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->mbo_fid1 = mdt->mdt_md_root_fid;
 284         repbody->mbo_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->mbo_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->mbo_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->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 387         b->mbo_size = ma->ma_som->msd_size;
 388         b->mbo_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->mbo_atime      = attr->la_atime;
 399         b->mbo_mtime      = attr->la_mtime;
 400         b->mbo_ctime      = attr->la_ctime;
 401         b->mbo_mode       = attr->la_mode;
 402         b->mbo_size       = attr->la_size;
 403         b->mbo_blocks     = attr->la_blocks;
 404         b->mbo_uid      = attr->la_uid;
 405         b->mbo_gid      = attr->la_gid;
 406         b->mbo_flags      = attr->la_flags;
 407         b->mbo_nlink      = attr->la_nlink;
 408         b->mbo_rdev       = attr->la_rdev;
 409
 410         /* XXX should pack the reply body according to lu_valid */
 411         b->mbo_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->mbo_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->mbo_blocks = 0;
 423                 /* if no object is allocated on osts, the size on mds is valid.
 424                  * b=22272 */
 425                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 426         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL &&
 427                    ma->ma_lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) {
 428                 /* A released file stores its size on MDS. */
 429                 /* But return 1 block for released file, unless tools like tar
 430                  * will consider it fully sparse. (LU-3864)
 431                  */
 432                 if (unlikely(b->mbo_size == 0))
 433                         b->mbo_blocks = 0;
 434                 else
 435                         b->mbo_blocks = 1;
 436                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 437         }
 438
 439         if (fid) {
 440                 b->mbo_fid1 = *fid;
 441                 b->mbo_valid |= OBD_MD_FLID;
 442                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n",
 443                                 PFID(fid), b->mbo_nlink,
 444                        b->mbo_mode, b->mbo_size);
 445         }
 446
 447         if (info)
 448                 mdt_body_reverse_idmap(info, b);
 449
 450         if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE))
 451                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 452                        PFID(fid), (unsigned long long)b->mbo_size);
 453 }
 454
 455 static inline int mdt_body_has_lov(const struct lu_attr *la,
 456                                    const struct mdt_body *body)
 457 {
 458         return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
 459                (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
 460 }
 461
 462 void mdt_client_compatibility(struct mdt_thread_info *info)
 463 {
 464         struct mdt_body       *body;
 465         struct ptlrpc_request *req = mdt_info_req(info);
 466         struct obd_export     *exp = req->rq_export;
 467         struct md_attr        *ma = &info->mti_attr;
 468         struct lu_attr        *la = &ma->ma_attr;
 469         ENTRY;
 470
 471         if (exp_connect_layout(exp))
 472                 /* the client can deal with 16-bit lmm_stripe_count */
 473                 RETURN_EXIT;
 474
 475         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 476
 477         if (!mdt_body_has_lov(la, body))
 478                 RETURN_EXIT;
 479
 480         /* now we have a reply with a lov for a client not compatible with the
 481          * layout lock so we have to clean the layout generation number */
 482         if (S_ISREG(la->la_mode))
 483                 ma->ma_lmm->lmm_layout_gen = 0;
 484         EXIT;
 485 }
 486
 487 int mdt_attr_get_eabuf_size(struct mdt_thread_info *info, struct mdt_object *o)
 488 {
 489         const struct lu_env *env = info->mti_env;
 490         int rc, rc2;
 491
 492         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 493                           XATTR_NAME_LOV);
 494
 495         if (rc == -ENODATA)
 496                 rc = 0;
 497
 498         if (rc < 0)
 499                 goto out;
 500
 501         /* Is it a directory? Let's check for the LMV as well */
 502         if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
 503                 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 504                                    XATTR_NAME_LMV);
 505
 506                 if (rc2 == -ENODATA)
 507                         rc2 = mo_xattr_get(env, mdt_object_child(o),
 508                                            &LU_BUF_NULL,
 509                                            XATTR_NAME_DEFAULT_LMV);
 510
 511                 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
 512                         rc = rc2;
 513         }
 514
 515 out:
 516         return rc;
 517 }
 518
 519 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
