&
 873             (ma->ma_need & MA_LOV_DEF && is_root) && ma->ma_need & MA_LOV) {
 874                 struct lu_fid      rootfid;
 875                 struct mdt_object *root;
 876                 struct mdt_device *mdt = info->mti_mdt;
 877
 878                 rc = dt_root_get(env, mdt->mdt_bottom, &rootfid);
 879                 if (rc)
 880                         RETURN(rc);
 881                 root = mdt_object_find(env, mdt, &rootfid);
 882                 if (IS_ERR(root))
 883                         RETURN(PTR_ERR(root));
 884                 rc = mdt_stripe_get(info, root, ma, XATTR_NAME_LOV);
 885                 mdt_object_put(info->mti_env, root);
 886                 if (unlikely(rc)) {
 887                         CERROR("%s: getattr error for "DFID": rc = %d\n",
 888                                mdt_obd_name(info->mti_mdt),
 889                                PFID(mdt_object_fid(o)), rc);
 890                         RETURN(rc);
 891                 }
 892         }
 893
 894         if (likely(ma->ma_valid & MA_INODE))
 895                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
 896         else
 897                 RETURN(-EFAULT);
 898
 899         if (mdt_body_has_lov(la, reqbody)) {
 900                 if (ma->ma_valid & MA_LOV) {
 901                         LASSERT(ma->ma_lmm_size);
 902                         repbody->eadatasize = ma->ma_lmm_size;
 903                         if (S_ISDIR(la->la_mode))
 904                                 repbody->valid |= OBD_MD_FLDIREA;
 905                         else
 906                                 repbody->valid |= OBD_MD_FLEASIZE;
 907                         mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->valid);
 908                 }
 909                 if (ma->ma_valid & MA_LMV) {
 910                         /* Return -ENOTSUPP for old client */
 911                         if (!mdt_is_striped_client(req->rq_export))
 912                                 RETURN(-ENOTSUPP);
 913
 914                         LASSERT(S_ISDIR(la->la_mode));
 915                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
 916                         repbody->eadatasize = ma->ma_lmv_size;
 917                         repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
 918                 }
 919                 if (ma->ma_valid & MA_LMV_DEF) {
 920                         /* Return -ENOTSUPP for old client */
 921                         if (!mdt_is_striped_client(req->rq_export))
 922                                 RETURN(-ENOTSUPP);
 923                         LASSERT(S_ISDIR(la->la_mode));
 924                         repbody->eadatasize = ma->ma_lmv_size;
 925                         repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_DEFAULT_MEA);
 926                 }
 927         } else if (S_ISLNK(la->la_mode) &&
 928                    reqbody->valid & OBD_MD_LINKNAME) {
 929                 buffer->lb_buf = ma->ma_lmm;
 930                 /* eadatasize from client includes NULL-terminator, so
 931                  * there is no need to read it */
 932                 buffer->lb_len = reqbody->eadatasize - 1;
 933                 rc = mo_readlink(env, next, buffer);
 934                 if (unlikely(rc <= 0)) {
 935                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
 936                                mdt_obd_name(info->mti_mdt),
 937                                PFID(mdt_object_fid(o)), rc);
 938                         rc = -EFAULT;
 939                 } else {
 940                         int print_limit = min_t(int, PAGE_CACHE_SIZE - 128, rc);
 941
 942                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
 943                                 rc -= 2;
 944                         repbody->valid |= OBD_MD_LINKNAME;
 945                         /* we need to report back size with NULL-terminator
 946                          * because client expects that */
 947                         repbody->eadatasize = rc + 1;
 948                         if (repbody->eadatasize != reqbody->eadatasize)
 949                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
 950                                        "on "DFID ", expected %d\n",
 951                                        mdt_obd_name(info->mti_mdt),
 952                                        rc, PFID(mdt_object_fid(o)),
 953                                        reqbody->eadatasize - 1);
 954                         /* NULL terminate */
 955                         ((char *)ma->ma_lmm)[rc] = 0;
 956
 957                         /* If the total CDEBUG() size is larger than a page, it
 958                          * will print a warning to the console, avoid this by
 959                          * printing just the last part of the symlink. */
 960                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
 961                                print_limit < rc ? "..." : "", print_limit,
 962                                (char *)ma->ma_lmm + rc - print_limit, rc);
 963                         rc = 0;
 964                 }
 965         }
 966
 967         if (reqbody->valid & OBD_MD_FLMODEASIZE) {
 968                 repbody->max_cookiesize = 0;
 969                 repbody->max_mdsize = info->mti_mdt->mdt_max_mdsize;
 970                 repbody->valid |= OBD_MD_FLMODEASIZE;
 971                 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
 972                        "MAX_COOKIE to : %d:%d\n", repbody->max_mdsize,
 973                        repbody->max_cookiesize);
 974         }
 975
 976         if (exp_connect_rmtclient(info->mti_exp) &&
 977             reqbody->valid & OBD_MD_FLRMTPERM) {
 978                 void *buf = req_capsule_server_get(pill, &RMF_ACL);
 979
 980                 /* mdt_getattr_lock only */
 981                 rc = mdt_pack_remote_perm(info, o, buf);
 982                 if (rc) {
 983                         repbody->valid &= ~OBD_MD_FLRMTPERM;
 984                         repbody->aclsize = 0;
 985                         RETURN(rc);
 986                 } else {
 987                         repbody->valid |= OBD_MD_FLRMTPERM;
 988                         repbody->aclsize = sizeof(struct mdt_remote_perm);
 989                 }
 990         }
 991 #ifdef CONFIG_FS_POSIX_ACL
 992         else if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
 993                  (reqbody->valid & OBD_MD_FLACL)) {
 994                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
 995                 buffer->lb_len = req_capsule_get_size(pill,
 996                                                       &RMF_ACL, RCL_SERVER);
 997                 if (buffer->lb_len > 0) {
 998                         rc = mo_xattr_get(env, next, buffer,
 999                                           XATTR_NAME_ACL_ACCESS);
1000                         if (rc < 0) {
1001                                 if (rc == -ENODATA) {
1002                                         repbody->aclsize = 0;
1003                                         repbody->valid |= OBD_MD_FLACL;
1004                                         rc = 0;
1005                                 } else if (rc == -EOPNOTSUPP) {
1006                                         rc = 0;
1007                                 } else {
1008                                         CERROR("%s: unable to read "DFID
1009                                                " ACL: rc = %d\n",
1010                                                mdt_obd_name(info->mti_mdt),
1011                                                PFID(mdt_object_fid(o)), rc);
1012                                 }
1013                         } else {
1014                                 repbody->aclsize = rc;
1015                                 repbody->valid |= OBD_MD_FLACL;
1016                                 rc = 0;
1017                         }
1018                 }
1019         }
1020 #endif
1021
1022         if (reqbody->valid & OBD_MD_FLMDSCAPA &&
1023             info->mti_mdt->mdt_lut.lut_mds_capa &&
1024             exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) {
1025                 struct lustre_capa *capa;
1026
1027                 capa = req_capsule_server_get(pill, &RMF_CAPA1);
1028                 LASSERT(capa);
1029                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
1030                 rc = mo_capa_get(env, next, capa, 0);
1031                 if (rc)
1032                         RETURN(rc);
1033                 repbody->valid |= OBD_MD_FLMDSCAPA;
1034         }
1035
1036 out:
1037         if (rc == 0)
1038                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
1039
1040         RETURN(rc);
1041 }
1042
1043 static int mdt_renew_capa(struct mdt_thread_info *info)
1044 {
1045         struct mdt_object  *obj = info->mti_object;
1046         struct mdt_body    *body;
1047         struct lustre_capa *capa, *c;
1048         int rc;
1049         ENTRY;
1050
1051         /* if object doesn't exist, or server has disabled capability,
1052          * return directly, client will find body->valid OBD_MD_FLOSSCAPA
1053          * flag not set.
1054          */
1055         if (!obj || !info->mti_mdt->mdt_lut.lut_oss_capa ||
1056             !(exp_connect_flags(info->mti_exp) & OBD_CONNECT_OSS_CAPA))
1057                 RETURN(0);
1058
1059         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1060         LASSERT(body != NULL);
1061
1062         c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
1063         LASSERT(c);
1064
1065         capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
1066         LASSERT(capa);
1067
1068         *capa = *c;
1069         rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
1070         if (rc == 0)
1071                 body->valid |= OBD_MD_FLOSSCAPA;
1072         RETURN(rc);
1073 }
1074
1075 static int mdt_getattr(struct tgt_session_info *tsi)
1076 {
1077         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1078         struct mdt_object       *obj = info->mti_object;
1079         struct req_capsule      *pill = info->mti_pill;
1080         struct mdt_body         *reqbody;
1081         struct mdt_body         *repbody;
1082         int rc, rc2;
1083         ENTRY;
1084
1085         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1086         LASSERT(reqbody);
1087
1088         if (reqbody->valid & OBD_MD_FLOSSCAPA) {
1089                 rc = req_capsule_server_pack(pill);
1090                 if (unlikely(rc))
1091                         RETURN(err_serious(rc));
1092                 rc = mdt_renew_capa(info);
1093                 GOTO(out_shrink, rc);
1094         }
1095
1096         LASSERT(obj != NULL);
1097         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1098
1099         /* Unlike intent case where we need to pre-fill out buffers early on
1100          * in intent policy for ldlm reasons, here we can have a much better
1101          * guess at EA size by just reading it from disk.
1102          * Exceptions are readdir and (missing) directory striping */
1103         /* Readlink */
1104         if (reqbody->valid & OBD_MD_LINKNAME) {
1105                 /* No easy way to know how long is the symlink, but it cannot
1106                  * be more than PATH_MAX, so we allocate +1 */
1107                 rc = PATH_MAX + 1;
1108
1109         /* A special case for fs ROOT: getattr there might fetch
1110          * default EA for entire fs, not just for this dir!
1111          */
1112         } else if (lu_fid_eq(mdt_object_fid(obj),
1113                              &info->mti_mdt->mdt_md_root_fid) &&
1114                    (reqbody->valid & OBD_MD_FLDIREA) &&
1115                    (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1116                                                                  MDS_GETATTR)) {
1117                 /* Should the default strping be bigger, mdt_fix_reply
1118                  * will reallocate */
1119                 rc = DEF_REP_MD_SIZE;
1120         } else {
1121                 /* Read the actual EA size from disk */
1122                 rc = mdt_attr_get_eabuf_size(info, obj);
1123         }
1124
1125         if (rc < 0)
1126                 GOTO(out_shrink, rc);
1127
1128         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1129
1130         rc = req_capsule_server_pack(pill);
1131         if (unlikely(rc != 0))
1132                 GOTO(out, rc = err_serious(rc));
1133
1134         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1135         LASSERT(repbody != NULL);
1136         repbody->eadatasize = 0;
1137         repbody->aclsize = 0;
1138
1139         if (reqbody->valid & OBD_MD_FLRMTPERM)
1140                 rc = mdt_init_ucred(info, reqbody);
1141         else
1142                 rc = mdt_check_ucred(info);
1143         if (unlikely(rc))
1144                 GOTO(out_shrink, rc);
1145
1146         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1147
1148         /*
1149          * Don't check capability at all, because rename might getattr for
1150          * remote obj, and at that time no capability is available.
1151          */
1152         mdt_set_capainfo(info, 1, &reqbody->fid1, BYPASS_CAPA);
1153         rc = mdt_getattr_internal(info, obj, 0);
1154         if (reqbody->valid & OBD_MD_FLRMTPERM)
1155                 mdt_exit_ucred(info);
1156         EXIT;
1157 out_shrink:
1158         mdt_client_compatibility(info);
1159         rc2 = mdt_fix_reply(info);
1160         if (rc == 0)
1161                 rc = rc2;
1162 out:
1163         mdt_thread_info_fini(info);
1164         return rc;
1165 }
1166
1167 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1168 {
1169         struct mdt_thread_info  *info;
1170         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1171         struct obd_export       *exp = req->rq_export;
1172         struct mdt_object       *o1, *o2, *o;
1173         struct mdt_lock_handle  *lh1, *lh2;
1174         struct mdc_swap_layouts *msl;
1175         int                      rc;
1176         ENTRY;
1177
1178         /* client does not support layout lock, so layout swaping
1179          * is disabled.
1180          * FIXME: there is a problem for old clients which don't support
1181          * layout lock yet. If those clients have already opened the file
1182          * they won't be notified at all so that old layout may still be
1183          * used to do IO. This can be fixed after file release is landed by
1184          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1185         if (!exp_connect_layout(exp))
1186                 RETURN(-EOPNOTSUPP);
1187
1188         info = tsi2mdt_info(tsi);
1189
1190         if (info->mti_dlm_req != NULL)
1191                 ldlm_request_cancel(req, info->mti_dlm_req, 0);
1192
1193         if (req_capsule_get_size(info->mti_pill, &RMF_CAPA1, RCL_CLIENT))
1194                 mdt_set_capainfo(info, 0, &info->mti_body->fid1,
1195                                  req_capsule_client_get(info->mti_pill,
1196                                                         &RMF_CAPA1));
1197
1198         if (req_capsule_get_size(info->mti_pill, &RMF_CAPA2, RCL_CLIENT))
1199                 mdt_set_capainfo(info, 1, &info->mti_body->fid2,
1200                                  req_capsule_client_get(info->mti_pill,
1201                                                         &RMF_CAPA2));
1202
1203         o1 = info->mti_object;
1204         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1205                                 &info->mti_body->fid2);
1206         if (IS_ERR(o))
1207                 GOTO(out, rc = PTR_ERR(o));
1208
1209         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1210                 GOTO(put, rc = -ENOENT);
1211
1212         rc = lu_fid_cmp(&info->mti_body->fid1, &info->mti_body->fid2);
1213         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1214                 GOTO(put, rc);
1215
1216         if (rc < 0)
1217                 swap(o1, o2);
1218
1219         /* permission check. Make sure the calling process having permission
1220          * to write both files. */
1221         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1222                                 MAY_WRITE);
1223         if (rc < 0)
1224                 GOTO(put, rc);
1225
1226         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1227                                 MAY_WRITE);
1228         if (rc < 0)
1229                 GOTO(put, rc);
1230
1231         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1232         if (msl == NULL)
1233                 GOTO(put, rc = -EPROTO);
1234
1235         lh1 = &info->mti_lh[MDT_LH_NEW];
1236         mdt_lock_reg_init(lh1, LCK_EX);
1237         lh2 = &info->mti_lh[MDT_LH_OLD];
1238         mdt_lock_reg_init(lh2, LCK_EX);
1239
1240         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1241                              MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK);
1242         if (rc < 0)
1243                 GOTO(put, rc);
1244
1245         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1246                              MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK);
1247         if (rc < 0)
1248                 GOTO(unlock1, rc);
1249
1250         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1251                              mdt_object_child(o2), msl->msl_flags);
1252         GOTO(unlock2, rc);
1253 unlock2:
1254         mdt_object_unlock(info, o2, lh2, rc);
1255 unlock1:
1256         mdt_object_unlock(info, o1, lh1, rc);
1257 put:
1258         mdt_object_put(info->mti_env, o);
1259 out:
1260         mdt_thread_info_fini(info);
1261         RETURN(rc);
1262 }
1263
1264 static int mdt_raw_lookup(struct mdt_thread_info *info,
1265                           struct mdt_object *parent,
1266                           const struct lu_name *lname,
1267                           struct ldlm_reply *ldlm_rep)
1268 {
1269         struct md_object *next = mdt_object_child(info->mti_object);
1270         const struct mdt_body *reqbody = info->mti_body;
1271         struct lu_fid *child_fid = &info->mti_tmp_fid1;
1272         struct mdt_body *repbody;
1273         int rc;
1274         ENTRY;
1275
1276         if (reqbody->valid != OBD_MD_FLID)
1277                 RETURN(0);
1278
1279         LASSERT(!info->mti_cross_ref);
1280
1281         /* Only got the fid of this obj by name */
1282         fid_zero(child_fid);
1283         rc = mdo_lookup(info->mti_env, next, lname, child_fid,
1284                         &info->mti_spec);
1285 #if 0
1286         /* XXX is raw_lookup possible as intent operation? */
1287         if (rc != 0) {
1288                 if (rc == -ENOENT)
1289                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1290                 RETURN(rc);
1291         } else
1292                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1293
1294         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1295 #endif
1296         if (rc == 0) {
1297                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1298                 repbody->fid1 = *child_fid;
1299                 repbody->valid = OBD_MD_FLID;
1300         }
1301         RETURN(1);
1302 }
1303
1304 /*
1305  * UPDATE lock should be taken against parent, and be release before exit;
1306  * child_bits lock should be taken against child, and be returned back:
1307  *            (1)normal request should release the child lock;
1308  *            (2)intent request will grant the lock to client.
1309  */
1310 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1311                                  struct mdt_lock_handle *lhc,
1312                                  __u64 child_bits,
1313                                  struct ldlm_reply *ldlm_rep)
1314 {
1315         struct ptlrpc_request  *req       = mdt_info_req(info);
1316         struct mdt_body        *reqbody   = NULL;
1317         struct mdt_object      *parent    = info->mti_object;
1318         struct mdt_object      *child;
1319         struct md_object       *next      = mdt_object_child(parent);
1320         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1321         struct lu_name         *lname     = NULL;
1322         struct mdt_lock_handle *lhp       = NULL;
1323         struct ldlm_lock       *lock;
1324         bool                    is_resent;
1325         bool                    try_layout;
1326         int                     ma_need = 0;
1327         int                     rc;
1328         ENTRY;
1329
1330         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1331         LASSERT(ergo(is_resent,
1332                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1333
1334         LASSERT(parent != NULL);
1335
1336         if (info->mti_cross_ref) {
1337                 /* Only getattr on the child. Parent is on another node. */
1338                 mdt_set_disposition(info, ldlm_rep,
1339                                     DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1340                 child = parent;
1341                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1342                        "ldlm_rep = %p\n",
1343                        PFID(mdt_object_fid(child)), ldlm_rep);
1344
1345                 if (is_resent) {
1346                         /* Do not take lock for resent case. */
1347                         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1348                         LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
1349                                  lhc->mlh_reg_lh.cookie);
1350                         LASSERT(fid_res_name_eq(mdt_object_fid(child),
1351                                                 &lock->l_resource->lr_name));
1352                         LDLM_LOCK_PUT(lock);
1353                 } else {
1354                         mdt_lock_handle_init(lhc);
1355                         mdt_lock_reg_init(lhc, LCK_PR);
1356
1357                         /*
1358                          * Object's name is on another MDS, no lookup or layout
1359                          * lock is needed here but update lock is.
1360                          */
1361                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
1362                                         MDS_INODELOCK_LAYOUT);
1363                         child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1364
1365                         rc = mdt_object_lock(info, child, lhc, child_bits,
1366                                              MDT_LOCAL_LOCK);
1367                         if (rc < 0)
1368                                 RETURN(rc);
1369                 }
1370
1371                 /* Finally, we can get attr for child. */
1372                 if (!mdt_object_exists(child)) {
1373                         LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1374                                         &child->mot_obj,
1375                                         "remote object doesn't exist.\n");
1376                         mdt_object_unlock(info, child, lhc, 1);
1377                         RETURN(-ENOENT);
1378                 }
1379
1380                 mdt_set_capainfo(info, 0, mdt_object_fid(child), BYPASS_CAPA);
1381                 rc = mdt_getattr_internal(info, child, 0);
1382                 if (unlikely(rc != 0))
1383                         mdt_object_unlock(info, child, lhc, 1);
1384
1385                 RETURN(rc);
1386         }
1387
1388         lname = &info->mti_name;
1389         mdt_name_unpack(info->mti_pill, &RMF_NAME, lname, MNF_FIX_ANON);
1390
1391         if (lu_name_is_valid(lname)) {
1392                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1393                        "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1394                        PNAME(lname), ldlm_rep);
1395         } else {
1396                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1397                 if (unlikely(reqbody == NULL))
1398                         RETURN(err_serious(-EPROTO));
1399
1400                 *child_fid = reqbody->fid2;
1401
1402                 if (unlikely(!fid_is_sane(child_fid)))
1403                         RETURN(err_serious(-EINVAL));
1404
1405                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1406                        "ldlm_rep = %p\n",
1407                        PFID(mdt_object_fid(parent)),
1408                        PFID(&reqbody->fid2), ldlm_rep);
1409         }
1410
1411         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1412
1413         if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1414                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1415                                 &parent->mot_obj,
1416                                 "Parent doesn't exist!\n");
1417                 RETURN(-ESTALE);
1418         }
1419
1420         if (mdt_object_remote(parent)) {
1421                 CERROR("%s: parent "DFID" is on remote target\n",
1422                        mdt_obd_name(info->mti_mdt),
1423                        PFID(mdt_object_fid(parent)));
1424                 RETURN(-EIO);
1425         }
1426
1427         if (lu_name_is_valid(lname)) {
1428                 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1429                 if (rc != 0) {
1430                         if (rc > 0)
1431                                 rc = 0;
1432                         RETURN(rc);
1433                 }
1434
1435                 /* step 1: lock parent only if parent is a directory */
1436                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1437                         lhp = &info->mti_lh[MDT_LH_PARENT];
1438                         mdt_lock_pdo_init(lhp, LCK_PR, lname);
1439                         rc = mdt_object_lock(info, parent, lhp,
1440                                              MDS_INODELOCK_UPDATE,
1441                                              MDT_LOCAL_LOCK);
1442                         if (unlikely(rc != 0))
1443                                 RETURN(rc);
1444                 }
1445
1446                 /* step 2: lookup child's fid by name */
1447                 fid_zero(child_fid);
1448                 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
1449                                 &info->mti_spec);
1450                 if (rc == -ENOENT)
1451                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1452
1453                 if (rc != 0)
1454                         GOTO(out_parent, rc);
1455         }
1456
1457         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1458
1459         /*
1460          *step 3: find the child object by fid & lock it.
