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_is_subdir(struct tgt_session_info *tsi)
1168 {
1169         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1170         struct mdt_object     *o = info->mti_object;
1171         struct req_capsule    *pill = info->mti_pill;
1172         const struct mdt_body *body = info->mti_body;
1173         struct mdt_body       *repbody;
1174         int                    rc;
1175         ENTRY;
1176
1177         LASSERT(o != NULL);
1178
1179         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1180
1181         /*
1182          * We save last checked parent fid to @repbody->fid1 for remote
1183          * directory case.
1184          */
1185         LASSERT(fid_is_sane(&body->fid2));
1186         LASSERT(mdt_object_exists(o) && !mdt_object_remote(o));
1187         rc = mdo_is_subdir(info->mti_env, mdt_object_child(o),
1188                            &body->fid2, &repbody->fid1);
1189         if (rc == 0 || rc == -EREMOTE)
1190                 repbody->valid |= OBD_MD_FLID;
1191
1192         mdt_thread_info_fini(info);
1193         RETURN(rc);
1194 }
1195
1196 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1197 {
1198         struct mdt_thread_info  *info;
1199         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1200         struct obd_export       *exp = req->rq_export;
1201         struct mdt_object       *o1, *o2, *o;
1202         struct mdt_lock_handle  *lh1, *lh2;
1203         struct mdc_swap_layouts *msl;
1204         int                      rc;
1205         ENTRY;
1206
1207         /* client does not support layout lock, so layout swaping
1208          * is disabled.
1209          * FIXME: there is a problem for old clients which don't support
1210          * layout lock yet. If those clients have already opened the file
1211          * they won't be notified at all so that old layout may still be
1212          * used to do IO. This can be fixed after file release is landed by
1213          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1214         if (!exp_connect_layout(exp))
1215                 RETURN(-EOPNOTSUPP);
1216
1217         info = tsi2mdt_info(tsi);
1218
1219         if (info->mti_dlm_req != NULL)
1220                 ldlm_request_cancel(req, info->mti_dlm_req, 0);
1221
1222         if (req_capsule_get_size(info->mti_pill, &RMF_CAPA1, RCL_CLIENT))
1223                 mdt_set_capainfo(info, 0, &info->mti_body->fid1,
1224                                  req_capsule_client_get(info->mti_pill,
1225                                                         &RMF_CAPA1));
1226
1227         if (req_capsule_get_size(info->mti_pill, &RMF_CAPA2, RCL_CLIENT))
1228                 mdt_set_capainfo(info, 1, &info->mti_body->fid2,
1229                                  req_capsule_client_get(info->mti_pill,
1230                                                         &RMF_CAPA2));
1231
1232         o1 = info->mti_object;
1233         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1234                                 &info->mti_body->fid2);
1235         if (IS_ERR(o))
1236                 GOTO(out, rc = PTR_ERR(o));
1237
1238         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1239                 GOTO(put, rc = -ENOENT);
1240
1241         rc = lu_fid_cmp(&info->mti_body->fid1, &info->mti_body->fid2);
1242         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1243                 GOTO(put, rc);
1244
1245         if (rc < 0)
1246                 swap(o1, o2);
1247
1248         /* permission check. Make sure the calling process having permission
1249          * to write both files. */
1250         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1251                                 MAY_WRITE);
1252         if (rc < 0)
1253                 GOTO(put, rc);
1254
1255         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1256                                 MAY_WRITE);
1257         if (rc < 0)
1258                 GOTO(put, rc);
1259
1260         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1261         if (msl == NULL)
1262                 GOTO(put, rc = -EPROTO);
1263
1264         lh1 = &info->mti_lh[MDT_LH_NEW];
1265         mdt_lock_reg_init(lh1, LCK_EX);
1266         lh2 = &info->mti_lh[MDT_LH_OLD];
1267         mdt_lock_reg_init(lh2, LCK_EX);
1268
1269         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1270                              MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK);
1271         if (rc < 0)
1272                 GOTO(put, rc);
1273
1274         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1275                              MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK);
1276         if (rc < 0)
1277                 GOTO(unlock1, rc);
1278
1279         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1280                              mdt_object_child(o2), msl->msl_flags);
1281         GOTO(unlock2, rc);
1282 unlock2:
1283         mdt_object_unlock(info, o2, lh2, rc);
1284 unlock1:
1285         mdt_object_unlock(info, o1, lh1, rc);
1286 put:
1287         mdt_object_put(info->mti_env, o);
1288 out:
1289         mdt_thread_info_fini(info);
1290         RETURN(rc);
1291 }
1292
1293 static int mdt_raw_lookup(struct mdt_thread_info *info,
1294                           struct mdt_object *parent,
1295                           const struct lu_name *lname,
1296                           struct ldlm_reply *ldlm_rep)
1297 {
1298         struct md_object *next = mdt_object_child(info->mti_object);
1299         const struct mdt_body *reqbody = info->mti_body;
1300         struct lu_fid *child_fid = &info->mti_tmp_fid1;
1301         struct mdt_body *repbody;
1302         int rc;
1303         ENTRY;
1304
1305         if (reqbody->valid != OBD_MD_FLID)
1306                 RETURN(0);
1307
1308         LASSERT(!info->mti_cross_ref);
1309
1310         /* Only got the fid of this obj by name */
1311         fid_zero(child_fid);
1312         rc = mdo_lookup(info->mti_env, next, lname, child_fid,
1313                         &info->mti_spec);
1314 #if 0
1315         /* XXX is raw_lookup possible as intent operation? */
1316         if (rc != 0) {
1317                 if (rc == -ENOENT)
1318                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1319                 RETURN(rc);
1320         } else
1321                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1322
1323         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1324 #endif
1325         if (rc == 0) {
1326                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1327                 repbody->fid1 = *child_fid;
1328                 repbody->valid = OBD_MD_FLID;
1329         }
1330         RETURN(1);
1331 }
1332
1333 /*
1334  * UPDATE lock should be taken against parent, and be release before exit;
1335  * child_bits lock should be taken against child, and be returned back:
1336  *            (1)normal request should release the child lock;
1337  *            (2)intent request will grant the lock to client.
1338  */
1339 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1340                                  struct mdt_lock_handle *lhc,
1341                                  __u64 child_bits,
1342                                  struct ldlm_reply *ldlm_rep)
1343 {
1344         struct ptlrpc_request  *req       = mdt_info_req(info);
1345         struct mdt_body        *reqbody   = NULL;
1346         struct mdt_object      *parent    = info->mti_object;
1347         struct mdt_object      *child;
1348         struct md_object       *next      = mdt_object_child(parent);
1349         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1350         struct lu_name         *lname     = NULL;
1351         struct mdt_lock_handle *lhp       = NULL;
1352         struct ldlm_lock       *lock;
1353         bool                    is_resent;
1354         bool                    try_layout;
1355         int                     ma_need = 0;
1356         int                     rc;
1357         ENTRY;
1358
1359         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1360         LASSERT(ergo(is_resent,
1361                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1362
1363         LASSERT(parent != NULL);
1364
1365         if (info->mti_cross_ref) {
1366                 /* Only getattr on the child. Parent is on another node. */
1367                 mdt_set_disposition(info, ldlm_rep,
1368                                     DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1369                 child = parent;
1370                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1371                        "ldlm_rep = %p\n",
1372                        PFID(mdt_object_fid(child)), ldlm_rep);
1373
1374                 if (is_resent) {
1375                         /* Do not take lock for resent case. */
1376                         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1377                         LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
1378                                  lhc->mlh_reg_lh.cookie);
1379                         LASSERT(fid_res_name_eq(mdt_object_fid(child),
1380                                                 &lock->l_resource->lr_name));
1381                         LDLM_LOCK_PUT(lock);
1382                 } else {
1383                         mdt_lock_handle_init(lhc);
1384                         mdt_lock_reg_init(lhc, LCK_PR);
1385
1386                         /*
1387                          * Object's name is on another MDS, no lookup or layout
1388                          * lock is needed here but update lock is.
1389                          */
1390                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
1391                                         MDS_INODELOCK_LAYOUT);
1392                         child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1393
1394                         rc = mdt_object_lock(info, child, lhc, child_bits,
1395                                              MDT_LOCAL_LOCK);
1396                         if (rc < 0)
1397                                 RETURN(rc);
1398                 }
1399
1400                 /* Finally, we can get attr for child. */
1401                 if (!mdt_object_exists(child)) {
1402                         LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1403                                         &child->mot_obj,
1404                                         "remote object doesn't exist.\n");
1405                         mdt_object_unlock(info, child, lhc, 1);
1406                         RETURN(-ENOENT);
1407                 }
1408
1409                 mdt_set_capainfo(info, 0, mdt_object_fid(child), BYPASS_CAPA);
1410                 rc = mdt_getattr_internal(info, child, 0);
1411                 if (unlikely(rc != 0))
1412                         mdt_object_unlock(info, child, lhc, 1);
1413
1414                 RETURN(rc);
1415         }
1416
1417         lname = &info->mti_name;
1418         mdt_name_unpack(info->mti_pill, &RMF_NAME, lname, MNF_FIX_ANON);
1419
1420         if (lu_name_is_valid(lname)) {
1421                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1422                        "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1423                        PNAME(lname), ldlm_rep);
1424         } else {
1425                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1426                 if (unlikely(reqbody == NULL))
1427                         RETURN(err_serious(-EPROTO));
1428
1429                 *child_fid = reqbody->fid2;
1430
1431                 if (unlikely(!fid_is_sane(child_fid)))
1432                         RETURN(err_serious(-EINVAL));
1433
1434                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1435                        "ldlm_rep = %p\n",
1436                        PFID(mdt_object_fid(parent)),
1437                        PFID(&reqbody->fid2), ldlm_rep);
1438         }
1439
1440         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1441
1442         if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1443                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1444                                 &parent->mot_obj,
1445                                 "Parent doesn't exist!\n");
1446                 RETURN(-ESTALE);
1447         }
1448
1449         if (mdt_object_remote(parent)) {
1450                 CERROR("%s: parent "DFID" is on remote target\n",
1451                        mdt_obd_name(info->mti_mdt),
1452                        PFID(mdt_object_fid(parent)));
1453                 RETURN(-EIO);
1454         }
1455
1456         if (lu_name_is_valid(lname)) {
1457                 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1458                 if (rc != 0) {
1459                         if (rc > 0)
1460                                 rc = 0;
1461                         RETURN(rc);
1462                 }
1463
1464                 /* step 1: lock parent only if parent is a directory */
1465                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1466                         lhp = &info->mti_lh[MDT_LH_PARENT];
1467                         mdt_lock_pdo_init(lhp, LCK_PR, lname);
1468                         rc = mdt_object_lock(info, parent, lhp,
1469                                              MDS_INODELOCK_UPDATE,
1470                                              MDT_LOCAL_LOCK);
1471                         if (unlikely(rc != 0))
1472                                 RETURN(rc);
1473                 }
1474
1475                 /* step 2: lookup child's fid by name */
1476                 fid_zero(child_fid);
1477                 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
1478                                 &info->mti_spec);
1479                 if (rc == -ENOENT)
1480                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1481
1482                 if (rc != 0)
1483                         GOTO(out_parent, rc);
1484         }
1485
1486         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1487
1488         /*
1489          *step 3: find the child object by fid & lock it.
1490          *        regardless if it is local or remote.
1491          *
1492          *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1493          *      set parent dir fid the same as child fid in getattr by fid case
1494          *      we should not lu_object_find() the object again, could lead
1495          *      to hung if there is a concurrent unlink destroyed the object.
1496          */
1497         if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1498                 mdt_object_get(info->mti_env, parent);
1499                 child = parent;
1500         } else {
1501                 child = mdt_object_find(info->mti_env, info->mti_mdt,
1502                                         child_fid);
1503         }
1504
1505         if (unlikely(IS_ERR(child)))
1506                 GOTO(out_parent, rc = PTR_ERR(child));
1507         if (is_resent) {
1508                 /* Do not take lock for resent case. */
1509                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1510                 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
1511                          lhc->mlh_reg_lh.cookie);
1512
1513                 if (!fid_res_name_eq(mdt_object_fid(child),
1514                                      &lock->l_resource->lr_name)) {
1515                         LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
1516                                                  &lock->l_resource->lr_name),
1517                                  "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1518                                  PLDLMRES(lock->l_resource),
1519                                  PFID(mdt_object_fid(parent)));
1520                         CWARN("Although resent, but still not get child lock"
1521                               "parent:"DFID" child:"DFID"\n",
1522                               PFID(mdt_object_fid(parent)),
1523                               PFID(mdt_object_fid(child)));
1524                         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
1525                         LDLM_LOCK_PUT(lock);
1526                         GOTO(relock, 0);
1527                 }
1528                 LDLM_LOCK_PUT(lock);
1529         } else {
1530 relock:
1531                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout*2);
1532                 mdt_lock_handle_init(lhc);
1533                 mdt_lock_reg_init(lhc, LCK_PR);
1534                 try_layout = false;
1535
1536                 if (!mdt_object_exists(child)) {
1537                         LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1538                                         &child->mot_obj,
1539                                         "Object doesn't exist!\n");
1540                         GOTO(out_child, rc = -ENOENT);
1541                 }
1542
1543                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1544                       mdt_object_exists(child) && !mdt_object_remote(child)) {
1545                         struct md_attr *ma = &info->mti_attr;
1546
1547                         ma->ma_valid = 0;
1548                         ma->ma_need = MA_INODE;
1549                         rc = mdt_attr_get_complex(info, child, ma);
1550                         if (unlikely(rc != 0))
1551                                 GOTO(out_child, rc);
1552
1553                         /* If the file has not been changed for some time, we
1554                          * return not only a LOOKUP lock, but also an UPDATE
1555                          * lock and this might save us RPC on later STAT. For
1556                          * directories, it also let negative dentry cache start
1557                          * working for this dir. */
1558                         if (ma->ma_valid & MA_INODE &&
1559                             ma->ma_attr.la_valid & LA_CTIME &&
1560                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1561                                 ma->ma_attr.la_ctime < cfs_time_current_sec())
1562                                 child_bits |= MDS_INODELOCK_UPDATE;
1563                 }
1564
1565                 /* layout lock must be granted in a best-effort way
1566                  * for IT operations */
1567                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1568                 if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1569                     exp_connect_layout(info->mti_exp) &&
1570                     S_ISREG(lu_object_attr(&child->mot_obj)) &&
1571                     !mdt_object_remote(child) && ldlm_rep != NULL) {
1572                         /* try to grant layout lock for regular file. */
1573                         try_layout = true;
1574                 }
1575
1576                 rc = 0;
1577                 if (try_layout) {
1578                         child_bits |= MDS_INODELOCK_LAYOUT;
1579                         /* try layout lock, it may fail to be granted due to
1580                          * contention at LOOKUP or UPDATE */
1581                         if (!mdt_object_lock_try(info, child, lhc, child_bits,
1582                                                  MDT_CROSS_LOCK)) {
1583                                 child_bits &= ~MDS_INODELOCK_LAYOUT;
1584                                 LASSERT(child_bits != 0);
1585                                 rc = mdt_object_lock(info, child, lhc,
1586                                                 child_bits, MDT_CROSS_LOCK);
1587                         } else {
1588                                 ma_need |= MA_LOV;
1589                         }
1590                 } else {
1591                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1592                          * client will enqueue the lock to the remote MDT */
1593                         if (mdt_object_remote(child))
1594                                 child_bits &= ~MDS_INODELOCK_UPDATE;
1595                         rc = mdt_object_lock(info, child, lhc, child_bits,
1596                                                 MDT_CROSS_LOCK);
1597                 }
1598                 if (unlikely(rc != 0))
1599                         GOTO(out_child, rc);
1600         }
1601
1602         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1603         /* Get MA_SOM attributes if update lock is given. */
1604         if (lock &&
1605             lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_UPDATE &&
1606             S_ISREG(lu_object_attr(&mdt_object_child(child)->mo_lu)))
1607                 ma_need |= MA_SOM;
1608
1609         /* finally, we can get attr for child. */
1610         mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
1611         rc = mdt_getattr_internal(info, child, ma_need);
1612         if (unlikely(rc != 0)) {
1613                 mdt_object_unlock(info, child, lhc, 1);
1614         } else if (lock) {
1615                 /* Debugging code. */
1616                 LDLM_DEBUG(lock, "Returning lock to client");
1617                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1618                                          &lock->l_resource->lr_name),
1619                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1620                          PLDLMRES(lock->l_resource),
1621                          PFID(mdt_object_fid(child)));
1622                 if (mdt_object_exists(child) && !mdt_object_remote(child))
1623                         mdt_pack_size2body(info, child);
1624         }
1625         if (lock)
1626                 LDLM_LOCK_PUT(lock);
1627
1628         EXIT;
1629 out_child:
1630         mdt_object_put(info->mti_env, child);
1631 out_parent:
1632         if (lhp)
1633                 mdt_object_unlock(info, parent, lhp, 1);
1634         return rc;
1635 }
1636
1637 /* normal handler: should release the child lock */
1638 static int mdt_getattr_name(struct tgt_session_info *tsi)
1639 {
1640         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1641         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1642         struct mdt_body        *reqbody;
1643         struct mdt_body        *repbody;
1644         int rc, rc2;
1645         ENTRY;
1646
1647         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1648         LASSERT(reqbody != NULL);
1649         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1650         LASSERT(repbody != NULL);
1651
1652         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1653         repbody->eadatasize = 0;
1654         repbody->aclsize = 0;
1655
1656         rc = mdt_init_ucred(info, reqbody);
1657         if (unlikely(rc))
1658                 GOTO(out_shrink, rc);
1659
1660         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1661         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1662                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1663                 lhc->mlh_reg_lh.cookie = 0;
1664         }
1665         mdt_exit_ucred(info);
1666         EXIT;
1667 out_shrink:
1668         mdt_client_compatibility(info);
1669         rc2 = mdt_fix_reply(info);
1670         if (rc == 0)
1671                 rc = rc2;
1672         mdt_thread_info_fini(info);
1673         return rc;
1674 }
1675
1676 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1677                          void *karg, void *uarg);
1678
1679 static int mdt_set_info(struct tgt_session_info *tsi)
1680 {
1681         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1682         char                    *key;
1683         void                    *val;
1684         int                      keylen, vallen, rc = 0;
1685
1686         ENTRY;
1687
1688         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
1689         if (key == NULL) {
1690                 DEBUG_REQ(D_HA, req, "no set_info key");
1691                 RETURN(err_serious(-EFAULT));
1692         }
1693
1694         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
1695                                       RCL_CLIENT);
1696
1697         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
1698         if (val == NULL) {
1699                 DEBUG_REQ(D_HA, req, "no set_info val");
1700                 RETURN(err_serious(-EFAULT));
1701         }
1702
1703         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
1704                                       RCL_CLIENT);
1705
1706         /* Swab any part of val you need to here */
1707         if (KEY_IS(KEY_READ_ONLY)) {
1708                 spin_lock(&req->rq_export->exp_lock);
1709                 if (*(__u32 *)val)
1710                         *exp_connect_flags_ptr(req->rq_export) |=
1711                                 OBD_CONNECT_RDONLY;
1712                 else
1713                         *exp_connect_flags_ptr(req->rq_export) &=
1714                                 ~OBD_CONNECT_RDONLY;
1715                 spin_unlock(&req->rq_export->exp_lock);
1716         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1717                 struct changelog_setinfo *cs = val;
1718
1719                 if (vallen != sizeof(*cs)) {
1720                         CERROR("%s: bad changelog_clear setinfo size %d\n",
1721                                tgt_name(tsi->tsi_tgt), vallen);
1722                         RETURN(-EINVAL);
1723                 }
1724                 if (ptlrpc_req_need_swab(req)) {
1725                         __swab64s(&cs->cs_recno);
1726                         __swab32s(&cs->cs_id);
1727                 }
1728
1729                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
1730                                    vallen, val, NULL);
1731         } else {
1732                 RETURN(-EINVAL);
1733         }
1734         RETURN(rc);
1735 }
1736
1737 static int mdt_readpage(struct tgt_session_info *tsi)
1738 {
1739         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
1740         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
1741         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
1742         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
1743         struct mdt_body         *repbody;
1744         int                      rc;
1745         int                      i;
1746
1747         ENTRY;
1748
1749         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1750                 RETURN(err_serious(-ENOMEM));
1751
1752         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
1753         if (repbody == NULL || reqbody == NULL)
1754                 RETURN(err_serious(-EFAULT));
1755
1756         /*
1757          * prepare @rdpg before calling lower layers and transfer itself. Here
1758          * reqbody->size contains offset of where to start to read and
1759          * reqbody->nlink contains number bytes to read.