 520                       const char *name)
 521 {
 522         const struct lu_env *env = info->mti_env;
 523         int rc;
 524         ENTRY;
 525
 526         LASSERT(info->mti_big_lmm_used == 0);
 527         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
 528         if (rc < 0)
 529                 RETURN(rc);
 530
 531         /* big_lmm may need to be grown */
 532         if (info->mti_big_lmmsize < rc) {
 533                 int size = size_roundup_power2(rc);
 534
 535                 if (info->mti_big_lmmsize > 0) {
 536                         /* free old buffer */
 537                         LASSERT(info->mti_big_lmm);
 538                         OBD_FREE_LARGE(info->mti_big_lmm,
 539                                        info->mti_big_lmmsize);
 540                         info->mti_big_lmm = NULL;
 541                         info->mti_big_lmmsize = 0;
 542                 }
 543
 544                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
 545                 if (info->mti_big_lmm == NULL)
 546                         RETURN(-ENOMEM);
 547                 info->mti_big_lmmsize = size;
 548         }
 549         LASSERT(info->mti_big_lmmsize >= rc);
 550
 551         info->mti_buf.lb_buf = info->mti_big_lmm;
 552         info->mti_buf.lb_len = info->mti_big_lmmsize;
 553         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
 554
 555         RETURN(rc);
 556 }
 557
 558 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
 559                    struct md_attr *ma, const char *name)
 560 {
 561         struct md_object *next = mdt_object_child(o);
 562         struct lu_buf    *buf = &info->mti_buf;
 563         int rc;
 564
 565         if (strcmp(name, XATTR_NAME_LOV) == 0) {
 566                 buf->lb_buf = ma->ma_lmm;
 567                 buf->lb_len = ma->ma_lmm_size;
 568                 LASSERT(!(ma->ma_valid & MA_LOV));
 569         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 570                 buf->lb_buf = ma->ma_lmv;
 571                 buf->lb_len = ma->ma_lmv_size;
 572                 LASSERT(!(ma->ma_valid & MA_LMV));
 573         } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 574                 buf->lb_buf = ma->ma_lmv;
 575                 buf->lb_len = ma->ma_lmv_size;
 576                 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
 577         } else {
 578                 return -EINVAL;
 579         }
 580
 581         rc = mo_xattr_get(info->mti_env, next, buf, name);
 582         if (rc > 0) {
 583
 584 got:
 585                 if (strcmp(name, XATTR_NAME_LOV) == 0) {
 586                         if (info->mti_big_lmm_used)
 587                                 ma->ma_lmm = info->mti_big_lmm;
 588
 589                         /* NOT return LOV EA with hole to old client. */
 590                         if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
 591                                      LOV_PATTERN_F_HOLE) &&
 592                             !(exp_connect_flags(info->mti_exp) &
 593                               OBD_CONNECT_LFSCK)) {
 594                                 return -EIO;
 595                         } else {
 596                                 ma->ma_lmm_size = rc;
 597                                 ma->ma_valid |= MA_LOV;
 598                         }
 599                 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 600                         if (info->mti_big_lmm_used)
 601                                 ma->ma_lmv = info->mti_big_lmm;
 602
 603                         ma->ma_lmv_size = rc;
 604                         ma->ma_valid |= MA_LMV;
 605                 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 606                         ma->ma_lmv_size = rc;
 607                         ma->ma_valid |= MA_LMV_DEF;
 608                 }
 609
 610                 /* Update mdt_max_mdsize so all clients will be aware that */
 611                 if (info->mti_mdt->mdt_max_mdsize < rc)
 612                         info->mti_mdt->mdt_max_mdsize = rc;
 613
 614                 rc = 0;
 615         } else if (rc == -ENODATA) {
 616                 /* no LOV EA */
 617                 rc = 0;
 618         } else if (rc == -ERANGE) {
 619                 /* Default LMV has fixed size, so it must be able to fit
 620                  * in the original buffer */
 621                 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
 622                         return rc;
 623                 rc = mdt_big_xattr_get(info, o, name);
 624                 if (rc > 0) {
 625                         info->mti_big_lmm_used = 1;
 626                         goto got;
 627                 }
 628         }
 629
 630         return rc;
 631 }
 632
 633 static int mdt_attr_get_pfid(struct mdt_thread_info *info,
 634                              struct mdt_object *o, struct lu_fid *pfid)
 635 {
 636         struct lu_buf           *buf = &info->mti_buf;
 637         struct link_ea_header   *leh;
 638         struct link_ea_entry    *lee;
 639         int                      rc;
 640         ENTRY;
 641
 642         buf->lb_buf = info->mti_big_lmm;