1461          *        regardless if it is local or remote.
1462          *
1463          *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1464          *      set parent dir fid the same as child fid in getattr by fid case
1465          *      we should not lu_object_find() the object again, could lead
1466          *      to hung if there is a concurrent unlink destroyed the object.
1467          */
1468         if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1469                 mdt_object_get(info->mti_env, parent);
1470                 child = parent;
1471         } else {
1472                 child = mdt_object_find(info->mti_env, info->mti_mdt,
1473                                         child_fid);
1474         }
1475
1476         if (unlikely(IS_ERR(child)))
1477                 GOTO(out_parent, rc = PTR_ERR(child));
1478         if (is_resent) {
1479                 /* Do not take lock for resent case. */
1480                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1481                 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
1482                          lhc->mlh_reg_lh.cookie);
1483
1484                 if (!fid_res_name_eq(mdt_object_fid(child),
1485                                      &lock->l_resource->lr_name)) {
1486                         LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
1487                                                  &lock->l_resource->lr_name),
1488                                  "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1489                                  PLDLMRES(lock->l_resource),
1490                                  PFID(mdt_object_fid(parent)));
1491                         CWARN("Although resent, but still not get child lock"
1492                               "parent:"DFID" child:"DFID"\n",
1493                               PFID(mdt_object_fid(parent)),
1494                               PFID(mdt_object_fid(child)));
1495                         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
1496                         LDLM_LOCK_PUT(lock);
1497                         GOTO(relock, 0);
1498                 }
1499                 LDLM_LOCK_PUT(lock);
1500         } else {
1501 relock:
1502                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout*2);
1503                 mdt_lock_handle_init(lhc);
1504                 mdt_lock_reg_init(lhc, LCK_PR);
1505                 try_layout = false;
1506
1507                 if (!mdt_object_exists(child)) {
1508                         LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1509                                         &child->mot_obj,
1510                                         "Object doesn't exist!\n");
1511                         GOTO(out_child, rc = -ENOENT);
1512                 }
1513
1514                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1515                       mdt_object_exists(child) && !mdt_object_remote(child)) {
1516                         struct md_attr *ma = &info->mti_attr;
1517
1518                         ma->ma_valid = 0;
1519                         ma->ma_need = MA_INODE;
1520                         rc = mdt_attr_get_complex(info, child, ma);
1521                         if (unlikely(rc != 0))
1522                                 GOTO(out_child, rc);
1523
1524                         /* If the file has not been changed for some time, we
1525                          * return not only a LOOKUP lock, but also an UPDATE
1526                          * lock and this might save us RPC on later STAT. For
1527                          * directories, it also let negative dentry cache start
1528                          * working for this dir. */
1529                         if (ma->ma_valid & MA_INODE &&
1530                             ma->ma_attr.la_valid & LA_CTIME &&
1531                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1532                                 ma->ma_attr.la_ctime < cfs_time_current_sec())
1533                                 child_bits |= MDS_INODELOCK_UPDATE;
1534                 }
1535
1536                 /* layout lock must be granted in a best-effort way
1537                  * for IT operations */
1538                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1539                 if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1540                     exp_connect_layout(info->mti_exp) &&
1541                     S_ISREG(lu_object_attr(&child->mot_obj)) &&
1542                     !mdt_object_remote(child) && ldlm_rep != NULL) {
1543                         /* try to grant layout lock for regular file. */
1544                         try_layout = true;
1545                 }
1546
1547                 rc = 0;
1548                 if (try_layout) {
1549                         child_bits |= MDS_INODELOCK_LAYOUT;
1550                         /* try layout lock, it may fail to be granted due to
1551                          * contention at LOOKUP or UPDATE */
1552                         if (!mdt_object_lock_try(info, child, lhc, child_bits,
1553                                                  MDT_CROSS_LOCK)) {
1554                                 child_bits &= ~MDS_INODELOCK_LAYOUT;
1555                                 LASSERT(child_bits != 0);
1556                                 rc = mdt_object_lock(info, child, lhc,
1557                                                 child_bits, MDT_CROSS_LOCK);
1558                         } else {
1559                                 ma_need |= MA_LOV;
1560                         }
1561                 } else {
1562                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1563                          * client will enqueue the lock to the remote MDT */
1564                         if (mdt_object_remote(child))
1565                                 child_bits &= ~MDS_INODELOCK_UPDATE;
1566                         rc = mdt_object_lock(info, child, lhc, child_bits,
1567                                                 MDT_CROSS_LOCK);
1568                 }
1569                 if (unlikely(rc != 0))
1570                         GOTO(out_child, rc);
1571         }
1572
1573         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1574         /* Get MA_SOM attributes if update lock is given. */
1575         if (lock &&
1576             lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_UPDATE &&
1577             S_ISREG(lu_object_attr(&mdt_object_child(child)->mo_lu)))
1578                 ma_need |= MA_SOM;
1579
1580         /* finally, we can get attr for child. */
1581         mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
1582         rc = mdt_getattr_internal(info, child, ma_need);
1583         if (unlikely(rc != 0)) {
1584                 mdt_object_unlock(info, child, lhc, 1);
1585         } else if (lock) {
1586                 /* Debugging code. */
1587                 LDLM_DEBUG(lock, "Returning lock to client");
1588                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1589                                          &lock->l_resource->lr_name),
1590                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1591                          PLDLMRES(lock->l_resource),
1592                          PFID(mdt_object_fid(child)));
1593                 if (mdt_object_exists(child) && !mdt_object_remote(child))
1594                         mdt_pack_size2body(info, child);
1595         }
1596         if (lock)
1597                 LDLM_LOCK_PUT(lock);
1598
1599         EXIT;
1600 out_child:
1601         mdt_object_put(info->mti_env, child);
1602 out_parent:
1603         if (lhp)
1604                 mdt_object_unlock(info, parent, lhp, 1);
1605         return rc;
1606 }
1607
1608 /* normal handler: should release the child lock */
1609 static int mdt_getattr_name(struct tgt_session_info *tsi)
1610 {
1611         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1612         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1613         struct mdt_body        *reqbody;
1614         struct mdt_body        *repbody;
1615         int rc, rc2;
1616         ENTRY;
1617
1618         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1619         LASSERT(reqbody != NULL);
1620         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1621         LASSERT(repbody != NULL);
1622
1623         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1624         repbody->eadatasize = 0;
1625         repbody->aclsize = 0;
1626
1627         rc = mdt_init_ucred(info, reqbody);
1628         if (unlikely(rc))
1629                 GOTO(out_shrink, rc);
1630
1631         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1632         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1633                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1634                 lhc->mlh_reg_lh.cookie = 0;
1635         }
1636         mdt_exit_ucred(info);
1637         EXIT;
1638 out_shrink:
1639         mdt_client_compatibility(info);
1640         rc2 = mdt_fix_reply(info);
1641         if (rc == 0)
1642                 rc = rc2;
1643         mdt_thread_info_fini(info);
1644         return rc;
1645 }
1646
1647 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1648                          void *karg, void *uarg);
1649
1650 static int mdt_set_info(struct tgt_session_info *tsi)
1651 {
1652         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1653         char                    *key;
1654         void                    *val;
1655         int                      keylen, vallen, rc = 0;
1656
1657         ENTRY;
1658
1659         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
1660         if (key == NULL) {
1661                 DEBUG_REQ(D_HA, req, "no set_info key");
1662                 RETURN(err_serious(-EFAULT));
1663         }
1664
1665         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
1666                                       RCL_CLIENT);
1667
1668         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
1669         if (val == NULL) {
1670                 DEBUG_REQ(D_HA, req, "no set_info val");
1671                 RETURN(err_serious(-EFAULT));
1672         }
1673
1674         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
1675                                       RCL_CLIENT);
1676
1677         /* Swab any part of val you need to here */
1678         if (KEY_IS(KEY_READ_ONLY)) {
1679                 spin_lock(&req->rq_export->exp_lock);
1680                 if (*(__u32 *)val)
1681                         *exp_connect_flags_ptr(req->rq_export) |=
1682                                 OBD_CONNECT_RDONLY;
1683                 else
1684                         *exp_connect_flags_ptr(req->rq_export) &=
1685                                 ~OBD_CONNECT_RDONLY;
1686                 spin_unlock(&req->rq_export->exp_lock);
1687         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1688                 struct changelog_setinfo *cs = val;
1689
1690                 if (vallen != sizeof(*cs)) {
1691                         CERROR("%s: bad changelog_clear setinfo size %d\n",
1692                                tgt_name(tsi->tsi_tgt), vallen);
1693                         RETURN(-EINVAL);
1694                 }
1695                 if (ptlrpc_req_need_swab(req)) {
1696                         __swab64s(&cs->cs_recno);
1697                         __swab32s(&cs->cs_id);
1698                 }
1699
1700                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
1701                                    vallen, val, NULL);
1702         } else {
1703                 RETURN(-EINVAL);
1704         }
1705         RETURN(rc);
1706 }
1707
1708 static int mdt_readpage(struct tgt_session_info *tsi)
1709 {
1710         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
1711         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
1712         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
1713         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
1714         struct mdt_body         *repbody;
1715         int                      rc;
1716         int                      i;
1717
1718         ENTRY;
1719
1720         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1721                 RETURN(err_serious(-ENOMEM));
1722
1723         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
1724         if (repbody == NULL || reqbody == NULL)
1725                 RETURN(err_serious(-EFAULT));
1726
1727         /*
1728          * prepare @rdpg before calling lower layers and transfer itself. Here
1729          * reqbody->size contains offset of where to start to read and
1730          * reqbody->nlink contains number bytes to read.
1731          */
1732         rdpg->rp_hash = reqbody->size;
1733         if (rdpg->rp_hash != reqbody->size) {
1734                 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1735                        rdpg->rp_hash, reqbody->size);
1736                 RETURN(-EFAULT);
1737         }
1738
1739         rdpg->rp_attrs = reqbody->mode;
1740         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
1741                 rdpg->rp_attrs |= LUDA_64BITHASH;
1742         rdpg->rp_count  = min_t(unsigned int, reqbody->nlink,
1743                                 exp_max_brw_size(tsi->tsi_exp));
1744         rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE - 1) >>
1745                           PAGE_CACHE_SHIFT;
1746         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1747         if (rdpg->rp_pages == NULL)
1748                 RETURN(-ENOMEM);
1749
1750         for (i = 0; i < rdpg->rp_npages; ++i) {
1751                 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1752                 if (rdpg->rp_pages[i] == NULL)
1753                         GOTO(free_rdpg, rc = -ENOMEM);
1754         }
1755
1756         /* call lower layers to fill allocated pages with directory data */
1757         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
1758         if (rc < 0)
1759                 GOTO(free_rdpg, rc);
1760
1761         /* send pages to client */
1762         rc = tgt_sendpage(tsi, rdpg, rc);
1763
1764         EXIT;
1765 free_rdpg:
1766
1767         for (i = 0; i < rdpg->rp_npages; i++)
1768                 if (rdpg->rp_pages[i] != NULL)
1769                         __free_page(rdpg->rp_pages[i]);
1770         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1771
1772         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1773                 RETURN(0);
1774
1775         return rc;
1776 }
1777
1778 static int mdt_reint_internal(struct mdt_thread_info *info,
1779                               struct mdt_lock_handle *lhc,
1780                               __u32 op)
1781 {
1782         struct req_capsule      *pill = info->mti_pill;
1783         struct mdt_body         *repbody;
1784         int                      rc = 0, rc2;
1785
1786         ENTRY;
1787
1788         rc = mdt_reint_unpack(info, op);
1789         if (rc != 0) {
1790                 CERROR("Can't unpack reint, rc %d\n", rc);
1791                 RETURN(err_serious(rc));
1792         }
1793
1794         /* for replay (no_create) lmm is not needed, client has it already */
1795         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1796                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1797                                      DEF_REP_MD_SIZE);
1798
1799         /* llog cookies are always 0, the field is kept for compatibility */
1800         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1801                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
1802
1803         rc = req_capsule_server_pack(pill);
1804         if (rc != 0) {
1805                 CERROR("Can't pack response, rc %d\n", rc);
1806                 RETURN(err_serious(rc));
1807         }
1808
1809         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1810                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1811                 LASSERT(repbody);
1812                 repbody->eadatasize = 0;
1813                 repbody->aclsize = 0;
1814         }
1815
1816         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1817
1818         /* for replay no cookkie / lmm need, because client have this already */
1819         if (info->mti_spec.no_create)
1820                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1821                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1822
1823         rc = mdt_init_ucred_reint(info);
1824         if (rc)
1825                 GOTO(out_shrink, rc);
1826
1827         rc = mdt_fix_attr_ucred(info, op);
1828         if (rc != 0)
1829                 GOTO(out_ucred, rc = err_serious(rc));
1830
1831         if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1832                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
1833                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1834                 GOTO(out_ucred, rc);
1835         }
1836         rc = mdt_reint_rec(info, lhc);
1837         EXIT;
1838 out_ucred:
1839         mdt_exit_ucred(info);
1840 out_shrink:
1841         mdt_client_compatibility(info);
1842         rc2 = mdt_fix_reply(info);
1843         if (rc == 0)
1844                 rc = rc2;
1845         return rc;
1846 }
1847
1848 static long mdt_reint_opcode(struct ptlrpc_request *req,
1849                              const struct req_format **fmt)
1850 {
1851         struct mdt_device       *mdt;
1852         struct mdt_rec_reint    *rec;
1853         long                     opc;
1854
1855         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
1856         if (rec != NULL) {
1857                 opc = rec->rr_opcode;
1858                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
1859                 if (opc < REINT_MAX && fmt[opc] != NULL)
1860                         req_capsule_extend(&req->rq_pill, fmt[opc]);
1861                 else {
1862                         mdt = mdt_exp2dev(req->rq_export);
1863                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
1864                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
1865                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
1866                         opc = err_serious(-EFAULT);
1867                 }
1868         } else {
1869                 opc = err_serious(-EFAULT);
1870         }
1871         return opc;
1872 }
1873
1874 static int mdt_reint(struct tgt_session_info *tsi)
1875 {
1876         long opc;
1877         int  rc;
1878         static const struct req_format *reint_fmts[REINT_MAX] = {
1879                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
1880                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
1881                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
1882                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
1883                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
1884                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
1885                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
1886                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
1887                 [REINT_MIGRATE]  = &RQF_MDS_REINT_RENAME
1888         };
1889
1890         ENTRY;
1891
1892         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
1893         if (opc >= 0) {
1894                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1895                 /*
1896                  * No lock possible here from client to pass it to reint code
1897                  * path.
1898                  */
1899                 rc = mdt_reint_internal(info, NULL, opc);
1900                 mdt_thread_info_fini(info);
1901         } else {
1902                 rc = opc;
1903         }
1904
1905         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1906         RETURN(rc);
1907 }
1908
1909 /* this should sync the whole device */
1910 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1911 {
1912         struct dt_device *dt = mdt->mdt_bottom;
1913         int rc;
1914         ENTRY;
1915
1916         rc = dt->dd_ops->dt_sync(env, dt);
1917         RETURN(rc);
1918 }
1919
1920 /* this should sync this object */
1921 static int mdt_object_sync(struct mdt_thread_info *info)
1922 {
1923         struct md_object *next;
1924         int rc;
1925         ENTRY;
1926
1927         if (!mdt_object_exists(info->mti_object)) {
1928                 CWARN("Non existing object  "DFID"!\n",
1929                       PFID(mdt_object_fid(info->mti_object)));
1930                 RETURN(-ESTALE);
1931         }
1932         next = mdt_object_child(info->mti_object);
1933         rc = mo_object_sync(info->mti_env, next);
1934
1935         RETURN(rc);
1936 }
1937
1938 static int mdt_sync(struct tgt_session_info *tsi)
1939 {
1940         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1941         struct req_capsule      *pill = tsi->tsi_pill;
1942         struct mdt_body         *body;
1943         int                      rc;
1944
1945         ENTRY;
1946
1947         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1948                 RETURN(err_serious(-ENOMEM));
1949
1950         if (fid_seq(&tsi->tsi_mdt_body->fid1) == 0) {
1951                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
1952         } else {
1953                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1954
1955                 /* sync an object */
1956                 rc = mdt_object_sync(info);
1957                 if (rc == 0) {
1958                         const struct lu_fid *fid;
1959                         struct lu_attr *la = &info->mti_attr.ma_attr;
1960
1961                         info->mti_attr.ma_need = MA_INODE;
1962                         info->mti_attr.ma_valid = 0;
1963                         rc = mdt_attr_get_complex(info, info->mti_object,
1964                                                   &info->mti_attr);
1965                         if (rc == 0) {
1966                                 body = req_capsule_server_get(pill,
1967                                                               &RMF_MDT_BODY);
1968                                 fid = mdt_object_fid(info->mti_object);
1969                                 mdt_pack_attr2body(info, body, la, fid);
1970                         }
1971                 }
1972                 mdt_thread_info_fini(info);
1973         }
1974         if (rc == 0)
1975                 mdt_counter_incr(req, LPROC_MDT_SYNC);
1976
1977         RETURN(rc);
1978 }
1979
1980 /*
1981  * Handle quota control requests to consult current usage/limit, but also
1982  * to configure quota enforcement
1983  */
1984 static int mdt_quotactl(struct tgt_session_info *tsi)
1985 {
1986         struct obd_export       *exp  = tsi->tsi_exp;
1987         struct req_capsule      *pill = tsi->tsi_pill;
1988         struct obd_quotactl     *oqctl, *repoqc;
1989         int                      id, rc;
1990         struct mdt_device       *mdt = mdt_exp2dev(exp);
1991         struct lu_device        *qmt = mdt->mdt_qmt_dev;
1992         ENTRY;
1993
1994         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1995         if (oqctl == NULL)
1996                 RETURN(err_serious(-EPROTO));
1997
1998         rc = req_capsule_server_pack(pill);
1999         if (rc)
2000                 RETURN(err_serious(rc));
2001
2002         switch (oqctl->qc_cmd) {
2003         case Q_QUOTACHECK:
2004         case LUSTRE_Q_INVALIDATE:
2005         case LUSTRE_Q_FINVALIDATE:
2006         case Q_QUOTAON:
2007         case Q_QUOTAOFF:
2008         case Q_INITQUOTA:
2009                 /* deprecated, not used any more */
2010                 RETURN(-EOPNOTSUPP);
2011                 /* master quotactl */
2012         case Q_GETINFO:
2013         case Q_SETINFO:
2014         case Q_SETQUOTA:
2015         case Q_GETQUOTA:
2016                 if (qmt == NULL)
2017                         RETURN(-EOPNOTSUPP);
2018                 /* slave quotactl */
2019         case Q_GETOINFO:
2020         case Q_GETOQUOTA:
2021                 break;
2022         default:
2023                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2024                 RETURN(-EFAULT);
2025         }
2026
2027         /* map uid/gid for remote client */
2028         id = oqctl->qc_id;
2029         if (exp_connect_rmtclient(exp)) {
2030                 struct lustre_idmap_table *idmap;
2031
2032                 idmap = exp->exp_mdt_data.med_idmap;
2033
2034                 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
2035                              oqctl->qc_cmd != Q_GETINFO))
2036                         RETURN(-EPERM);
2037
2038                 if (oqctl->qc_type == USRQUOTA)
2039                         id = lustre_idmap_lookup_uid(NULL, idmap, 0,
2040                                                      oqctl->qc_id);
2041                 else if (oqctl->qc_type == GRPQUOTA)
2042                         id = lustre_idmap_lookup_gid(NULL, idmap, 0,
2043                                                      oqctl->qc_id);
2044                 else
2045                         RETURN(-EINVAL);
2046
2047                 if (id == CFS_IDMAP_NOTFOUND) {
2048                         CDEBUG(D_QUOTA, "no mapping for id %u\n", oqctl->qc_id);
2049                         RETURN(-EACCES);
2050                 }
2051         }
2052
2053         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2054         if (repoqc == NULL)
2055                 RETURN(err_serious(-EFAULT));
2056
2057         if (oqctl->qc_id != id)
2058                 swap(oqctl->qc_id, id);
2059
2060         switch (oqctl->qc_cmd) {
2061
2062         case Q_GETINFO:
2063         case Q_SETINFO:
2064         case Q_SETQUOTA:
2065         case Q_GETQUOTA:
2066                 /* forward quotactl request to QMT */
2067                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2068                 break;
2069
2070         case Q_GETOINFO:
2071         case Q_GETOQUOTA:
2072                 /* slave quotactl */
2073                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2074                                    oqctl);
2075                 break;
2076
2077         default:
2078                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2079                 RETURN(-EFAULT);
2080         }
2081
2082         if (oqctl->qc_id != id)
2083                 swap(oqctl->qc_id, id);
2084
2085         *repoqc = *oqctl;
2086         RETURN(rc);
2087 }
2088
2089 /** clone llog ctxt from child (mdd)
2090  * This allows remote llog (replicator) access.
2091  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2092  * context was originally set up, or we can handle them directly.
2093  * I choose the latter, but that means I need any llog
2094  * contexts set up by child to be accessable by the mdt.  So we clone the
2095  * context into our context list here.