1760          */
1761         rdpg->rp_hash = reqbody->size;
1762         if (rdpg->rp_hash != reqbody->size) {
1763                 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1764                        rdpg->rp_hash, reqbody->size);
1765                 RETURN(-EFAULT);
1766         }
1767
1768         rdpg->rp_attrs = reqbody->mode;
1769         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
1770                 rdpg->rp_attrs |= LUDA_64BITHASH;
1771         rdpg->rp_count  = min_t(unsigned int, reqbody->nlink,
1772                                 exp_max_brw_size(tsi->tsi_exp));
1773         rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE - 1) >>
1774                           PAGE_CACHE_SHIFT;
1775         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1776         if (rdpg->rp_pages == NULL)
1777                 RETURN(-ENOMEM);
1778
1779         for (i = 0; i < rdpg->rp_npages; ++i) {
1780                 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1781                 if (rdpg->rp_pages[i] == NULL)
1782                         GOTO(free_rdpg, rc = -ENOMEM);
1783         }
1784
1785         /* call lower layers to fill allocated pages with directory data */
1786         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
1787         if (rc < 0)
1788                 GOTO(free_rdpg, rc);
1789
1790         /* send pages to client */
1791         rc = tgt_sendpage(tsi, rdpg, rc);
1792
1793         EXIT;
1794 free_rdpg:
1795
1796         for (i = 0; i < rdpg->rp_npages; i++)
1797                 if (rdpg->rp_pages[i] != NULL)
1798                         __free_page(rdpg->rp_pages[i]);
1799         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1800
1801         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1802                 RETURN(0);
1803
1804         return rc;
1805 }
1806
1807 static int mdt_reint_internal(struct mdt_thread_info *info,
1808                               struct mdt_lock_handle *lhc,
1809                               __u32 op)
1810 {
1811         struct req_capsule      *pill = info->mti_pill;
1812         struct mdt_body         *repbody;
1813         int                      rc = 0, rc2;
1814
1815         ENTRY;
1816
1817         rc = mdt_reint_unpack(info, op);
1818         if (rc != 0) {
1819                 CERROR("Can't unpack reint, rc %d\n", rc);
1820                 RETURN(err_serious(rc));
1821         }
1822
1823         /* for replay (no_create) lmm is not needed, client has it already */
1824         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1825                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1826                                      DEF_REP_MD_SIZE);
1827
1828         /* llog cookies are always 0, the field is kept for compatibility */
1829         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1830                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
1831
1832         rc = req_capsule_server_pack(pill);
1833         if (rc != 0) {
1834                 CERROR("Can't pack response, rc %d\n", rc);
1835                 RETURN(err_serious(rc));
1836         }
1837
1838         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1839                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1840                 LASSERT(repbody);
1841                 repbody->eadatasize = 0;
1842                 repbody->aclsize = 0;
1843         }
1844
1845         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1846
1847         /* for replay no cookkie / lmm need, because client have this already */
1848         if (info->mti_spec.no_create)
1849                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1850                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1851
1852         rc = mdt_init_ucred_reint(info);
1853         if (rc)
1854                 GOTO(out_shrink, rc);
1855
1856         rc = mdt_fix_attr_ucred(info, op);
1857         if (rc != 0)
1858                 GOTO(out_ucred, rc = err_serious(rc));
1859
1860         if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1861                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
1862                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1863                 GOTO(out_ucred, rc);
1864         }
1865         rc = mdt_reint_rec(info, lhc);
1866         EXIT;
1867 out_ucred:
1868         mdt_exit_ucred(info);
1869 out_shrink:
1870         mdt_client_compatibility(info);
1871         rc2 = mdt_fix_reply(info);
1872         if (rc == 0)
1873                 rc = rc2;
1874         return rc;
1875 }
1876
1877 static long mdt_reint_opcode(struct ptlrpc_request *req,
1878                              const struct req_format **fmt)
1879 {
1880         struct mdt_device       *mdt;
1881         struct mdt_rec_reint    *rec;
1882         long                     opc;
1883
1884         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
1885         if (rec != NULL) {
1886                 opc = rec->rr_opcode;
1887                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
1888                 if (opc < REINT_MAX && fmt[opc] != NULL)
1889                         req_capsule_extend(&req->rq_pill, fmt[opc]);
1890                 else {
1891                         mdt = mdt_exp2dev(req->rq_export);
1892                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
1893                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
1894                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
1895                         opc = err_serious(-EFAULT);
1896                 }
1897         } else {
1898                 opc = err_serious(-EFAULT);
1899         }
1900         return opc;
1901 }
1902
1903 static int mdt_reint(struct tgt_session_info *tsi)
1904 {
1905         long opc;
1906         int  rc;
1907         static const struct req_format *reint_fmts[REINT_MAX] = {
1908                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
1909                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
1910                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
1911                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
1912                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
1913                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
1914                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
1915                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
1916                 [REINT_MIGRATE]  = &RQF_MDS_REINT_RENAME
1917         };
1918
1919         ENTRY;
1920
1921         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
1922         if (opc >= 0) {
1923                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1924                 /*
1925                  * No lock possible here from client to pass it to reint code
1926                  * path.
1927                  */
1928                 rc = mdt_reint_internal(info, NULL, opc);
1929                 mdt_thread_info_fini(info);
1930         } else {
1931                 rc = opc;
1932         }
1933
1934         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1935         RETURN(rc);
1936 }
1937
1938 /* this should sync the whole device */
1939 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1940 {
1941         struct dt_device *dt = mdt->mdt_bottom;
1942         int rc;
1943         ENTRY;
1944
1945         rc = dt->dd_ops->dt_sync(env, dt);
1946         RETURN(rc);
1947 }
1948
1949 /* this should sync this object */
1950 static int mdt_object_sync(struct mdt_thread_info *info)
1951 {
1952         struct md_object *next;
1953         int rc;
1954         ENTRY;
1955
1956         if (!mdt_object_exists(info->mti_object)) {
1957                 CWARN("Non existing object  "DFID"!\n",
1958                       PFID(mdt_object_fid(info->mti_object)));
1959                 RETURN(-ESTALE);
1960         }
1961         next = mdt_object_child(info->mti_object);
1962         rc = mo_object_sync(info->mti_env, next);
1963
1964         RETURN(rc);
1965 }
1966
1967 static int mdt_sync(struct tgt_session_info *tsi)
1968 {
1969         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1970         struct req_capsule      *pill = tsi->tsi_pill;
1971         struct mdt_body         *body;
1972         int                      rc;
1973
1974         ENTRY;
1975
1976         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1977                 RETURN(err_serious(-ENOMEM));
1978
1979         if (fid_seq(&tsi->tsi_mdt_body->fid1) == 0) {
1980                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
1981         } else {
1982                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1983
1984                 /* sync an object */
1985                 rc = mdt_object_sync(info);
1986                 if (rc == 0) {
1987                         const struct lu_fid *fid;
1988                         struct lu_attr *la = &info->mti_attr.ma_attr;
1989
1990                         info->mti_attr.ma_need = MA_INODE;
1991                         info->mti_attr.ma_valid = 0;
1992                         rc = mdt_attr_get_complex(info, info->mti_object,
1993                                                   &info->mti_attr);
1994                         if (rc == 0) {
1995                                 body = req_capsule_server_get(pill,
1996                                                               &RMF_MDT_BODY);
1997                                 fid = mdt_object_fid(info->mti_object);
1998                                 mdt_pack_attr2body(info, body, la, fid);
1999                         }
2000                 }
2001                 mdt_thread_info_fini(info);
2002         }
2003         if (rc == 0)
2004                 mdt_counter_incr(req, LPROC_MDT_SYNC);
2005
2006         RETURN(rc);
2007 }
2008
2009 /*
2010  * Handle quota control requests to consult current usage/limit, but also
2011  * to configure quota enforcement
2012  */
2013 static int mdt_quotactl(struct tgt_session_info *tsi)
2014 {
2015         struct obd_export       *exp  = tsi->tsi_exp;
2016         struct req_capsule      *pill = tsi->tsi_pill;
2017         struct obd_quotactl     *oqctl, *repoqc;
2018         int                      id, rc;
2019         struct mdt_device       *mdt = mdt_exp2dev(exp);
2020         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2021         ENTRY;
2022
2023         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
2024         if (oqctl == NULL)
2025                 RETURN(err_serious(-EPROTO));
2026
2027         rc = req_capsule_server_pack(pill);
2028         if (rc)
2029                 RETURN(err_serious(rc));
2030
2031         switch (oqctl->qc_cmd) {
2032         case Q_QUOTACHECK:
2033         case LUSTRE_Q_INVALIDATE:
2034         case LUSTRE_Q_FINVALIDATE:
2035         case Q_QUOTAON:
2036         case Q_QUOTAOFF:
2037         case Q_INITQUOTA:
2038                 /* deprecated, not used any more */
2039                 RETURN(-EOPNOTSUPP);
2040                 /* master quotactl */
2041         case Q_GETINFO:
2042         case Q_SETINFO:
2043         case Q_SETQUOTA:
2044         case Q_GETQUOTA:
2045                 if (qmt == NULL)
2046                         RETURN(-EOPNOTSUPP);
2047                 /* slave quotactl */
2048         case Q_GETOINFO:
2049         case Q_GETOQUOTA:
2050                 break;
2051         default:
2052                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2053                 RETURN(-EFAULT);
2054         }
2055
2056         /* map uid/gid for remote client */
2057         id = oqctl->qc_id;
2058         if (exp_connect_rmtclient(exp)) {
2059                 struct lustre_idmap_table *idmap;
2060
2061                 idmap = exp->exp_mdt_data.med_idmap;
2062
2063                 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
2064                              oqctl->qc_cmd != Q_GETINFO))
2065                         RETURN(-EPERM);
2066
2067                 if (oqctl->qc_type == USRQUOTA)
2068                         id = lustre_idmap_lookup_uid(NULL, idmap, 0,
2069                                                      oqctl->qc_id);
2070                 else if (oqctl->qc_type == GRPQUOTA)
2071                         id = lustre_idmap_lookup_gid(NULL, idmap, 0,
2072                                                      oqctl->qc_id);
2073                 else
2074                         RETURN(-EINVAL);
2075
2076                 if (id == CFS_IDMAP_NOTFOUND) {
2077                         CDEBUG(D_QUOTA, "no mapping for id %u\n", oqctl->qc_id);
2078                         RETURN(-EACCES);
2079                 }
2080         }
2081
2082         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2083         if (repoqc == NULL)
2084                 RETURN(err_serious(-EFAULT));
2085
2086         if (oqctl->qc_id != id)
2087                 swap(oqctl->qc_id, id);
2088
2089         switch (oqctl->qc_cmd) {
2090
2091         case Q_GETINFO:
2092         case Q_SETINFO:
2093         case Q_SETQUOTA:
2094         case Q_GETQUOTA:
2095                 /* forward quotactl request to QMT */
2096                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2097                 break;
2098
2099         case Q_GETOINFO:
2100         case Q_GETOQUOTA:
2101                 /* slave quotactl */
2102                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2103                                    oqctl);
2104                 break;
2105
2106         default:
2107                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2108                 RETURN(-EFAULT);
2109         }
2110
2111         if (oqctl->qc_id != id)
2112                 swap(oqctl->qc_id, id);
2113
2114         *repoqc = *oqctl;
2115         RETURN(rc);
2116 }
2117
2118 /** clone llog ctxt from child (mdd)
2119  * This allows remote llog (replicator) access.
2120  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2121  * context was originally set up, or we can handle them directly.
2122  * I choose the latter, but that means I need any llog
2123  * contexts set up by child to be accessable by the mdt.  So we clone the
2124  * context into our context list here.
2125  */
2126 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2127                                int idx)
2128 {
2129         struct md_device  *next = mdt->mdt_child;
2130         struct llog_ctxt *ctxt;
2131         int rc;
2132
2133         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2134                 return 0;
2135
2136         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2137         if (rc || ctxt == NULL) {
2138                 return 0;
2139         }
2140
2141         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2142         if (rc)
2143                 CERROR("Can't set mdt ctxt %d\n", rc);
2144
2145         return rc;
2146 }
2147
2148 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2149                                  struct mdt_device *mdt, int idx)
2150 {
2151         struct llog_ctxt *ctxt;
2152
2153         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2154         if (ctxt == NULL)
2155                 return 0;
2156         /* Put once for the get we just did, and once for the clone */
2157         llog_ctxt_put(ctxt);
2158         llog_ctxt_put(ctxt);
2159         return 0;
2160 }
2161
2162 /*
2163  * sec context handlers
2164  */
2165 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2166 {
2167         int rc;
2168
2169         rc = mdt_handle_idmap(tsi);
2170         if (unlikely(rc)) {
2171                 struct ptlrpc_request   *req = tgt_ses_req(tsi);
2172                 __u32                    opc;
2173
2174                 opc = lustre_msg_get_opc(req->rq_reqmsg);
2175                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2176                         sptlrpc_svc_ctx_invalidate(req);
2177         }
2178
2179         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2180
2181         return rc;
2182 }
2183
2184 /*
2185  * quota request handlers
2186  */
2187 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2188 {
2189         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
2190         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2191         int                      rc;
2192         ENTRY;
2193
2194         if (qmt == NULL)
2195                 RETURN(err_serious(-EOPNOTSUPP));
2196
2197         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2198         RETURN(rc);
2199 }
2200
2201 struct mdt_object *mdt_object_new(const struct lu_env *env,
2202                                   struct mdt_device *d,
2203                                   const struct lu_fid *f)
2204 {
2205         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2206         struct lu_object *o;
2207         struct mdt_object *m;
2208         ENTRY;
2209
2210         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2211         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2212         if (unlikely(IS_ERR(o)))
2213                 m = (struct mdt_object *)o;
2214         else
2215                 m = mdt_obj(o);
2216         RETURN(m);
2217 }
2218
2219 struct mdt_object *mdt_object_find(const struct lu_env *env,
2220                                    struct mdt_device *d,
2221                                    const struct lu_fid *f)
2222 {
2223         struct lu_object *o;
2224         struct mdt_object *m;
2225         ENTRY;
2226
2227         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2228         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2229         if (unlikely(IS_ERR(o)))
2230                 m = (struct mdt_object *)o;
2231         else
2232                 m = mdt_obj(o);
2233
2234         RETURN(m);
2235 }
2236
2237 /**
2238  * Asyncronous commit for mdt device.
2239  *
2240  * Pass asynchonous commit call down the MDS stack.
2241  *
2242  * \param env environment
2243  * \param mdt the mdt device
2244  */
2245 static void mdt_device_commit_async(const struct lu_env *env,
2246                                     struct mdt_device *mdt)
2247 {
2248         struct dt_device *dt = mdt->mdt_bottom;
2249         int rc;
2250
2251         rc = dt->dd_ops->dt_commit_async(env, dt);
2252         if (unlikely(rc != 0))
2253                 CWARN("async commit start failed with rc = %d", rc);
2254 }
2255
2256 /**
2257  * Mark the lock as "synchonous".
2258  *
2259  * Mark the lock to deffer transaction commit to the unlock time.
2260  *
2261  * \param lock the lock to mark as "synchonous"
2262  *
2263  * \see mdt_is_lock_sync
2264  * \see mdt_save_lock
2265  */
2266 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2267 {
2268         lock->l_ast_data = (void*)1;
2269 }
2270
2271 /**
2272  * Check whehter the lock "synchonous" or not.
2273  *
2274  * \param lock the lock to check
2275  * \retval 1 the lock is "synchonous"
2276  * \retval 0 the lock isn't "synchronous"
2277  *
2278  * \see mdt_set_lock_sync
2279  * \see mdt_save_lock
2280  */
2281 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2282 {
2283         return lock->l_ast_data != NULL;
2284 }
2285
2286 /**
2287  * Blocking AST for mdt locks.
2288  *
2289  * Starts transaction commit if in case of COS lock conflict or
2290  * deffers such a commit to the mdt_save_lock.
2291  *
2292  * \param lock the lock which blocks a request or cancelling lock
2293  * \param desc unused
2294  * \param data unused
2295  * \param flag indicates whether this cancelling or blocking callback
2296  * \retval 0
2297  * \see ldlm_blocking_ast_nocheck
2298  */
2299 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2300                      void *data, int flag)
2301 {
2302         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2303         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2304         int rc;
2305         ENTRY;
2306
2307         if (flag == LDLM_CB_CANCELING)
2308                 RETURN(0);
2309         lock_res_and_lock(lock);
2310         if (lock->l_blocking_ast != mdt_blocking_ast) {
2311                 unlock_res_and_lock(lock);
2312                 RETURN(0);
2313         }
2314         if (mdt_cos_is_enabled(mdt) &&
2315             lock->l_req_mode & (LCK_PW | LCK_EX) &&
2316             lock->l_blocking_lock != NULL &&
2317             lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
2318                 mdt_set_lock_sync(lock);
2319         }
2320         rc = ldlm_blocking_ast_nocheck(lock);
2321
2322         /* There is no lock conflict if l_blocking_lock == NULL,
2323          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2324          * when the last reference to a local lock was released */
2325         if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2326                 struct lu_env env;
2327
2328                 rc = lu_env_init(&env, LCT_LOCAL);
2329                 if (unlikely(rc != 0))
2330                         CWARN("lu_env initialization failed with rc = %d,"
2331                               "cannot start asynchronous commit\n", rc);
2332                 else
2333                         mdt_device_commit_async(&env, mdt);
2334                 lu_env_fini(&env);
2335         }
2336         RETURN(rc);
2337 }
2338
2339 /* Used for cross-MDT lock */
2340 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2341                             void *data, int flag)
2342 {
2343         struct lustre_handle lockh;
2344         int               rc;
2345
2346         switch (flag) {
2347         case LDLM_CB_BLOCKING:
2348                 ldlm_lock2handle(lock, &lockh);
2349                 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
2350                 if (rc < 0) {
2351                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2352                         RETURN(rc);
2353                 }
2354                 break;
2355         case LDLM_CB_CANCELING:
2356                 LDLM_DEBUG(lock, "Revoke remote lock\n");
2357                 break;
2358         default:
2359                 LBUG();
2360         }
2361         RETURN(0);
2362 }
2363
2364 int mdt_remote_object_lock(struct mdt_thread_info *mti,
2365                            struct mdt_object *o, const struct lu_fid *fid,
2366                            struct lustre_handle *lh, ldlm_mode_t mode,
2367                            __u64 ibits)
2368 {
2369         struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
2370         ldlm_policy_data_t *policy = &mti->mti_policy;
2371         struct ldlm_res_id *res_id = &mti->mti_res_id;
2372         int rc = 0;
2373         ENTRY;
2374
2375         LASSERT(mdt_object_remote(o));
2376
2377         LASSERT(ibits == MDS_INODELOCK_UPDATE);
2378
2379         fid_build_reg_res_name(fid, res_id);
2380
2381         memset(einfo, 0, sizeof(*einfo));
2382         einfo->ei_type = LDLM_IBITS;
2383         einfo->ei_mode = mode;
2384         einfo->ei_cb_bl = mdt_remote_blocking_ast;
2385         einfo->ei_cb_cp = ldlm_completion_ast;
2386         einfo->ei_enq_slave = 0;
2387         einfo->ei_res_id = res_id;
2388
2389         memset(policy, 0, sizeof(*policy));
2390         policy->l_inodebits.bits = ibits;
2391
2392         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2393                             policy);
2394         RETURN(rc);
2395 }
2396
2397 static int mdt_object_local_lock(struct mdt_thread_info *info,
2398                                  struct mdt_object *o,
2399                                  struct mdt_lock_handle *lh, __u64 ibits,
2400                                  bool nonblock, int locality)
2401 {
2402         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2403         ldlm_policy_data_t *policy = &info->mti_policy;
2404         struct ldlm_res_id *res_id = &info->mti_res_id;
2405         __u64 dlmflags;
2406         int rc;
2407         ENTRY;
2408
2409         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2410         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2411         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2412         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2413
2414         /* Only enqueue LOOKUP lock for remote object */
2415         if (mdt_object_remote(o))
2416                 LASSERT(ibits == MDS_INODELOCK_LOOKUP);
2417
2418         if (lh->mlh_type == MDT_PDO_LOCK) {
2419                 /* check for exists after object is locked */
2420                 if (mdt_object_exists(o) == 0) {
2421                         /* Non-existent object shouldn't have PDO lock */
2422                         RETURN(-ESTALE);
2423                 } else {
2424                         /* Non-dir object shouldn't have PDO lock */
2425                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2426                                 RETURN(-ENOTDIR);
2427                 }
2428         }
2429
2430         memset(policy, 0, sizeof(*policy));
2431         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2432
2433         dlmflags = LDLM_FL_ATOMIC_CB;
2434         if (nonblock)
2435                 dlmflags |= LDLM_FL_BLOCK_NOWAIT;
2436
2437         /*
2438          * Take PDO lock on whole directory and build correct @res_id for lock
2439          * on part of directory.