 643         buf->lb_len = info->mti_big_lmmsize;
 644         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
 645                           buf, XATTR_NAME_LINK);
 646         /* ignore errors, MA_PFID won't be set and it is
 647          * up to the caller to treat this as an error */
 648         if (rc == -ERANGE || buf->lb_len == 0) {
 649                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
 650                 buf->lb_buf = info->mti_big_lmm;
 651                 buf->lb_len = info->mti_big_lmmsize;
 652         }
 653
 654         if (rc < 0)
 655                 RETURN(rc);
 656         if (rc < sizeof(*leh)) {
 657                 CERROR("short LinkEA on "DFID": rc = %d\n",
 658                        PFID(mdt_object_fid(o)), rc);
 659                 RETURN(-ENODATA);
 660         }
 661
 662         leh = (struct link_ea_header *) buf->lb_buf;
 663         lee = (struct link_ea_entry *)(leh + 1);
 664         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
 665                 leh->leh_magic = LINK_EA_MAGIC;
 666                 leh->leh_reccount = __swab32(leh->leh_reccount);
 667                 leh->leh_len = __swab64(leh->leh_len);
 668         }
 669         if (leh->leh_magic != LINK_EA_MAGIC)
 670                 RETURN(-EINVAL);
 671         if (leh->leh_reccount == 0)
 672                 RETURN(-ENODATA);
 673
 674         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
 675         fid_be_to_cpu(pfid, pfid);
 676
 677         RETURN(0);
 678 }
 679
 680 int mdt_attr_get_complex(struct mdt_thread_info *info,
 681                          struct mdt_object *o, struct md_attr *ma)
 682 {
 683         const struct lu_env *env = info->mti_env;
 684         struct md_object    *next = mdt_object_child(o);
 685         struct lu_buf       *buf = &info->mti_buf;
 686         int                  need = ma->ma_need;
 687         int                  rc = 0, rc2;
 688         u32                  mode;
 689         ENTRY;
 690
 691         ma->ma_valid = 0;
 692
 693         if (mdt_object_exists(o) == 0)
 694                 GOTO(out, rc = -ENOENT);
 695         mode = lu_object_attr(&next->mo_lu);
 696
 697         if (need & MA_INODE) {
 698                 ma->ma_need = MA_INODE;
 699                 rc = mo_attr_get(env, next, ma);
 700                 if (rc)
 701                         GOTO(out, rc);
 702                 ma->ma_valid |= MA_INODE;
 703         }
 704
 705         if (need & MA_PFID) {
 706                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
 707                 if (rc == 0)
 708                         ma->ma_valid |= MA_PFID;
 709                 /* ignore this error, parent fid is not mandatory */
 710                 rc = 0;
 711         }
 712
 713         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
 714                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
 715                 if (rc)
 716                         GOTO(out, rc);
 717         }
 718
 719         if (need & MA_LMV && S_ISDIR(mode)) {
 720                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
 721                 if (rc != 0)
 722                         GOTO(out, rc);
 723         }
 724
 725         if (need & MA_LMV_DEF && S_ISDIR(mode)) {
 726                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
 727                 if (rc != 0)
 728                         GOTO(out, rc);
 729         }
 730
 731         if (need & MA_SOM && S_ISREG(mode)) {
 732                 buf->lb_buf = info->mti_xattr_buf;
 733                 buf->lb_len = sizeof(info->mti_xattr_buf);
 734                 CLASSERT(sizeof(struct som_attrs) <=
 735                          sizeof(info->mti_xattr_buf));
 736                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_SOM);
 737                 rc2 = lustre_buf2som(info->mti_xattr_buf, rc2, ma->ma_som);
 738                 if (rc2 == 0)
 739                         ma->ma_valid |= MA_SOM;
 740                 else if (rc2 < 0 && rc2 != -ENODATA)
 741                         GOTO(out, rc = rc2);
 742         }
 743
 744         if (need & MA_HSM && S_ISREG(mode)) {
 745                 buf->lb_buf = info->mti_xattr_buf;
 746                 buf->lb_len = sizeof(info->mti_xattr_buf);
 747                 CLASSERT(sizeof(struct hsm_attrs) <=
 748                          sizeof(info->mti_xattr_buf));
 749                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
 750                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
 751                 if (rc2 == 0)
 752                         ma->ma_valid |= MA_HSM;
 753                 else if (rc2 < 0 && rc2 != -ENODATA)
 754                         GOTO(out, rc = rc2);
 755         }
 756
 757 #ifdef CONFIG_FS_POSIX_ACL
 758         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
 759                 buf->lb_buf = ma->ma_acl;
 760                 buf->lb_len = ma->ma_acl_size;
 761                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
 762                 if (rc2 > 0) {
 763                         ma->ma_acl_size = rc2;
 764                         ma->ma_valid |= MA_ACL_DEF;