2096  */
2097 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2098                                int idx)
2099 {
2100         struct md_device  *next = mdt->mdt_child;
2101         struct llog_ctxt *ctxt;
2102         int rc;
2103
2104         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2105                 return 0;
2106
2107         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2108         if (rc || ctxt == NULL) {
2109                 return 0;
2110         }
2111
2112         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2113         if (rc)
2114                 CERROR("Can't set mdt ctxt %d\n", rc);
2115
2116         return rc;
2117 }
2118
2119 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2120                                  struct mdt_device *mdt, int idx)
2121 {
2122         struct llog_ctxt *ctxt;
2123
2124         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2125         if (ctxt == NULL)
2126                 return 0;
2127         /* Put once for the get we just did, and once for the clone */
2128         llog_ctxt_put(ctxt);
2129         llog_ctxt_put(ctxt);
2130         return 0;
2131 }
2132
2133 /*
2134  * sec context handlers
2135  */
2136 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2137 {
2138         int rc;
2139
2140         rc = mdt_handle_idmap(tsi);
2141         if (unlikely(rc)) {
2142                 struct ptlrpc_request   *req = tgt_ses_req(tsi);
2143                 __u32                    opc;
2144
2145                 opc = lustre_msg_get_opc(req->rq_reqmsg);
2146                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2147                         sptlrpc_svc_ctx_invalidate(req);
2148         }
2149
2150         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2151
2152         return rc;
2153 }
2154
2155 /*
2156  * quota request handlers
2157  */
2158 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2159 {
2160         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
2161         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2162         int                      rc;
2163         ENTRY;
2164
2165         if (qmt == NULL)
2166                 RETURN(err_serious(-EOPNOTSUPP));
2167
2168         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2169         RETURN(rc);
2170 }
2171
2172 struct mdt_object *mdt_object_new(const struct lu_env *env,
2173                                   struct mdt_device *d,
2174                                   const struct lu_fid *f)
2175 {
2176         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2177         struct lu_object *o;
2178         struct mdt_object *m;
2179         ENTRY;
2180
2181         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2182         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2183         if (unlikely(IS_ERR(o)))
2184                 m = (struct mdt_object *)o;
2185         else
2186                 m = mdt_obj(o);
2187         RETURN(m);
2188 }
2189
2190 struct mdt_object *mdt_object_find(const struct lu_env *env,
2191                                    struct mdt_device *d,
2192                                    const struct lu_fid *f)
2193 {
2194         struct lu_object *o;
2195         struct mdt_object *m;
2196         ENTRY;
2197
2198         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2199         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2200         if (unlikely(IS_ERR(o)))
2201                 m = (struct mdt_object *)o;
2202         else
2203                 m = mdt_obj(o);
2204
2205         RETURN(m);
2206 }
2207
2208 /**
2209  * Asyncronous commit for mdt device.
2210  *
2211  * Pass asynchonous commit call down the MDS stack.
2212  *
2213  * \param env environment
2214  * \param mdt the mdt device
2215  */
2216 static void mdt_device_commit_async(const struct lu_env *env,
2217                                     struct mdt_device *mdt)
2218 {
2219         struct dt_device *dt = mdt->mdt_bottom;
2220         int rc;
2221
2222         rc = dt->dd_ops->dt_commit_async(env, dt);
2223         if (unlikely(rc != 0))
2224                 CWARN("async commit start failed with rc = %d", rc);
2225 }
2226
2227 /**
2228  * Mark the lock as "synchonous".
2229  *
2230  * Mark the lock to deffer transaction commit to the unlock time.
2231  *
2232  * \param lock the lock to mark as "synchonous"
2233  *
2234  * \see mdt_is_lock_sync
2235  * \see mdt_save_lock
2236  */
2237 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2238 {
2239         lock->l_ast_data = (void*)1;
2240 }
2241
2242 /**
2243  * Check whehter the lock "synchonous" or not.
2244  *
2245  * \param lock the lock to check
2246  * \retval 1 the lock is "synchonous"
2247  * \retval 0 the lock isn't "synchronous"
2248  *
2249  * \see mdt_set_lock_sync
2250  * \see mdt_save_lock
2251  */
2252 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2253 {
2254         return lock->l_ast_data != NULL;
2255 }
2256
2257 /**
2258  * Blocking AST for mdt locks.
2259  *
2260  * Starts transaction commit if in case of COS lock conflict or
2261  * deffers such a commit to the mdt_save_lock.
2262  *
2263  * \param lock the lock which blocks a request or cancelling lock
2264  * \param desc unused
2265  * \param data unused
2266  * \param flag indicates whether this cancelling or blocking callback
2267  * \retval 0
2268  * \see ldlm_blocking_ast_nocheck
2269  */
2270 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2271                      void *data, int flag)
2272 {
2273         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2274         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2275         int rc;
2276         ENTRY;
2277
2278         if (flag == LDLM_CB_CANCELING)
2279                 RETURN(0);
2280         lock_res_and_lock(lock);
2281         if (lock->l_blocking_ast != mdt_blocking_ast) {
2282                 unlock_res_and_lock(lock);
2283                 RETURN(0);
2284         }
2285         if (mdt_cos_is_enabled(mdt) &&
2286             lock->l_req_mode & (LCK_PW | LCK_EX) &&
2287             lock->l_blocking_lock != NULL &&
2288             lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
2289                 mdt_set_lock_sync(lock);
2290         }
2291         rc = ldlm_blocking_ast_nocheck(lock);
2292
2293         /* There is no lock conflict if l_blocking_lock == NULL,
2294          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2295          * when the last reference to a local lock was released */
2296         if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2297                 struct lu_env env;
2298
2299                 rc = lu_env_init(&env, LCT_LOCAL);
2300                 if (unlikely(rc != 0))
2301                         CWARN("lu_env initialization failed with rc = %d,"
2302                               "cannot start asynchronous commit\n", rc);
2303                 else
2304                         mdt_device_commit_async(&env, mdt);
2305                 lu_env_fini(&env);
2306         }
2307         RETURN(rc);
2308 }
2309
2310 /* Used for cross-MDT lock */
2311 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2312                             void *data, int flag)
2313 {
2314         struct lustre_handle lockh;
2315         int               rc;
2316
2317         switch (flag) {
2318         case LDLM_CB_BLOCKING:
2319                 ldlm_lock2handle(lock, &lockh);
2320                 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
2321                 if (rc < 0) {
2322                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2323                         RETURN(rc);
2324                 }
2325                 break;
2326         case LDLM_CB_CANCELING:
2327                 LDLM_DEBUG(lock, "Revoke remote lock\n");
2328                 break;
2329         default:
2330                 LBUG();
2331         }
2332         RETURN(0);
2333 }
2334
2335 int mdt_remote_object_lock(struct mdt_thread_info *mti,
2336                            struct mdt_object *o, const struct lu_fid *fid,
2337                            struct lustre_handle *lh, ldlm_mode_t mode,
2338                            __u64 ibits)
2339 {
2340         struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
2341         ldlm_policy_data_t *policy = &mti->mti_policy;
2342         struct ldlm_res_id *res_id = &mti->mti_res_id;
2343         int rc = 0;
2344         ENTRY;
2345
2346         LASSERT(mdt_object_remote(o));
2347
2348         LASSERT(ibits == MDS_INODELOCK_UPDATE);
2349
2350         fid_build_reg_res_name(fid, res_id);
2351
2352         memset(einfo, 0, sizeof(*einfo));
2353         einfo->ei_type = LDLM_IBITS;
2354         einfo->ei_mode = mode;
2355         einfo->ei_cb_bl = mdt_remote_blocking_ast;
2356         einfo->ei_cb_cp = ldlm_completion_ast;
2357         einfo->ei_enq_slave = 0;
2358         einfo->ei_res_id = res_id;
2359
2360         memset(policy, 0, sizeof(*policy));
2361         policy->l_inodebits.bits = ibits;
2362
2363         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2364                             policy);
2365         RETURN(rc);
2366 }
2367
2368 static int mdt_object_local_lock(struct mdt_thread_info *info,
2369                                  struct mdt_object *o,
2370                                  struct mdt_lock_handle *lh, __u64 ibits,
2371                                  bool nonblock, int locality)
2372 {
2373         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2374         ldlm_policy_data_t *policy = &info->mti_policy;
2375         struct ldlm_res_id *res_id = &info->mti_res_id;
2376         __u64 dlmflags;
2377         int rc;
2378         ENTRY;
2379
2380         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2381         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2382         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2383         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2384
2385         /* Only enqueue LOOKUP lock for remote object */
2386         if (mdt_object_remote(o))
2387                 LASSERT(ibits == MDS_INODELOCK_LOOKUP);
2388
2389         if (lh->mlh_type == MDT_PDO_LOCK) {
2390                 /* check for exists after object is locked */
2391                 if (mdt_object_exists(o) == 0) {
2392                         /* Non-existent object shouldn't have PDO lock */
2393                         RETURN(-ESTALE);
2394                 } else {
2395                         /* Non-dir object shouldn't have PDO lock */
2396                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2397                                 RETURN(-ENOTDIR);
2398                 }
2399         }
2400
2401         memset(policy, 0, sizeof(*policy));
2402         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2403
2404         dlmflags = LDLM_FL_ATOMIC_CB;
2405         if (nonblock)
2406                 dlmflags |= LDLM_FL_BLOCK_NOWAIT;
2407
2408         /*
2409          * Take PDO lock on whole directory and build correct @res_id for lock
2410          * on part of directory.
2411          */
2412         if (lh->mlh_pdo_hash != 0) {
2413                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2414                 mdt_lock_pdo_mode(info, o, lh);
2415                 if (lh->mlh_pdo_mode != LCK_NL) {
2416                         /*
2417                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2418                          * is never going to be sent to client and we do not
2419                          * want it slowed down due to possible cancels.
2420                          */
2421                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2422                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2423                                           policy, res_id, dlmflags,
2424                                           info->mti_exp == NULL ? NULL :
2425                                           &info->mti_exp->exp_handle.h_cookie);
2426                         if (unlikely(rc))
2427                                 RETURN(rc);
2428                 }
2429
2430                 /*
2431                  * Finish res_id initializing by name hash marking part of
2432                  * directory which is taking modification.
2433                  */
2434                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2435         }
2436
2437         policy->l_inodebits.bits = ibits;
2438
2439         /*
2440          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2441          * going to be sent to client. If it is - mdt_intent_policy() path will
2442          * fix it up and turn FL_LOCAL flag off.
2443          */
2444         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2445                           res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2446                           info->mti_exp == NULL ? NULL :
2447                           &info->mti_exp->exp_handle.h_cookie);
2448         if (rc)
2449                 mdt_object_unlock(info, o, lh, 1);
2450         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2451                  lh->mlh_pdo_hash != 0 &&
2452                  (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX)) {
2453                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2454         }
2455
2456         RETURN(rc);
2457 }
2458
2459 static int
2460 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
2461                          struct mdt_lock_handle *lh, __u64 ibits,
2462                          bool nonblock, int locality)
2463 {
2464         int rc;
2465         ENTRY;
2466
2467         if (!mdt_object_remote(o))
2468                 return mdt_object_local_lock(info, o, lh, ibits, nonblock,
2469                                              locality);
2470
2471         if (locality == MDT_LOCAL_LOCK) {
2472                 CERROR("%s: try to get local lock for remote object"
2473                        DFID".\n", mdt_obd_name(info->mti_mdt),
2474                        PFID(mdt_object_fid(o)));
2475                 RETURN(-EPROTO);
2476         }
2477
2478         /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
2479         ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
2480                    MDS_INODELOCK_XATTR);
2481         if (ibits & MDS_INODELOCK_UPDATE) {
2482                 /* Sigh, PDO needs to enqueue 2 locks right now, but
2483                  * enqueue RPC can only request 1 lock, to avoid extra
2484                  * RPC, so it will instead enqueue EX lock for remote
2485                  * object anyway XXX*/
2486                 if (lh->mlh_type == MDT_PDO_LOCK &&
2487                     lh->mlh_pdo_hash != 0) {
2488                         CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
2489                                "EX lock.\n", mdt_obd_name(info->mti_mdt),
2490                                PFID(mdt_object_fid(o)));
2491                         lh->mlh_pdo_hash = 0;
2492                         lh->mlh_rreg_mode = LCK_EX;
2493                         lh->mlh_type = MDT_REG_LOCK;
2494                 }
2495                 rc = mdt_remote_object_lock(info, o, mdt_object_fid(o),
2496                                             &lh->mlh_rreg_lh,
2497                                             lh->mlh_rreg_mode,
2498                                             MDS_INODELOCK_UPDATE);
2499                 if (rc != ELDLM_OK)
2500                         RETURN(rc);
2501         }
2502
2503         /* Only enqueue LOOKUP lock for remote object */
2504         if (ibits & MDS_INODELOCK_LOOKUP) {
2505                 rc = mdt_object_local_lock(info, o, lh,
2506                                            MDS_INODELOCK_LOOKUP,
2507                                            nonblock, locality);
2508                 if (rc != ELDLM_OK)
2509                         RETURN(rc);
2510         }
2511
2512         RETURN(0);
2513 }
2514
2515 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2516                     struct mdt_lock_handle *lh, __u64 ibits, int locality)
2517 {
2518         return mdt_object_lock_internal(info, o, lh, ibits, false, locality);
2519 }
2520
2521 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
2522                         struct mdt_lock_handle *lh, __u64 ibits, int locality)
2523 {
2524         struct mdt_lock_handle tmp = *lh;
2525         int rc;
2526
2527         rc = mdt_object_lock_internal(info, o, &tmp, ibits, true, locality);
2528         if (rc == 0)
2529                 *lh = tmp;
2530
2531         return rc == 0;
2532 }
2533
2534 /**
2535  * Save a lock within request object.
2536  *
2537  * Keep the lock referenced until whether client ACK or transaction
2538  * commit happens or release the lock immediately depending on input
2539  * parameters. If COS is ON, a write lock is converted to COS lock
2540  * before saving.
2541  *
2542  * \param info thead info object
2543  * \param h lock handle
2544  * \param mode lock mode
2545  * \param decref force immediate lock releasing
2546  */
2547 static
2548 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2549                    ldlm_mode_t mode, int decref)
2550 {
2551         ENTRY;
2552
2553         if (lustre_handle_is_used(h)) {
2554                 if (decref || !info->mti_has_trans ||
2555                     !(mode & (LCK_PW | LCK_EX))){
2556                         mdt_fid_unlock(h, mode);
2557                 } else {
2558                         struct mdt_device *mdt = info->mti_mdt;
2559                         struct ldlm_lock *lock = ldlm_handle2lock(h);
2560                         struct ptlrpc_request *req = mdt_info_req(info);
2561                         int no_ack = 0;
2562
2563                         LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2564                                  h->cookie);
2565                         /* there is no request if mdt_object_unlock() is called
2566                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
2567                         if (likely(req != NULL)) {
2568                                 CDEBUG(D_HA, "request = %p reply state = %p"
2569                                        " transno = "LPD64"\n", req,
2570                                        req->rq_reply_state, req->rq_transno);
2571                                 if (mdt_cos_is_enabled(mdt)) {
2572                                         no_ack = 1;
2573                                         ldlm_lock_downgrade(lock, LCK_COS);
2574                                         mode = LCK_COS;
2575                                 }
2576                                 ptlrpc_save_lock(req, h, mode, no_ack);
2577                         } else {
2578                                 ldlm_lock_decref(h, mode);
2579                         }
2580                         if (mdt_is_lock_sync(lock)) {
2581                                 CDEBUG(D_HA, "found sync-lock,"
2582                                        " async commit started\n");
2583                                 mdt_device_commit_async(info->mti_env,
2584                                                         mdt);
2585                         }
2586                         LDLM_LOCK_PUT(lock);
2587                 }
2588                 h->cookie = 0ull;
2589         }
2590
2591         EXIT;
2592 }
2593
2594 /**
2595  * Unlock mdt object.
2596  *
2597  * Immeditely release the regular lock and the PDO lock or save the
2598  * lock in reqeuest and keep them referenced until client ACK or
2599  * transaction commit.
2600  *
2601  * \param info thread info object
2602  * \param o mdt object
2603  * \param lh mdt lock handle referencing regular and PDO locks
2604  * \param decref force immediate lock releasing
2605  */
2606 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2607                        struct mdt_lock_handle *lh, int decref)
2608 {
2609         ENTRY;
2610
2611         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2612         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2613
2614         if (lustre_handle_is_used(&lh->mlh_rreg_lh))
2615                 ldlm_lock_decref(&lh->mlh_rreg_lh, lh->mlh_rreg_mode);
2616
2617         EXIT;
2618 }
2619
2620 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2621                                         const struct lu_fid *f,
2622                                         struct mdt_lock_handle *lh,
2623                                         __u64 ibits)
2624 {
2625         struct mdt_object *o;
2626
2627         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2628         if (!IS_ERR(o)) {
2629                 int rc;
2630
2631                 rc = mdt_object_lock(info, o, lh, ibits,
2632                                      MDT_LOCAL_LOCK);
2633                 if (rc != 0) {
2634                         mdt_object_put(info->mti_env, o);
2635                         o = ERR_PTR(rc);
2636                 }
2637         }
2638         return o;
2639 }
2640
2641 void mdt_object_unlock_put(struct mdt_thread_info * info,
2642                            struct mdt_object * o,
2643                            struct mdt_lock_handle *lh,
2644                            int decref)
2645 {
2646         mdt_object_unlock(info, o, lh, decref);
2647         mdt_object_put(info->mti_env, o);
2648 }
2649
2650 /*
2651  * Generic code handling requests that have struct mdt_body passed in:
2652  *
2653  *  - extract mdt_body from request and save it in @info, if present;
2654  *
2655  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
2656  *  @info;
2657  *
2658  *  - if HABEO_CORPUS flag is set for this request type check whether object
2659  *  actually exists on storage (lu_object_exists()).
2660  *
2661  */
2662 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2663 {
2664         const struct mdt_body    *body;
2665         struct mdt_object        *obj;
2666         const struct lu_env      *env;
2667         struct req_capsule       *pill;
2668         int                       rc;
2669         ENTRY;
2670
2671         env = info->mti_env;
2672         pill = info->mti_pill;
2673
2674         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2675         if (body == NULL)
2676                 RETURN(-EFAULT);
2677
2678         if (!(body->valid & OBD_MD_FLID))
2679                 RETURN(0);
2680
2681         if (!fid_is_sane(&body->fid1)) {
2682                 CERROR("Invalid fid: "DFID"\n", PFID(&body->fid1));
2683                 RETURN(-EINVAL);
2684         }
2685
2686         /*
2687          * Do not get size or any capa fields before we check that request
2688          * contains capa actually. There are some requests which do not, for
2689          * instance MDS_IS_SUBDIR.
2690          */
2691         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2692             req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
2693                 mdt_set_capainfo(info, 0, &body->fid1,
2694                                  req_capsule_client_get(pill, &RMF_CAPA1));
2695
2696         obj = mdt_object_find(env, info->mti_mdt, &body->fid1);
2697         if (!IS_ERR(obj)) {
2698                 if ((flags & HABEO_CORPUS) &&
2699                     !mdt_object_exists(obj)) {
2700                         mdt_object_put(env, obj);
2701                         /* for capability renew ENOENT will be handled in
2702                          * mdt_renew_capa */
2703                         if (body->valid & OBD_MD_FLOSSCAPA)
2704                                 rc = 0;
2705                         else
2706                                 rc = -ENOENT;
2707                 } else {
2708                         info->mti_object = obj;
2709                         rc = 0;
2710                 }
2711         } else
2712                 rc = PTR_ERR(obj);
2713
2714         RETURN(rc);
2715 }
2716
2717 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2718 {
2719         struct req_capsule *pill = info->mti_pill;
2720         int rc;
2721         ENTRY;
2722
2723         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2724                 rc = mdt_body_unpack(info, flags);
2725         else
2726                 rc = 0;
2727
2728         if (rc == 0 && (flags & HABEO_REFERO)) {
2729                 /* Pack reply. */
2730                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2731                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2732                                              DEF_REP_MD_SIZE);
2733                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2734                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
2735                                              RCL_SERVER, 0);
2736
2737                 rc = req_capsule_server_pack(pill);
2738         }
2739         RETURN(rc);
2740 }
2741
2742 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2743 {
2744         struct md_device *next = m->mdt_child;
2745
2746         return next->md_ops->mdo_init_capa_ctxt(env, next,
2747                                                 m->mdt_lut.lut_mds_capa,
2748                                                 m->mdt_capa_timeout,
2749                                                 m->mdt_capa_alg,
2750                                                 m->mdt_capa_keys);
2751 }
2752
2753 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2754 {
2755         lh->mlh_type = MDT_NUL_LOCK;
2756         lh->mlh_reg_lh.cookie = 0ull;
2757         lh->mlh_reg_mode = LCK_MINMODE;
2758         lh->mlh_pdo_lh.cookie = 0ull;
2759         lh->mlh_pdo_mode = LCK_MINMODE;
2760         lh->mlh_rreg_lh.cookie = 0ull;
2761         lh->mlh_rreg_mode = LCK_MINMODE;
2762 }
2763
2764 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2765 {
2766         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2767         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2768 }
2769
2770 /*
2771  * Initialize fields of struct mdt_thread_info. Other fields are left in
2772  * uninitialized state, because it's too expensive to zero out whole
2773  * mdt_thread_info (> 1K) on each request arrival.