2440          */
2441         if (lh->mlh_pdo_hash != 0) {
2442                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2443                 mdt_lock_pdo_mode(info, o, lh);
2444                 if (lh->mlh_pdo_mode != LCK_NL) {
2445                         /*
2446                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2447                          * is never going to be sent to client and we do not
2448                          * want it slowed down due to possible cancels.
2449                          */
2450                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2451                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2452                                           policy, res_id, dlmflags,
2453                                           info->mti_exp == NULL ? NULL :
2454                                           &info->mti_exp->exp_handle.h_cookie);
2455                         if (unlikely(rc))
2456                                 RETURN(rc);
2457                 }
2458
2459                 /*
2460                  * Finish res_id initializing by name hash marking part of
2461                  * directory which is taking modification.
2462                  */
2463                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2464         }
2465
2466         policy->l_inodebits.bits = ibits;
2467
2468         /*
2469          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2470          * going to be sent to client. If it is - mdt_intent_policy() path will
2471          * fix it up and turn FL_LOCAL flag off.
2472          */
2473         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2474                           res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2475                           info->mti_exp == NULL ? NULL :
2476                           &info->mti_exp->exp_handle.h_cookie);
2477         if (rc)
2478                 mdt_object_unlock(info, o, lh, 1);
2479         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2480                  lh->mlh_pdo_hash != 0 &&
2481                  (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX)) {
2482                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2483         }
2484
2485         RETURN(rc);
2486 }
2487
2488 static int
2489 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
2490                          struct mdt_lock_handle *lh, __u64 ibits,
2491                          bool nonblock, int locality)
2492 {
2493         int rc;
2494         ENTRY;
2495
2496         if (!mdt_object_remote(o))
2497                 return mdt_object_local_lock(info, o, lh, ibits, nonblock,
2498                                              locality);
2499
2500         if (locality == MDT_LOCAL_LOCK) {
2501                 CERROR("%s: try to get local lock for remote object"
2502                        DFID".\n", mdt_obd_name(info->mti_mdt),
2503                        PFID(mdt_object_fid(o)));
2504                 RETURN(-EPROTO);
2505         }
2506
2507         /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
2508         ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
2509                    MDS_INODELOCK_XATTR);
2510         if (ibits & MDS_INODELOCK_UPDATE) {
2511                 /* Sigh, PDO needs to enqueue 2 locks right now, but
2512                  * enqueue RPC can only request 1 lock, to avoid extra
2513                  * RPC, so it will instead enqueue EX lock for remote
2514                  * object anyway XXX*/
2515                 if (lh->mlh_type == MDT_PDO_LOCK &&
2516                     lh->mlh_pdo_hash != 0) {
2517                         CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
2518                                "EX lock.\n", mdt_obd_name(info->mti_mdt),
2519                                PFID(mdt_object_fid(o)));
2520                         lh->mlh_pdo_hash = 0;
2521                         lh->mlh_rreg_mode = LCK_EX;
2522                         lh->mlh_type = MDT_REG_LOCK;
2523                 }
2524                 rc = mdt_remote_object_lock(info, o, mdt_object_fid(o),
2525                                             &lh->mlh_rreg_lh,
2526                                             lh->mlh_rreg_mode,
2527                                             MDS_INODELOCK_UPDATE);
2528                 if (rc != ELDLM_OK)
2529                         RETURN(rc);
2530         }
2531
2532         /* Only enqueue LOOKUP lock for remote object */
2533         if (ibits & MDS_INODELOCK_LOOKUP) {
2534                 rc = mdt_object_local_lock(info, o, lh,
2535                                            MDS_INODELOCK_LOOKUP,
2536                                            nonblock, locality);
2537                 if (rc != ELDLM_OK)
2538                         RETURN(rc);
2539         }
2540
2541         RETURN(0);
2542 }
2543
2544 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2545                     struct mdt_lock_handle *lh, __u64 ibits, int locality)
2546 {
2547         return mdt_object_lock_internal(info, o, lh, ibits, false, locality);
2548 }
2549
2550 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
2551                         struct mdt_lock_handle *lh, __u64 ibits, int locality)
2552 {
2553         struct mdt_lock_handle tmp = *lh;
2554         int rc;
2555
2556         rc = mdt_object_lock_internal(info, o, &tmp, ibits, true, locality);
2557         if (rc == 0)
2558                 *lh = tmp;
2559
2560         return rc == 0;
2561 }
2562
2563 /**
2564  * Save a lock within request object.
2565  *
2566  * Keep the lock referenced until whether client ACK or transaction
2567  * commit happens or release the lock immediately depending on input
2568  * parameters. If COS is ON, a write lock is converted to COS lock
2569  * before saving.
2570  *
2571  * \param info thead info object
2572  * \param h lock handle
2573  * \param mode lock mode
2574  * \param decref force immediate lock releasing
2575  */
2576 static
2577 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2578                    ldlm_mode_t mode, int decref)
2579 {
2580         ENTRY;
2581
2582         if (lustre_handle_is_used(h)) {
2583                 if (decref || !info->mti_has_trans ||
2584                     !(mode & (LCK_PW | LCK_EX))){
2585                         mdt_fid_unlock(h, mode);
2586                 } else {
2587                         struct mdt_device *mdt = info->mti_mdt;
2588                         struct ldlm_lock *lock = ldlm_handle2lock(h);
2589                         struct ptlrpc_request *req = mdt_info_req(info);
2590                         int no_ack = 0;
2591
2592                         LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2593                                  h->cookie);
2594                         /* there is no request if mdt_object_unlock() is called
2595                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
2596                         if (likely(req != NULL)) {
2597                                 CDEBUG(D_HA, "request = %p reply state = %p"
2598                                        " transno = "LPD64"\n", req,
2599                                        req->rq_reply_state, req->rq_transno);
2600                                 if (mdt_cos_is_enabled(mdt)) {
2601                                         no_ack = 1;
2602                                         ldlm_lock_downgrade(lock, LCK_COS);
2603                                         mode = LCK_COS;
2604                                 }
2605                                 ptlrpc_save_lock(req, h, mode, no_ack);
2606                         } else {
2607                                 ldlm_lock_decref(h, mode);
2608                         }
2609                         if (mdt_is_lock_sync(lock)) {
2610                                 CDEBUG(D_HA, "found sync-lock,"
2611                                        " async commit started\n");
2612                                 mdt_device_commit_async(info->mti_env,
2613                                                         mdt);
2614                         }
2615                         LDLM_LOCK_PUT(lock);
2616                 }
2617                 h->cookie = 0ull;
2618         }
2619
2620         EXIT;
2621 }
2622
2623 /**
2624  * Unlock mdt object.
2625  *
2626  * Immeditely release the regular lock and the PDO lock or save the
2627  * lock in reqeuest and keep them referenced until client ACK or
2628  * transaction commit.
2629  *
2630  * \param info thread info object
2631  * \param o mdt object
2632  * \param lh mdt lock handle referencing regular and PDO locks
2633  * \param decref force immediate lock releasing
2634  */
2635 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2636                        struct mdt_lock_handle *lh, int decref)
2637 {
2638         ENTRY;
2639
2640         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2641         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2642
2643         if (lustre_handle_is_used(&lh->mlh_rreg_lh))
2644                 ldlm_lock_decref(&lh->mlh_rreg_lh, lh->mlh_rreg_mode);
2645
2646         EXIT;
2647 }
2648
2649 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2650                                         const struct lu_fid *f,
2651                                         struct mdt_lock_handle *lh,
2652                                         __u64 ibits)
2653 {
2654         struct mdt_object *o;
2655
2656         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2657         if (!IS_ERR(o)) {
2658                 int rc;
2659
2660                 rc = mdt_object_lock(info, o, lh, ibits,
2661                                      MDT_LOCAL_LOCK);
2662                 if (rc != 0) {
2663                         mdt_object_put(info->mti_env, o);
2664                         o = ERR_PTR(rc);
2665                 }
2666         }
2667         return o;
2668 }
2669
2670 void mdt_object_unlock_put(struct mdt_thread_info * info,
2671                            struct mdt_object * o,
2672                            struct mdt_lock_handle *lh,
2673                            int decref)
2674 {
2675         mdt_object_unlock(info, o, lh, decref);
2676         mdt_object_put(info->mti_env, o);
2677 }
2678
2679 /*
2680  * Generic code handling requests that have struct mdt_body passed in:
2681  *
2682  *  - extract mdt_body from request and save it in @info, if present;
2683  *
2684  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
2685  *  @info;
2686  *
2687  *  - if HABEO_CORPUS flag is set for this request type check whether object
2688  *  actually exists on storage (lu_object_exists()).
2689  *
2690  */
2691 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2692 {
2693         const struct mdt_body    *body;
2694         struct mdt_object        *obj;
2695         const struct lu_env      *env;
2696         struct req_capsule       *pill;
2697         int                       rc;
2698         ENTRY;
2699
2700         env = info->mti_env;
2701         pill = info->mti_pill;
2702
2703         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2704         if (body == NULL)
2705                 RETURN(-EFAULT);
2706
2707         if (!(body->valid & OBD_MD_FLID))
2708                 RETURN(0);
2709
2710         if (!fid_is_sane(&body->fid1)) {
2711                 CERROR("Invalid fid: "DFID"\n", PFID(&body->fid1));
2712                 RETURN(-EINVAL);
2713         }
2714
2715         /*
2716          * Do not get size or any capa fields before we check that request
2717          * contains capa actually. There are some requests which do not, for
2718          * instance MDS_IS_SUBDIR.
2719          */
2720         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2721             req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
2722                 mdt_set_capainfo(info, 0, &body->fid1,
2723                                  req_capsule_client_get(pill, &RMF_CAPA1));
2724
2725         obj = mdt_object_find(env, info->mti_mdt, &body->fid1);
2726         if (!IS_ERR(obj)) {
2727                 if ((flags & HABEO_CORPUS) &&
2728                     !mdt_object_exists(obj)) {
2729                         mdt_object_put(env, obj);
2730                         /* for capability renew ENOENT will be handled in
2731                          * mdt_renew_capa */
2732                         if (body->valid & OBD_MD_FLOSSCAPA)
2733                                 rc = 0;
2734                         else
2735                                 rc = -ENOENT;
2736                 } else {
2737                         info->mti_object = obj;
2738                         rc = 0;
2739                 }
2740         } else
2741                 rc = PTR_ERR(obj);
2742
2743         RETURN(rc);
2744 }
2745
2746 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2747 {
2748         struct req_capsule *pill = info->mti_pill;
2749         int rc;
2750         ENTRY;
2751
2752         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2753                 rc = mdt_body_unpack(info, flags);
2754         else
2755                 rc = 0;
2756
2757         if (rc == 0 && (flags & HABEO_REFERO)) {
2758                 /* Pack reply. */
2759                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2760                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2761                                              DEF_REP_MD_SIZE);
2762                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2763                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
2764                                              RCL_SERVER, 0);
2765
2766                 rc = req_capsule_server_pack(pill);
2767         }
2768         RETURN(rc);
2769 }
2770
2771 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2772 {
2773         struct md_device *next = m->mdt_child;
2774
2775         return next->md_ops->mdo_init_capa_ctxt(env, next,
2776                                                 m->mdt_lut.lut_mds_capa,
2777                                                 m->mdt_capa_timeout,
2778                                                 m->mdt_capa_alg,
2779                                                 m->mdt_capa_keys);
2780 }
2781
2782 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2783 {
2784         lh->mlh_type = MDT_NUL_LOCK;
2785         lh->mlh_reg_lh.cookie = 0ull;
2786         lh->mlh_reg_mode = LCK_MINMODE;
2787         lh->mlh_pdo_lh.cookie = 0ull;
2788         lh->mlh_pdo_mode = LCK_MINMODE;
2789         lh->mlh_rreg_lh.cookie = 0ull;
2790         lh->mlh_rreg_mode = LCK_MINMODE;
2791 }
2792
2793 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2794 {
2795         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2796         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2797 }
2798
2799 /*
2800  * Initialize fields of struct mdt_thread_info. Other fields are left in
2801  * uninitialized state, because it's too expensive to zero out whole
2802  * mdt_thread_info (> 1K) on each request arrival.
2803  */
2804 void mdt_thread_info_init(struct ptlrpc_request *req,
2805                           struct mdt_thread_info *info)
2806 {
2807         int i;
2808
2809         info->mti_pill = &req->rq_pill;
2810
2811         /* lock handle */
2812         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2813                 mdt_lock_handle_init(&info->mti_lh[i]);
2814
2815         /* mdt device: it can be NULL while CONNECT */
2816         if (req->rq_export) {
2817                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2818                 info->mti_exp = req->rq_export;
2819         } else
2820                 info->mti_mdt = NULL;
2821         info->mti_env = req->rq_svc_thread->t_env;
2822         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2823
2824         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2825         info->mti_big_buf = LU_BUF_NULL;
2826         info->mti_body = NULL;
2827         info->mti_object = NULL;
2828         info->mti_dlm_req = NULL;
2829         info->mti_has_trans = 0;
2830         info->mti_cross_ref = 0;
2831         info->mti_opdata = 0;
2832         info->mti_big_lmm_used = 0;
2833
2834         info->mti_spec.no_create = 0;
2835         info->mti_spec.sp_rm_entry = 0;
2836
2837         info->mti_spec.u.sp_ea.eadata = NULL;
2838         info->mti_spec.u.sp_ea.eadatalen = 0;
2839 }
2840
2841 void mdt_thread_info_fini(struct mdt_thread_info *info)
2842 {
2843         int i;
2844
2845         if (info->mti_object != NULL) {
2846                 mdt_object_put(info->mti_env, info->mti_object);
2847                 info->mti_object = NULL;
2848         }
2849
2850         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2851                 mdt_lock_handle_fini(&info->mti_lh[i]);
2852         info->mti_env = NULL;
2853         info->mti_pill = NULL;
2854         info->mti_exp = NULL;
2855
2856         if (unlikely(info->mti_big_buf.lb_buf != NULL))
2857                 lu_buf_free(&info->mti_big_buf);
2858 }
2859
2860 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
2861 {
2862         struct mdt_thread_info  *mti;
2863         struct lustre_capa      *lc;
2864
2865         mti = mdt_th_info(tsi->tsi_env);
2866         LASSERT(mti != NULL);
2867
2868         mdt_thread_info_init(tgt_ses_req(tsi), mti);
2869         if (tsi->tsi_corpus != NULL) {
2870                 struct req_capsule *pill = tsi->tsi_pill;
2871
2872                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
2873                 lu_object_get(tsi->tsi_corpus);
2874
2875                 /*
2876                  * XXX: must be part of tgt_mdt_body_unpack but moved here
2877                  * due to mdt_set_capainfo().
2878                  */
2879                 if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2880                     req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT) > 0) {
2881                         lc = req_capsule_client_get(pill, &RMF_CAPA1);
2882                         mdt_set_capainfo(mti, 0, &tsi->tsi_mdt_body->fid1, lc);
2883                 }
2884         }
2885         mti->mti_body = tsi->tsi_mdt_body;
2886         mti->mti_dlm_req = tsi->tsi_dlm_req;
2887
2888         return mti;
2889 }
2890
2891 static int mdt_tgt_connect(struct tgt_session_info *tsi)
2892 {
2893         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2894         int                      rc;
2895
2896         ENTRY;
2897
2898         rc = tgt_connect(tsi);
2899         if (rc != 0)
2900                 RETURN(rc);
2901
2902         rc = mdt_init_idmap(tsi);
2903         if (rc != 0)
2904                 GOTO(err, rc);
2905         RETURN(0);
2906 err:
2907         obd_disconnect(class_export_get(req->rq_export));
2908         return rc;
2909 }
2910
2911 enum mdt_it_code {
2912         MDT_IT_OPEN,
2913         MDT_IT_OCREAT,
2914         MDT_IT_CREATE,
2915         MDT_IT_GETATTR,
2916         MDT_IT_READDIR,
2917         MDT_IT_LOOKUP,
2918         MDT_IT_UNLINK,
2919         MDT_IT_TRUNC,
2920         MDT_IT_GETXATTR,
2921         MDT_IT_LAYOUT,
2922         MDT_IT_QUOTA,
2923         MDT_IT_NR
2924 };
2925
2926 static int mdt_intent_getattr(enum mdt_it_code opcode,
2927                               struct mdt_thread_info *info,
2928                               struct ldlm_lock **,
2929                               __u64);
2930
2931 static int mdt_intent_getxattr(enum mdt_it_code opcode,
2932                                 struct mdt_thread_info *info,
2933                                 struct ldlm_lock **lockp,
2934                                 __u64 flags);
2935
2936 static int mdt_intent_layout(enum mdt_it_code opcode,
2937                              struct mdt_thread_info *info,
2938                              struct ldlm_lock **,
2939                              __u64);
2940 static int mdt_intent_reint(enum mdt_it_code opcode,
2941                             struct mdt_thread_info *info,
2942                             struct ldlm_lock **,
2943                             __u64);
2944
2945 static struct mdt_it_flavor {
2946         const struct req_format *it_fmt;
2947         __u32                    it_flags;
2948         int                    (*it_act)(enum mdt_it_code ,
2949                                          struct mdt_thread_info *,
2950                                          struct ldlm_lock **,
2951                                          __u64);
2952         long                     it_reint;
2953 } mdt_it_flavor[] = {
2954         [MDT_IT_OPEN]     = {
2955                 .it_fmt   = &RQF_LDLM_INTENT,
2956                 /*.it_flags = HABEO_REFERO,*/
2957                 .it_flags = 0,
2958                 .it_act   = mdt_intent_reint,
2959                 .it_reint = REINT_OPEN
2960         },
2961         [MDT_IT_OCREAT]   = {
2962                 .it_fmt   = &RQF_LDLM_INTENT,
2963                 /*
2964                  * OCREAT is not a MUTABOR request as if the file
2965                  * already exists.
2966                  * We do the extra check of OBD_CONNECT_RDONLY in
2967                  * mdt_reint_open() when we really need to create
2968                  * the object.
2969                  */
2970                 .it_flags = 0,
2971                 .it_act   = mdt_intent_reint,
2972                 .it_reint = REINT_OPEN
2973         },
2974         [MDT_IT_CREATE]   = {
2975                 .it_fmt   = &RQF_LDLM_INTENT,
2976                 .it_flags = MUTABOR,
2977                 .it_act   = mdt_intent_reint,
2978                 .it_reint = REINT_CREATE
2979         },
2980         [MDT_IT_GETATTR]  = {
2981                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2982                 .it_flags = HABEO_REFERO,
2983                 .it_act   = mdt_intent_getattr
2984         },
2985         [MDT_IT_READDIR]  = {
2986                 .it_fmt   = NULL,
2987                 .it_flags = 0,
2988                 .it_act   = NULL
2989         },
2990         [MDT_IT_LOOKUP]   = {
2991                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2992                 .it_flags = HABEO_REFERO,
2993                 .it_act   = mdt_intent_getattr
2994         },
2995         [MDT_IT_UNLINK]   = {
2996                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
2997                 .it_flags = MUTABOR,
2998                 .it_act   = NULL,
2999                 .it_reint = REINT_UNLINK
3000         },
3001         [MDT_IT_TRUNC]    = {
3002                 .it_fmt   = NULL,
3003                 .it_flags = MUTABOR,
3004                 .it_act   = NULL
3005         },
3006         [MDT_IT_GETXATTR] = {
3007                 .it_fmt   = &RQF_LDLM_INTENT_GETXATTR,
3008                 .it_flags = HABEO_CORPUS,
3009                 .it_act   = mdt_intent_getxattr
3010         },
3011         [MDT_IT_LAYOUT] = {
3012                 .it_fmt   = &RQF_LDLM_INTENT_LAYOUT,
3013                 .it_flags = 0,
3014                 .it_act   = mdt_intent_layout
3015         }
3016 };
3017
3018 static int
3019 mdt_intent_lock_replace(struct mdt_thread_info *info, struct ldlm_lock **lockp,
3020                         struct ldlm_lock *new_lock, struct mdt_lock_handle *lh,
3021                         __u64 flags)
3022 {
3023         struct ptlrpc_request  *req = mdt_info_req(info);
3024         struct ldlm_lock       *lock = *lockp;
3025
3026         /*
3027          * Get new lock only for cases when possible resent did not find any
3028          * lock.