2774  */
2775 void mdt_thread_info_init(struct ptlrpc_request *req,
2776                           struct mdt_thread_info *info)
2777 {
2778         int i;
2779
2780         info->mti_pill = &req->rq_pill;
2781
2782         /* lock handle */
2783         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2784                 mdt_lock_handle_init(&info->mti_lh[i]);
2785
2786         /* mdt device: it can be NULL while CONNECT */
2787         if (req->rq_export) {
2788                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2789                 info->mti_exp = req->rq_export;
2790         } else
2791                 info->mti_mdt = NULL;
2792         info->mti_env = req->rq_svc_thread->t_env;
2793         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2794
2795         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2796         info->mti_big_buf = LU_BUF_NULL;
2797         info->mti_body = NULL;
2798         info->mti_object = NULL;
2799         info->mti_dlm_req = NULL;
2800         info->mti_has_trans = 0;
2801         info->mti_cross_ref = 0;
2802         info->mti_opdata = 0;
2803         info->mti_big_lmm_used = 0;
2804
2805         info->mti_spec.no_create = 0;
2806         info->mti_spec.sp_rm_entry = 0;
2807
2808         info->mti_spec.u.sp_ea.eadata = NULL;
2809         info->mti_spec.u.sp_ea.eadatalen = 0;
2810 }
2811
2812 void mdt_thread_info_fini(struct mdt_thread_info *info)
2813 {
2814         int i;
2815
2816         if (info->mti_object != NULL) {
2817                 mdt_object_put(info->mti_env, info->mti_object);
2818                 info->mti_object = NULL;
2819         }
2820
2821         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2822                 mdt_lock_handle_fini(&info->mti_lh[i]);
2823         info->mti_env = NULL;
2824         info->mti_pill = NULL;
2825         info->mti_exp = NULL;
2826
2827         if (unlikely(info->mti_big_buf.lb_buf != NULL))
2828                 lu_buf_free(&info->mti_big_buf);
2829 }
2830
2831 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
2832 {
2833         struct mdt_thread_info  *mti;
2834         struct lustre_capa      *lc;
2835
2836         mti = mdt_th_info(tsi->tsi_env);
2837         LASSERT(mti != NULL);
2838
2839         mdt_thread_info_init(tgt_ses_req(tsi), mti);
2840         if (tsi->tsi_corpus != NULL) {
2841                 struct req_capsule *pill = tsi->tsi_pill;
2842
2843                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
2844                 lu_object_get(tsi->tsi_corpus);
2845
2846                 /*
2847                  * XXX: must be part of tgt_mdt_body_unpack but moved here
2848                  * due to mdt_set_capainfo().
2849                  */
2850                 if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2851                     req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT) > 0) {
2852                         lc = req_capsule_client_get(pill, &RMF_CAPA1);
2853                         mdt_set_capainfo(mti, 0, &tsi->tsi_mdt_body->fid1, lc);
2854                 }
2855         }
2856         mti->mti_body = tsi->tsi_mdt_body;
2857         mti->mti_dlm_req = tsi->tsi_dlm_req;
2858
2859         return mti;
2860 }
2861
2862 static int mdt_tgt_connect(struct tgt_session_info *tsi)
2863 {
2864         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2865         int                      rc;
2866
2867         ENTRY;
2868
2869         rc = tgt_connect(tsi);
2870         if (rc != 0)
2871                 RETURN(rc);
2872
2873         rc = mdt_init_idmap(tsi);
2874         if (rc != 0)
2875                 GOTO(err, rc);
2876         RETURN(0);
2877 err:
2878         obd_disconnect(class_export_get(req->rq_export));
2879         return rc;
2880 }
2881
2882 enum mdt_it_code {
2883         MDT_IT_OPEN,
2884         MDT_IT_OCREAT,
2885         MDT_IT_CREATE,
2886         MDT_IT_GETATTR,
2887         MDT_IT_READDIR,
2888         MDT_IT_LOOKUP,
2889         MDT_IT_UNLINK,
2890         MDT_IT_TRUNC,
2891         MDT_IT_GETXATTR,
2892         MDT_IT_LAYOUT,
2893         MDT_IT_QUOTA,
2894         MDT_IT_NR
2895 };
2896
2897 static int mdt_intent_getattr(enum mdt_it_code opcode,
2898                               struct mdt_thread_info *info,
2899                               struct ldlm_lock **,
2900                               __u64);
2901
2902 static int mdt_intent_getxattr(enum mdt_it_code opcode,
2903                                 struct mdt_thread_info *info,
2904                                 struct ldlm_lock **lockp,
2905                                 __u64 flags);
2906
2907 static int mdt_intent_layout(enum mdt_it_code opcode,
2908                              struct mdt_thread_info *info,
2909                              struct ldlm_lock **,
2910                              __u64);
2911 static int mdt_intent_reint(enum mdt_it_code opcode,
2912                             struct mdt_thread_info *info,
2913                             struct ldlm_lock **,
2914                             __u64);
2915
2916 static struct mdt_it_flavor {
2917         const struct req_format *it_fmt;
2918         __u32                    it_flags;
2919         int                    (*it_act)(enum mdt_it_code ,
2920                                          struct mdt_thread_info *,
2921                                          struct ldlm_lock **,
2922                                          __u64);
2923         long                     it_reint;
2924 } mdt_it_flavor[] = {
2925         [MDT_IT_OPEN]     = {
2926                 .it_fmt   = &RQF_LDLM_INTENT,
2927                 /*.it_flags = HABEO_REFERO,*/
2928                 .it_flags = 0,
2929                 .it_act   = mdt_intent_reint,
2930                 .it_reint = REINT_OPEN
2931         },
2932         [MDT_IT_OCREAT]   = {
2933                 .it_fmt   = &RQF_LDLM_INTENT,
2934                 /*
2935                  * OCREAT is not a MUTABOR request as if the file
2936                  * already exists.
2937                  * We do the extra check of OBD_CONNECT_RDONLY in
2938                  * mdt_reint_open() when we really need to create
2939                  * the object.
2940                  */
2941                 .it_flags = 0,
2942                 .it_act   = mdt_intent_reint,
2943                 .it_reint = REINT_OPEN
2944         },
2945         [MDT_IT_CREATE]   = {
2946                 .it_fmt   = &RQF_LDLM_INTENT,
2947                 .it_flags = MUTABOR,
2948                 .it_act   = mdt_intent_reint,
2949                 .it_reint = REINT_CREATE
2950         },
2951         [MDT_IT_GETATTR]  = {
2952                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2953                 .it_flags = HABEO_REFERO,
2954                 .it_act   = mdt_intent_getattr
2955         },
2956         [MDT_IT_READDIR]  = {
2957                 .it_fmt   = NULL,
2958                 .it_flags = 0,
2959                 .it_act   = NULL
2960         },
2961         [MDT_IT_LOOKUP]   = {
2962                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2963                 .it_flags = HABEO_REFERO,
2964                 .it_act   = mdt_intent_getattr
2965         },
2966         [MDT_IT_UNLINK]   = {
2967                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
2968                 .it_flags = MUTABOR,
2969                 .it_act   = NULL,
2970                 .it_reint = REINT_UNLINK
2971         },
2972         [MDT_IT_TRUNC]    = {
2973                 .it_fmt   = NULL,
2974                 .it_flags = MUTABOR,
2975                 .it_act   = NULL
2976         },
2977         [MDT_IT_GETXATTR] = {
2978                 .it_fmt   = &RQF_LDLM_INTENT_GETXATTR,
2979                 .it_flags = HABEO_CORPUS,
2980                 .it_act   = mdt_intent_getxattr
2981         },
2982         [MDT_IT_LAYOUT] = {
2983                 .it_fmt   = &RQF_LDLM_INTENT_LAYOUT,
2984                 .it_flags = 0,
2985                 .it_act   = mdt_intent_layout
2986         }
2987 };
2988
2989 static int
2990 mdt_intent_lock_replace(struct mdt_thread_info *info, struct ldlm_lock **lockp,
2991                         struct ldlm_lock *new_lock, struct mdt_lock_handle *lh,
2992                         __u64 flags)
2993 {
2994         struct ptlrpc_request  *req = mdt_info_req(info);
2995         struct ldlm_lock       *lock = *lockp;
2996
2997         /*
2998          * Get new lock only for cases when possible resent did not find any
2999          * lock.
3000          */
3001         if (new_lock == NULL)
3002                 new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3003
3004         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3005                 lh->mlh_reg_lh.cookie = 0;
3006                 RETURN(0);
3007         }
3008
3009         LASSERTF(new_lock != NULL,
3010                  "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
3011
3012         /*
3013          * If we've already given this lock to a client once, then we should
3014          * have no readers or writers.  Otherwise, we should have one reader
3015          * _or_ writer ref (which will be zeroed below) before returning the
3016          * lock to a client.
3017          */
3018         if (new_lock->l_export == req->rq_export) {
3019                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3020         } else {
3021                 LASSERT(new_lock->l_export == NULL);
3022                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3023         }
3024
3025         *lockp = new_lock;
3026
3027         if (new_lock->l_export == req->rq_export) {
3028                 /*
3029                  * Already gave this to the client, which means that we
3030                  * reconstructed a reply.
3031                  */
3032                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3033                         MSG_RESENT);
3034                 lh->mlh_reg_lh.cookie = 0;
3035                 RETURN(ELDLM_LOCK_REPLACED);
3036         }
3037
3038         /*
3039          * Fixup the lock to be given to the client.
3040          */
3041         lock_res_and_lock(new_lock);
3042         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3043          * possible blocking AST. */
3044         while (new_lock->l_readers > 0) {
3045                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3046                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3047                 new_lock->l_readers--;
3048         }
3049         while (new_lock->l_writers > 0) {
3050                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3051                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3052                 new_lock->l_writers--;
3053         }
3054
3055         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3056         new_lock->l_blocking_ast = lock->l_blocking_ast;
3057         new_lock->l_completion_ast = lock->l_completion_ast;
3058         new_lock->l_remote_handle = lock->l_remote_handle;
3059         new_lock->l_flags &= ~LDLM_FL_LOCAL;
3060
3061         unlock_res_and_lock(new_lock);
3062
3063         cfs_hash_add(new_lock->l_export->exp_lock_hash,
3064                      &new_lock->l_remote_handle,
3065                      &new_lock->l_exp_hash);
3066
3067         LDLM_LOCK_RELEASE(new_lock);
3068         lh->mlh_reg_lh.cookie = 0;
3069
3070         RETURN(ELDLM_LOCK_REPLACED);
3071 }
3072
3073 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3074                                     struct ldlm_lock *new_lock,
3075                                     struct ldlm_lock **old_lock,
3076                                     struct mdt_lock_handle *lh,
3077                                     enum mdt_it_code opcode)
3078 {
3079         struct ptlrpc_request  *req = mdt_info_req(info);
3080         struct obd_export      *exp = req->rq_export;
3081         struct lustre_handle    remote_hdl;
3082         struct ldlm_request    *dlmreq;
3083         struct ldlm_lock       *lock;
3084
3085         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3086                 return;
3087
3088         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3089         remote_hdl = dlmreq->lock_handle[0];
3090         /* If the client does not require open lock, it does not need to
3091          * search lock in exp_lock_hash, since the server thread will
3092          * make sure the lock will be released, and the resend request
3093          * can always re-enqueue the lock */
3094         if ((opcode != MDT_IT_OPEN) || (opcode == MDT_IT_OPEN &&
3095             info->mti_spec.sp_cr_flags & MDS_OPEN_LOCK)) {
3096                 /* In the function below, .hs_keycmp resolves to
3097                  * ldlm_export_lock_keycmp() */
3098                 /* coverity[overrun-buffer-val] */
3099                 lock = cfs_hash_lookup(exp->exp_lock_hash, &remote_hdl);
3100                 if (lock) {
3101                         lock_res_and_lock(lock);
3102                         if (lock != new_lock) {
3103                                 lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie;
3104                                 lh->mlh_reg_mode = lock->l_granted_mode;
3105
3106                                 LDLM_DEBUG(lock, "Restoring lock cookie");
3107                                 DEBUG_REQ(D_DLMTRACE, req,
3108                                           "restoring lock cookie "LPX64,
3109                                           lh->mlh_reg_lh.cookie);
3110                                 if (old_lock)
3111                                         *old_lock = LDLM_LOCK_GET(lock);
3112                                 cfs_hash_put(exp->exp_lock_hash,
3113                                              &lock->l_exp_hash);
3114                                 unlock_res_and_lock(lock);
3115                                 return;
3116                         }
3117                         cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
3118                         unlock_res_and_lock(lock);
3119                 }
3120         }
3121         /*
3122          * If the xid matches, then we know this is a resent request, and allow
3123          * it. (It's probably an OPEN, for which we don't send a lock.
3124          */
3125         if (req_xid_is_last(req))
3126                 return;
3127
3128         /*
3129          * This remote handle isn't enqueued, so we never received or processed
3130          * this request.  Clear MSG_RESENT, because it can be handled like any
3131          * normal request now.
3132          */
3133         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3134
3135         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
3136                   remote_hdl.cookie);
3137 }
3138
3139 static int mdt_intent_getxattr(enum mdt_it_code opcode,
3140                                 struct mdt_thread_info *info,
3141                                 struct ldlm_lock **lockp,
3142                                 __u64 flags)
3143 {
3144         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3145         struct ldlm_reply      *ldlm_rep = NULL;
3146         int rc, grc;
3147
3148         /*
3149          * Initialize lhc->mlh_reg_lh either from a previously granted lock
3150          * (for the resend case) or a new lock. Below we will use it to
3151          * replace the original lock.
3152          */
3153         mdt_intent_fixup_resent(info, *lockp, NULL, lhc, opcode);
3154         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3155                 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3156                 rc = mdt_object_lock(info, info->mti_object, lhc,
3157                                         MDS_INODELOCK_XATTR,
3158                                         MDT_LOCAL_LOCK);
3159                 if (rc)
3160                         return rc;
3161         }
3162
3163         grc = mdt_getxattr(info);
3164
3165         rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3166
3167         if (mdt_info_req(info)->rq_repmsg != NULL)
3168                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3169         if (ldlm_rep == NULL)
3170                 RETURN(err_serious(-EFAULT));
3171
3172         ldlm_rep->lock_policy_res2 = grc;
3173
3174         return rc;
3175 }
3176
3177 static int mdt_intent_getattr(enum mdt_it_code opcode,
3178                               struct mdt_thread_info *info,
3179                               struct ldlm_lock **lockp,
3180                               __u64 flags)
3181 {
3182         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3183         struct ldlm_lock       *new_lock = NULL;
3184         __u64                   child_bits;
3185         struct ldlm_reply      *ldlm_rep;
3186         struct mdt_body        *reqbody;
3187         struct mdt_body        *repbody;
3188         int                     rc, rc2;
3189         ENTRY;
3190
3191         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3192         LASSERT(reqbody);
3193
3194         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3195         LASSERT(repbody);
3196
3197         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
3198         repbody->eadatasize = 0;
3199         repbody->aclsize = 0;
3200
3201         switch (opcode) {
3202         case MDT_IT_LOOKUP:
3203                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3204                 break;
3205         case MDT_IT_GETATTR:
3206                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3207                              MDS_INODELOCK_PERM;
3208                 break;
3209         default:
3210                 CERROR("Unsupported intent (%d)\n", opcode);
3211                 GOTO(out_shrink, rc = -EINVAL);
3212         }
3213
3214         rc = mdt_init_ucred(info, reqbody);
3215         if (rc)
3216                 GOTO(out_shrink, rc);
3217
3218         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3219         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3220
3221         /* Get lock from request for possible resent case. */
3222         mdt_intent_fixup_resent(info, *lockp, &new_lock, lhc, opcode);
3223
3224         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3225         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3226
3227         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3228                 ldlm_rep->lock_policy_res2 = 0;
3229         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3230             ldlm_rep->lock_policy_res2) {
3231                 lhc->mlh_reg_lh.cookie = 0ull;
3232                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3233         }
3234
3235         rc = mdt_intent_lock_replace(info, lockp, new_lock, lhc, flags);
3236         EXIT;
3237 out_ucred:
3238         mdt_exit_ucred(info);
3239 out_shrink:
3240         mdt_client_compatibility(info);
3241         rc2 = mdt_fix_reply(info);
3242         if (rc == 0)
3243                 rc = rc2;
3244         return rc;
3245 }
3246
3247 static int mdt_intent_layout(enum mdt_it_code opcode,
3248                              struct mdt_thread_info *info,
3249                              struct ldlm_lock **lockp,
3250                              __u64 flags)
3251 {
3252         struct layout_intent *layout;
3253         struct lu_fid *fid;
3254         struct mdt_object *obj = NULL;
3255         int rc = 0;
3256         ENTRY;
3257
3258         if (opcode != MDT_IT_LAYOUT) {
3259                 CERROR("%s: Unknown intent (%d)\n", mdt_obd_name(info->mti_mdt),
3260                         opcode);
3261                 RETURN(-EINVAL);
3262         }
3263
3264         fid = &info->mti_tmp_fid2;
3265         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3266
3267         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3268         if (IS_ERR(obj))
3269                 RETURN(PTR_ERR(obj));
3270
3271         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3272                 /* get the length of lsm */
3273                 rc = mdt_attr_get_eabuf_size(info, obj);
3274                 if (rc < 0)
3275                         RETURN(rc);
3276
3277                 if (rc > info->mti_mdt->mdt_max_mdsize)
3278                         info->mti_mdt->mdt_max_mdsize = rc;
3279         }
3280
3281         mdt_object_put(info->mti_env, obj);
3282
3283         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3284         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER, rc);
3285         rc = req_capsule_server_pack(info->mti_pill);
3286         if (rc != 0)
3287                 RETURN(-EINVAL);
3288
3289         layout = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3290         LASSERT(layout != NULL);
3291         if (layout->li_opc == LAYOUT_INTENT_ACCESS)
3292                 /* return to normal ldlm handling */
3293                 RETURN(0);
3294
3295         CERROR("%s: Unsupported layout intent (%d)\n",
3296                 mdt_obd_name(info->mti_mdt), layout->li_opc);
3297         RETURN(-EINVAL);
3298 }
3299
3300 static int mdt_intent_reint(enum mdt_it_code opcode,
3301                             struct mdt_thread_info *info,
3302                             struct ldlm_lock **lockp,
3303                             __u64 flags)
3304 {
3305         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3306         struct ldlm_reply      *rep = NULL;
3307         long                    opc;
3308         int                     rc;
3309
3310         static const struct req_format *intent_fmts[REINT_MAX] = {
3311                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3312                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3313         };
3314
3315         ENTRY;
3316
3317         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
3318         if (opc < 0)
3319                 RETURN(opc);
3320
3321         if (mdt_it_flavor[opcode].it_reint != opc) {
3322                 CERROR("Reint code %ld doesn't match intent: %d\n",
3323                        opc, opcode);
3324                 RETURN(err_serious(-EPROTO));
3325         }
3326
3327         /* Get lock from request for possible resent case. */
3328         mdt_intent_fixup_resent(info, *lockp, NULL, lhc, opcode);
3329
3330         rc = mdt_reint_internal(info, lhc, opc);
3331
3332         /* Check whether the reply has been packed successfully. */
3333         if (mdt_info_req(info)->rq_repmsg != NULL)
3334                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3335         if (rep == NULL)
3336                 RETURN(err_serious(-EFAULT));
3337
3338         /* MDC expects this in any case */
3339         if (rc != 0)
3340                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3341
3342         /* the open lock or the lock for cross-ref object should be
3343          * returned to the client */
3344         if (rc == -EREMOTE || mdt_get_disposition(rep, DISP_OPEN_LOCK)) {
3345                 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
3346                 rep->lock_policy_res2 = 0;
3347                 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3348                 RETURN(rc);
3349         }
3350
3351         rep->lock_policy_res2 = clear_serious(rc);
3352
3353         if (rep->lock_policy_res2 == -ENOENT &&
3354             mdt_get_disposition(rep, DISP_LOOKUP_NEG))
3355                 rep->lock_policy_res2 = 0;
3356
3357         if (rc == -ENOTCONN || rc == -ENODEV ||
3358             rc == -EOVERFLOW) { /**< if VBR failure then return error */
3359                 /*
3360                  * If it is the disconnect error (ENODEV & ENOCONN), the error
3361                  * will be returned by rq_status, and client at ptlrpc layer
3362                  * will detect this, then disconnect, reconnect the import
3363                  * immediately, instead of impacting the following the rpc.
3364                  */
3365                 lhc->mlh_reg_lh.cookie = 0ull;
3366                 RETURN(rc);
3367         } else {
3368                 /*
3369                  * For other cases, the error will be returned by intent.
3370                  * and client will retrieve the result from intent.
3371                  */
3372                  /*
3373                   * FIXME: when open lock is finished, that should be
3374                   * checked here.
3375                   */
3376                 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3377                         LASSERTF(rc == 0, "Error occurred but lock handle "
3378                                  "is still in use, rc = %d\n", rc);
3379                         rep->lock_policy_res2 = 0;
3380                         rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3381                         RETURN(rc);
3382                 } else {
3383                         lhc->mlh_reg_lh.cookie = 0ull;
3384                         RETURN(ELDLM_LOCK_ABORTED);
3385                 }
3386         }
3387 }
3388
3389 static int mdt_intent_code(long itcode)
3390 {
3391         int rc;
3392
3393         switch(itcode) {
3394         case IT_OPEN:
3395                 rc = MDT_IT_OPEN;
3396                 break;
3397         case IT_OPEN|IT_CREAT:
3398                 rc = MDT_IT_OCREAT;
3399                 break;
3400         case IT_CREAT:
3401                 rc = MDT_IT_CREATE;
3402                 break;
3403         case IT_READDIR:
3404                 rc = MDT_IT_READDIR;
3405                 break;
3406         case IT_GETATTR:
3407                 rc = MDT_IT_GETATTR;
3408                 break;
3409         case IT_LOOKUP:
3410                 rc = MDT_IT_LOOKUP;
3411                 break;
3412         case IT_UNLINK:
3413                 rc = MDT_IT_UNLINK;
3414                 break;
3415         case IT_TRUNC:
3416                 rc = MDT_IT_TRUNC;
3417                 break;
3418         case IT_GETXATTR:
3419                 rc = MDT_IT_GETXATTR;
3420                 break;
3421         case IT_LAYOUT:
3422                 rc = MDT_IT_LAYOUT;
3423                 break;
3424         case IT_QUOTA_DQACQ:
3425         case IT_QUOTA_CONN:
3426                 rc = MDT_IT_QUOTA;
3427                 break;
3428         default:
3429                 CERROR("Unknown intent opcode: %ld\n", itcode);
3430                 rc = -EINVAL;
3431                 break;
3432         }
3433         return rc;
3434 }
3435
3436 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
3437                           struct ldlm_lock **lockp, __u64 flags)
3438 {
3439         struct req_capsule   *pill;
3440         struct mdt_it_flavor *flv;
3441         int opc;
3442         int rc;
3443         ENTRY;
3444
3445         opc = mdt_intent_code(itopc);
3446         if (opc < 0)
3447                 RETURN(-EINVAL);
3448
3449         pill = info->mti_pill;
3450
3451         if (opc == MDT_IT_QUOTA) {
3452                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
3453
3454                 if (qmt == NULL)
3455                         RETURN(-EOPNOTSUPP);
3456
3457                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
3458                 /* pass the request to quota master */
3459                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
3460                                                  mdt_info_req(info), lockp,
3461                                                  flags);
3462                 RETURN(rc);
3463         }
3464
3465         flv  = &mdt_it_flavor[opc];
3466         if (flv->it_fmt != NULL)
3467                 req_capsule_extend(pill, flv->it_fmt);
3468
3469         rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3470         if (rc == 0) {
3471                 struct ptlrpc_request *req = mdt_info_req(info);
3472                 if (flv->it_flags & MUTABOR &&
3473                     exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
3474                         RETURN(-EROFS);
3475         }
3476         if (rc == 0 && flv->it_act != NULL) {
3477                 struct ldlm_reply *rep;
3478
3479                 /* execute policy */
3480                 rc = flv->it_act(opc, info, lockp, flags);
3481
3482                 /* Check whether the reply has been packed successfully. */
3483                 if (mdt_info_req(info)->rq_repmsg != NULL) {
3484                         rep = req_capsule_server_get(info->mti_pill,
3485                                                      &RMF_DLM_REP);
3486                         rep->lock_policy_res2 =
3487                                 ptlrpc_status_hton(rep->lock_policy_res2);
3488                 }
3489         } else {
3490                 rc = -EPROTO;
3491         }
3492         RETURN(rc);
3493 }
3494
3495 static int mdt_intent_policy(struct ldlm_namespace *ns,
3496                              struct ldlm_lock **lockp, void *req_cookie,
3497                              ldlm_mode_t mode, __u64 flags, void *data)
3498 {
3499         struct tgt_session_info *tsi;
3500         struct mdt_thread_info  *info;
3501         struct ptlrpc_request   *req  =  req_cookie;
3502         struct ldlm_intent      *it;
3503         struct req_capsule      *pill;
3504         int rc;
3505
3506         ENTRY;
3507
3508         LASSERT(req != NULL);
3509
3510         tsi = tgt_ses_info(req->rq_svc_thread->t_env);
3511
3512         info = tsi2mdt_info(tsi);
3513         LASSERT(info != NULL);
3514         pill = info->mti_pill;
3515         LASSERT(pill->rc_req == req);
3516
3517         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3518                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
3519                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3520                 if (it != NULL) {
3521                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
3522                         if (rc == 0)
3523                                 rc = ELDLM_OK;
3524
3525                         /* Lock without inodebits makes no sense and will oops
3526                          * later in ldlm. Let's check it now to see if we have
3527                          * ibits corrupted somewhere in mdt_intent_opc().
3528                          * The case for client miss to set ibits has been
3529                          * processed by others. */
3530                         LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
3531                                         lr_type == LDLM_IBITS,
3532                                      info->mti_dlm_req->lock_desc.\
3533                                         l_policy_data.l_inodebits.bits != 0));
3534                 } else
3535                         rc = err_serious(-EFAULT);
3536         } else {
3537                 /* No intent was provided */
3538                 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3539                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
3540                 rc = req_capsule_server_pack(pill);
3541                 if (rc)
3542                         rc = err_serious(rc);
3543         }
3544         mdt_thread_info_fini(info);
3545         RETURN(rc);
3546 }
3547
3548 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
3549 {
3550         struct seq_server_site  *ss = mdt_seq_site(mdt);
3551
3552         if (ss->ss_node_id == 0)
3553                 return;
3554
3555         if (ss->ss_client_seq != NULL) {
3556                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3557                 ss->ss_client_seq->lcs_exp = NULL;
3558         }
3559
3560         if (ss->ss_server_fld != NULL) {
3561                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
3562                 ss->ss_server_fld->lsf_control_exp = NULL;
3563         }
3564 }
3565
3566 static void mdt_seq_fini_cli(struct mdt_device *mdt)
3567 {
3568         struct seq_server_site *ss = mdt_seq_site(mdt);
3569
3570         if (ss == NULL)
3571                 return;
3572
3573         if (ss->ss_server_seq != NULL)
3574                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
3575 }
3576
3577 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
3578 {
3579         mdt_seq_fini_cli(mdt);
3580         mdt_deregister_seq_exp(mdt);
3581
3582         return seq_site_fini(env, mdt_seq_site(mdt));
3583 }
3584
3585 /**
3586  * It will retrieve its FLDB entries from MDT0, and it only happens
3587  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
3588  * and it needs to refresh FLDB from the MDT0.
3589  **/
3590 static int mdt_register_lwp_callback(void *data)
3591 {
3592         struct lu_env           env;
3593         struct mdt_device       *mdt = data;
3594         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
3595         int                     rc;
3596         ENTRY;
3597
3598         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
3599
3600         if (!likely(fld->lsf_new))
3601                 RETURN(0);
3602
3603         rc = lu_env_init(&env, LCT_MD_THREAD);
3604         if (rc) {
3605                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
3606                 RETURN(rc);
3607         }
3608
3609         rc = fld_update_from_controller(&env, fld);
3610         if (rc != 0) {
3611                 CERROR("%s: cannot update controller: rc = %d\n",
3612                        mdt_obd_name(mdt), rc);
3613                 GOTO(out, rc);
3614         }
3615 out:
3616         lu_env_fini(&env);
3617         RETURN(rc);
3618 }
3619
3620 static int mdt_register_seq_exp(struct mdt_device *mdt)
3621 {
3622         struct seq_server_site  *ss = mdt_seq_site(mdt);
3623         char                    *lwp_name = NULL;
3624         int                     rc;
3625
3626         if (ss->ss_node_id == 0)
3627                 return 0;
3628
3629         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
3630         if (lwp_name == NULL)
3631                 GOTO(out_free, rc = -ENOMEM);
3632
3633         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
3634         if (rc != 0)
3635                 GOTO(out_free, rc);
3636
3637         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
3638                                       NULL, NULL);
3639         if (rc != 0)
3640                 GOTO(out_free, rc);
3641
3642         rc = lustre_register_lwp_item(lwp_name,
3643                                       &ss->ss_server_fld->lsf_control_exp,
3644                                       mdt_register_lwp_callback, mdt);
3645         if (rc != 0) {
3646                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3647                 ss->ss_client_seq->lcs_exp = NULL;
3648                 GOTO(out_free, rc);
3649         }
3650 out_free:
3651         if (lwp_name != NULL)
3652                 OBD_FREE(lwp_name, MAX_OBD_NAME);
3653
3654         return rc;
3655 }
3656
3657 /*
3658  * Init client sequence manager which is used by local MDS to talk to sequence
3659  * controller on remote node.
3660  */
3661 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
3662 {
3663         struct seq_server_site  *ss = mdt_seq_site(mdt);
3664         int                     rc;
3665         char                    *prefix;
3666         ENTRY;
3667
3668         /* check if this is adding the first MDC and controller is not yet
3669          * initialized. */
3670         OBD_ALLOC_PTR(ss->ss_client_seq);
3671         if (ss->ss_client_seq == NULL)
3672                 RETURN(-ENOMEM);
3673
3674         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3675         if (prefix == NULL) {
3676                 OBD_FREE_PTR(ss->ss_client_seq);
3677                 ss->ss_client_seq = NULL;
3678                 RETURN(-ENOMEM);
3679         }
3680
3681         /* Note: seq_client_fini will be called in seq_site_fini */
3682         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
3683         rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
3684                              prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
3685                                                             NULL);
3686         OBD_FREE(prefix, MAX_OBD_NAME + 5);
3687         if (rc != 0) {
3688                 OBD_FREE_PTR(ss->ss_client_seq);
3689                 ss->ss_client_seq = NULL;
3690                 RETURN(rc);
3691         }
3692
3693         rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
3694
3695         RETURN(rc);
3696 }
3697
3698 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
3699 {
3700         struct seq_server_site  *ss;
3701         int                     rc;
3702         ENTRY;
3703
3704         ss = mdt_seq_site(mdt);
3705         /* init sequence controller server(MDT0) */
3706         if (ss->ss_node_id == 0) {
3707                 OBD_ALLOC_PTR(ss->ss_control_seq);
3708                 if (ss->ss_control_seq == NULL)
3709                         RETURN(-ENOMEM);
3710
3711                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
3712                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
3713                                      ss);
3714                 if (rc)
3715                         GOTO(out_seq_fini, rc);
3716         }
3717
3718         /* Init normal sequence server */
3719         OBD_ALLOC_PTR(ss->ss_server_seq);
3720         if (ss->ss_server_seq == NULL)
3721                 GOTO(out_seq_fini, rc = -ENOMEM);
3722
3723         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
3724                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
3725         if (rc)
3726                 GOTO(out_seq_fini, rc);
3727
3728         /* init seq client for seq server to talk to seq controller(MDT0) */
3729         rc = mdt_seq_init_cli(env, mdt);
3730         if (rc != 0)
3731                 GOTO(out_seq_fini, rc);
3732
3733         if (ss->ss_node_id != 0)
3734                 /* register controler export through lwp */
3735                 rc = mdt_register_seq_exp(mdt);
3736
3737         EXIT;
3738 out_seq_fini:
3739         if (rc)
3740                 mdt_seq_fini(env, mdt);
3741
3742         return rc;
3743 }
3744
3745 /*
3746  * FLD wrappers
3747  */
3748 static int mdt_fld_fini(const struct lu_env *env,
3749                         struct mdt_device *m)
3750 {
3751         struct seq_server_site *ss = mdt_seq_site(m);
3752         ENTRY;
3753
3754         if (ss && ss->ss_server_fld) {
3755                 fld_server_fini(env, ss->ss_server_fld);
3756                 OBD_FREE_PTR(ss->ss_server_fld);
3757                 ss->ss_server_fld = NULL;
3758         }
3759
3760         RETURN(0);
3761 }
3762
3763 static int mdt_fld_init(const struct lu_env *env,
3764                         const char *uuid,
3765                         struct mdt_device *m)
3766 {
3767         struct seq_server_site *ss;
3768         int rc;
3769         ENTRY;
3770
3771         ss = mdt_seq_site(m);
3772
3773         OBD_ALLOC_PTR(ss->ss_server_fld);
3774         if (ss->ss_server_fld == NULL)
3775                 RETURN(rc = -ENOMEM);
3776
3777         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
3778                              LU_SEQ_RANGE_MDT);
3779         if (rc) {
3780                 OBD_FREE_PTR(ss->ss_server_fld);
3781                 ss->ss_server_fld = NULL;
3782                 RETURN(rc);
3783         }
3784
3785         RETURN(0);
3786 }
3787
3788 static void mdt_stack_pre_fini(const struct lu_env *env,
3789                            struct mdt_device *m, struct lu_device *top)
3790 {
3791         struct lustre_cfg_bufs  *bufs;
3792         struct lustre_cfg       *lcfg;
3793         struct mdt_thread_info  *info;
3794         ENTRY;
3795
3796         LASSERT(top);
3797
3798         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3799         LASSERT(info != NULL);
3800
3801         bufs = &info->mti_u.bufs;
3802
3803         LASSERT(m->mdt_child_exp);
3804         LASSERT(m->mdt_child_exp->exp_obd);
3805
3806         /* process cleanup, pass mdt obd name to get obd umount flags */
3807         /* XXX: this is needed because all layers are referenced by
3808          * objects (some of them are pinned by osd, for example *
3809          * the proper solution should be a model where object used
3810          * by osd only doesn't have mdt/mdd slices -bzzz */
3811         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
3812         lustre_cfg_bufs_set_string(bufs, 1, NULL);
3813         lcfg = lustre_cfg_new(LCFG_PRE_CLEANUP, bufs);
3814         if (!lcfg) {
3815                 CERROR("%s:Cannot alloc lcfg!\n", mdt_obd_name(m));
3816                 return;
3817         }
3818         top->ld_ops->ldo_process_config(env, top, lcfg);
3819         lustre_cfg_free(lcfg);
3820         EXIT;
3821 }
3822
3823 static void mdt_stack_fini(const struct lu_env *env,
3824                            struct mdt_device *m, struct lu_device *top)
3825 {
3826         struct obd_device       *obd = mdt2obd_dev(m);
3827         struct lustre_cfg_bufs  *bufs;
3828         struct lustre_cfg       *lcfg;
3829         struct mdt_thread_info  *info;
3830         char                     flags[3] = "";
3831         ENTRY;
3832
3833         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3834         LASSERT(info != NULL);
3835
3836         lu_dev_del_linkage(top->ld_site, top);
3837
3838         lu_site_purge(env, top->ld_site, -1);
3839
3840         bufs = &info->mti_u.bufs;
3841         /* process cleanup, pass mdt obd name to get obd umount flags */
3842         /* another purpose is to let all layers to release their objects */
3843         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
3844         if (obd->obd_force)
3845                 strcat(flags, "F");
3846         if (obd->obd_fail)
3847                 strcat(flags, "A");
3848         lustre_cfg_bufs_set_string(bufs, 1, flags);
3849         lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
3850         if (!lcfg) {
3851                 CERROR("Cannot alloc lcfg!\n");
3852                 return;
3853         }
3854         LASSERT(top);
3855         top->ld_ops->ldo_process_config(env, top, lcfg);
3856         lustre_cfg_free(lcfg);
3857
3858         lu_site_purge(env, top->ld_site, -1);
3859
3860         m->mdt_child = NULL;
3861         m->mdt_bottom = NULL;
3862
3863         obd_disconnect(m->mdt_child_exp);
3864         m->mdt_child_exp = NULL;
3865
3866         obd_disconnect(m->mdt_bottom_exp);
3867         m->mdt_child_exp = NULL;
3868 }
3869
3870 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
3871                                const char *next, struct obd_export **exp)
3872 {
3873         struct obd_connect_data *data = NULL;
3874         struct obd_device       *obd;
3875         int                      rc;
3876         ENTRY;
3877
3878         OBD_ALLOC_PTR(data);
3879         if (data == NULL)
3880                 GOTO(out, rc = -ENOMEM);
3881
3882         obd = class_name2obd(next);
3883         if (obd == NULL) {
3884                 CERROR("%s: can't locate next device: %s\n",
3885                        mdt_obd_name(m), next);
3886                 GOTO(out, rc = -ENOTCONN);
3887         }
3888
3889         data->ocd_connect_flags = OBD_CONNECT_VERSION;
3890         data->ocd_version = LUSTRE_VERSION_CODE;
3891
3892         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
3893         if (rc) {
3894                 CERROR("%s: cannot connect to next dev %s (%d)\n",
3895                        mdt_obd_name(m), next, rc);
3896                 GOTO(out, rc);
3897         }
3898
3899 out:
3900         if (data)
3901                 OBD_FREE_PTR(data);
3902         RETURN(rc);
3903 }
3904
3905 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
3906                           struct lustre_cfg *cfg)
3907 {
3908         char                   *dev = lustre_cfg_string(cfg, 0);
3909         int                     rc, name_size, uuid_size;
3910         char                   *name, *uuid, *p;
3911         struct lustre_cfg_bufs *bufs;
3912         struct lustre_cfg      *lcfg;
3913         struct obd_device      *obd;
3914         struct lustre_profile  *lprof;
3915         struct lu_site         *site;
3916         ENTRY;
3917
3918         /* in 1.8 we had the only device in the stack - MDS.