3029          */
3030         if (new_lock == NULL)
3031                 new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3032
3033         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3034                 lh->mlh_reg_lh.cookie = 0;
3035                 RETURN(0);
3036         }
3037
3038         LASSERTF(new_lock != NULL,
3039                  "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
3040
3041         /*
3042          * If we've already given this lock to a client once, then we should
3043          * have no readers or writers.  Otherwise, we should have one reader
3044          * _or_ writer ref (which will be zeroed below) before returning the
3045          * lock to a client.
3046          */
3047         if (new_lock->l_export == req->rq_export) {
3048                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3049         } else {
3050                 LASSERT(new_lock->l_export == NULL);
3051                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3052         }
3053
3054         *lockp = new_lock;
3055
3056         if (new_lock->l_export == req->rq_export) {
3057                 /*
3058                  * Already gave this to the client, which means that we
3059                  * reconstructed a reply.
3060                  */
3061                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3062                         MSG_RESENT);
3063                 lh->mlh_reg_lh.cookie = 0;
3064                 RETURN(ELDLM_LOCK_REPLACED);
3065         }
3066
3067         /*
3068          * Fixup the lock to be given to the client.
3069          */
3070         lock_res_and_lock(new_lock);
3071         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3072          * possible blocking AST. */
3073         while (new_lock->l_readers > 0) {
3074                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3075                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3076                 new_lock->l_readers--;
3077         }
3078         while (new_lock->l_writers > 0) {
3079                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3080                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3081                 new_lock->l_writers--;
3082         }
3083
3084         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3085         new_lock->l_blocking_ast = lock->l_blocking_ast;
3086         new_lock->l_completion_ast = lock->l_completion_ast;
3087         new_lock->l_remote_handle = lock->l_remote_handle;
3088         new_lock->l_flags &= ~LDLM_FL_LOCAL;
3089
3090         unlock_res_and_lock(new_lock);
3091
3092         cfs_hash_add(new_lock->l_export->exp_lock_hash,
3093                      &new_lock->l_remote_handle,
3094                      &new_lock->l_exp_hash);
3095
3096         LDLM_LOCK_RELEASE(new_lock);
3097         lh->mlh_reg_lh.cookie = 0;
3098
3099         RETURN(ELDLM_LOCK_REPLACED);
3100 }
3101
3102 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3103                                     struct ldlm_lock *new_lock,
3104                                     struct ldlm_lock **old_lock,
3105                                     struct mdt_lock_handle *lh,
3106                                     enum mdt_it_code opcode)
3107 {
3108         struct ptlrpc_request  *req = mdt_info_req(info);
3109         struct obd_export      *exp = req->rq_export;
3110         struct lustre_handle    remote_hdl;
3111         struct ldlm_request    *dlmreq;
3112         struct ldlm_lock       *lock;
3113
3114         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3115                 return;
3116
3117         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3118         remote_hdl = dlmreq->lock_handle[0];
3119         /* If the client does not require open lock, it does not need to
3120          * search lock in exp_lock_hash, since the server thread will
3121          * make sure the lock will be released, and the resend request
3122          * can always re-enqueue the lock */
3123         if ((opcode != MDT_IT_OPEN) || (opcode == MDT_IT_OPEN &&
3124             info->mti_spec.sp_cr_flags & MDS_OPEN_LOCK)) {
3125                 /* In the function below, .hs_keycmp resolves to
3126                  * ldlm_export_lock_keycmp() */
3127                 /* coverity[overrun-buffer-val] */
3128                 lock = cfs_hash_lookup(exp->exp_lock_hash, &remote_hdl);
3129                 if (lock) {
3130                         lock_res_and_lock(lock);
3131                         if (lock != new_lock) {
3132                                 lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie;
3133                                 lh->mlh_reg_mode = lock->l_granted_mode;
3134
3135                                 LDLM_DEBUG(lock, "Restoring lock cookie");
3136                                 DEBUG_REQ(D_DLMTRACE, req,
3137                                           "restoring lock cookie "LPX64,
3138                                           lh->mlh_reg_lh.cookie);
3139                                 if (old_lock)
3140                                         *old_lock = LDLM_LOCK_GET(lock);
3141                                 cfs_hash_put(exp->exp_lock_hash,
3142                                              &lock->l_exp_hash);
3143                                 unlock_res_and_lock(lock);
3144                                 return;
3145                         }
3146                         cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
3147                         unlock_res_and_lock(lock);
3148                 }
3149         }
3150         /*
3151          * If the xid matches, then we know this is a resent request, and allow
3152          * it. (It's probably an OPEN, for which we don't send a lock.
3153          */
3154         if (req_xid_is_last(req))
3155                 return;
3156
3157         /*
3158          * This remote handle isn't enqueued, so we never received or processed
3159          * this request.  Clear MSG_RESENT, because it can be handled like any
3160          * normal request now.
3161          */
3162         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3163
3164         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
3165                   remote_hdl.cookie);
3166 }
3167
3168 static int mdt_intent_getxattr(enum mdt_it_code opcode,
3169                                 struct mdt_thread_info *info,
3170                                 struct ldlm_lock **lockp,
3171                                 __u64 flags)
3172 {
3173         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3174         struct ldlm_reply      *ldlm_rep = NULL;
3175         int rc, grc;
3176
3177         /*
3178          * Initialize lhc->mlh_reg_lh either from a previously granted lock
3179          * (for the resend case) or a new lock. Below we will use it to
3180          * replace the original lock.
3181          */
3182         mdt_intent_fixup_resent(info, *lockp, NULL, lhc, opcode);
3183         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3184                 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3185                 rc = mdt_object_lock(info, info->mti_object, lhc,
3186                                         MDS_INODELOCK_XATTR,
3187                                         MDT_LOCAL_LOCK);
3188                 if (rc)
3189                         return rc;
3190         }
3191
3192         grc = mdt_getxattr(info);
3193
3194         rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3195
3196         if (mdt_info_req(info)->rq_repmsg != NULL)
3197                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3198         if (ldlm_rep == NULL)
3199                 RETURN(err_serious(-EFAULT));
3200
3201         ldlm_rep->lock_policy_res2 = grc;
3202
3203         return rc;
3204 }
3205
3206 static int mdt_intent_getattr(enum mdt_it_code opcode,
3207                               struct mdt_thread_info *info,
3208                               struct ldlm_lock **lockp,
3209                               __u64 flags)
3210 {
3211         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3212         struct ldlm_lock       *new_lock = NULL;
3213         __u64                   child_bits;
3214         struct ldlm_reply      *ldlm_rep;
3215         struct mdt_body        *reqbody;
3216         struct mdt_body        *repbody;
3217         int                     rc, rc2;
3218         ENTRY;
3219
3220         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3221         LASSERT(reqbody);
3222
3223         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3224         LASSERT(repbody);
3225
3226         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
3227         repbody->eadatasize = 0;
3228         repbody->aclsize = 0;
3229
3230         switch (opcode) {
3231         case MDT_IT_LOOKUP:
3232                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3233                 break;
3234         case MDT_IT_GETATTR:
3235                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3236                              MDS_INODELOCK_PERM;
3237                 break;
3238         default:
3239                 CERROR("Unsupported intent (%d)\n", opcode);
3240                 GOTO(out_shrink, rc = -EINVAL);
3241         }
3242
3243         rc = mdt_init_ucred(info, reqbody);
3244         if (rc)
3245                 GOTO(out_shrink, rc);
3246
3247         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3248         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3249
3250         /* Get lock from request for possible resent case. */
3251         mdt_intent_fixup_resent(info, *lockp, &new_lock, lhc, opcode);
3252
3253         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3254         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3255
3256         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3257                 ldlm_rep->lock_policy_res2 = 0;
3258         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3259             ldlm_rep->lock_policy_res2) {
3260                 lhc->mlh_reg_lh.cookie = 0ull;
3261                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3262         }
3263
3264         rc = mdt_intent_lock_replace(info, lockp, new_lock, lhc, flags);
3265         EXIT;
3266 out_ucred:
3267         mdt_exit_ucred(info);
3268 out_shrink:
3269         mdt_client_compatibility(info);
3270         rc2 = mdt_fix_reply(info);
3271         if (rc == 0)
3272                 rc = rc2;
3273         return rc;
3274 }
3275
3276 static int mdt_intent_layout(enum mdt_it_code opcode,
3277                              struct mdt_thread_info *info,
3278                              struct ldlm_lock **lockp,
3279                              __u64 flags)
3280 {
3281         struct layout_intent *layout;
3282         struct lu_fid *fid;
3283         struct mdt_object *obj = NULL;
3284         int rc = 0;
3285         ENTRY;
3286
3287         if (opcode != MDT_IT_LAYOUT) {
3288                 CERROR("%s: Unknown intent (%d)\n", mdt_obd_name(info->mti_mdt),
3289                         opcode);
3290                 RETURN(-EINVAL);
3291         }
3292
3293         fid = &info->mti_tmp_fid2;
3294         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3295
3296         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3297         if (IS_ERR(obj))
3298                 RETURN(PTR_ERR(obj));
3299
3300         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3301                 /* get the length of lsm */
3302                 rc = mdt_attr_get_eabuf_size(info, obj);
3303                 if (rc < 0)
3304                         RETURN(rc);
3305
3306                 if (rc > info->mti_mdt->mdt_max_mdsize)
3307                         info->mti_mdt->mdt_max_mdsize = rc;
3308         }
3309
3310         mdt_object_put(info->mti_env, obj);
3311
3312         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3313         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER, rc);
3314         rc = req_capsule_server_pack(info->mti_pill);
3315         if (rc != 0)
3316                 RETURN(-EINVAL);
3317
3318         layout = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3319         LASSERT(layout != NULL);
3320         if (layout->li_opc == LAYOUT_INTENT_ACCESS)
3321                 /* return to normal ldlm handling */
3322                 RETURN(0);
3323
3324         CERROR("%s: Unsupported layout intent (%d)\n",
3325                 mdt_obd_name(info->mti_mdt), layout->li_opc);
3326         RETURN(-EINVAL);
3327 }
3328
3329 static int mdt_intent_reint(enum mdt_it_code opcode,
3330                             struct mdt_thread_info *info,
3331                             struct ldlm_lock **lockp,
3332                             __u64 flags)
3333 {
3334         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3335         struct ldlm_reply      *rep = NULL;
3336         long                    opc;
3337         int                     rc;
3338
3339         static const struct req_format *intent_fmts[REINT_MAX] = {
3340                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3341                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3342         };
3343
3344         ENTRY;
3345
3346         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
3347         if (opc < 0)
3348                 RETURN(opc);
3349
3350         if (mdt_it_flavor[opcode].it_reint != opc) {
3351                 CERROR("Reint code %ld doesn't match intent: %d\n",
3352                        opc, opcode);
3353                 RETURN(err_serious(-EPROTO));
3354         }
3355
3356         /* Get lock from request for possible resent case. */
3357         mdt_intent_fixup_resent(info, *lockp, NULL, lhc, opcode);
3358
3359         rc = mdt_reint_internal(info, lhc, opc);
3360
3361         /* Check whether the reply has been packed successfully. */
3362         if (mdt_info_req(info)->rq_repmsg != NULL)
3363                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3364         if (rep == NULL)
3365                 RETURN(err_serious(-EFAULT));
3366
3367         /* MDC expects this in any case */
3368         if (rc != 0)
3369                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3370
3371         /* the open lock or the lock for cross-ref object should be
3372          * returned to the client */
3373         if (rc == -EREMOTE || mdt_get_disposition(rep, DISP_OPEN_LOCK)) {
3374                 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
3375                 rep->lock_policy_res2 = 0;
3376                 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3377                 RETURN(rc);
3378         }
3379
3380         rep->lock_policy_res2 = clear_serious(rc);
3381
3382         if (rep->lock_policy_res2 == -ENOENT &&
3383             mdt_get_disposition(rep, DISP_LOOKUP_NEG))
3384                 rep->lock_policy_res2 = 0;
3385
3386         if (rc == -ENOTCONN || rc == -ENODEV ||
3387             rc == -EOVERFLOW) { /**< if VBR failure then return error */
3388                 /*
3389                  * If it is the disconnect error (ENODEV & ENOCONN), the error
3390                  * will be returned by rq_status, and client at ptlrpc layer
3391                  * will detect this, then disconnect, reconnect the import
3392                  * immediately, instead of impacting the following the rpc.
3393                  */
3394                 lhc->mlh_reg_lh.cookie = 0ull;
3395                 RETURN(rc);
3396         } else {
3397                 /*
3398                  * For other cases, the error will be returned by intent.
3399                  * and client will retrieve the result from intent.
3400                  */
3401                  /*
3402                   * FIXME: when open lock is finished, that should be
3403                   * checked here.
3404                   */
3405                 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3406                         LASSERTF(rc == 0, "Error occurred but lock handle "
3407                                  "is still in use, rc = %d\n", rc);
3408                         rep->lock_policy_res2 = 0;
3409                         rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3410                         RETURN(rc);
3411                 } else {
3412                         lhc->mlh_reg_lh.cookie = 0ull;
3413                         RETURN(ELDLM_LOCK_ABORTED);
3414                 }
3415         }
3416 }
3417
3418 static int mdt_intent_code(long itcode)
3419 {
3420         int rc;
3421
3422         switch(itcode) {
3423         case IT_OPEN:
3424                 rc = MDT_IT_OPEN;
3425                 break;
3426         case IT_OPEN|IT_CREAT:
3427                 rc = MDT_IT_OCREAT;
3428                 break;
3429         case IT_CREAT:
3430                 rc = MDT_IT_CREATE;
3431                 break;
3432         case IT_READDIR:
3433                 rc = MDT_IT_READDIR;
3434                 break;
3435         case IT_GETATTR:
3436                 rc = MDT_IT_GETATTR;
3437                 break;
3438         case IT_LOOKUP:
3439                 rc = MDT_IT_LOOKUP;
3440                 break;
3441         case IT_UNLINK:
3442                 rc = MDT_IT_UNLINK;
3443                 break;
3444         case IT_TRUNC:
3445                 rc = MDT_IT_TRUNC;
3446                 break;
3447         case IT_GETXATTR:
3448                 rc = MDT_IT_GETXATTR;
3449                 break;
3450         case IT_LAYOUT:
3451                 rc = MDT_IT_LAYOUT;
3452                 break;
3453         case IT_QUOTA_DQACQ:
3454         case IT_QUOTA_CONN:
3455                 rc = MDT_IT_QUOTA;
3456                 break;
3457         default:
3458                 CERROR("Unknown intent opcode: %ld\n", itcode);
3459                 rc = -EINVAL;
3460                 break;
3461         }
3462         return rc;
3463 }
3464
3465 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
3466                           struct ldlm_lock **lockp, __u64 flags)
3467 {
3468         struct req_capsule   *pill;
3469         struct mdt_it_flavor *flv;
3470         int opc;
3471         int rc;
3472         ENTRY;
3473
3474         opc = mdt_intent_code(itopc);
3475         if (opc < 0)
3476                 RETURN(-EINVAL);
3477
3478         pill = info->mti_pill;
3479
3480         if (opc == MDT_IT_QUOTA) {
3481                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
3482
3483                 if (qmt == NULL)
3484                         RETURN(-EOPNOTSUPP);
3485
3486                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
3487                 /* pass the request to quota master */
3488                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
3489                                                  mdt_info_req(info), lockp,
3490                                                  flags);
3491                 RETURN(rc);
3492         }
3493
3494         flv  = &mdt_it_flavor[opc];
3495         if (flv->it_fmt != NULL)
3496                 req_capsule_extend(pill, flv->it_fmt);
3497
3498         rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3499         if (rc == 0) {
3500                 struct ptlrpc_request *req = mdt_info_req(info);
3501                 if (flv->it_flags & MUTABOR &&
3502                     exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
3503                         RETURN(-EROFS);
3504         }
3505         if (rc == 0 && flv->it_act != NULL) {
3506                 struct ldlm_reply *rep;
3507
3508                 /* execute policy */
3509                 rc = flv->it_act(opc, info, lockp, flags);
3510
3511                 /* Check whether the reply has been packed successfully. */
3512                 if (mdt_info_req(info)->rq_repmsg != NULL) {
3513                         rep = req_capsule_server_get(info->mti_pill,
3514                                                      &RMF_DLM_REP);
3515                         rep->lock_policy_res2 =
3516                                 ptlrpc_status_hton(rep->lock_policy_res2);
3517                 }
3518         } else {
3519                 rc = -EPROTO;
3520         }
3521         RETURN(rc);
3522 }
3523
3524 static int mdt_intent_policy(struct ldlm_namespace *ns,
3525                              struct ldlm_lock **lockp, void *req_cookie,
3526                              ldlm_mode_t mode, __u64 flags, void *data)
3527 {
3528         struct tgt_session_info *tsi;
3529         struct mdt_thread_info  *info;
3530         struct ptlrpc_request   *req  =  req_cookie;
3531         struct ldlm_intent      *it;
3532         struct req_capsule      *pill;
3533         int rc;
3534
3535         ENTRY;
3536
3537         LASSERT(req != NULL);
3538
3539         tsi = tgt_ses_info(req->rq_svc_thread->t_env);
3540
3541         info = tsi2mdt_info(tsi);
3542         LASSERT(info != NULL);
3543         pill = info->mti_pill;
3544         LASSERT(pill->rc_req == req);
3545
3546         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3547                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
3548                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3549                 if (it != NULL) {
3550                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
3551                         if (rc == 0)
3552                                 rc = ELDLM_OK;
3553
3554                         /* Lock without inodebits makes no sense and will oops
3555                          * later in ldlm. Let's check it now to see if we have
3556                          * ibits corrupted somewhere in mdt_intent_opc().
3557                          * The case for client miss to set ibits has been
3558                          * processed by others. */
3559                         LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
3560                                         lr_type == LDLM_IBITS,
3561                                      info->mti_dlm_req->lock_desc.\
3562                                         l_policy_data.l_inodebits.bits != 0));
3563                 } else
3564                         rc = err_serious(-EFAULT);
3565         } else {
3566                 /* No intent was provided */
3567                 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3568                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
3569                 rc = req_capsule_server_pack(pill);
3570                 if (rc)
3571                         rc = err_serious(rc);
3572         }
3573         mdt_thread_info_fini(info);
3574         RETURN(rc);
3575 }
3576
3577 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
3578 {
3579         struct seq_server_site  *ss = mdt_seq_site(mdt);
3580
3581         if (ss->ss_node_id == 0)
3582                 return;
3583
3584         if (ss->ss_client_seq != NULL) {
3585                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3586                 ss->ss_client_seq->lcs_exp = NULL;
3587         }
3588
3589         if (ss->ss_server_fld != NULL) {
3590                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
3591                 ss->ss_server_fld->lsf_control_exp = NULL;
3592         }
3593 }
3594
3595 static void mdt_seq_fini_cli(struct mdt_device *mdt)
3596 {
3597         struct seq_server_site *ss = mdt_seq_site(mdt);
3598
3599         if (ss == NULL)
3600                 return;
3601
3602         if (ss->ss_server_seq != NULL)
3603                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
3604 }
3605
3606 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
3607 {
3608         mdt_seq_fini_cli(mdt);
3609         mdt_deregister_seq_exp(mdt);
3610
3611         return seq_site_fini(env, mdt_seq_site(mdt));
3612 }
3613
3614 /**
3615  * It will retrieve its FLDB entries from MDT0, and it only happens
3616  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
3617  * and it needs to refresh FLDB from the MDT0.