3919          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
3920          * in 2.3 OSD is instantiated by obd_mount.c, so we need
3921          * to generate names and setup MDT, MDD. MDT will be using
3922          * generated name to connect to MDD. for MDD the next device
3923          * will be LOD with name taken from so called "profile" which
3924          * is generated by mount_option line
3925          *
3926          * 1.8 MGS generates config. commands like this:
3927          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
3928          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
3929          * 2.0 MGS generates config. commands like this:
3930          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
3931          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
3932          *                    3:lustre-MDT0000-mdtlov  4:f
3933          *
3934          * we generate MDD name from MDT one, just replacing T with D
3935          *
3936          * after all the preparations, the logical equivalent will be
3937          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
3938          *                    3:lustre-MDT0000-mdtlov  4:f
3939          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
3940          *                    3:lustre-MDD0000  4:f
3941          *
3942          *  notice we build the stack from down to top: MDD first, then MDT */
3943
3944         name_size = MAX_OBD_NAME;
3945         uuid_size = MAX_OBD_NAME;
3946
3947         OBD_ALLOC(name, name_size);
3948         OBD_ALLOC(uuid, uuid_size);
3949         if (name == NULL || uuid == NULL)
3950                 GOTO(cleanup_mem, rc = -ENOMEM);
3951
3952         OBD_ALLOC_PTR(bufs);
3953         if (!bufs)
3954                 GOTO(cleanup_mem, rc = -ENOMEM);
3955
3956         strcpy(name, dev);
3957         p = strstr(name, "-MDT");
3958         if (p == NULL)
3959                 GOTO(free_bufs, rc = -ENOMEM);
3960         p[3] = 'D';
3961
3962         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
3963
3964         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
3965         if (lprof == NULL || lprof->lp_dt == NULL) {
3966                 CERROR("can't find the profile: %s\n",
3967                        lustre_cfg_string(cfg, 0));
3968                 GOTO(free_bufs, rc = -EINVAL);
3969         }
3970
3971         lustre_cfg_bufs_reset(bufs, name);
3972         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
3973         lustre_cfg_bufs_set_string(bufs, 2, uuid);
3974         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
3975
3976         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
3977         if (!lcfg)
3978                 GOTO(free_bufs, rc = -ENOMEM);
3979
3980         rc = class_attach(lcfg);
3981         if (rc)
3982                 GOTO(lcfg_cleanup, rc);
3983
3984         obd = class_name2obd(name);
3985         if (!obd) {
3986                 CERROR("Can not find obd %s (%s in config)\n",
3987                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
3988                 GOTO(class_detach, rc = -EINVAL);
3989         }
3990
3991         lustre_cfg_free(lcfg);
3992
3993         lustre_cfg_bufs_reset(bufs, name);
3994         lustre_cfg_bufs_set_string(bufs, 1, uuid);
3995         lustre_cfg_bufs_set_string(bufs, 2, dev);
3996         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
3997
3998         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
3999
4000         rc = class_setup(obd, lcfg);
4001         if (rc)
4002                 GOTO(class_detach, rc);
4003
4004         /* connect to MDD we just setup */
4005         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
4006         if (rc)
4007                 GOTO(class_detach, rc);
4008
4009         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
4010         LASSERT(site);
4011         LASSERT(mdt_lu_site(mdt) == NULL);
4012         mdt->mdt_lu_dev.ld_site = site;
4013         site->ls_top_dev = &mdt->mdt_lu_dev;
4014         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
4015
4016         /* now connect to bottom OSD */
4017         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
4018         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
4019         if (rc)
4020                 GOTO(class_detach, rc);
4021         mdt->mdt_bottom =
4022                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
4023
4024         rc = lu_env_refill((struct lu_env *)env);
4025         if (rc != 0)
4026                 CERROR("Failure to refill session: '%d'\n", rc);
4027
4028         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
4029
4030         EXIT;
4031 class_detach:
4032         if (rc)
4033                 class_detach(obd, lcfg);
4034 lcfg_cleanup:
4035         lustre_cfg_free(lcfg);
4036 free_bufs:
4037         OBD_FREE_PTR(bufs);
4038 cleanup_mem:
4039         if (name)
4040                 OBD_FREE(name, name_size);
4041         if (uuid)
4042                 OBD_FREE(uuid, uuid_size);
4043         RETURN(rc);
4044 }
4045
4046 /* setup quota master target on MDT0 */
4047 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
4048                           struct lustre_cfg *cfg)
4049 {
4050         struct obd_device       *obd;
4051         char                    *dev = lustre_cfg_string(cfg, 0);
4052         char                    *qmtname, *uuid, *p;
4053         struct lustre_cfg_bufs  *bufs;
4054         struct lustre_cfg       *lcfg;
4055         struct lustre_profile   *lprof;
4056         struct obd_connect_data *data;
4057         int                      rc;
4058         ENTRY;
4059
4060         LASSERT(mdt->mdt_qmt_exp == NULL);
4061         LASSERT(mdt->mdt_qmt_dev == NULL);
4062
4063         /* quota master is on MDT0 only for now */
4064         if (mdt->mdt_seq_site.ss_node_id != 0)
4065                 RETURN(0);
4066
4067         /* MGS generates config commands which look as follows:
4068          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4069          *                    3:lustre-MDT0000-mdtlov  4:f
4070          *
4071          * We generate the QMT name from the MDT one, just replacing MD with QM
4072          * after all the preparations, the logical equivalent will be:
4073          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
4074          *                    3:lustre-MDT0000-osd  4:f */
4075         OBD_ALLOC(qmtname, MAX_OBD_NAME);
4076         OBD_ALLOC(uuid, UUID_MAX);
4077         OBD_ALLOC_PTR(bufs);
4078         OBD_ALLOC_PTR(data);
4079         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
4080                 GOTO(cleanup_mem, rc = -ENOMEM);
4081
4082         strcpy(qmtname, dev);
4083         p = strstr(qmtname, "-MDT");
4084         if (p == NULL)
4085                 GOTO(cleanup_mem, rc = -ENOMEM);
4086         /* replace MD with QM */
4087         p[1] = 'Q';
4088         p[2] = 'M';
4089
4090         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
4091
4092         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4093         if (lprof == NULL || lprof->lp_dt == NULL) {
4094                 CERROR("can't find profile for %s\n",
4095                        lustre_cfg_string(cfg, 0));
4096                 GOTO(cleanup_mem, rc = -EINVAL);
4097         }
4098
4099         lustre_cfg_bufs_reset(bufs, qmtname);
4100         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
4101         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4102         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4103
4104         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4105         if (!lcfg)
4106                 GOTO(cleanup_mem, rc = -ENOMEM);
4107
4108         rc = class_attach(lcfg);
4109         if (rc)
4110                 GOTO(lcfg_cleanup, rc);
4111
4112         obd = class_name2obd(qmtname);
4113         if (!obd) {
4114                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
4115                        lustre_cfg_string(cfg, 0));
4116                 GOTO(class_detach, rc = -EINVAL);
4117         }
4118
4119         lustre_cfg_free(lcfg);
4120
4121         lustre_cfg_bufs_reset(bufs, qmtname);
4122         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4123         lustre_cfg_bufs_set_string(bufs, 2, dev);
4124
4125         /* for quota, the next device should be the OSD device */
4126         lustre_cfg_bufs_set_string(bufs, 3,
4127                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
4128
4129         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4130
4131         rc = class_setup(obd, lcfg);
4132         if (rc)
4133                 GOTO(class_detach, rc);
4134
4135         mdt->mdt_qmt_dev = obd->obd_lu_dev;
4136
4137         /* configure local quota objects */
4138         rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
4139                                                    &mdt->mdt_lu_dev,
4140                                                    mdt->mdt_qmt_dev);
4141         if (rc)
4142                 GOTO(class_cleanup, rc);
4143
4144         /* connect to quota master target */
4145         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4146         data->ocd_version = LUSTRE_VERSION_CODE;
4147         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
4148                          data, NULL);
4149         if (rc) {
4150                 CERROR("cannot connect to quota master device %s (%d)\n",
4151                        qmtname, rc);
4152                 GOTO(class_cleanup, rc);
4153         }
4154
4155         EXIT;
4156 class_cleanup:
4157         if (rc) {
4158                 class_manual_cleanup(obd);
4159                 mdt->mdt_qmt_dev = NULL;
4160         }
4161 class_detach:
4162         if (rc)
4163                 class_detach(obd, lcfg);
4164 lcfg_cleanup:
4165         lustre_cfg_free(lcfg);
4166 cleanup_mem:
4167         if (bufs)
4168                 OBD_FREE_PTR(bufs);
4169         if (qmtname)
4170                 OBD_FREE(qmtname, MAX_OBD_NAME);
4171         if (uuid)
4172                 OBD_FREE(uuid, UUID_MAX);
4173         if (data)
4174                 OBD_FREE_PTR(data);
4175         return rc;
4176 }
4177
4178 /* Shutdown quota master target associated with mdt */
4179 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
4180 {
4181         ENTRY;
4182
4183         if (mdt->mdt_qmt_exp == NULL)
4184                 RETURN_EXIT;
4185         LASSERT(mdt->mdt_qmt_dev != NULL);
4186
4187         /* the qmt automatically shuts down when the mdt disconnects */
4188         obd_disconnect(mdt->mdt_qmt_exp);
4189         mdt->mdt_qmt_exp = NULL;
4190         mdt->mdt_qmt_dev = NULL;
4191         EXIT;
4192 }
4193
4194 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
4195  * for now. This will be removed along with converting rest of MDT code
4196  * to use tgt_session_info */
4197 int mdt_tgt_getxattr(struct tgt_session_info *tsi)
4198 {
4199         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
4200         int                      rc;
4201
4202         rc = mdt_getxattr(info);
4203
4204         mdt_thread_info_fini(info);
4205         return rc;
4206 }
4207
4208 static struct tgt_handler mdt_tgt_handlers[] = {
4209 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4210                 0,                      MDS_CONNECT,    mdt_tgt_connect,
4211                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
4212 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4213                 0,                      MDS_DISCONNECT, tgt_disconnect,
4214                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
4215 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4216                 HABEO_REFERO,           MDS_SET_INFO,   mdt_set_info,
4217                 &RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION),
4218 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
4219 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_GETSTATUS,  mdt_getstatus),
4220 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETATTR,    mdt_getattr),
4221 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_GETATTR_NAME,
4222                                                         mdt_getattr_name),
4223 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETXATTR,   mdt_tgt_getxattr),
4224 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_STATFS,     mdt_statfs),
4225 TGT_MDT_HDL(0           | MUTABOR,      MDS_REINT,      mdt_reint),
4226 TGT_MDT_HDL(HABEO_CORPUS,               MDS_CLOSE,      mdt_close),
4227 TGT_MDT_HDL(HABEO_CORPUS,               MDS_DONE_WRITING,
4228                                                         mdt_done_writing),
4229 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_READPAGE,   mdt_readpage),
4230 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_SYNC,       mdt_sync),
4231 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
4232 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_PROGRESS,
4233                                                         mdt_hsm_progress),
4234 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_REGISTER,
4235                                                         mdt_hsm_ct_register),
4236 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_UNREGISTER,
4237                                                         mdt_hsm_ct_unregister),
4238 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_STATE_GET,
4239                                                         mdt_hsm_state_get),
4240 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_STATE_SET,
4241                                                         mdt_hsm_state_set),
4242 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_ACTION, mdt_hsm_action),
4243 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_REQUEST,
4244                                                         mdt_hsm_request),
4245 TGT_MDT_HDL(HABEO_CLAVIS | HABEO_CORPUS | HABEO_REFERO | MUTABOR,
4246             MDS_SWAP_LAYOUTS,
4247             mdt_swap_layouts),
4248 };
4249
4250 static struct tgt_handler mdt_sec_ctx_ops[] = {
4251 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
4252 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
4253 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
4254 };
4255
4256 static struct tgt_handler mdt_quota_ops[] = {
4257 TGT_QUOTA_HDL(HABEO_REFERO,             QUOTA_DQACQ,      mdt_quota_dqacq),
4258 };
4259
4260 static struct tgt_opc_slice mdt_common_slice[] = {
4261         {
4262                 .tos_opc_start  = MDS_FIRST_OPC,
4263                 .tos_opc_end    = MDS_LAST_OPC,
4264                 .tos_hs         = mdt_tgt_handlers
4265         },
4266         {
4267                 .tos_opc_start  = OBD_FIRST_OPC,
4268                 .tos_opc_end    = OBD_LAST_OPC,
4269                 .tos_hs         = tgt_obd_handlers
4270         },
4271         {
4272                 .tos_opc_start  = LDLM_FIRST_OPC,
4273                 .tos_opc_end    = LDLM_LAST_OPC,
4274                 .tos_hs         = tgt_dlm_handlers
4275         },
4276         {
4277                 .tos_opc_start  = SEC_FIRST_OPC,
4278                 .tos_opc_end    = SEC_LAST_OPC,
4279                 .tos_hs         = mdt_sec_ctx_ops
4280         },
4281         {
4282                 .tos_opc_start  = OUT_UPDATE_FIRST_OPC,
4283                 .tos_opc_end    = OUT_UPDATE_LAST_OPC,
4284                 .tos_hs         = tgt_out_handlers
4285         },
4286         {
4287                 .tos_opc_start  = FLD_FIRST_OPC,
4288                 .tos_opc_end    = FLD_LAST_OPC,
4289                 .tos_hs         = fld_handlers
4290         },
4291         {
4292                 .tos_opc_start  = SEQ_FIRST_OPC,
4293                 .tos_opc_end    = SEQ_LAST_OPC,
4294                 .tos_hs         = seq_handlers
4295         },
4296         {
4297                 .tos_opc_start  = QUOTA_DQACQ,
4298                 .tos_opc_end    = QUOTA_LAST_OPC,
4299                 .tos_hs         = mdt_quota_ops
4300         },
4301         {
4302                 .tos_opc_start  = LLOG_FIRST_OPC,
4303                 .tos_opc_end    = LLOG_LAST_OPC,
4304                 .tos_hs         = tgt_llog_handlers
4305         },
4306         {
4307                 .tos_opc_start  = LFSCK_FIRST_OPC,
4308                 .tos_opc_end    = LFSCK_LAST_OPC,
4309                 .tos_hs         = tgt_lfsck_handlers
4310         },
4311
4312         {
4313                 .tos_hs         = NULL
4314         }
4315 };
4316
4317 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4318 {
4319         struct md_device        *next = m->mdt_child;
4320         struct lu_device        *d    = &m->mdt_lu_dev;
4321         struct obd_device       *obd  = mdt2obd_dev(m);
4322         struct lfsck_stop        stop;
4323         ENTRY;
4324
4325         stop.ls_status = LS_PAUSED;
4326         stop.ls_flags = 0;
4327         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
4328
4329         target_recovery_fini(obd);
4330         ping_evictor_stop();
4331         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
4332
4333         if (m->mdt_opts.mo_coordinator)
4334                 mdt_hsm_cdt_stop(m);
4335
4336         mdt_hsm_cdt_fini(m);
4337
4338         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
4339         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4340         obd_exports_barrier(obd);
4341         obd_zombie_barrier();
4342
4343         mdt_procfs_fini(m);
4344
4345         tgt_fini(env, &m->mdt_lut);
4346         mdt_fs_cleanup(env, m);
4347         upcall_cache_cleanup(m->mdt_identity_cache);
4348         m->mdt_identity_cache = NULL;
4349
4350         if (m->mdt_namespace != NULL) {
4351                 ldlm_namespace_free(m->mdt_namespace, NULL,
4352                                     d->ld_obd->obd_force);
4353                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4354         }
4355
4356         mdt_quota_fini(env, m);
4357
4358         cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
4359
4360         mdt_seq_fini(env, m);
4361         mdt_fld_fini(env, m);
4362
4363         next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL);
4364         cfs_timer_disarm(&m->mdt_ck_timer);
4365         mdt_ck_thread_stop(m);
4366
4367         /*
4368          * Finish the stack
4369          */
4370         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4371
4372         LASSERT(atomic_read(&d->ld_ref) == 0);
4373
4374         server_put_mount(mdt_obd_name(m), true);
4375
4376         EXIT;
4377 }
4378
4379 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4380
4381 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4382                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
4383 {
4384         struct mdt_thread_info    *info;
4385         struct obd_device         *obd;
4386         const char                *dev = lustre_cfg_string(cfg, 0);
4387         const char                *num = lustre_cfg_string(cfg, 2);
4388         struct lustre_mount_info  *lmi = NULL;
4389         struct lustre_sb_info     *lsi;
4390         struct lu_site            *s;
4391         struct seq_server_site    *ss_site;
4392         const char                *identity_upcall = "NONE";
4393         struct md_device          *next;
4394         int                        rc;
4395         long                       node_id;
4396         mntopt_t                   mntopts;
4397         ENTRY;
4398
4399         lu_device_init(&m->mdt_lu_dev, ldt);
4400         /*
4401          * Environment (env) might be missing mdt_thread_key values at that
4402          * point, if device is allocated when mdt_thread_key is in QUIESCENT
4403          * mode.
4404          *
4405          * Usually device allocation path doesn't use module key values, but
4406          * mdt has to do a lot of work here, so allocate key value.
4407          */
4408         rc = lu_env_refill((struct lu_env *)env);
4409         if (rc != 0)
4410                 RETURN(rc);
4411
4412         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4413         LASSERT(info != NULL);
4414
4415         obd = class_name2obd(dev);
4416         LASSERT(obd != NULL);
4417
4418         m->mdt_max_mdsize = MAX_MD_SIZE; /* 4 stripes */
4419
4420         m->mdt_som_conf = 0;
4421
4422         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4423
4424         /* default is coordinator off, it is started through conf_param
4425          * or /proc */
4426         m->mdt_opts.mo_coordinator = 0;
4427
4428         lmi = server_get_mount(dev);
4429         if (lmi == NULL) {
4430                 CERROR("Cannot get mount info for %s!\n", dev);
4431                 RETURN(-EFAULT);
4432         } else {
4433                 lsi = s2lsi(lmi->lmi_sb);
4434                 /* CMD is supported only in IAM mode */
4435                 LASSERT(num);
4436                 node_id = simple_strtol(num, NULL, 10);
4437                 obd->u.obt.obt_magic = OBT_MAGIC;
4438         }
4439
4440         spin_lock_init(&m->mdt_ioepoch_lock);
4441         m->mdt_capa_timeout = CAPA_TIMEOUT;
4442         m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1;
4443         m->mdt_ck_timeout = CAPA_KEY_TIMEOUT;
4444         m->mdt_squash.rsi_uid = 0;
4445         m->mdt_squash.rsi_gid = 0;
4446         INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
4447         init_rwsem(&m->mdt_squash.rsi_sem);
4448         spin_lock_init(&m->mdt_osfs_lock);
4449         m->mdt_osfs_age = cfs_time_shift_64(-1000);
4450         m->mdt_enable_remote_dir = 0;
4451         m->mdt_enable_remote_dir_gid = 0;
4452
4453         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
4454         m->mdt_lu_dev.ld_obd = obd;
4455         /* Set this lu_device to obd for error handling purposes. */
4456         obd->obd_lu_dev = &m->mdt_lu_dev;
4457
4458         /* init the stack */
4459         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
4460         if (rc) {
4461                 CERROR("%s: Can't init device stack, rc %d\n",
4462                        mdt_obd_name(m), rc);
4463                 GOTO(err_lmi, rc);
4464         }
4465
4466         s = mdt_lu_site(m);
4467         ss_site = mdt_seq_site(m);
4468         s->ld_seq_site = ss_site;
4469         ss_site->ss_lu = s;
4470
4471         /* set server index */
4472         ss_site->ss_node_id = node_id;
4473
4474         /* failover is the default
4475          * FIXME: we do not failout mds0/mgs, which may cause some problems.
4476          * assumed whose ss_node_id == 0 XXX
4477          * */
4478         obd->obd_replayable = 1;
4479         /* No connection accepted until configurations will finish */
4480         obd->obd_no_conn = 1;
4481
4482         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4483                 char *str = lustre_cfg_string(cfg, 4);
4484                 if (strchr(str, 'n')) {
4485                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
4486                         obd->obd_replayable = 0;
4487                 }
4488         }
4489
4490         rc = mdt_fld_init(env, mdt_obd_name(m), m);
4491         if (rc)
4492                 GOTO(err_fini_stack, rc);
4493
4494         rc = mdt_seq_init(env, m);
4495         if (rc)
4496                 GOTO(err_fini_fld, rc);
4497
4498         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
4499                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
4500         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4501                                               LDLM_NAMESPACE_SERVER,
4502                                               LDLM_NAMESPACE_GREEDY,
4503                                               LDLM_NS_TYPE_MDT);
4504         if (m->mdt_namespace == NULL)
4505                 GOTO(err_fini_seq, rc = -ENOMEM);
4506
4507         m->mdt_namespace->ns_lvbp = m;
4508         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
4509
4510         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4511         /* set obd_namespace for compatibility with old code */
4512         obd->obd_namespace = m->mdt_namespace;
4513
4514         cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m);
4515
4516         rc = mdt_hsm_cdt_init(m);
4517         if (rc != 0) {
4518                 CERROR("%s: error initializing coordinator, rc %d\n",
4519                        mdt_obd_name(m), rc);
4520                 GOTO(err_free_ns, rc);
4521         }
4522
4523         rc = mdt_ck_thread_start(m);
4524         if (rc)
4525                 GOTO(err_free_hsm, rc);
4526
4527         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
4528                       OBD_FAIL_MDS_ALL_REQUEST_NET,
4529                       OBD_FAIL_MDS_ALL_REPLY_NET);
4530         if (rc)
4531                 GOTO(err_capa, rc);
4532
4533         rc = mdt_fs_setup(env, m, obd, lsi);
4534         if (rc)
4535                 GOTO(err_tgt, rc);
4536
4537         tgt_adapt_sptlrpc_conf(&m->mdt_lut, 1);
4538
4539         next = m->mdt_child;
4540         rc = next->md_ops->mdo_iocontrol(env, next, OBD_IOC_GET_MNTOPT, 0,
4541                                          &mntopts);
4542         if (rc)
4543                 GOTO(err_fs_cleanup, rc);
4544
4545         if (mntopts & MNTOPT_USERXATTR)
4546                 m->mdt_opts.mo_user_xattr = 1;
4547         else
4548                 m->mdt_opts.mo_user_xattr = 0;
4549
4550         rc = next->md_ops->mdo_maxeasize_get(env, next, &m->mdt_max_ea_size);
4551         if (rc)
4552                 GOTO(err_fs_cleanup, rc);
4553
4554         if (mntopts & MNTOPT_ACL)
4555                 m->mdt_opts.mo_acl = 1;
4556         else
4557                 m->mdt_opts.mo_acl = 0;
4558
4559         /* XXX: to support suppgid for ACL, we enable identity_upcall
4560          * by default, otherwise, maybe got unexpected -EACCESS. */
4561         if (m->mdt_opts.mo_acl)
4562                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4563
4564         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
4565                                                 identity_upcall,
4566                                                 &mdt_identity_upcall_cache_ops);
4567         if (IS_ERR(m->mdt_identity_cache)) {
4568                 rc = PTR_ERR(m->mdt_identity_cache);
4569                 m->mdt_identity_cache = NULL;
4570                 GOTO(err_fs_cleanup, rc);
4571         }
4572
4573         rc = mdt_procfs_init(m, dev);
4574         if (rc) {
4575                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4576                 GOTO(err_recovery, rc);
4577         }
4578
4579         rc = mdt_quota_init(env, m, cfg);
4580         if (rc)
4581                 GOTO(err_procfs, rc);
4582
4583         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
4584         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
4585                            "mdt_ldlm_client", m->mdt_ldlm_client);
4586
4587         ping_evictor_start();
4588
4589         /* recovery will be started upon mdt_prepare()
4590          * when the whole stack is complete and ready
4591          * to serve the requests */
4592
4593         mdt_init_capa_ctxt(env, m);
4594
4595         /* Reduce the initial timeout on an MDS because it doesn't need such
4596          * a long timeout as an OST does. Adaptive timeouts will adjust this
4597          * value appropriately. */
4598         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4599                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4600
4601         RETURN(0);
4602 err_procfs:
4603         mdt_procfs_fini(m);
4604 err_recovery:
4605         target_recovery_fini(obd);
4606         upcall_cache_cleanup(m->mdt_identity_cache);
4607         m->mdt_identity_cache = NULL;
4608 err_fs_cleanup:
4609         mdt_fs_cleanup(env, m);
4610 err_tgt:
4611         tgt_fini(env, &m->mdt_lut);
4612 err_capa:
4613         cfs_timer_disarm(&m->mdt_ck_timer);
4614         mdt_ck_thread_stop(m);
4615 err_free_hsm:
4616         mdt_hsm_cdt_fini(m);
4617 err_free_ns:
4618         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4619         obd->obd_namespace = m->mdt_namespace = NULL;
4620 err_fini_seq:
4621         mdt_seq_fini(env, m);
4622 err_fini_fld:
4623         mdt_fld_fini(env, m);
4624 err_fini_stack:
4625         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4626 err_lmi:
4627         if (lmi)
4628                 server_put_mount(dev, true);
4629         return(rc);
4630 }
4631
4632 /* For interoperability, the left element is old parameter, the right one
4633  * is the new version of the parameter, if some parameter is deprecated,
4634  * the new version should be set as NULL. */
4635 static struct cfg_interop_param mdt_interop_param[] = {
4636         { "mdt.group_upcall",   NULL },
4637         { "mdt.quota_type",     NULL },
4638         { "mdd.quota_type",     NULL },
4639         { "mdt.rootsquash",     "mdt.root_squash" },
4640         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
4641         { NULL }
4642 };
4643
4644 /* used by MGS to process specific configurations */
4645 static int mdt_process_config(const struct lu_env *env,
4646                               struct lu_device *d, struct lustre_cfg *cfg)
4647 {
4648         struct mdt_device *m = mdt_dev(d);
4649         struct md_device *md_next = m->mdt_child;
4650         struct lu_device *next = md2lu_dev(md_next);
4651         int rc;
4652         ENTRY;
4653
4654         switch (cfg->lcfg_command) {
4655         case LCFG_PARAM: {
4656                 struct obd_device          *obd = d->ld_obd;
4657
4658                 /* For interoperability */
4659                 struct cfg_interop_param   *ptr = NULL;
4660                 struct lustre_cfg          *old_cfg = NULL;
4661                 char                       *param = NULL;
4662
4663                 param = lustre_cfg_string(cfg, 1);
4664                 if (param == NULL) {
4665                         CERROR("param is empty\n");
4666                         rc = -EINVAL;
4667                         break;
4668                 }
4669
4670                 ptr = class_find_old_param(param, mdt_interop_param);
4671                 if (ptr != NULL) {
4672                         if (ptr->new_param == NULL) {
4673                                 rc = 0;