3618  **/
3619 static int mdt_register_lwp_callback(void *data)
3620 {
3621         struct lu_env           env;
3622         struct mdt_device       *mdt = data;
3623         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
3624         int                     rc;
3625         ENTRY;
3626
3627         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
3628
3629         if (!likely(fld->lsf_new))
3630                 RETURN(0);
3631
3632         rc = lu_env_init(&env, LCT_MD_THREAD);
3633         if (rc) {
3634                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
3635                 RETURN(rc);
3636         }
3637
3638         rc = fld_update_from_controller(&env, fld);
3639         if (rc != 0) {
3640                 CERROR("%s: cannot update controller: rc = %d\n",
3641                        mdt_obd_name(mdt), rc);
3642                 GOTO(out, rc);
3643         }
3644 out:
3645         lu_env_fini(&env);
3646         RETURN(rc);
3647 }
3648
3649 static int mdt_register_seq_exp(struct mdt_device *mdt)
3650 {
3651         struct seq_server_site  *ss = mdt_seq_site(mdt);
3652         char                    *lwp_name = NULL;
3653         int                     rc;
3654
3655         if (ss->ss_node_id == 0)
3656                 return 0;
3657
3658         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
3659         if (lwp_name == NULL)
3660                 GOTO(out_free, rc = -ENOMEM);
3661
3662         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
3663         if (rc != 0)
3664                 GOTO(out_free, rc);
3665
3666         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
3667                                       NULL, NULL);
3668         if (rc != 0)
3669                 GOTO(out_free, rc);
3670
3671         rc = lustre_register_lwp_item(lwp_name,
3672                                       &ss->ss_server_fld->lsf_control_exp,
3673                                       mdt_register_lwp_callback, mdt);
3674         if (rc != 0) {
3675                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3676                 ss->ss_client_seq->lcs_exp = NULL;
3677                 GOTO(out_free, rc);
3678         }
3679 out_free:
3680         if (lwp_name != NULL)
3681                 OBD_FREE(lwp_name, MAX_OBD_NAME);
3682
3683         return rc;
3684 }
3685
3686 /*
3687  * Init client sequence manager which is used by local MDS to talk to sequence
3688  * controller on remote node.
3689  */
3690 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
3691 {
3692         struct seq_server_site  *ss = mdt_seq_site(mdt);
3693         int                     rc;
3694         char                    *prefix;
3695         ENTRY;
3696
3697         /* check if this is adding the first MDC and controller is not yet
3698          * initialized. */
3699         OBD_ALLOC_PTR(ss->ss_client_seq);
3700         if (ss->ss_client_seq == NULL)
3701                 RETURN(-ENOMEM);
3702
3703         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3704         if (prefix == NULL) {
3705                 OBD_FREE_PTR(ss->ss_client_seq);
3706                 ss->ss_client_seq = NULL;
3707                 RETURN(-ENOMEM);
3708         }
3709
3710         /* Note: seq_client_fini will be called in seq_site_fini */
3711         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
3712         rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
3713                              prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
3714                                                             NULL);
3715         OBD_FREE(prefix, MAX_OBD_NAME + 5);
3716         if (rc != 0) {
3717                 OBD_FREE_PTR(ss->ss_client_seq);
3718                 ss->ss_client_seq = NULL;
3719                 RETURN(rc);
3720         }
3721
3722         rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
3723
3724         RETURN(rc);
3725 }
3726
3727 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
3728 {
3729         struct seq_server_site  *ss;
3730         int                     rc;
3731         ENTRY;
3732
3733         ss = mdt_seq_site(mdt);
3734         /* init sequence controller server(MDT0) */
3735         if (ss->ss_node_id == 0) {
3736                 OBD_ALLOC_PTR(ss->ss_control_seq);
3737                 if (ss->ss_control_seq == NULL)
3738                         RETURN(-ENOMEM);
3739
3740                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
3741                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
3742                                      ss);
3743                 if (rc)
3744                         GOTO(out_seq_fini, rc);
3745         }
3746
3747         /* Init normal sequence server */
3748         OBD_ALLOC_PTR(ss->ss_server_seq);
3749         if (ss->ss_server_seq == NULL)
3750                 GOTO(out_seq_fini, rc = -ENOMEM);
3751
3752         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
3753                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
3754         if (rc)
3755                 GOTO(out_seq_fini, rc);
3756
3757         /* init seq client for seq server to talk to seq controller(MDT0) */
3758         rc = mdt_seq_init_cli(env, mdt);
3759         if (rc != 0)
3760                 GOTO(out_seq_fini, rc);
3761
3762         if (ss->ss_node_id != 0)
3763                 /* register controler export through lwp */
3764                 rc = mdt_register_seq_exp(mdt);
3765
3766         EXIT;
3767 out_seq_fini:
3768         if (rc)
3769                 mdt_seq_fini(env, mdt);
3770
3771         return rc;
3772 }
3773
3774 /*
3775  * FLD wrappers
3776  */
3777 static int mdt_fld_fini(const struct lu_env *env,
3778                         struct mdt_device *m)
3779 {
3780         struct seq_server_site *ss = mdt_seq_site(m);
3781         ENTRY;
3782
3783         if (ss && ss->ss_server_fld) {
3784                 fld_server_fini(env, ss->ss_server_fld);
3785                 OBD_FREE_PTR(ss->ss_server_fld);
3786                 ss->ss_server_fld = NULL;
3787         }
3788
3789         RETURN(0);
3790 }
3791
3792 static int mdt_fld_init(const struct lu_env *env,
3793                         const char *uuid,
3794                         struct mdt_device *m)
3795 {
3796         struct seq_server_site *ss;
3797         int rc;
3798         ENTRY;
3799
3800         ss = mdt_seq_site(m);
3801
3802         OBD_ALLOC_PTR(ss->ss_server_fld);
3803         if (ss->ss_server_fld == NULL)
3804                 RETURN(rc = -ENOMEM);
3805
3806         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
3807                              LU_SEQ_RANGE_MDT);
3808         if (rc) {
3809                 OBD_FREE_PTR(ss->ss_server_fld);
3810                 ss->ss_server_fld = NULL;
3811                 RETURN(rc);
3812         }
3813
3814         RETURN(0);
3815 }
3816
3817 static void mdt_stack_pre_fini(const struct lu_env *env,
3818                            struct mdt_device *m, struct lu_device *top)
3819 {
3820         struct lustre_cfg_bufs  *bufs;
3821         struct lustre_cfg       *lcfg;
3822         struct mdt_thread_info  *info;
3823         ENTRY;
3824
3825         LASSERT(top);
3826
3827         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3828         LASSERT(info != NULL);
3829
3830         bufs = &info->mti_u.bufs;
3831
3832         LASSERT(m->mdt_child_exp);
3833         LASSERT(m->mdt_child_exp->exp_obd);
3834
3835         /* process cleanup, pass mdt obd name to get obd umount flags */
3836         /* XXX: this is needed because all layers are referenced by
3837          * objects (some of them are pinned by osd, for example *
3838          * the proper solution should be a model where object used
3839          * by osd only doesn't have mdt/mdd slices -bzzz */
3840         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
3841         lustre_cfg_bufs_set_string(bufs, 1, NULL);
3842         lcfg = lustre_cfg_new(LCFG_PRE_CLEANUP, bufs);
3843         if (!lcfg) {
3844                 CERROR("%s:Cannot alloc lcfg!\n", mdt_obd_name(m));
3845                 return;
3846         }
3847         top->ld_ops->ldo_process_config(env, top, lcfg);
3848         lustre_cfg_free(lcfg);
3849         EXIT;
3850 }
3851
3852 static void mdt_stack_fini(const struct lu_env *env,
3853                            struct mdt_device *m, struct lu_device *top)
3854 {
3855         struct obd_device       *obd = mdt2obd_dev(m);
3856         struct lustre_cfg_bufs  *bufs;
3857         struct lustre_cfg       *lcfg;
3858         struct mdt_thread_info  *info;
3859         char                     flags[3] = "";
3860         ENTRY;
3861
3862         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3863         LASSERT(info != NULL);
3864
3865         lu_dev_del_linkage(top->ld_site, top);
3866
3867         lu_site_purge(env, top->ld_site, -1);
3868
3869         bufs = &info->mti_u.bufs;
3870         /* process cleanup, pass mdt obd name to get obd umount flags */
3871         /* another purpose is to let all layers to release their objects */
3872         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
3873         if (obd->obd_force)
3874                 strcat(flags, "F");
3875         if (obd->obd_fail)
3876                 strcat(flags, "A");
3877         lustre_cfg_bufs_set_string(bufs, 1, flags);
3878         lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
3879         if (!lcfg) {
3880                 CERROR("Cannot alloc lcfg!\n");
3881                 return;
3882         }
3883         LASSERT(top);
3884         top->ld_ops->ldo_process_config(env, top, lcfg);
3885         lustre_cfg_free(lcfg);
3886
3887         lu_site_purge(env, top->ld_site, -1);
3888
3889         m->mdt_child = NULL;
3890         m->mdt_bottom = NULL;
3891
3892         obd_disconnect(m->mdt_child_exp);
3893         m->mdt_child_exp = NULL;
3894
3895         obd_disconnect(m->mdt_bottom_exp);
3896         m->mdt_child_exp = NULL;
3897 }
3898
3899 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
3900                                const char *next, struct obd_export **exp)
3901 {
3902         struct obd_connect_data *data = NULL;
3903         struct obd_device       *obd;
3904         int                      rc;
3905         ENTRY;
3906
3907         OBD_ALLOC_PTR(data);
3908         if (data == NULL)
3909                 GOTO(out, rc = -ENOMEM);
3910
3911         obd = class_name2obd(next);
3912         if (obd == NULL) {
3913                 CERROR("%s: can't locate next device: %s\n",
3914                        mdt_obd_name(m), next);
3915                 GOTO(out, rc = -ENOTCONN);
3916         }
3917
3918         data->ocd_connect_flags = OBD_CONNECT_VERSION;
3919         data->ocd_version = LUSTRE_VERSION_CODE;
3920
3921         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
3922         if (rc) {
3923                 CERROR("%s: cannot connect to next dev %s (%d)\n",
3924                        mdt_obd_name(m), next, rc);
3925                 GOTO(out, rc);
3926         }
3927
3928 out:
3929         if (data)
3930                 OBD_FREE_PTR(data);
3931         RETURN(rc);
3932 }
3933
3934 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
3935                           struct lustre_cfg *cfg)
3936 {
3937         char                   *dev = lustre_cfg_string(cfg, 0);
3938         int                     rc, name_size, uuid_size;
3939         char                   *name, *uuid, *p;
3940         struct lustre_cfg_bufs *bufs;
3941         struct lustre_cfg      *lcfg;
3942         struct obd_device      *obd;
3943         struct lustre_profile  *lprof;
3944         struct lu_site         *site;
3945         ENTRY;
3946
3947         /* in 1.8 we had the only device in the stack - MDS.
3948          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
3949          * in 2.3 OSD is instantiated by obd_mount.c, so we need
3950          * to generate names and setup MDT, MDD. MDT will be using
3951          * generated name to connect to MDD. for MDD the next device
3952          * will be LOD with name taken from so called "profile" which
3953          * is generated by mount_option line
3954          *
3955          * 1.8 MGS generates config. commands like this:
3956          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
3957          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
3958          * 2.0 MGS generates config. commands like this:
3959          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
3960          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
3961          *                    3:lustre-MDT0000-mdtlov  4:f
3962          *
3963          * we generate MDD name from MDT one, just replacing T with D
3964          *
3965          * after all the preparations, the logical equivalent will be
3966          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
3967          *                    3:lustre-MDT0000-mdtlov  4:f
3968          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
3969          *                    3:lustre-MDD0000  4:f
3970          *
3971          *  notice we build the stack from down to top: MDD first, then MDT */
3972
3973         name_size = MAX_OBD_NAME;
3974         uuid_size = MAX_OBD_NAME;
3975
3976         OBD_ALLOC(name, name_size);
3977         OBD_ALLOC(uuid, uuid_size);
3978         if (name == NULL || uuid == NULL)
3979                 GOTO(cleanup_mem, rc = -ENOMEM);
3980
3981         OBD_ALLOC_PTR(bufs);
3982         if (!bufs)
3983                 GOTO(cleanup_mem, rc = -ENOMEM);
3984
3985         strcpy(name, dev);
3986         p = strstr(name, "-MDT");
3987         if (p == NULL)
3988                 GOTO(free_bufs, rc = -ENOMEM);
3989         p[3] = 'D';
3990
3991         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
3992
3993         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
3994         if (lprof == NULL || lprof->lp_dt == NULL) {
3995                 CERROR("can't find the profile: %s\n",
3996                        lustre_cfg_string(cfg, 0));
3997                 GOTO(free_bufs, rc = -EINVAL);
3998         }
3999
4000         lustre_cfg_bufs_reset(bufs, name);
4001         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
4002         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4003         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4004
4005         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4006         if (!lcfg)
4007                 GOTO(free_bufs, rc = -ENOMEM);
4008
4009         rc = class_attach(lcfg);
4010         if (rc)
4011                 GOTO(lcfg_cleanup, rc);
4012
4013         obd = class_name2obd(name);
4014         if (!obd) {
4015                 CERROR("Can not find obd %s (%s in config)\n",
4016                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
4017                 GOTO(class_detach, rc = -EINVAL);
4018         }
4019
4020         lustre_cfg_free(lcfg);
4021
4022         lustre_cfg_bufs_reset(bufs, name);
4023         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4024         lustre_cfg_bufs_set_string(bufs, 2, dev);
4025         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4026
4027         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4028
4029         rc = class_setup(obd, lcfg);
4030         if (rc)
4031                 GOTO(class_detach, rc);
4032
4033         /* connect to MDD we just setup */
4034         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
4035         if (rc)
4036                 GOTO(class_detach, rc);
4037
4038         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
4039         LASSERT(site);
4040         LASSERT(mdt_lu_site(mdt) == NULL);
4041         mdt->mdt_lu_dev.ld_site = site;
4042         site->ls_top_dev = &mdt->mdt_lu_dev;
4043         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
4044
4045         /* now connect to bottom OSD */
4046         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
4047         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
4048         if (rc)
4049                 GOTO(class_detach, rc);
4050         mdt->mdt_bottom =
4051                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
4052
4053         rc = lu_env_refill((struct lu_env *)env);
4054         if (rc != 0)
4055                 CERROR("Failure to refill session: '%d'\n", rc);
4056
4057         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
4058
4059         EXIT;
4060 class_detach:
4061         if (rc)
4062                 class_detach(obd, lcfg);
4063 lcfg_cleanup:
4064         lustre_cfg_free(lcfg);
4065 free_bufs:
4066         OBD_FREE_PTR(bufs);
4067 cleanup_mem:
4068         if (name)
4069                 OBD_FREE(name, name_size);
4070         if (uuid)
4071                 OBD_FREE(uuid, uuid_size);
4072         RETURN(rc);
4073 }
4074
4075 /* setup quota master target on MDT0 */
4076 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
4077                           struct lustre_cfg *cfg)
4078 {
4079         struct obd_device       *obd;
4080         char                    *dev = lustre_cfg_string(cfg, 0);
4081         char                    *qmtname, *uuid, *p;
4082         struct lustre_cfg_bufs  *bufs;
4083         struct lustre_cfg       *lcfg;
4084         struct lustre_profile   *lprof;
4085         struct obd_connect_data *data;
4086         int                      rc;
4087         ENTRY;
4088
4089         LASSERT(mdt->mdt_qmt_exp == NULL);
4090         LASSERT(mdt->mdt_qmt_dev == NULL);
4091
4092         /* quota master is on MDT0 only for now */
4093         if (mdt->mdt_seq_site.ss_node_id != 0)
4094                 RETURN(0);
4095
4096         /* MGS generates config commands which look as follows:
4097          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4098          *                    3:lustre-MDT0000-mdtlov  4:f
4099          *
4100          * We generate the QMT name from the MDT one, just replacing MD with QM
4101          * after all the preparations, the logical equivalent will be:
4102          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
4103          *                    3:lustre-MDT0000-osd  4:f */
4104         OBD_ALLOC(qmtname, MAX_OBD_NAME);
4105         OBD_ALLOC(uuid, UUID_MAX);
4106         OBD_ALLOC_PTR(bufs);
4107         OBD_ALLOC_PTR(data);
4108         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
4109                 GOTO(cleanup_mem, rc = -ENOMEM);
4110
4111         strcpy(qmtname, dev);
4112         p = strstr(qmtname, "-MDT");
4113         if (p == NULL)
4114                 GOTO(cleanup_mem, rc = -ENOMEM);
4115         /* replace MD with QM */
4116         p[1] = 'Q';
4117         p[2] = 'M';
4118
4119         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
4120
4121         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4122         if (lprof == NULL || lprof->lp_dt == NULL) {
4123                 CERROR("can't find profile for %s\n",
4124                        lustre_cfg_string(cfg, 0));
4125                 GOTO(cleanup_mem, rc = -EINVAL);
4126         }
4127
4128         lustre_cfg_bufs_reset(bufs, qmtname);
4129         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
4130         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4131         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4132
4133         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4134         if (!lcfg)
4135                 GOTO(cleanup_mem, rc = -ENOMEM);
4136
4137         rc = class_attach(lcfg);
4138         if (rc)
4139                 GOTO(lcfg_cleanup, rc);
4140
4141         obd = class_name2obd(qmtname);
4142         if (!obd) {
4143                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
4144                        lustre_cfg_string(cfg, 0));
4145                 GOTO(class_detach, rc = -EINVAL);
4146         }
4147
4148         lustre_cfg_free(lcfg);
4149
4150         lustre_cfg_bufs_reset(bufs, qmtname);
4151         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4152         lustre_cfg_bufs_set_string(bufs, 2, dev);
4153
4154         /* for quota, the next device should be the OSD device */
4155         lustre_cfg_bufs_set_string(bufs, 3,
4156                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
4157
4158         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4159
4160         rc = class_setup(obd, lcfg);
4161         if (rc)
4162                 GOTO(class_detach, rc);
4163
4164         mdt->mdt_qmt_dev = obd->obd_lu_dev;
4165
4166         /* configure local quota objects */
4167         rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
4168                                                    &mdt->mdt_lu_dev,
4169                                                    mdt->mdt_qmt_dev);
4170         if (rc)
4171                 GOTO(class_cleanup, rc);
4172
4173         /* connect to quota master target */
4174         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4175         data->ocd_version = LUSTRE_VERSION_CODE;
4176         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
4177                          data, NULL);
4178         if (rc) {
4179                 CERROR("cannot connect to quota master device %s (%d)\n",
4180                        qmtname, rc);
4181                 GOTO(class_cleanup, rc);
4182         }
4183
4184         EXIT;
4185 class_cleanup:
4186         if (rc) {
4187                 class_manual_cleanup(obd);
4188                 mdt->mdt_qmt_dev = NULL;
4189         }
4190 class_detach:
4191         if (rc)
4192                 class_detach(obd, lcfg);
4193 lcfg_cleanup:
4194         lustre_cfg_free(lcfg);
4195 cleanup_mem:
4196         if (bufs)
4197                 OBD_FREE_PTR(bufs);
4198         if (qmtname)
4199                 OBD_FREE(qmtname, MAX_OBD_NAME);
4200         if (uuid)
4201                 OBD_FREE(uuid, UUID_MAX);
4202         if (data)
4203                 OBD_FREE_PTR(data);
4204         return rc;
4205 }
4206
4207 /* Shutdown quota master target associated with mdt */
4208 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
4209 {
4210         ENTRY;
4211
4212         if (mdt->mdt_qmt_exp == NULL)
4213                 RETURN_EXIT;
4214         LASSERT(mdt->mdt_qmt_dev != NULL);
4215
4216         /* the qmt automatically shuts down when the mdt disconnects */
4217         obd_disconnect(mdt->mdt_qmt_exp);
4218         mdt->mdt_qmt_exp = NULL;
4219         mdt->mdt_qmt_dev = NULL;
4220         EXIT;
4221 }
4222
4223 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
4224  * for now. This will be removed along with converting rest of MDT code
4225  * to use tgt_session_info */
4226 int mdt_tgt_getxattr(struct tgt_session_info *tsi)
4227 {
4228         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
4229         int                      rc;
4230
4231         rc = mdt_getxattr(info);
4232
4233         mdt_thread_info_fini(info);
4234         return rc;
4235 }
4236
4237 static struct tgt_handler mdt_tgt_handlers[] = {
4238 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4239                 0,                      MDS_CONNECT,    mdt_tgt_connect,
4240                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
4241 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4242                 0,                      MDS_DISCONNECT, tgt_disconnect,
4243                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
4244 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4245                 HABEO_REFERO,           MDS_SET_INFO,   mdt_set_info,
4246                 &RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION),
4247 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
4248 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_GETSTATUS,  mdt_getstatus),
4249 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETATTR,    mdt_getattr),
4250 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_GETATTR_NAME,
4251                                                         mdt_getattr_name),
4252 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETXATTR,   mdt_tgt_getxattr),
4253 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_STATFS,     mdt_statfs),
4254 TGT_MDT_HDL(0           | MUTABOR,      MDS_REINT,      mdt_reint),
4255 TGT_MDT_HDL(HABEO_CORPUS,               MDS_CLOSE,      mdt_close),
4256 TGT_MDT_HDL(HABEO_CORPUS,               MDS_DONE_WRITING,
4257                                                         mdt_done_writing),
4258 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_READPAGE,   mdt_readpage),
4259 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_SYNC,       mdt_sync),
4260 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_IS_SUBDIR,  mdt_is_subdir),
4261 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
4262 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_PROGRESS,
4263                                                         mdt_hsm_progress),
4264 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_REGISTER,
4265                                                         mdt_hsm_ct_register),
4266 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_UNREGISTER,
4267                                                         mdt_hsm_ct_unregister),
4268 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_STATE_GET,
4269                                                         mdt_hsm_state_get),
4270 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_STATE_SET,
4271                                                         mdt_hsm_state_set),
4272 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_ACTION, mdt_hsm_action),
4273 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_REQUEST,
4274                                                         mdt_hsm_request),
4275 TGT_MDT_HDL(HABEO_CLAVIS | HABEO_CORPUS | HABEO_REFERO | MUTABOR,
4276             MDS_SWAP_LAYOUTS,
4277             mdt_swap_layouts),
4278 };
4279
4280 static struct tgt_handler mdt_sec_ctx_ops[] = {
4281 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
4282 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
4283 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
4284 };
4285
4286 static struct tgt_handler mdt_quota_ops[] = {
4287 TGT_QUOTA_HDL(HABEO_REFERO,             QUOTA_DQACQ,      mdt_quota_dqacq),
4288 };
4289
4290 static struct tgt_opc_slice mdt_common_slice[] = {
4291         {
4292                 .tos_opc_start  = MDS_FIRST_OPC,
4293                 .tos_opc_end    = MDS_LAST_OPC,
4294                 .tos_hs         = mdt_tgt_handlers
4295         },
4296         {
4297                 .tos_opc_start  = OBD_FIRST_OPC,
4298                 .tos_opc_end    = OBD_LAST_OPC,
4299                 .tos_hs         = tgt_obd_handlers
4300         },
4301         {
4302                 .tos_opc_start  = LDLM_FIRST_OPC,
4303                 .tos_opc_end    = LDLM_LAST_OPC,
4304                 .tos_hs         = tgt_dlm_handlers
4305         },
4306         {
4307                 .tos_opc_start  = SEC_FIRST_OPC,
4308                 .tos_opc_end    = SEC_LAST_OPC,
4309                 .tos_hs         = mdt_sec_ctx_ops
4310         },
4311         {
4312                 .tos_opc_start  = OUT_UPDATE_FIRST_OPC,
4313                 .tos_opc_end    = OUT_UPDATE_LAST_OPC,
4314                 .tos_hs         = tgt_out_handlers
4315         },
4316         {
4317                 .tos_opc_start  = FLD_FIRST_OPC,
4318                 .tos_opc_end    = FLD_LAST_OPC,
4319                 .tos_hs         = fld_handlers
4320         },
4321         {
4322                 .tos_opc_start  = SEQ_FIRST_OPC,
4323                 .tos_opc_end    = SEQ_LAST_OPC,
4324                 .tos_hs         = seq_handlers
4325         },
4326         {
4327                 .tos_opc_start  = QUOTA_DQACQ,
4328                 .tos_opc_end    = QUOTA_LAST_OPC,
4329                 .tos_hs         = mdt_quota_ops
4330         },
4331         {
4332                 .tos_opc_start  = LLOG_FIRST_OPC,
4333                 .tos_opc_end    = LLOG_LAST_OPC,
4334                 .tos_hs         = tgt_llog_handlers
4335         },
4336         {
4337                 .tos_opc_start  = LFSCK_FIRST_OPC,
4338                 .tos_opc_end    = LFSCK_LAST_OPC,
4339                 .tos_hs         = tgt_lfsck_handlers
4340         },
4341
4342         {
4343                 .tos_hs         = NULL
4344         }
4345 };
4346
4347 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4348 {
4349         struct md_device        *next = m->mdt_child;
4350         struct lu_device        *d    = &m->mdt_lu_dev;
4351         struct obd_device       *obd  = mdt2obd_dev(m);
4352         struct lfsck_stop        stop;
4353         ENTRY;
4354
4355         stop.ls_status = LS_PAUSED;
4356         stop.ls_flags = 0;
4357         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
4358
4359         target_recovery_fini(obd);
4360         ping_evictor_stop();
4361         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
4362
4363         if (m->mdt_opts.mo_coordinator)
4364                 mdt_hsm_cdt_stop(m);
4365
4366         mdt_hsm_cdt_fini(m);
4367
4368         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
4369         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4370         obd_exports_barrier(obd);
4371         obd_zombie_barrier();
4372
4373         mdt_procfs_fini(m);
4374
4375         tgt_fini(env, &m->mdt_lut);
4376         mdt_fs_cleanup(env, m);
4377         upcall_cache_cleanup(m->mdt_identity_cache);
4378         m->mdt_identity_cache = NULL;
4379
4380         if (m->mdt_namespace != NULL) {
4381                 ldlm_namespace_free(m->mdt_namespace, NULL,
4382                                     d->ld_obd->obd_force);
4383                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4384         }
4385
4386         mdt_quota_fini(env, m);
4387
4388         cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
4389
4390         mdt_seq_fini(env, m);
4391         mdt_fld_fini(env, m);
4392
4393         next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL);
4394         cfs_timer_disarm(&m->mdt_ck_timer);
4395         mdt_ck_thread_stop(m);
4396
4397         /*
4398          * Finish the stack
4399          */
4400         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4401
4402         LASSERT(atomic_read(&d->ld_ref) == 0);
4403
4404         server_put_mount(mdt_obd_name(m), true);
4405
4406         EXIT;
4407 }
4408
4409 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4410
4411 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4412                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
4413 {
4414         struct mdt_thread_info    *info;
4415         struct obd_device         *obd;
4416         const char                *dev = lustre_cfg_string(cfg, 0);
4417         const char                *num = lustre_cfg_string(cfg, 2);
4418         struct lustre_mount_info  *lmi = NULL;
4419         struct lustre_sb_info     *lsi;
4420         struct lu_site            *s;
4421         struct seq_server_site    *ss_site;
4422         const char                *identity_upcall = "NONE";
4423         struct md_device          *next;
4424         int                        rc;
4425         long                       node_id;
4426         mntopt_t                   mntopts;
4427         ENTRY;
4428
4429         lu_device_init(&m->mdt_lu_dev, ldt);
4430         /*
4431          * Environment (env) might be missing mdt_thread_key values at that
4432          * point, if device is allocated when mdt_thread_key is in QUIESCENT
4433          * mode.
4434          *
4435          * Usually device allocation path doesn't use module key values, but
4436          * mdt has to do a lot of work here, so allocate key value.
4437          */
4438         rc = lu_env_refill((struct lu_env *)env);
4439         if (rc != 0)
4440                 RETURN(rc);
4441
4442         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4443         LASSERT(info != NULL);
4444
4445         obd = class_name2obd(dev);
4446         LASSERT(obd != NULL);
4447
4448         m->mdt_max_mdsize = MAX_MD_SIZE; /* 4 stripes */
4449
4450         m->mdt_som_conf = 0;
4451
4452         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4453
4454         /* default is coordinator off, it is started through conf_param
4455          * or /proc */
4456         m->mdt_opts.mo_coordinator = 0;
4457
4458         lmi = server_get_mount(dev);
4459         if (lmi == NULL) {
4460                 CERROR("Cannot get mount info for %s!\n", dev);
4461                 RETURN(-EFAULT);
4462         } else {
4463                 lsi = s2lsi(lmi->lmi_sb);
4464                 /* CMD is supported only in IAM mode */
4465                 LASSERT(num);
4466                 node_id = simple_strtol(num, NULL, 10);
4467                 obd->u.obt.obt_magic = OBT_MAGIC;
4468         }
4469
4470         spin_lock_init(&m->mdt_ioepoch_lock);
4471         m->mdt_capa_timeout = CAPA_TIMEOUT;
4472         m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1;
4473         m->mdt_ck_timeout = CAPA_KEY_TIMEOUT;
4474         m->mdt_squash.rsi_uid = 0;
4475         m->mdt_squash.rsi_gid = 0;
4476         INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
4477         init_rwsem(&m->mdt_squash.rsi_sem);
4478         spin_lock_init(&m->mdt_osfs_lock);
4479         m->mdt_osfs_age = cfs_time_shift_64(-1000);
4480         m->mdt_enable_remote_dir = 0;
4481         m->mdt_enable_remote_dir_gid = 0;
4482
4483         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
4484         m->mdt_lu_dev.ld_obd = obd;
4485         /* Set this lu_device to obd for error handling purposes. */
4486         obd->obd_lu_dev = &m->mdt_lu_dev;
4487
4488         /* init the stack */
4489         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
4490         if (rc) {
4491                 CERROR("%s: Can't init device stack, rc %d\n",
4492                        mdt_obd_name(m), rc);
4493                 GOTO(err_lmi, rc);
4494         }
4495
4496         s = mdt_lu_site(m);
4497         ss_site = mdt_seq_site(m);
4498         s->ld_seq_site = ss_site;
4499         ss_site->ss_lu = s;
4500
4501         /* set server index */
4502         ss_site->ss_node_id = node_id;
4503
4504         /* failover is the default
4505          * FIXME: we do not failout mds0/mgs, which may cause some problems.
4506          * assumed whose ss_node_id == 0 XXX
4507          * */
4508         obd->obd_replayable = 1;
4509         /* No connection accepted until configurations will finish */
4510         obd->obd_no_conn = 1;
4511
4512         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4513                 char *str = lustre_cfg_string(cfg, 4);
4514                 if (strchr(str, 'n')) {
4515                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
4516                         obd->obd_replayable = 0;
4517                 }
4518         }
4519
4520         rc = mdt_fld_init(env, mdt_obd_name(m), m);
4521         if (rc)
4522                 GOTO(err_fini_stack, rc);
4523
4524         rc = mdt_seq_init(env, m);
4525         if (rc)
4526                 GOTO(err_fini_fld, rc);
4527
4528         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
4529                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
4530         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4531                                               LDLM_NAMESPACE_SERVER,
4532                                               LDLM_NAMESPACE_GREEDY,
4533                                               LDLM_NS_TYPE_MDT);
4534         if (m->mdt_namespace == NULL)
4535                 GOTO(err_fini_seq, rc = -ENOMEM);
4536
4537         m->mdt_namespace->ns_lvbp = m;
4538         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
4539
4540         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4541         /* set obd_namespace for compatibility with old code */
4542         obd->obd_namespace = m->mdt_namespace;
4543
4544         cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m);
4545
4546         rc = mdt_hsm_cdt_init(m);
4547         if (rc != 0) {
4548                 CERROR("%s: error initializing coordinator, rc %d\n",
4549                        mdt_obd_name(m), rc);
4550                 GOTO(err_free_ns, rc);
4551         }
4552
4553         rc = mdt_ck_thread_start(m);
4554         if (rc)
4555                 GOTO(err_free_hsm, rc);
4556
4557         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
4558                       OBD_FAIL_MDS_ALL_REQUEST_NET,
4559                       OBD_FAIL_MDS_ALL_REPLY_NET);
4560         if (rc)
4561                 GOTO(err_capa, rc);
4562
4563         rc = mdt_fs_setup(env, m, obd, lsi);
4564         if (rc)
4565                 GOTO(err_tgt, rc);
4566
4567         tgt_adapt_sptlrpc_conf(&m->mdt_lut, 1);
4568
4569         next = m->mdt_child;
4570         rc = next->md_ops->mdo_iocontrol(env, next, OBD_IOC_GET_MNTOPT, 0,
4571                                          &mntopts);
4572         if (rc)
4573                 GOTO(err_fs_cleanup, rc);
4574
4575         if (mntopts & MNTOPT_USERXATTR)
4576                 m->mdt_opts.mo_user_xattr = 1;
4577         else
4578                 m->mdt_opts.mo_user_xattr = 0;
4579
4580         rc = next->md_ops->mdo_maxeasize_get(env, next, &m->mdt_max_ea_size);
4581         if (rc)
4582                 GOTO(err_fs_cleanup, rc);
4583
4584         if (mntopts & MNTOPT_ACL)
4585                 m->mdt_opts.mo_acl = 1;
4586         else
4587                 m->mdt_opts.mo_acl = 0;
4588
4589         /* XXX: to support suppgid for ACL, we enable identity_upcall
4590          * by default, otherwise, maybe got unexpected -EACCESS. */
4591         if (m->mdt_opts.mo_acl)
4592                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4593
4594         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
4595                                                 identity_upcall,
4596                                                 &mdt_identity_upcall_cache_ops);
4597         if (IS_ERR(m->mdt_identity_cache)) {
4598                 rc = PTR_ERR(m->mdt_identity_cache);
4599                 m->mdt_identity_cache = NULL;
4600                 GOTO(err_fs_cleanup, rc);
4601         }
4602
4603         rc = mdt_procfs_init(m, dev);
4604         if (rc) {
4605                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4606                 GOTO(err_recovery, rc);
4607         }
4608
4609         rc = mdt_quota_init(env, m, cfg);
4610         if (rc)
4611                 GOTO(err_procfs, rc);
4612
4613         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
4614         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
4615                            "mdt_ldlm_client", m->mdt_ldlm_client);
4616
4617         ping_evictor_start();
4618
4619         /* recovery will be started upon mdt_prepare()
4620          * when the whole stack is complete and ready
4621          * to serve the requests */
4622
4623         mdt_init_capa_ctxt(env, m);
4624
4625         /* Reduce the initial timeout on an MDS because it doesn't need such
4626          * a long timeout as an OST does. Adaptive timeouts will adjust this
4627          * value appropriately. */
4628         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4629                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4630
4631         RETURN(0);
4632 err_procfs:
4633         mdt_procfs_fini(m);
4634 err_recovery:
4635         target_recovery_fini(obd);
4636         upcall_cache_cleanup(m->mdt_identity_cache);
4637         m->mdt_identity_cache = NULL;
4638 err_fs_cleanup:
4639         mdt_fs_cleanup(env, m);
4640 err_tgt:
4641         tgt_fini(env, &m->mdt_lut);
4642 err_capa:
4643         cfs_timer_disarm(&m->mdt_ck_timer);
4644         mdt_ck_thread_stop(m);
4645 err_free_hsm:
4646         mdt_hsm_cdt_fini(m);
4647 err_free_ns:
4648         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4649         obd->obd_namespace = m->mdt_namespace = NULL;
4650 err_fini_seq:
4651         mdt_seq_fini(env, m);
4652 err_fini_fld:
4653         mdt_fld_fini(env, m);
4654 err_fini_stack:
4655         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4656 err_lmi:
4657         if (lmi)
4658                 server_put_mount(dev, true);
4659         return(rc);
4660 }
4661
4662 /* For interoperability, the left element is old parameter, the right one
4663  * is the new version of the parameter, if some parameter is deprecated,
4664  * the new version should be set as NULL. */
4665 static struct cfg_interop_param mdt_interop_param[] = {
4666         { "mdt.group_upcall",   NULL },
4667         { "mdt.quota_type",     NULL },
4668         { "mdd.quota_type",     NULL },
4669         { "mdt.rootsquash",     "mdt.root_squash" },
4670         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
4671         { NULL }
4672 };
4673
4674 /* used by MGS to process specific configurations */
4675 static int mdt_process_config(const struct lu_env *env,
4676                               struct lu_device *d, struct lustre_cfg *cfg)
4677 {
4678         struct mdt_device *m = mdt_dev(d);
4679         struct md_device *md_next = m->mdt_child;
4680         struct lu_device *next = md2lu_dev(md_next);
4681         int rc;
4682         ENTRY;
4683
4684         switch (cfg->lcfg_command) {
4685         case LCFG_PARAM: {
4686                 struct obd_device          *obd = d->ld_obd;
4687
4688                 /* For interoperability */
4689                 struct cfg_interop_param   *ptr = NULL;
4690                 struct lustre_cfg          *old_cfg = NULL;
4691                 char                       *param = NULL;
4692
4693                 param = lustre_cfg_string(cfg, 1);
4694                 if (param == NULL) {
4695                         CERROR("param is empty\n");
4696                         rc = -EINVAL;
4697                         break;
4698                 }
4699
4700                 ptr = class_find_old_param(param, mdt_interop_param);
4701                 if (ptr != NULL) {
4702                         if (ptr->new_param == NULL) {
4703                                 rc = 0;
4704                                 CWARN("For interoperability, skip this %s."
4705                                       " It is obsolete.\n", ptr->old_param);
4706                                 break;
4707                         }
4708
4709                         CWARN("Found old param %s, changed it to %s.\n",
4710                               ptr->old_param, ptr->new_param);
4711
4712                         old_cfg = cfg;
4713                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
4714                         if (IS_ERR(cfg)) {
4715                                 rc = PTR_ERR(cfg);
4716                                 break;
4717                         }
4718                 }
4719
4720                 rc = class_process_proc_seq_param(PARAM_MDT, obd->obd_vars,
4721                                                         cfg, obd);
4722                 if (rc > 0 || rc == -ENOSYS) {
4723                         /* is it an HSM var ? */
4724                         rc = class_process_proc_seq_param(PARAM_HSM,
4725                                                         hsm_cdt_get_proc_vars(),
4726                                                         cfg, obd);
4727                         if (rc > 0 || rc == -ENOSYS)
4728                                 /* we don't understand; pass it on */
4729                                 rc = next->ld_ops->ldo_process_config(env, next,
4730                                                                       cfg);
4731                 }
4732
4733                 if (old_cfg != NULL)
4734                         lustre_cfg_free(cfg);
4735
4736                 break;
4737         }
4738         default:
4739                 /* others are passed further */
4740                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4741                 break;
4742         }
4743         RETURN(rc);
4744 }
4745
4746 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4747                                           const struct lu_object_header *hdr,
4748                                           struct lu_device *d)
4749 {
4750         struct mdt_object *mo;
4751
4752         ENTRY;
4753
4754         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
4755         if (mo != NULL) {
4756                 struct lu_object *o;
4757                 struct lu_object_header *h;
4758
4759                 o = &mo->mot_obj;
4760                 h = &mo->mot_header;
4761                 lu_object_header_init(h);
4762                 lu_object_init(o, h, d);
4763                 lu_object_add_top(h, o);
4764                 o->lo_ops = &mdt_obj_ops;
4765                 mutex_init(&mo->mot_ioepoch_mutex);
4766                 mutex_init(&mo->mot_lov_mutex);
4767                 init_rwsem(&mo->mot_open_sem);
4768                 RETURN(o);
4769         }
4770         RETURN(NULL);
4771 }
4772
4773 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
4774                            const struct lu_object_conf *unused)
4775 {
4776         struct mdt_device *d = mdt_dev(o->lo_dev);
4777         struct lu_device  *under;
4778         struct lu_object  *below;
4779         int                rc = 0;
4780         ENTRY;
4781
4782         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4783                PFID(lu_object_fid(o)));
4784
4785         under = &d->mdt_child->md_lu_dev;
4786         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4787         if (below != NULL) {
4788                 lu_object_add(o, below);
4789         } else
4790                 rc = -ENOMEM;
4791
4792         RETURN(rc);
4793 }
4794
4795 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4796 {
4797         struct mdt_object *mo = mdt_obj(o);
4798         struct lu_object_header *h;
4799         ENTRY;
4800
4801         h = o->lo_header;
4802         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4803                PFID(lu_object_fid(o)));
4804
4805         LASSERT(atomic_read(&mo->mot_open_count) == 0);
4806         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
4807
4808         lu_object_fini(o);
4809         lu_object_header_fini(h);
4810         OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
4811
4812         EXIT;
4813 }
4814
4815 static int mdt_object_print(const struct lu_env *env, void *cookie,
4816                             lu_printer_t p, const struct lu_object *o)
4817 {
4818         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
4819         return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p(ioepoch="LPU64" "
4820                     "flags="LPX64", epochcount=%d, writecount=%d)",
4821                     mdto, mdto->mot_ioepoch, mdto->mot_flags,
4822                     mdto->mot_ioepoch_count, mdto->mot_writecount);
4823 }
4824
4825 static int mdt_prepare(const struct lu_env *env,
4826                 struct lu_device *pdev,
4827                 struct lu_device *cdev)
4828 {
4829         struct mdt_device *mdt = mdt_dev(cdev);
4830         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
4831         struct obd_device *obd = cdev->ld_obd;
4832         int rc;
4833
4834         ENTRY;
4835
4836         LASSERT(obd);
4837
4838         rc = next->ld_ops->ldo_prepare(env, cdev, next);
4839         if (rc)
4840                 RETURN(rc);
4841
4842         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
4843         if (rc)
4844                 RETURN(rc);
4845
4846         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
4847         if (rc)
4848                 RETURN(rc);
4849
4850         rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
4851         /* The LFSCK instance is registered just now, so it must be there when
4852          * register the namespace to such instance. */
4853         LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
4854
4855         if (mdt->mdt_seq_site.ss_node_id == 0) {
4856                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
4857                                                          &mdt->mdt_md_root_fid);
4858                 if (rc)
4859                         RETURN(rc);
4860         }
4861
4862         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
4863         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
4864         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
4865         LASSERT(obd->obd_no_conn);
4866         spin_lock(&obd->obd_dev_lock);
4867         obd->obd_no_conn = 0;
4868         spin_unlock(&obd->obd_dev_lock);
4869
4870         if (obd->obd_recovering == 0)
4871                 mdt_postrecov(env, mdt);
4872
4873         RETURN(rc);
4874 }
4875
4876 const struct lu_device_operations mdt_lu_ops = {
4877         .ldo_object_alloc   = mdt_object_alloc,
4878         .ldo_process_config = mdt_process_config,
4879         .ldo_prepare        = mdt_prepare,
4880 };
4881
4882 static const struct lu_object_operations mdt_obj_ops = {
4883         .loo_object_init    = mdt_object_init,
4884         .loo_object_free    = mdt_object_free,
4885         .loo_object_print   = mdt_object_print
4886 };
4887
4888 static int mdt_obd_set_info_async(const struct lu_env *env,
4889                                   struct obd_export *exp,
4890                                   __u32 keylen, void *key,
4891                                   __u32 vallen, void *val,
4892                                   struct ptlrpc_request_set *set)
4893 {
4894         int rc;
4895
4896         ENTRY;
4897
4898         if (KEY_IS(KEY_SPTLRPC_CONF)) {
4899                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp), 0);
4900                 RETURN(rc);
4901         }
4902
4903         RETURN(0);
4904 }
4905
4906 /**
4907  * Match client and server connection feature flags.