4674                                 CWARN("For interoperability, skip this %s."
4675                                       " It is obsolete.\n", ptr->old_param);
4676                                 break;
4677                         }
4678
4679                         CWARN("Found old param %s, changed it to %s.\n",
4680                               ptr->old_param, ptr->new_param);
4681
4682                         old_cfg = cfg;
4683                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
4684                         if (IS_ERR(cfg)) {
4685                                 rc = PTR_ERR(cfg);
4686                                 break;
4687                         }
4688                 }
4689
4690                 rc = class_process_proc_seq_param(PARAM_MDT, obd->obd_vars,
4691                                                         cfg, obd);
4692                 if (rc > 0 || rc == -ENOSYS) {
4693                         /* is it an HSM var ? */
4694                         rc = class_process_proc_seq_param(PARAM_HSM,
4695                                                         hsm_cdt_get_proc_vars(),
4696                                                         cfg, obd);
4697                         if (rc > 0 || rc == -ENOSYS)
4698                                 /* we don't understand; pass it on */
4699                                 rc = next->ld_ops->ldo_process_config(env, next,
4700                                                                       cfg);
4701                 }
4702
4703                 if (old_cfg != NULL)
4704                         lustre_cfg_free(cfg);
4705
4706                 break;
4707         }
4708         default:
4709                 /* others are passed further */
4710                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4711                 break;
4712         }
4713         RETURN(rc);
4714 }
4715
4716 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4717                                           const struct lu_object_header *hdr,
4718                                           struct lu_device *d)
4719 {
4720         struct mdt_object *mo;
4721
4722         ENTRY;
4723
4724         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
4725         if (mo != NULL) {
4726                 struct lu_object *o;
4727                 struct lu_object_header *h;
4728
4729                 o = &mo->mot_obj;
4730                 h = &mo->mot_header;
4731                 lu_object_header_init(h);
4732                 lu_object_init(o, h, d);
4733                 lu_object_add_top(h, o);
4734                 o->lo_ops = &mdt_obj_ops;
4735                 mutex_init(&mo->mot_ioepoch_mutex);
4736                 mutex_init(&mo->mot_lov_mutex);
4737                 init_rwsem(&mo->mot_open_sem);
4738                 RETURN(o);
4739         }
4740         RETURN(NULL);
4741 }
4742
4743 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
4744                            const struct lu_object_conf *unused)
4745 {
4746         struct mdt_device *d = mdt_dev(o->lo_dev);
4747         struct lu_device  *under;
4748         struct lu_object  *below;
4749         int                rc = 0;
4750         ENTRY;
4751
4752         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4753                PFID(lu_object_fid(o)));
4754
4755         under = &d->mdt_child->md_lu_dev;
4756         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4757         if (below != NULL) {
4758                 lu_object_add(o, below);
4759         } else
4760                 rc = -ENOMEM;
4761
4762         RETURN(rc);
4763 }
4764
4765 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4766 {
4767         struct mdt_object *mo = mdt_obj(o);
4768         struct lu_object_header *h;
4769         ENTRY;
4770
4771         h = o->lo_header;
4772         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4773                PFID(lu_object_fid(o)));
4774
4775         LASSERT(atomic_read(&mo->mot_open_count) == 0);
4776         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
4777
4778         lu_object_fini(o);
4779         lu_object_header_fini(h);
4780         OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
4781
4782         EXIT;
4783 }
4784
4785 static int mdt_object_print(const struct lu_env *env, void *cookie,
4786                             lu_printer_t p, const struct lu_object *o)
4787 {
4788         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
4789         return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p(ioepoch="LPU64" "
4790                     "flags="LPX64", epochcount=%d, writecount=%d)",
4791                     mdto, mdto->mot_ioepoch, mdto->mot_flags,
4792                     mdto->mot_ioepoch_count, mdto->mot_writecount);
4793 }
4794
4795 static int mdt_prepare(const struct lu_env *env,
4796                 struct lu_device *pdev,
4797                 struct lu_device *cdev)
4798 {
4799         struct mdt_device *mdt = mdt_dev(cdev);
4800         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
4801         struct obd_device *obd = cdev->ld_obd;
4802         int rc;
4803
4804         ENTRY;
4805
4806         LASSERT(obd);
4807
4808         rc = next->ld_ops->ldo_prepare(env, cdev, next);
4809         if (rc)
4810                 RETURN(rc);
4811
4812         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
4813         if (rc)
4814                 RETURN(rc);
4815
4816         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
4817         if (rc)
4818                 RETURN(rc);
4819
4820         rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
4821         /* The LFSCK instance is registered just now, so it must be there when
4822          * register the namespace to such instance. */
4823         LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
4824
4825         if (mdt->mdt_seq_site.ss_node_id == 0) {
4826                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
4827                                                          &mdt->mdt_md_root_fid);
4828                 if (rc)
4829                         RETURN(rc);
4830         }
4831
4832         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
4833         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
4834         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
4835         LASSERT(obd->obd_no_conn);
4836         spin_lock(&obd->obd_dev_lock);
4837         obd->obd_no_conn = 0;
4838         spin_unlock(&obd->obd_dev_lock);
4839
4840         if (obd->obd_recovering == 0)
4841                 mdt_postrecov(env, mdt);
4842
4843         RETURN(rc);
4844 }
4845
4846 const struct lu_device_operations mdt_lu_ops = {
4847         .ldo_object_alloc   = mdt_object_alloc,
4848         .ldo_process_config = mdt_process_config,
4849         .ldo_prepare        = mdt_prepare,
4850 };
4851
4852 static const struct lu_object_operations mdt_obj_ops = {
4853         .loo_object_init    = mdt_object_init,
4854         .loo_object_free    = mdt_object_free,
4855         .loo_object_print   = mdt_object_print
4856 };
4857
4858 static int mdt_obd_set_info_async(const struct lu_env *env,
4859                                   struct obd_export *exp,
4860                                   __u32 keylen, void *key,
4861                                   __u32 vallen, void *val,
4862                                   struct ptlrpc_request_set *set)
4863 {
4864         int rc;
4865
4866         ENTRY;
4867
4868         if (KEY_IS(KEY_SPTLRPC_CONF)) {
4869                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp), 0);
4870                 RETURN(rc);
4871         }
4872
4873         RETURN(0);
4874 }
4875
4876 /**
4877  * Match client and server connection feature flags.
4878  *
4879  * Compute the compatibility flags for a connection request based on
4880  * features mutually supported by client and server.
4881  *
4882  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
4883  * must not be updated here, otherwise a partially initialized value may
4884  * be exposed. After the connection request is successfully processed,
4885  * the top-level MDT connect request handler atomically updates the export
4886  * connect flags from the obd_connect_data::ocd_connect_flags field of the
4887  * reply. \see mdt_connect().
4888  *
4889  * \param exp   the obd_export associated with this client/target pair
4890  * \param mdt   the target device for the connection
4891  * \param data  stores data for this connect request
4892  *
4893  * \retval 0       success
4894  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
4895  * \retval -EBADE  client and server feature requirements are incompatible
4896  */
4897 static int mdt_connect_internal(struct obd_export *exp,
4898                                 struct mdt_device *mdt,
4899                                 struct obd_connect_data *data)
4900 {
4901         LASSERT(data != NULL);
4902
4903         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
4904         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
4905
4906         /* If no known bits (which should not happen, probably,
4907            as everybody should support LOOKUP and UPDATE bits at least)
4908            revert to compat mode with plain locks. */
4909         if (!data->ocd_ibits_known &&
4910             data->ocd_connect_flags & OBD_CONNECT_IBITS)
4911                 data->ocd_connect_flags &= ~OBD_CONNECT_IBITS;
4912
4913         if (!mdt->mdt_opts.mo_acl)
4914                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
4915
4916         if (!mdt->mdt_opts.mo_user_xattr)
4917                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
4918
4919         if (!mdt->mdt_som_conf)
4920                 data->ocd_connect_flags &= ~OBD_CONNECT_SOM;
4921
4922         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
4923                 data->ocd_brw_size = min(data->ocd_brw_size,
4924                                          (__u32)MD_MAX_BRW_SIZE);
4925                 if (data->ocd_brw_size == 0) {
4926                         CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
4927                                " ocd_version: %x ocd_grant: %d "
4928                                "ocd_index: %u ocd_brw_size is "
4929                                "unexpectedly zero, network data "
4930                                "corruption? Refusing connection of this"
4931                                " client\n",
4932                                mdt_obd_name(mdt),
4933                                exp->exp_client_uuid.uuid,
4934                                exp, data->ocd_connect_flags, data->ocd_version,
4935                                data->ocd_grant, data->ocd_index);
4936                         return -EPROTO;
4937                 }
4938         }
4939
4940         /* NB: Disregard the rule against updating
4941          * exp_connect_data.ocd_connect_flags in this case, since
4942          * tgt_client_new() needs to know if this is a lightweight
4943          * connection, and it is safe to expose this flag before
4944          * connection processing completes. */
4945         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
4946                 spin_lock(&exp->exp_lock);
4947                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
4948                 spin_unlock(&exp->exp_lock);
4949         }
4950
4951         data->ocd_version = LUSTRE_VERSION_CODE;
4952
4953         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
4954                 CWARN("%s: MDS requires FID support, but client not\n",
4955                       mdt_obd_name(mdt));
4956                 return -EBADE;
4957         }
4958
4959         if (mdt->mdt_som_conf &&
4960             !(data->ocd_connect_flags & (OBD_CONNECT_LIGHTWEIGHT |
4961                                          OBD_CONNECT_MDS_MDS |
4962                                          OBD_CONNECT_SOM))) {
4963                 CWARN("%s: MDS has SOM enabled, but client does not support "
4964                       "it\n", mdt_obd_name(mdt));
4965                 return -EBADE;
4966         }
4967
4968         if (OCD_HAS_FLAG(data, PINGLESS)) {
4969                 if (ptlrpc_pinger_suppress_pings()) {
4970                         spin_lock(&exp->exp_obd->obd_dev_lock);
4971                         list_del_init(&exp->exp_obd_chain_timed);
4972                         spin_unlock(&exp->exp_obd->obd_dev_lock);
4973                 } else {
4974                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
4975                 }
4976         }
4977
4978         data->ocd_max_easize = mdt->mdt_max_ea_size;
4979
4980         return 0;
4981 }
4982
4983 /* mds_connect copy */
4984 static int mdt_obd_connect(const struct lu_env *env,
4985                            struct obd_export **exp, struct obd_device *obd,
4986                            struct obd_uuid *cluuid,
4987                            struct obd_connect_data *data,
4988                            void *localdata)
4989 {
4990         struct obd_export      *lexp;
4991         struct lustre_handle    conn = { 0 };
4992         struct mdt_device      *mdt;
4993         int                     rc;
4994         ENTRY;
4995
4996         LASSERT(env != NULL);
4997         if (!exp || !obd || !cluuid)
4998                 RETURN(-EINVAL);
4999
5000         mdt = mdt_dev(obd->obd_lu_dev);
5001
5002         /*
5003          * first, check whether the stack is ready to handle requests
5004          * XXX: probably not very appropriate method is used now
5005          *      at some point we should find a better one
5006          */
5007         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) && data != NULL &&
5008             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
5009                 rc = obd_get_info(env, mdt->mdt_child_exp,
5010                                   sizeof(KEY_OSP_CONNECTED),
5011                                   KEY_OSP_CONNECTED, NULL, NULL, NULL);
5012                 if (rc)
5013                         RETURN(-EAGAIN);
5014                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
5015         }
5016
5017         rc = class_connect(&conn, obd, cluuid);
5018         if (rc)
5019                 RETURN(rc);
5020
5021         lexp = class_conn2export(&conn);
5022         LASSERT(lexp != NULL);
5023
5024         rc = mdt_connect_internal(lexp, mdt, data);
5025         if (rc == 0) {
5026                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5027
5028                 LASSERT(lcd);
5029                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5030                 rc = tgt_client_new(env, lexp);
5031                 if (rc == 0)
5032                         mdt_export_stats_init(obd, lexp, localdata);
5033
5034                 /* For phase I, sync for cross-ref operation. */
5035                 lexp->exp_keep_sync = 1;
5036         }
5037
5038         if (rc != 0) {
5039                 class_disconnect(lexp);
5040                 *exp = NULL;
5041         } else {
5042                 *exp = lexp;
5043         }
5044
5045         RETURN(rc);
5046 }
5047
5048 static int mdt_obd_reconnect(const struct lu_env *env,
5049                              struct obd_export *exp, struct obd_device *obd,
5050                              struct obd_uuid *cluuid,
5051                              struct obd_connect_data *data,
5052                              void *localdata)
5053 {
5054         int                     rc;
5055         ENTRY;
5056
5057         if (exp == NULL || obd == NULL || cluuid == NULL)
5058                 RETURN(-EINVAL);
5059
5060         rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
5061         if (rc == 0)
5062                 mdt_export_stats_init(obd, exp, localdata);
5063
5064         RETURN(rc);
5065 }
5066
5067 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
5068                                    struct mdt_thread_info *info,
5069                                    struct mdt_file_data *mfd)
5070 {
5071         struct lu_context ses;
5072         int rc;
5073         ENTRY;
5074
5075         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
5076         if (rc)
5077                 RETURN(rc);
5078
5079         env->le_ses = &ses;
5080         lu_context_enter(&ses);
5081
5082         mdt_ucred(info)->uc_valid = UCRED_OLD;
5083         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
5084
5085         lu_context_exit(&ses);
5086         lu_context_fini(&ses);
5087         env->le_ses = NULL;
5088
5089         RETURN(rc);
5090 }
5091
5092 static int mdt_export_cleanup(struct obd_export *exp)
5093 {
5094         struct mdt_export_data *med = &exp->exp_mdt_data;
5095         struct obd_device      *obd = exp->exp_obd;
5096         struct mdt_device      *mdt;
5097         struct mdt_thread_info *info;
5098         struct lu_env           env;
5099         CFS_LIST_HEAD(closing_list);
5100         struct mdt_file_data *mfd, *n;
5101         int rc = 0;
5102         ENTRY;
5103
5104         spin_lock(&med->med_open_lock);
5105         while (!cfs_list_empty(&med->med_open_head)) {
5106                 cfs_list_t *tmp = med->med_open_head.next;
5107                 mfd = cfs_list_entry(tmp, struct mdt_file_data, mfd_list);
5108
5109                 /* Remove mfd handle so it can't be found again.
5110                  * We are consuming the mfd_list reference here. */
5111                 class_handle_unhash(&mfd->mfd_handle);
5112                 cfs_list_move_tail(&mfd->mfd_list, &closing_list);
5113         }
5114         spin_unlock(&med->med_open_lock);
5115         mdt = mdt_dev(obd->obd_lu_dev);
5116         LASSERT(mdt != NULL);
5117
5118         rc = lu_env_init(&env, LCT_MD_THREAD);
5119         if (rc)
5120                 RETURN(rc);
5121
5122         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5123         LASSERT(info != NULL);
5124         memset(info, 0, sizeof *info);
5125         info->mti_env = &env;
5126         info->mti_mdt = mdt;
5127         info->mti_exp = exp;
5128
5129         if (!cfs_list_empty(&closing_list)) {
5130                 struct md_attr *ma = &info->mti_attr;
5131
5132                 /* Close any open files (which may also cause orphan unlinking). */
5133                 cfs_list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5134                         cfs_list_del_init(&mfd->mfd_list);
5135                         ma->ma_need = ma->ma_valid = 0;
5136
5137                         /* This file is being closed due to an eviction, it
5138                          * could have been modified and now dirty regarding to
5139                          * HSM archive, check this!
5140                          * The logic here is to mark a file dirty if there's a
5141                          * chance it was dirtied before the client was evicted,
5142                          * so that we don't have to wait for a release attempt
5143                          * before finding out the file was actually dirty and
5144                          * fail the release. Aggressively marking it dirty here
5145                          * will cause the policy engine to attempt to
5146                          * re-archive it; when rearchiving, we can compare the
5147                          * current version to the HSM data_version and make the
5148                          * archive request into a noop if it's not actually
5149                          * dirty.
5150                          */
5151                         if (mfd->mfd_mode & (FMODE_WRITE|MDS_FMODE_TRUNC))
5152                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
5153
5154                         /* Don't unlink orphan on failover umount, LU-184 */
5155                         if (exp->exp_flags & OBD_OPT_FAILOVER) {
5156                                 ma->ma_valid = MA_FLAGS;
5157                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5158                         }
5159                         mdt_mfd_close(info, mfd);
5160                 }
5161         }
5162         info->mti_mdt = NULL;
5163         /* cleanup client slot early */
5164         /* Do not erase record for recoverable client. */
5165         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5166                 tgt_client_del(&env, exp);
5167         lu_env_fini(&env);
5168
5169         RETURN(rc);
5170 }
5171
5172 static int mdt_obd_disconnect(struct obd_export *exp)
5173 {
5174         int rc;
5175         ENTRY;
5176
5177         LASSERT(exp);
5178         class_export_get(exp);
5179
5180         rc = server_disconnect_export(exp);
5181         if (rc != 0)
5182                 CDEBUG(D_IOCTL, "server disconnect error: %d\n", rc);
5183
5184         rc = mdt_export_cleanup(exp);
5185         class_export_put(exp);
5186         RETURN(rc);
5187 }
5188
5189 /* FIXME: Can we avoid using these two interfaces? */
5190 static int mdt_init_export(struct obd_export *exp)
5191 {
5192         struct mdt_export_data *med = &exp->exp_mdt_data;
5193         int                     rc;
5194         ENTRY;
5195
5196         CFS_INIT_LIST_HEAD(&med->med_open_head);
5197         spin_lock_init(&med->med_open_lock);
5198         mutex_init(&med->med_idmap_mutex);
5199         med->med_idmap = NULL;
5200         spin_lock(&exp->exp_lock);
5201         exp->exp_connecting = 1;
5202         spin_unlock(&exp->exp_lock);
5203
5204         /* self-export doesn't need client data and ldlm initialization */
5205         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5206                                      &exp->exp_client_uuid)))
5207                 RETURN(0);
5208
5209         rc = tgt_client_alloc(exp);
5210         if (rc)
5211                 GOTO(err, rc);
5212
5213         rc = ldlm_init_export(exp);
5214         if (rc)
5215                 GOTO(err_free, rc);
5216
5217         RETURN(rc);
5218
5219 err_free:
5220         tgt_client_free(exp);
5221 err:
5222         CERROR("%s: Failed to initialize export: rc = %d\n",
5223                exp->exp_obd->obd_name, rc);
5224         return rc;
5225 }
5226
5227 static int mdt_destroy_export(struct obd_export *exp)
5228 {
5229         ENTRY;
5230
5231         if (exp_connect_rmtclient(exp))
5232                 mdt_cleanup_idmap(&exp->exp_mdt_data);
5233
5234         target_destroy_export(exp);
5235         /* destroy can be called from failed obd_setup, so
5236          * checking uuid is safer than obd_self_export */
5237         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5238                                      &exp->exp_client_uuid)))
5239                 RETURN(0);
5240
5241         ldlm_destroy_export(exp);
5242         tgt_client_free(exp);
5243
5244         LASSERT(cfs_list_empty(&exp->exp_outstanding_replies));
5245         LASSERT(cfs_list_empty(&exp->exp_mdt_data.med_open_head));
5246
5247         RETURN(0);
5248 }
5249
5250 /** The maximum depth that fid2path() will search.
5251  * This is limited only because we want to store the fids for
5252  * historical path lookup purposes.