4908  *
4909  * Compute the compatibility flags for a connection request based on
4910  * features mutually supported by client and server.
4911  *
4912  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
4913  * must not be updated here, otherwise a partially initialized value may
4914  * be exposed. After the connection request is successfully processed,
4915  * the top-level MDT connect request handler atomically updates the export
4916  * connect flags from the obd_connect_data::ocd_connect_flags field of the
4917  * reply. \see mdt_connect().
4918  *
4919  * \param exp   the obd_export associated with this client/target pair
4920  * \param mdt   the target device for the connection
4921  * \param data  stores data for this connect request
4922  *
4923  * \retval 0       success
4924  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
4925  * \retval -EBADE  client and server feature requirements are incompatible
4926  */
4927 static int mdt_connect_internal(struct obd_export *exp,
4928                                 struct mdt_device *mdt,
4929                                 struct obd_connect_data *data)
4930 {
4931         LASSERT(data != NULL);
4932
4933         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
4934         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
4935
4936         /* If no known bits (which should not happen, probably,
4937            as everybody should support LOOKUP and UPDATE bits at least)
4938            revert to compat mode with plain locks. */
4939         if (!data->ocd_ibits_known &&
4940             data->ocd_connect_flags & OBD_CONNECT_IBITS)
4941                 data->ocd_connect_flags &= ~OBD_CONNECT_IBITS;
4942
4943         if (!mdt->mdt_opts.mo_acl)
4944                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
4945
4946         if (!mdt->mdt_opts.mo_user_xattr)
4947                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
4948
4949         if (!mdt->mdt_som_conf)
4950                 data->ocd_connect_flags &= ~OBD_CONNECT_SOM;
4951
4952         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
4953                 data->ocd_brw_size = min(data->ocd_brw_size,
4954                                          (__u32)MD_MAX_BRW_SIZE);
4955                 if (data->ocd_brw_size == 0) {
4956                         CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
4957                                " ocd_version: %x ocd_grant: %d "
4958                                "ocd_index: %u ocd_brw_size is "
4959                                "unexpectedly zero, network data "
4960                                "corruption? Refusing connection of this"
4961                                " client\n",
4962                                mdt_obd_name(mdt),
4963                                exp->exp_client_uuid.uuid,
4964                                exp, data->ocd_connect_flags, data->ocd_version,
4965                                data->ocd_grant, data->ocd_index);
4966                         return -EPROTO;
4967                 }
4968         }
4969
4970         /* NB: Disregard the rule against updating
4971          * exp_connect_data.ocd_connect_flags in this case, since
4972          * tgt_client_new() needs to know if this is a lightweight
4973          * connection, and it is safe to expose this flag before
4974          * connection processing completes. */
4975         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
4976                 spin_lock(&exp->exp_lock);
4977                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
4978                 spin_unlock(&exp->exp_lock);
4979         }
4980
4981         data->ocd_version = LUSTRE_VERSION_CODE;
4982
4983         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
4984                 CWARN("%s: MDS requires FID support, but client not\n",
4985                       mdt_obd_name(mdt));
4986                 return -EBADE;
4987         }
4988
4989         if (mdt->mdt_som_conf &&
4990             !(data->ocd_connect_flags & (OBD_CONNECT_LIGHTWEIGHT |
4991                                          OBD_CONNECT_MDS_MDS |
4992                                          OBD_CONNECT_SOM))) {
4993                 CWARN("%s: MDS has SOM enabled, but client does not support "
4994                       "it\n", mdt_obd_name(mdt));
4995                 return -EBADE;
4996         }
4997
4998         if (OCD_HAS_FLAG(data, PINGLESS)) {
4999                 if (ptlrpc_pinger_suppress_pings()) {
5000                         spin_lock(&exp->exp_obd->obd_dev_lock);
5001                         list_del_init(&exp->exp_obd_chain_timed);
5002                         spin_unlock(&exp->exp_obd->obd_dev_lock);
5003                 } else {
5004                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
5005                 }
5006         }
5007
5008         data->ocd_max_easize = mdt->mdt_max_ea_size;
5009
5010         return 0;
5011 }
5012
5013 /* mds_connect copy */
5014 static int mdt_obd_connect(const struct lu_env *env,
5015                            struct obd_export **exp, struct obd_device *obd,
5016                            struct obd_uuid *cluuid,
5017                            struct obd_connect_data *data,
5018                            void *localdata)
5019 {
5020         struct obd_export      *lexp;
5021         struct lustre_handle    conn = { 0 };
5022         struct mdt_device      *mdt;
5023         int                     rc;
5024         ENTRY;
5025
5026         LASSERT(env != NULL);
5027         if (!exp || !obd || !cluuid)
5028                 RETURN(-EINVAL);
5029
5030         mdt = mdt_dev(obd->obd_lu_dev);
5031
5032         /*
5033          * first, check whether the stack is ready to handle requests
5034          * XXX: probably not very appropriate method is used now
5035          *      at some point we should find a better one
5036          */
5037         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) && data != NULL &&
5038             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
5039                 rc = obd_get_info(env, mdt->mdt_child_exp,
5040                                   sizeof(KEY_OSP_CONNECTED),
5041                                   KEY_OSP_CONNECTED, NULL, NULL, NULL);
5042                 if (rc)
5043                         RETURN(-EAGAIN);
5044                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
5045         }
5046
5047         rc = class_connect(&conn, obd, cluuid);
5048         if (rc)
5049                 RETURN(rc);
5050
5051         lexp = class_conn2export(&conn);
5052         LASSERT(lexp != NULL);
5053
5054         rc = mdt_connect_internal(lexp, mdt, data);
5055         if (rc == 0) {
5056                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5057
5058                 LASSERT(lcd);
5059                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5060                 rc = tgt_client_new(env, lexp);
5061                 if (rc == 0)
5062                         mdt_export_stats_init(obd, lexp, localdata);
5063
5064                 /* For phase I, sync for cross-ref operation. */
5065                 lexp->exp_keep_sync = 1;
5066         }
5067
5068         if (rc != 0) {
5069                 class_disconnect(lexp);
5070                 *exp = NULL;
5071         } else {
5072                 *exp = lexp;
5073         }
5074
5075         RETURN(rc);
5076 }
5077
5078 static int mdt_obd_reconnect(const struct lu_env *env,
5079                              struct obd_export *exp, struct obd_device *obd,
5080                              struct obd_uuid *cluuid,
5081                              struct obd_connect_data *data,
5082                              void *localdata)
5083 {
5084         int                     rc;
5085         ENTRY;
5086
5087         if (exp == NULL || obd == NULL || cluuid == NULL)
5088                 RETURN(-EINVAL);
5089
5090         rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
5091         if (rc == 0)
5092                 mdt_export_stats_init(obd, exp, localdata);
5093
5094         RETURN(rc);
5095 }
5096
5097 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
5098                                    struct mdt_thread_info *info,
5099                                    struct mdt_file_data *mfd)
5100 {
5101         struct lu_context ses;
5102         int rc;
5103         ENTRY;
5104
5105         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
5106         if (rc)
5107                 RETURN(rc);
5108
5109         env->le_ses = &ses;
5110         lu_context_enter(&ses);
5111
5112         mdt_ucred(info)->uc_valid = UCRED_OLD;
5113         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
5114
5115         lu_context_exit(&ses);
5116         lu_context_fini(&ses);
5117         env->le_ses = NULL;
5118
5119         RETURN(rc);
5120 }
5121
5122 static int mdt_export_cleanup(struct obd_export *exp)
5123 {
5124         struct mdt_export_data *med = &exp->exp_mdt_data;
5125         struct obd_device      *obd = exp->exp_obd;
5126         struct mdt_device      *mdt;
5127         struct mdt_thread_info *info;
5128         struct lu_env           env;
5129         CFS_LIST_HEAD(closing_list);
5130         struct mdt_file_data *mfd, *n;
5131         int rc = 0;
5132         ENTRY;
5133
5134         spin_lock(&med->med_open_lock);
5135         while (!cfs_list_empty(&med->med_open_head)) {
5136                 cfs_list_t *tmp = med->med_open_head.next;
5137                 mfd = cfs_list_entry(tmp, struct mdt_file_data, mfd_list);
5138
5139                 /* Remove mfd handle so it can't be found again.
5140                  * We are consuming the mfd_list reference here. */
5141                 class_handle_unhash(&mfd->mfd_handle);
5142                 cfs_list_move_tail(&mfd->mfd_list, &closing_list);
5143         }
5144         spin_unlock(&med->med_open_lock);
5145         mdt = mdt_dev(obd->obd_lu_dev);
5146         LASSERT(mdt != NULL);
5147
5148         rc = lu_env_init(&env, LCT_MD_THREAD);
5149         if (rc)
5150                 RETURN(rc);
5151
5152         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5153         LASSERT(info != NULL);
5154         memset(info, 0, sizeof *info);
5155         info->mti_env = &env;
5156         info->mti_mdt = mdt;
5157         info->mti_exp = exp;
5158
5159         if (!cfs_list_empty(&closing_list)) {
5160                 struct md_attr *ma = &info->mti_attr;
5161
5162                 /* Close any open files (which may also cause orphan unlinking). */
5163                 cfs_list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5164                         cfs_list_del_init(&mfd->mfd_list);
5165                         ma->ma_need = ma->ma_valid = 0;
5166
5167                         /* This file is being closed due to an eviction, it
5168                          * could have been modified and now dirty regarding to
5169                          * HSM archive, check this!
5170                          * The logic here is to mark a file dirty if there's a
5171                          * chance it was dirtied before the client was evicted,
5172                          * so that we don't have to wait for a release attempt
5173                          * before finding out the file was actually dirty and
5174                          * fail the release. Aggressively marking it dirty here
5175                          * will cause the policy engine to attempt to
5176                          * re-archive it; when rearchiving, we can compare the
5177                          * current version to the HSM data_version and make the
5178                          * archive request into a noop if it's not actually
5179                          * dirty.
5180                          */
5181                         if (mfd->mfd_mode & (FMODE_WRITE|MDS_FMODE_TRUNC))
5182                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
5183
5184                         /* Don't unlink orphan on failover umount, LU-184 */
5185                         if (exp->exp_flags & OBD_OPT_FAILOVER) {
5186                                 ma->ma_valid = MA_FLAGS;
5187                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5188                         }
5189                         mdt_mfd_close(info, mfd);
5190                 }
5191         }
5192         info->mti_mdt = NULL;
5193         /* cleanup client slot early */
5194         /* Do not erase record for recoverable client. */
5195         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5196                 tgt_client_del(&env, exp);
5197         lu_env_fini(&env);
5198
5199         RETURN(rc);
5200 }
5201
5202 static int mdt_obd_disconnect(struct obd_export *exp)
5203 {
5204         int rc;
5205         ENTRY;
5206
5207         LASSERT(exp);
5208         class_export_get(exp);
5209
5210         rc = server_disconnect_export(exp);
5211         if (rc != 0)
5212                 CDEBUG(D_IOCTL, "server disconnect error: %d\n", rc);
5213
5214         rc = mdt_export_cleanup(exp);
5215         class_export_put(exp);
5216         RETURN(rc);
5217 }
5218
5219 /* FIXME: Can we avoid using these two interfaces? */
5220 static int mdt_init_export(struct obd_export *exp)
5221 {
5222         struct mdt_export_data *med = &exp->exp_mdt_data;
5223         int                     rc;
5224         ENTRY;
5225
5226         CFS_INIT_LIST_HEAD(&med->med_open_head);
5227         spin_lock_init(&med->med_open_lock);
5228         mutex_init(&med->med_idmap_mutex);
5229         med->med_idmap = NULL;
5230         spin_lock(&exp->exp_lock);
5231         exp->exp_connecting = 1;
5232         spin_unlock(&exp->exp_lock);
5233
5234         /* self-export doesn't need client data and ldlm initialization */
5235         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5236                                      &exp->exp_client_uuid)))
5237                 RETURN(0);
5238
5239         rc = tgt_client_alloc(exp);
5240         if (rc)
5241                 GOTO(err, rc);
5242
5243         rc = ldlm_init_export(exp);
5244         if (rc)
5245                 GOTO(err_free, rc);
5246
5247         RETURN(rc);
5248
5249 err_free:
5250         tgt_client_free(exp);
5251 err:
5252         CERROR("%s: Failed to initialize export: rc = %d\n",
5253                exp->exp_obd->obd_name, rc);
5254         return rc;
5255 }
5256
5257 static int mdt_destroy_export(struct obd_export *exp)
5258 {
5259         ENTRY;
5260
5261         if (exp_connect_rmtclient(exp))
5262                 mdt_cleanup_idmap(&exp->exp_mdt_data);
5263
5264         target_destroy_export(exp);
5265         /* destroy can be called from failed obd_setup, so
5266          * checking uuid is safer than obd_self_export */
5267         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5268                                      &exp->exp_client_uuid)))
5269                 RETURN(0);
5270
5271         ldlm_destroy_export(exp);
5272         tgt_client_free(exp);
5273
5274         LASSERT(cfs_list_empty(&exp->exp_outstanding_replies));
5275         LASSERT(cfs_list_empty(&exp->exp_mdt_data.med_open_head));
5276
5277         RETURN(0);
5278 }
5279
5280 /** The maximum depth that fid2path() will search.
5281  * This is limited only because we want to store the fids for
5282  * historical path lookup purposes.
5283  */
5284 #define MAX_PATH_DEPTH 100
5285
5286 /** mdt_path() lookup structure. */
5287 struct path_lookup_info {
5288         __u64                   pli_recno;      /**< history point */
5289         __u64                   pli_currec;     /**< current record */
5290         struct lu_fid           pli_fid;
5291         struct lu_fid           pli_fids[MAX_PATH_DEPTH]; /**< path, in fids */
5292         struct mdt_object       *pli_mdt_obj;
5293         char                    *pli_path;      /**< full path */
5294         int                     pli_pathlen;
5295         int                     pli_linkno;     /**< which hardlink to follow */
5296         int                     pli_fidcount;   /**< number of \a pli_fids */
5297 };
5298
5299 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
5300                    struct linkea_data *ldata)
5301 {
5302         int rc;
5303
5304         LASSERT(ldata->ld_buf->lb_buf != NULL);
5305
5306         if (!mdt_object_exists(mdt_obj))
5307                 return -ENODATA;
5308
5309         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5310                           ldata->ld_buf, XATTR_NAME_LINK);
5311         if (rc == -ERANGE) {
5312                 /* Buf was too small, figure out what we need. */
5313                 lu_buf_free(ldata->ld_buf);
5314                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5315                                   ldata->ld_buf, XATTR_NAME_LINK);
5316                 if (rc < 0)
5317                         return rc;
5318                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
5319                 if (ldata->ld_buf->lb_buf == NULL)
5320                         return -ENOMEM;
5321                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5322                                   ldata->ld_buf, XATTR_NAME_LINK);
5323         }
5324         if (rc < 0)
5325                 return rc;
5326
5327         return linkea_init(ldata);
5328 }
5329
5330 static int mdt_path_current(struct mdt_thread_info *info,
5331                             struct path_lookup_info *pli)
5332 {
5333         struct mdt_device       *mdt = info->mti_mdt;
5334         struct mdt_object       *mdt_obj;
5335         struct link_ea_header   *leh;
5336         struct link_ea_entry    *lee;
5337         struct lu_name          *tmpname = &info->mti_name;
5338         struct lu_fid           *tmpfid = &info->mti_tmp_fid1;
5339         struct lu_buf           *buf = &info->mti_big_buf;
5340         char                    *ptr;
5341         int                     reclen;
5342         struct linkea_data      ldata = { 0 };
5343         int                     rc = 0;
5344         ENTRY;
5345
5346         /* temp buffer for path element, the buffer will be finally freed
5347          * in mdt_thread_info_fini */
5348         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
5349         if (buf->lb_buf == NULL)
5350                 RETURN(-ENOMEM);
5351
5352         ldata.ld_buf = buf;
5353         ptr = pli->pli_path + pli->pli_pathlen - 1;
5354         *ptr = 0;
5355         --ptr;
5356         pli->pli_fidcount = 0;
5357         pli->pli_fids[0] = *(struct lu_fid *)mdt_object_fid(pli->pli_mdt_obj);
5358         *tmpfid = pli->pli_fids[0];
5359         /* root FID only exists on MDT0, and fid2path should also ends at MDT0,
5360          * so checking root_fid can only happen on MDT0. */
5361         while (!lu_fid_eq(&mdt->mdt_md_root_fid,
5362                           &pli->pli_fids[pli->pli_fidcount])) {
5363                 struct lu_buf           lmv_buf;
5364
5365                 mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
5366                 if (IS_ERR(mdt_obj))
5367                         GOTO(out, rc = PTR_ERR(mdt_obj));
5368
5369                 if (mdt_object_remote(mdt_obj)) {
5370                         mdt_object_put(info->mti_env, mdt_obj);
5371                         GOTO(remote_out, rc = -EREMOTE);
5372                 }
5373
5374                 lmv_buf.lb_buf = info->mti_xattr_buf;
5375                 lmv_buf.lb_len = sizeof(info->mti_xattr_buf);
5376
5377                 /* Check if it is slave stripes */
5378                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5379                                   &lmv_buf, XATTR_NAME_LMV);
5380                 if (rc > 0) {
5381                         union lmv_mds_md *lmm = lmv_buf.lb_buf;
5382
5383                         /* For slave stripes, get its master */
5384                         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE) {
5385                                 struct lmv_mds_md_v1 *lmm1 = &lmm->lmv_md_v1;
5386
5387                                 fid_le_to_cpu(tmpfid, &lmm1->lmv_master_fid);
5388                                 if (!fid_is_sane(tmpfid)) {
5389                                         mdt_object_put(info->mti_env, mdt_obj);
5390                                         GOTO(out, rc = -EINVAL);
5391                                 }
5392                                 mdt_object_put(info->mti_env, mdt_obj);
5393                                 pli->pli_fids[pli->pli_fidcount] = *tmpfid;
5394                                 continue;
5395                         }
5396                 }
5397
5398                 if (!mdt_object_exists(mdt_obj)) {
5399                         mdt_object_put(info->mti_env, mdt_obj);
5400                         GOTO(out, rc = -ENOENT);
5401                 }
5402
5403                 rc = mdt_links_read(info, mdt_obj, &ldata);
5404                 mdt_object_put(info->mti_env, mdt_obj);
5405                 if (rc != 0)
5406                         GOTO(out, rc);
5407
5408                 leh = buf->lb_buf;
5409                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
5410                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
5411                 /* If set, use link #linkno for path lookup, otherwise use
5412                    link #0.  Only do this for the final path element. */
5413                 if (pli->pli_fidcount == 0 &&
5414                     pli->pli_linkno < leh->leh_reccount) {
5415                         int count;
5416                         for (count = 0; count < pli->pli_linkno; count++) {
5417                                 lee = (struct link_ea_entry *)
5418                                      ((char *)lee + reclen);
5419                                 linkea_entry_unpack(lee, &reclen, tmpname,
5420                                                     tmpfid);
5421                         }
5422                         if (pli->pli_linkno < leh->leh_reccount - 1)
5423                                 /* indicate to user there are more links */
5424                                 pli->pli_linkno++;
5425                 }
5426
5427                 /* Pack the name in the end of the buffer */
5428                 ptr -= tmpname->ln_namelen;
5429                 if (ptr - 1 <= pli->pli_path)
5430                         GOTO(out, rc = -EOVERFLOW);
5431                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
5432                 *(--ptr) = '/';
5433
5434                 /* Store the parent fid for historic lookup */
5435                 if (++pli->pli_fidcount >= MAX_PATH_DEPTH)
5436                         GOTO(out, rc = -EOVERFLOW);
5437                 pli->pli_fids[pli->pli_fidcount] = *tmpfid;
5438         }
5439
5440 remote_out:
5441         ptr++; /* skip leading / */
5442         memmove(pli->pli_path, ptr, pli->pli_path + pli->pli_pathlen - ptr);
5443
5444         EXIT;
5445 out:
5446         return rc;
5447 }
5448
5449 /* Returns the full path to this fid, as of changelog record recno. */
5450 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
5451                     char *path, int pathlen, __u64 *recno, int *linkno,
5452                     struct lu_fid *fid)
5453 {
5454         struct mdt_device       *mdt = info->mti_mdt;
5455         struct path_lookup_info *pli;
5456         int                     tries = 3;
5457         int                     rc = -EAGAIN;
5458         ENTRY;
5459
5460         if (pathlen < 3)
5461                 RETURN(-EOVERFLOW);
5462
5463         if (lu_fid_eq(&mdt->mdt_md_root_fid, mdt_object_fid(obj))) {
5464                 path[0] = '\0';
5465                 RETURN(0);
5466         }
5467
5468         OBD_ALLOC_PTR(pli);
5469         if (pli == NULL)
5470                 RETURN(-ENOMEM);
5471
5472         pli->pli_mdt_obj = obj;
5473         pli->pli_recno = *recno;
5474         pli->pli_path = path;
5475         pli->pli_pathlen = pathlen;
5476         pli->pli_linkno = *linkno;
5477
5478         /* Retry multiple times in case file is being moved */
5479         while (tries-- && rc == -EAGAIN)
5480                 rc = mdt_path_current(info, pli);
5481
5482         /* return the last resolved fids to the client, so the client will
5483          * build the left path on another MDT for remote object */
5484         *fid = pli->pli_fids[pli->pli_fidcount];
5485
5486         *recno = pli->pli_currec;
5487         /* Return next link index to caller */
5488         *linkno = pli->pli_linkno;
5489
5490         OBD_FREE_PTR(pli);
5491
5492         RETURN(rc);
5493 }
5494
5495 static int mdt_fid2path(struct mdt_thread_info *info,
5496                         struct getinfo_fid2path *fp)
5497 {
5498         struct mdt_device *mdt = info->mti_mdt;
5499         struct mdt_object *obj;
5500         int    rc;
5501         ENTRY;
5502
5503         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
5504                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
5505
5506         if (!fid_is_sane(&fp->gf_fid))
5507                 RETURN(-EINVAL);
5508
5509         if (!fid_is_namespace_visible(&fp->gf_fid)) {
5510                 CWARN("%s: "DFID" is invalid, sequence should be "
5511                       ">= "LPX64"\n", mdt_obd_name(mdt),
5512                       PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
5513                 RETURN(-EINVAL);
5514         }
5515
5516         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
5517         if (obj == NULL || IS_ERR(obj)) {
5518                 CDEBUG(D_IOCTL, "no object "DFID": %ld\n", PFID(&fp->gf_fid),
5519                        PTR_ERR(obj));
5520                 RETURN(-EINVAL);
5521         }
5522
5523         if (mdt_object_remote(obj))
5524                 rc = -EREMOTE;
5525         else if (!mdt_object_exists(obj))
5526                 rc = -ENOENT;
5527         else
5528                 rc = 0;
5529
5530         if (rc < 0) {
5531                 mdt_object_put(info->mti_env, obj);
5532                 CDEBUG(D_IOCTL, "nonlocal object "DFID": %d\n",
5533                        PFID(&fp->gf_fid), rc);
5534                 RETURN(rc);
5535         }
5536
5537         rc = mdt_path(info, obj, fp->gf_path, fp->gf_pathlen, &fp->gf_recno,
5538                       &fp->gf_linkno, &fp->gf_fid);
5539
5540         CDEBUG(D_INFO, "fid "DFID", path %s recno "LPX64" linkno %u\n",
5541                PFID(&fp->gf_fid), fp->gf_path, fp->gf_recno, fp->gf_linkno);
5542
5543         mdt_object_put(info->mti_env, obj);
5544
5545         RETURN(rc);
5546 }
5547
5548 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key,
5549                             void *val, int vallen)
5550 {
5551         struct getinfo_fid2path *fpout, *fpin;
5552         int rc = 0;
5553
5554         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
5555         fpout = val;
5556
5557         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5558                 lustre_swab_fid2path(fpin);
5559
5560         memcpy(fpout, fpin, sizeof(*fpin));
5561         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
5562                 RETURN(-EINVAL);
5563
5564         rc = mdt_fid2path(info, fpout);
5565         RETURN(rc);
5566 }
5567
5568 int mdt_get_info(struct tgt_session_info *tsi)
5569 {
5570         char    *key;
5571         int      keylen;
5572         __u32   *vallen;
5573         void    *valout;
5574         int      rc;
5575
5576         ENTRY;
5577
5578         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
5579         if (key == NULL) {
5580                 CDEBUG(D_IOCTL, "No GETINFO key");
5581                 RETURN(err_serious(-EFAULT));
5582         }
5583         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
5584                                       RCL_CLIENT);
5585
5586         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
5587         if (vallen == NULL) {
5588                 CDEBUG(D_IOCTL, "Unable to get RMF_GETINFO_VALLEN buffer");
5589                 RETURN(err_serious(-EFAULT));
5590         }
5591
5592         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
5593                              *vallen);
5594         rc = req_capsule_server_pack(tsi->tsi_pill);
5595         if (rc)
5596                 RETURN(err_serious(rc));
5597
5598         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
5599         if (valout == NULL) {
5600                 CDEBUG(D_IOCTL, "Unable to get get-info RPC out buffer");
5601                 RETURN(err_serious(-EFAULT));
5602         }
5603
5604         if (KEY_IS(KEY_FID2PATH)) {
5605                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
5606
5607                 rc = mdt_rpc_fid2path(info, key, valout, *vallen);
5608                 mdt_thread_info_fini(info);
5609         } else {
5610                 rc = -EINVAL;
5611         }
5612         RETURN(rc);
5613 }
5614
5615 /* Pass the ioc down */
5616 static int mdt_ioc_child(struct lu_env *env, struct mdt_device *mdt,
5617                          unsigned int cmd, int len, void *data)
5618 {
5619         struct lu_context ioctl_session;
5620         struct md_device *next = mdt->mdt_child;
5621         int rc;
5622         ENTRY;
5623
5624         rc = lu_context_init(&ioctl_session, LCT_SERVER_SESSION);
5625         if (rc)
5626                 RETURN(rc);
5627         ioctl_session.lc_thread = (struct ptlrpc_thread *)current;
5628         lu_context_enter(&ioctl_session);
5629         env->le_ses = &ioctl_session;
5630
5631         LASSERT(next->md_ops->mdo_iocontrol);
5632         rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data);
5633
5634         lu_context_exit(&ioctl_session);
5635         lu_context_fini(&ioctl_session);
5636         RETURN(rc);
5637 }
5638
5639 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
5640 {
5641         struct obd_ioctl_data *data = karg;
5642         struct lu_fid *fid;
5643         __u64 version;
5644         struct mdt_object *obj;
5645         struct mdt_lock_handle  *lh;
5646         int rc;
5647         ENTRY;
5648
5649         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
5650             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
5651                 RETURN(-EINVAL);
5652
5653         fid = (struct lu_fid *)data->ioc_inlbuf1;
5654
5655         if (!fid_is_sane(fid))
5656                 RETURN(-EINVAL);
5657
5658         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
5659
5660         lh = &mti->mti_lh[MDT_LH_PARENT];
5661         mdt_lock_reg_init(lh, LCK_CR);
5662
5663         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
5664         if (IS_ERR(obj))
5665                 RETURN(PTR_ERR(obj));
5666
5667         if (mdt_object_remote(obj)) {
5668                 rc = -EREMOTE;
5669                 /**
5670                  * before calling version get the correct MDS should be
5671                  * fid, this is error to find remote object here
5672                  */
5673                 CERROR("nonlocal object "DFID"\n", PFID(fid));
5674         } else if (!mdt_object_exists(obj)) {
5675                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
5676                 rc = -ENOENT;
5677         } else {
5678                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
5679                *(__u64 *)data->ioc_inlbuf2 = version;
5680                 rc = 0;
5681         }
5682         mdt_object_unlock_put(mti, obj, lh, 1);
5683         RETURN(rc);
5684 }
5685
5686 /* ioctls on obd dev */
5687 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
5688                          void *karg, void *uarg)
5689 {
5690         struct lu_env      env;
5691         struct obd_device *obd = exp->exp_obd;
5692         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
5693         struct dt_device  *dt = mdt->mdt_bottom;
5694         int rc;
5695
5696         ENTRY;
5697         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
5698         rc = lu_env_init(&env, LCT_MD_THREAD);
5699         if (rc)
5700                 RETURN(rc);
5701
5702         switch (cmd) {
5703         case OBD_IOC_SYNC:
5704                 rc = mdt_device_sync(&env, mdt);
5705                 break;
5706         case OBD_IOC_SET_READONLY:
5707                 rc = dt->dd_ops->dt_ro(&env, dt);
5708                 break;
5709         case OBD_IOC_ABORT_RECOVERY:
5710                 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
5711                 target_stop_recovery_thread(obd);
5712                 rc = 0;
5713                 break;
5714         case OBD_IOC_CHANGELOG_REG:
5715         case OBD_IOC_CHANGELOG_DEREG:
5716         case OBD_IOC_CHANGELOG_CLEAR:
5717                 rc = mdt_ioc_child(&env, mdt, cmd, len, karg);
5718                 break;
5719         case OBD_IOC_START_LFSCK: {
5720                 struct md_device *next = mdt->mdt_child;
5721                 struct obd_ioctl_data *data = karg;
5722                 struct lfsck_start_param lsp;
5723
5724                 if (unlikely(data == NULL)) {
5725                         rc = -EINVAL;
5726                         break;
5727                 }
5728
5729                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
5730                 lsp.lsp_index_valid = 0;
5731                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
5732                 break;
5733         }
5734         case OBD_IOC_STOP_LFSCK: {
5735                 struct md_device        *next = mdt->mdt_child;
5736                 struct obd_ioctl_data   *data = karg;
5737                 struct lfsck_stop        stop;
5738
5739                 stop.ls_status = LS_STOPPED;
5740                 /* Old lfsck utils may pass NULL @stop. */
5741                 if (data->ioc_inlbuf1 == NULL)
5742                         stop.ls_flags = 0;
5743                 else
5744                         stop.ls_flags =
5745                         ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
5746
5747                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
5748                 break;
5749         }
5750         case OBD_IOC_GET_OBJ_VERSION: {
5751                 struct mdt_thread_info *mti;
5752                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5753                 memset(mti, 0, sizeof *mti);
5754                 mti->mti_env = &env;
5755                 mti->mti_mdt = mdt;
5756                 mti->mti_exp = exp;
5757
5758                 rc = mdt_ioc_version_get(mti, karg);
5759                 break;
5760         }
5761         case OBD_IOC_CATLOGLIST: {
5762                 struct mdt_thread_info *mti;
5763
5764                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5765                 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
5766                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
5767                                        &mti->mti_tmp_fid1);
5768                 break;
5769          }
5770         default:
5771                 rc = -EOPNOTSUPP;
5772                 CERROR("%s: Not supported cmd = %d, rc = %d\n",
5773                         mdt_obd_name(mdt), cmd, rc);
5774         }
5775
5776         lu_env_fini(&env);
5777         RETURN(rc);
5778 }
5779
5780 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
5781 {
5782         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
5783         struct lfsck_start_param lsp;
5784         int rc;
5785         ENTRY;
5786
5787         lsp.lsp_start = NULL;
5788         lsp.lsp_index_valid = 0;
5789         rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
5790                                                    OBD_IOC_START_LFSCK,
5791                                                    0, &lsp);
5792         if (rc != 0 && rc != -EALREADY)
5793                 CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
5794                       mdt_obd_name(mdt), rc);
5795
5796         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
5797         RETURN(rc);
5798 }
5799
5800 static int mdt_obd_postrecov(struct obd_device *obd)
5801 {
5802         struct lu_env env;
5803         int rc;
5804
5805         rc = lu_env_init(&env, LCT_MD_THREAD);
5806         if (rc)
5807                 RETURN(rc);
5808         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
5809         lu_env_fini(&env);
5810         return rc;
5811 }
5812
5813 static struct obd_ops mdt_obd_device_ops = {
5814         .o_owner          = THIS_MODULE,
5815         .o_set_info_async = mdt_obd_set_info_async,
5816         .o_connect        = mdt_obd_connect,
5817         .o_reconnect      = mdt_obd_reconnect,
5818         .o_disconnect     = mdt_obd_disconnect,
5819         .o_init_export    = mdt_init_export,
5820         .o_destroy_export = mdt_destroy_export,
5821         .o_iocontrol      = mdt_iocontrol,
5822         .o_postrecov      = mdt_obd_postrecov,
5823 };
5824
5825 static struct lu_device* mdt_device_fini(const struct lu_env *env,
5826                                          struct lu_device *d)
5827 {
5828         struct mdt_device *m = mdt_dev(d);
5829         ENTRY;
5830
5831         mdt_fini(env, m);
5832         RETURN(NULL);
5833 }
5834
5835 static struct lu_device *mdt_device_free(const struct lu_env *env,
5836                                          struct lu_device *d)
5837 {
5838         struct mdt_device *m = mdt_dev(d);
5839         ENTRY;
5840
5841         lu_device_fini(&m->mdt_lu_dev);
5842         OBD_FREE_PTR(m);
5843
5844         RETURN(NULL);
5845 }
5846
5847 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
5848                                           struct lu_device_type *t,
5849                                           struct lustre_cfg *cfg)
5850 {
5851         struct lu_device  *l;
5852         struct mdt_device *m;
5853
5854         OBD_ALLOC_PTR(m);
5855         if (m != NULL) {
5856                 int rc;
5857
5858                 l = &m->mdt_lu_dev;
5859                 rc = mdt_init0(env, m, t, cfg);
5860                 if (rc != 0) {
5861                         mdt_device_free(env, l);
5862                         l = ERR_PTR(rc);
5863                         return l;
5864                 }
5865         } else
5866                 l = ERR_PTR(-ENOMEM);
5867         return l;
5868 }
5869
5870 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
5871 LU_KEY_INIT(mdt, struct mdt_thread_info);
5872
5873 static void mdt_key_fini(const struct lu_context *ctx,
5874                          struct lu_context_key *key, void* data)
5875 {
5876         struct mdt_thread_info *info = data;
5877
5878         if (info->mti_big_lmm) {
5879                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
5880                 info->mti_big_lmm = NULL;
5881                 info->mti_big_lmmsize = 0;
5882         }
5883         OBD_FREE_PTR(info);
5884 }
5885
5886 /* context key: mdt_thread_key */
5887 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
5888
5889 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
5890 {
5891         return lu_ucred(info->mti_env);
5892 }
5893
5894 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
5895 {
5896         return lu_ucred_check(info->mti_env);
5897 }
5898
5899 /**
5900  * Enable/disable COS (Commit On Sharing).
5901  *
5902  * Set/Clear the COS flag in mdt options.
5903  *
5904  * \param mdt mdt device
5905  * \param val 0 disables COS, other values enable COS
5906  */
5907 void mdt_enable_cos(struct mdt_device *mdt, int val)
5908 {
5909         struct lu_env env;
5910         int rc;
5911
5912         mdt->mdt_opts.mo_cos = !!val;
5913         rc = lu_env_init(&env, LCT_LOCAL);
5914         if (unlikely(rc != 0)) {
5915                 CWARN("lu_env initialization failed with rc = %d,"
5916                       "cannot sync\n", rc);
5917                 return;
5918         }
5919         mdt_device_sync(&env, mdt);
5920         lu_env_fini(&env);
5921 }
5922
5923 /**
5924  * Check COS (Commit On Sharing) status.
5925  *
5926  * Return COS flag status.
5927  *
5928  * \param mdt mdt device
5929  */
5930 int mdt_cos_is_enabled(struct mdt_device *mdt)
5931 {
5932         return mdt->mdt_opts.mo_cos != 0;
5933 }
5934
5935 static struct lu_device_type_operations mdt_device_type_ops = {
5936         .ldto_device_alloc = mdt_device_alloc,
5937         .ldto_device_free  = mdt_device_free,
5938         .ldto_device_fini  = mdt_device_fini
5939 };
5940
5941 static struct lu_device_type mdt_device_type = {
5942         .ldt_tags     = LU_DEVICE_MD,
5943         .ldt_name     = LUSTRE_MDT_NAME,
5944         .ldt_ops      = &mdt_device_type_ops,
5945         .ldt_ctx_tags = LCT_MD_THREAD
5946 };
5947
5948 static int __init mdt_mod_init(void)
5949 {
5950         int rc;
5951
5952         CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") ==
5953                  FID_NOBRACE_LEN + 1);
5954         CLASSERT(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") ==
5955                  FID_LEN + 1);
5956         rc = lu_kmem_init(mdt_caches);
5957         if (rc)
5958                 return rc;
5959
5960         rc = mds_mod_init();
5961         if (rc)
5962                 GOTO(lu_fini, rc);
5963
5964         rc = class_register_type(&mdt_obd_device_ops, NULL, true, NULL,
5965 #ifndef HAVE_ONLY_PROCFS_SEQ
5966                                  NULL,
5967 #endif
5968                                  LUSTRE_MDT_NAME, &mdt_device_type);
5969         if (rc)
5970                 GOTO(mds_fini, rc);
5971 lu_fini:
5972         if (rc)
5973                 lu_kmem_fini(mdt_caches);
5974 mds_fini:
5975         if (rc)
5976                 mds_mod_exit();
5977         return rc;
5978 }
5979
5980 static void __exit mdt_mod_exit(void)
5981 {
5982         class_unregister_type(LUSTRE_MDT_NAME);
5983         mds_mod_exit();
5984         lu_kmem_fini(mdt_caches);
5985 }
5986
5987 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
5988 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
5989 MODULE_LICENSE("GPL");
5990
5991 cfs_module(mdt, LUSTRE_VERSION_STRING, mdt_mod_init, mdt_mod_exit);