5253  */
5254 #define MAX_PATH_DEPTH 100
5255
5256 /** mdt_path() lookup structure. */
5257 struct path_lookup_info {
5258         __u64                   pli_recno;      /**< history point */
5259         __u64                   pli_currec;     /**< current record */
5260         struct lu_fid           pli_fid;
5261         struct lu_fid           pli_fids[MAX_PATH_DEPTH]; /**< path, in fids */
5262         struct mdt_object       *pli_mdt_obj;
5263         char                    *pli_path;      /**< full path */
5264         int                     pli_pathlen;
5265         int                     pli_linkno;     /**< which hardlink to follow */
5266         int                     pli_fidcount;   /**< number of \a pli_fids */
5267 };
5268
5269 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
5270                    struct linkea_data *ldata)
5271 {
5272         int rc;
5273
5274         LASSERT(ldata->ld_buf->lb_buf != NULL);
5275
5276         if (!mdt_object_exists(mdt_obj))
5277                 return -ENODATA;
5278
5279         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5280                           ldata->ld_buf, XATTR_NAME_LINK);
5281         if (rc == -ERANGE) {
5282                 /* Buf was too small, figure out what we need. */
5283                 lu_buf_free(ldata->ld_buf);
5284                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5285                                   ldata->ld_buf, XATTR_NAME_LINK);
5286                 if (rc < 0)
5287                         return rc;
5288                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
5289                 if (ldata->ld_buf->lb_buf == NULL)
5290                         return -ENOMEM;
5291                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5292                                   ldata->ld_buf, XATTR_NAME_LINK);
5293         }
5294         if (rc < 0)
5295                 return rc;
5296
5297         return linkea_init(ldata);
5298 }
5299
5300 static int mdt_path_current(struct mdt_thread_info *info,
5301                             struct path_lookup_info *pli)
5302 {
5303         struct mdt_device       *mdt = info->mti_mdt;
5304         struct mdt_object       *mdt_obj;
5305         struct link_ea_header   *leh;
5306         struct link_ea_entry    *lee;
5307         struct lu_name          *tmpname = &info->mti_name;
5308         struct lu_fid           *tmpfid = &info->mti_tmp_fid1;
5309         struct lu_buf           *buf = &info->mti_big_buf;
5310         char                    *ptr;
5311         int                     reclen;
5312         struct linkea_data      ldata = { 0 };
5313         int                     rc = 0;
5314         ENTRY;
5315
5316         /* temp buffer for path element, the buffer will be finally freed
5317          * in mdt_thread_info_fini */
5318         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
5319         if (buf->lb_buf == NULL)
5320                 RETURN(-ENOMEM);
5321
5322         ldata.ld_buf = buf;
5323         ptr = pli->pli_path + pli->pli_pathlen - 1;
5324         *ptr = 0;
5325         --ptr;
5326         pli->pli_fidcount = 0;
5327         pli->pli_fids[0] = *(struct lu_fid *)mdt_object_fid(pli->pli_mdt_obj);
5328         *tmpfid = pli->pli_fids[0];
5329         /* root FID only exists on MDT0, and fid2path should also ends at MDT0,
5330          * so checking root_fid can only happen on MDT0. */
5331         while (!lu_fid_eq(&mdt->mdt_md_root_fid,
5332                           &pli->pli_fids[pli->pli_fidcount])) {
5333                 struct lu_buf           lmv_buf;
5334
5335                 mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
5336                 if (IS_ERR(mdt_obj))
5337                         GOTO(out, rc = PTR_ERR(mdt_obj));
5338
5339                 if (mdt_object_remote(mdt_obj)) {
5340                         mdt_object_put(info->mti_env, mdt_obj);
5341                         GOTO(remote_out, rc = -EREMOTE);
5342                 }
5343
5344                 lmv_buf.lb_buf = info->mti_xattr_buf;
5345                 lmv_buf.lb_len = sizeof(info->mti_xattr_buf);
5346
5347                 /* Check if it is slave stripes */
5348                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5349                                   &lmv_buf, XATTR_NAME_LMV);
5350                 if (rc > 0) {
5351                         union lmv_mds_md *lmm = lmv_buf.lb_buf;
5352
5353                         /* For slave stripes, get its master */
5354                         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE) {
5355                                 struct lmv_mds_md_v1 *lmm1 = &lmm->lmv_md_v1;
5356
5357                                 fid_le_to_cpu(tmpfid, &lmm1->lmv_master_fid);
5358                                 if (!fid_is_sane(tmpfid)) {
5359                                         mdt_object_put(info->mti_env, mdt_obj);
5360                                         GOTO(out, rc = -EINVAL);
5361                                 }
5362                                 mdt_object_put(info->mti_env, mdt_obj);
5363                                 pli->pli_fids[pli->pli_fidcount] = *tmpfid;
5364                                 continue;
5365                         }
5366                 }
5367
5368                 if (!mdt_object_exists(mdt_obj)) {
5369                         mdt_object_put(info->mti_env, mdt_obj);
5370                         GOTO(out, rc = -ENOENT);
5371                 }
5372
5373                 rc = mdt_links_read(info, mdt_obj, &ldata);
5374                 mdt_object_put(info->mti_env, mdt_obj);
5375                 if (rc != 0)
5376                         GOTO(out, rc);
5377
5378                 leh = buf->lb_buf;
5379                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
5380                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
5381                 /* If set, use link #linkno for path lookup, otherwise use
5382                    link #0.  Only do this for the final path element. */
5383                 if (pli->pli_fidcount == 0 &&
5384                     pli->pli_linkno < leh->leh_reccount) {
5385                         int count;
5386                         for (count = 0; count < pli->pli_linkno; count++) {
5387                                 lee = (struct link_ea_entry *)
5388                                      ((char *)lee + reclen);
5389                                 linkea_entry_unpack(lee, &reclen, tmpname,
5390                                                     tmpfid);
5391                         }
5392                         if (pli->pli_linkno < leh->leh_reccount - 1)
5393                                 /* indicate to user there are more links */
5394                                 pli->pli_linkno++;
5395                 }
5396
5397                 /* Pack the name in the end of the buffer */
5398                 ptr -= tmpname->ln_namelen;
5399                 if (ptr - 1 <= pli->pli_path)
5400                         GOTO(out, rc = -EOVERFLOW);
5401                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
5402                 *(--ptr) = '/';
5403
5404                 /* Store the parent fid for historic lookup */
5405                 if (++pli->pli_fidcount >= MAX_PATH_DEPTH)
5406                         GOTO(out, rc = -EOVERFLOW);
5407                 pli->pli_fids[pli->pli_fidcount] = *tmpfid;
5408         }
5409
5410 remote_out:
5411         ptr++; /* skip leading / */
5412         memmove(pli->pli_path, ptr, pli->pli_path + pli->pli_pathlen - ptr);
5413
5414         EXIT;
5415 out:
5416         return rc;
5417 }
5418
5419 /* Returns the full path to this fid, as of changelog record recno. */
5420 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
5421                     char *path, int pathlen, __u64 *recno, int *linkno,
5422                     struct lu_fid *fid)
5423 {
5424         struct mdt_device       *mdt = info->mti_mdt;
5425         struct path_lookup_info *pli;
5426         int                     tries = 3;
5427         int                     rc = -EAGAIN;
5428         ENTRY;
5429
5430         if (pathlen < 3)
5431                 RETURN(-EOVERFLOW);
5432
5433         if (lu_fid_eq(&mdt->mdt_md_root_fid, mdt_object_fid(obj))) {
5434                 path[0] = '\0';
5435                 RETURN(0);
5436         }
5437
5438         OBD_ALLOC_PTR(pli);
5439         if (pli == NULL)
5440                 RETURN(-ENOMEM);
5441
5442         pli->pli_mdt_obj = obj;
5443         pli->pli_recno = *recno;
5444         pli->pli_path = path;
5445         pli->pli_pathlen = pathlen;
5446         pli->pli_linkno = *linkno;
5447
5448         /* Retry multiple times in case file is being moved */
5449         while (tries-- && rc == -EAGAIN)
5450                 rc = mdt_path_current(info, pli);
5451
5452         /* return the last resolved fids to the client, so the client will
5453          * build the left path on another MDT for remote object */
5454         *fid = pli->pli_fids[pli->pli_fidcount];
5455
5456         *recno = pli->pli_currec;
5457         /* Return next link index to caller */
5458         *linkno = pli->pli_linkno;
5459
5460         OBD_FREE_PTR(pli);
5461
5462         RETURN(rc);
5463 }
5464
5465 static int mdt_fid2path(struct mdt_thread_info *info,
5466                         struct getinfo_fid2path *fp)
5467 {
5468         struct mdt_device *mdt = info->mti_mdt;
5469         struct mdt_object *obj;
5470         int    rc;
5471         ENTRY;
5472
5473         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
5474                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
5475
5476         if (!fid_is_sane(&fp->gf_fid))
5477                 RETURN(-EINVAL);
5478
5479         if (!fid_is_namespace_visible(&fp->gf_fid)) {
5480                 CWARN("%s: "DFID" is invalid, sequence should be "
5481                       ">= "LPX64"\n", mdt_obd_name(mdt),
5482                       PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
5483                 RETURN(-EINVAL);
5484         }
5485
5486         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
5487         if (obj == NULL || IS_ERR(obj)) {
5488                 CDEBUG(D_IOCTL, "no object "DFID": %ld\n", PFID(&fp->gf_fid),
5489                        PTR_ERR(obj));
5490                 RETURN(-EINVAL);
5491         }
5492
5493         if (mdt_object_remote(obj))
5494                 rc = -EREMOTE;
5495         else if (!mdt_object_exists(obj))
5496                 rc = -ENOENT;
5497         else
5498                 rc = 0;
5499
5500         if (rc < 0) {
5501                 mdt_object_put(info->mti_env, obj);
5502                 CDEBUG(D_IOCTL, "nonlocal object "DFID": %d\n",
5503                        PFID(&fp->gf_fid), rc);
5504                 RETURN(rc);
5505         }
5506
5507         rc = mdt_path(info, obj, fp->gf_path, fp->gf_pathlen, &fp->gf_recno,
5508                       &fp->gf_linkno, &fp->gf_fid);
5509
5510         CDEBUG(D_INFO, "fid "DFID", path %s recno "LPX64" linkno %u\n",
5511                PFID(&fp->gf_fid), fp->gf_path, fp->gf_recno, fp->gf_linkno);
5512
5513         mdt_object_put(info->mti_env, obj);
5514
5515         RETURN(rc);
5516 }
5517
5518 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key,
5519                             void *val, int vallen)
5520 {
5521         struct getinfo_fid2path *fpout, *fpin;
5522         int rc = 0;
5523
5524         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
5525         fpout = val;
5526
5527         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5528                 lustre_swab_fid2path(fpin);
5529
5530         memcpy(fpout, fpin, sizeof(*fpin));
5531         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
5532                 RETURN(-EINVAL);
5533
5534         rc = mdt_fid2path(info, fpout);
5535         RETURN(rc);
5536 }
5537
5538 int mdt_get_info(struct tgt_session_info *tsi)
5539 {
5540         char    *key;
5541         int      keylen;
5542         __u32   *vallen;
5543         void    *valout;
5544         int      rc;
5545
5546         ENTRY;
5547
5548         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
5549         if (key == NULL) {
5550                 CDEBUG(D_IOCTL, "No GETINFO key");
5551                 RETURN(err_serious(-EFAULT));
5552         }
5553         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
5554                                       RCL_CLIENT);
5555
5556         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
5557         if (vallen == NULL) {
5558                 CDEBUG(D_IOCTL, "Unable to get RMF_GETINFO_VALLEN buffer");
5559                 RETURN(err_serious(-EFAULT));
5560         }
5561
5562         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
5563                              *vallen);
5564         rc = req_capsule_server_pack(tsi->tsi_pill);
5565         if (rc)
5566                 RETURN(err_serious(rc));
5567
5568         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
5569         if (valout == NULL) {
5570                 CDEBUG(D_IOCTL, "Unable to get get-info RPC out buffer");
5571                 RETURN(err_serious(-EFAULT));
5572         }
5573
5574         if (KEY_IS(KEY_FID2PATH)) {
5575                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
5576
5577                 rc = mdt_rpc_fid2path(info, key, valout, *vallen);
5578                 mdt_thread_info_fini(info);
5579         } else {
5580                 rc = -EINVAL;
5581         }
5582         RETURN(rc);
5583 }
5584
5585 /* Pass the ioc down */
5586 static int mdt_ioc_child(struct lu_env *env, struct mdt_device *mdt,
5587                          unsigned int cmd, int len, void *data)
5588 {
5589         struct lu_context ioctl_session;
5590         struct md_device *next = mdt->mdt_child;
5591         int rc;
5592         ENTRY;
5593
5594         rc = lu_context_init(&ioctl_session, LCT_SERVER_SESSION);
5595         if (rc)
5596                 RETURN(rc);
5597         ioctl_session.lc_thread = (struct ptlrpc_thread *)current;
5598         lu_context_enter(&ioctl_session);
5599         env->le_ses = &ioctl_session;
5600
5601         LASSERT(next->md_ops->mdo_iocontrol);
5602         rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data);
5603
5604         lu_context_exit(&ioctl_session);
5605         lu_context_fini(&ioctl_session);
5606         RETURN(rc);
5607 }
5608
5609 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
5610 {
5611         struct obd_ioctl_data *data = karg;
5612         struct lu_fid *fid;
5613         __u64 version;
5614         struct mdt_object *obj;
5615         struct mdt_lock_handle  *lh;
5616         int rc;
5617         ENTRY;
5618
5619         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
5620             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
5621                 RETURN(-EINVAL);
5622
5623         fid = (struct lu_fid *)data->ioc_inlbuf1;
5624
5625         if (!fid_is_sane(fid))
5626                 RETURN(-EINVAL);
5627
5628         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
5629
5630         lh = &mti->mti_lh[MDT_LH_PARENT];
5631         mdt_lock_reg_init(lh, LCK_CR);
5632
5633         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
5634         if (IS_ERR(obj))
5635                 RETURN(PTR_ERR(obj));
5636
5637         if (mdt_object_remote(obj)) {
5638                 rc = -EREMOTE;
5639                 /**
5640                  * before calling version get the correct MDS should be
5641                  * fid, this is error to find remote object here
5642                  */
5643                 CERROR("nonlocal object "DFID"\n", PFID(fid));
5644         } else if (!mdt_object_exists(obj)) {
5645                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
5646                 rc = -ENOENT;
5647         } else {
5648                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
5649                *(__u64 *)data->ioc_inlbuf2 = version;
5650                 rc = 0;
5651         }
5652         mdt_object_unlock_put(mti, obj, lh, 1);
5653         RETURN(rc);
5654 }
5655
5656 /* ioctls on obd dev */
5657 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
5658                          void *karg, void *uarg)
5659 {
5660         struct lu_env      env;
5661         struct obd_device *obd = exp->exp_obd;
5662         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
5663         struct dt_device  *dt = mdt->mdt_bottom;
5664         int rc;
5665
5666         ENTRY;
5667         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
5668         rc = lu_env_init(&env, LCT_MD_THREAD);
5669         if (rc)
5670                 RETURN(rc);
5671
5672         switch (cmd) {
5673         case OBD_IOC_SYNC:
5674                 rc = mdt_device_sync(&env, mdt);
5675                 break;
5676         case OBD_IOC_SET_READONLY:
5677                 rc = dt->dd_ops->dt_ro(&env, dt);
5678                 break;
5679         case OBD_IOC_ABORT_RECOVERY:
5680                 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
5681                 target_stop_recovery_thread(obd);
5682                 rc = 0;
5683                 break;
5684         case OBD_IOC_CHANGELOG_REG:
5685         case OBD_IOC_CHANGELOG_DEREG:
5686         case OBD_IOC_CHANGELOG_CLEAR:
5687                 rc = mdt_ioc_child(&env, mdt, cmd, len, karg);
5688                 break;
5689         case OBD_IOC_START_LFSCK: {
5690                 struct md_device *next = mdt->mdt_child;
5691                 struct obd_ioctl_data *data = karg;
5692                 struct lfsck_start_param lsp;
5693
5694                 if (unlikely(data == NULL)) {
5695                         rc = -EINVAL;
5696                         break;
5697                 }
5698
5699                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
5700                 lsp.lsp_index_valid = 0;
5701                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
5702                 break;
5703         }
5704         case OBD_IOC_STOP_LFSCK: {
5705                 struct md_device        *next = mdt->mdt_child;
5706                 struct obd_ioctl_data   *data = karg;
5707                 struct lfsck_stop        stop;
5708
5709                 stop.ls_status = LS_STOPPED;
5710                 /* Old lfsck utils may pass NULL @stop. */
5711                 if (data->ioc_inlbuf1 == NULL)
5712                         stop.ls_flags = 0;
5713                 else
5714                         stop.ls_flags =
5715                         ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
5716
5717                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
5718                 break;
5719         }
5720         case OBD_IOC_GET_OBJ_VERSION: {
5721                 struct mdt_thread_info *mti;
5722                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5723                 memset(mti, 0, sizeof *mti);
5724                 mti->mti_env = &env;
5725                 mti->mti_mdt = mdt;
5726                 mti->mti_exp = exp;
5727
5728                 rc = mdt_ioc_version_get(mti, karg);
5729                 break;
5730         }
5731         case OBD_IOC_CATLOGLIST: {
5732                 struct mdt_thread_info *mti;
5733
5734                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5735                 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
5736                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
5737                                        &mti->mti_tmp_fid1);
5738                 break;
5739          }
5740         default:
5741                 rc = -EOPNOTSUPP;
5742                 CERROR("%s: Not supported cmd = %d, rc = %d\n",
5743                         mdt_obd_name(mdt), cmd, rc);
5744         }
5745
5746         lu_env_fini(&env);
5747         RETURN(rc);
5748 }
5749
5750 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
5751 {
5752         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
5753         struct lfsck_start_param lsp;
5754         int rc;
5755         ENTRY;
5756
5757         lsp.lsp_start = NULL;
5758         lsp.lsp_index_valid = 0;
5759         rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
5760                                                    OBD_IOC_START_LFSCK,
5761                                                    0, &lsp);
5762         if (rc != 0 && rc != -EALREADY)
5763                 CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
5764                       mdt_obd_name(mdt), rc);
5765
5766         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
5767         RETURN(rc);
5768 }
5769
5770 static int mdt_obd_postrecov(struct obd_device *obd)
5771 {
5772         struct lu_env env;
5773         int rc;
5774
5775         rc = lu_env_init(&env, LCT_MD_THREAD);
5776         if (rc)
5777                 RETURN(rc);
5778         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
5779         lu_env_fini(&env);
5780         return rc;
5781 }
5782
5783 static struct obd_ops mdt_obd_device_ops = {
5784         .o_owner          = THIS_MODULE,
5785         .o_set_info_async = mdt_obd_set_info_async,
5786         .o_connect        = mdt_obd_connect,
5787         .o_reconnect      = mdt_obd_reconnect,
5788         .o_disconnect     = mdt_obd_disconnect,
5789         .o_init_export    = mdt_init_export,
5790         .o_destroy_export = mdt_destroy_export,
5791         .o_iocontrol      = mdt_iocontrol,
5792         .o_postrecov      = mdt_obd_postrecov,
5793 };
5794
5795 static struct lu_device* mdt_device_fini(const struct lu_env *env,
5796                                          struct lu_device *d)
5797 {
5798         struct mdt_device *m = mdt_dev(d);
5799         ENTRY;
5800
5801         mdt_fini(env, m);
5802         RETURN(NULL);
5803 }
5804
5805 static struct lu_device *mdt_device_free(const struct lu_env *env,
5806                                          struct lu_device *d)
5807 {
5808         struct mdt_device *m = mdt_dev(d);
5809         ENTRY;
5810
5811         lu_device_fini(&m->mdt_lu_dev);
5812         OBD_FREE_PTR(m);
5813
5814         RETURN(NULL);
5815 }
5816
5817 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
5818                                           struct lu_device_type *t,
5819                                           struct lustre_cfg *cfg)
5820 {
5821         struct lu_device  *l;
5822         struct mdt_device *m;
5823
5824         OBD_ALLOC_PTR(m);
5825         if (m != NULL) {
5826                 int rc;
5827
5828                 l = &m->mdt_lu_dev;
5829                 rc = mdt_init0(env, m, t, cfg);
5830                 if (rc != 0) {
5831                         mdt_device_free(env, l);
5832                         l = ERR_PTR(rc);
5833                         return l;
5834                 }
5835         } else
5836                 l = ERR_PTR(-ENOMEM);
5837         return l;
5838 }
5839
5840 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
5841 LU_KEY_INIT(mdt, struct mdt_thread_info);
5842
5843 static void mdt_key_fini(const struct lu_context *ctx,
5844                          struct lu_context_key *key, void* data)
5845 {
5846         struct mdt_thread_info *info = data;
5847
5848         if (info->mti_big_lmm) {
5849                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
5850                 info->mti_big_lmm = NULL;
5851                 info->mti_big_lmmsize = 0;
5852         }
5853         OBD_FREE_PTR(info);
5854 }
5855
5856 /* context key: mdt_thread_key */
5857 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
5858
5859 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
5860 {
5861         return lu_ucred(info->mti_env);
5862 }
5863
5864 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
5865 {
5866         return lu_ucred_check(info->mti_env);
5867 }
5868
5869 /**
5870  * Enable/disable COS (Commit On Sharing).
5871  *
5872  * Set/Clear the COS flag in mdt options.
5873  *
5874  * \param mdt mdt device
5875  * \param val 0 disables COS, other values enable COS
5876  */
5877 void mdt_enable_cos(struct mdt_device *mdt, int val)
5878 {
5879         struct lu_env env;
5880         int rc;
5881
5882         mdt->mdt_opts.mo_cos = !!val;
5883         rc = lu_env_init(&env, LCT_LOCAL);
5884         if (unlikely(rc != 0)) {
5885                 CWARN("lu_env initialization failed with rc = %d,"
5886                       "cannot sync\n", rc);
5887                 return;
5888         }
5889         mdt_device_sync(&env, mdt);
5890         lu_env_fini(&env);
5891 }
5892
5893 /**
5894  * Check COS (Commit On Sharing) status.
5895  *
5896  * Return COS flag status.
5897  *
5898  * \param mdt mdt device
5899  */
5900 int mdt_cos_is_enabled(struct mdt_device *mdt)
5901 {
5902         return mdt->mdt_opts.mo_cos != 0;
5903 }
5904
5905 static struct lu_device_type_operations mdt_device_type_ops = {
5906         .ldto_device_alloc = mdt_device_alloc,
5907         .ldto_device_free  = mdt_device_free,
5908         .ldto_device_fini  = mdt_device_fini
5909 };
5910
5911 static struct lu_device_type mdt_device_type = {
5912         .ldt_tags     = LU_DEVICE_MD,
5913         .ldt_name     = LUSTRE_MDT_NAME,
5914         .ldt_ops      = &mdt_device_type_ops,
5915         .ldt_ctx_tags = LCT_MD_THREAD
5916 };
5917
5918 static int __init mdt_mod_init(void)
5919 {
5920         int rc;
5921
5922         CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") ==
5923                  FID_NOBRACE_LEN + 1);
5924         CLASSERT(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") ==
5925                  FID_LEN + 1);
5926         rc = lu_kmem_init(mdt_caches);
5927         if (rc)
5928                 return rc;
5929
5930         rc = mds_mod_init();
5931         if (rc)
5932                 GOTO(lu_fini, rc);
5933
5934         rc = class_register_type(&mdt_obd_device_ops, NULL, true, NULL,
5935 #ifndef HAVE_ONLY_PROCFS_SEQ
5936                                  NULL,
5937 #endif
5938                                  LUSTRE_MDT_NAME, &mdt_device_type);
5939         if (rc)
5940                 GOTO(mds_fini, rc);
5941 lu_fini:
5942         if (rc)
5943                 lu_kmem_fini(mdt_caches);
5944 mds_fini:
5945         if (rc)
5946                 mds_mod_exit();
5947         return rc;
5948 }
5949
5950 static void __exit mdt_mod_exit(void)
5951 {
5952         class_unregister_type(LUSTRE_MDT_NAME);
5953         mds_mod_exit();
5954         lu_kmem_fini(mdt_caches);
5955 }
5956
5957 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
5958 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
5959 MODULE_LICENSE("GPL");
5960
5961 cfs_module(mdt, LUSTRE_VERSION_STRING, mdt_mod_init, mdt_mod_exit);