 765                 } else if (rc2 == -ENODATA) {
 766                         /* no ACLs */
 767                         ma->ma_acl_size = 0;
 768                 } else
 769                         GOTO(out, rc = rc2);
 770         }
 771 #endif
 772 out:
 773         ma->ma_need = need;
 774         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
 775                rc, ma->ma_valid, ma->ma_lmm);
 776         RETURN(rc);
 777 }
 778
 779 static int mdt_getattr_internal(struct mdt_thread_info *info,
 780                                 struct mdt_object *o, int ma_need)
 781 {
 782         struct md_object        *next = mdt_object_child(o);
 783         const struct mdt_body   *reqbody = info->mti_body;
 784         struct ptlrpc_request   *req = mdt_info_req(info);
 785         struct md_attr          *ma = &info->mti_attr;
 786         struct lu_attr          *la = &ma->ma_attr;
 787         struct req_capsule      *pill = info->mti_pill;
 788         const struct lu_env     *env = info->mti_env;
 789         struct mdt_body         *repbody;
 790         struct lu_buf           *buffer = &info->mti_buf;
 791         int                     rc;
 792         int                     is_root;
 793         ENTRY;
 794
 795         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
 796                 RETURN(err_serious(-ENOMEM));
 797
 798         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 799
 800         ma->ma_valid = 0;
 801
 802         if (mdt_object_remote(o)) {
 803                 /* This object is located on remote node.*/
 804                 /* Return -ENOTSUPP for old client */
 805                 if (!mdt_is_dne_client(req->rq_export))
 806                         GOTO(out, rc = -ENOTSUPP);
 807
 808                 repbody->mbo_fid1 = *mdt_object_fid(o);
 809                 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
 810                 GOTO(out, rc = 0);
 811         }
 812
 813         if (reqbody->mbo_eadatasize > 0) {
 814                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
 815                 if (buffer->lb_buf == NULL)
 816                         GOTO(out, rc = -EPROTO);
 817                 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
 818                                                       RCL_SERVER);
 819         } else {
 820                 buffer->lb_buf = NULL;
 821                 buffer->lb_len = 0;
 822                 ma_need &= ~(MA_LOV | MA_LMV);
 823                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
 824                        mdt_obd_name(info->mti_mdt),
 825                        req->rq_export->exp_client_uuid.uuid);
 826         }
 827
 828         /* If it is dir object and client require MEA, then we got MEA */
 829         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 830             (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
 831                 /* Assumption: MDT_MD size is enough for lmv size. */
 832                 ma->ma_lmv = buffer->lb_buf;
 833                 ma->ma_lmv_size = buffer->lb_len;
 834                 ma->ma_need = MA_INODE;
 835                 if (ma->ma_lmv_size > 0) {
 836                         if (reqbody->mbo_valid & OBD_MD_MEA)
 837                                 ma->ma_need |= MA_LMV;
 838                         else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA)
 839                                 ma->ma_need |= MA_LMV_DEF;
 840                 }
 841         } else {
 842                 ma->ma_lmm = buffer->lb_buf;
 843                 ma->ma_lmm_size = buffer->lb_len;
 844                 ma->ma_need = MA_INODE | MA_HSM;
 845                 if (ma->ma_lmm_size > 0)
 846                         ma->ma_need |= MA_LOV;
 847         }
 848
 849         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 850             reqbody->mbo_valid & OBD_MD_FLDIREA  &&
 851             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
 852                 /* get default stripe info for this dir. */
 853                 ma->ma_need |= MA_LOV_DEF;
 854         }
 855         ma->ma_need |= ma_need;
 856         if (ma->ma_need & MA_SOM)
 857                 ma->ma_som = &info->mti_u.som.data;
 858
 859         rc = mdt_attr_get_complex(info, o, ma);
 860         if (unlikely(rc)) {
 861                 CERROR("%s: getattr error for "DFID": rc = %d\n",
 862                        mdt_obd_name(info->mti_mdt),
 863                        PFID(mdt_object_fid(o)), rc);
 864                 RETURN(rc);
 865         }
 866
 867         /* if file is released, check if a restore is running */
 868         if ((ma->ma_valid & MA_HSM) && (ma->ma_hsm.mh_flags & HS_RELEASED) &&
 869             mdt_hsm_restore_is_running(info, mdt_object_fid(o))) {
 870                 repbody->mbo_t_state = MS_RESTORE;
 871                 repbody->mbo_valid |= OBD_MD_TSTATE;
 872         }
 873
 874         is_root = lu_fid_eq(mdt_object_fid(o), &info->mti_mdt->mdt_md_root_fid);
 875
 876         /* the Lustre protocol supposes to return default striping
 877          * on the user-visible root if explicitly requested */
 878         if ((ma->ma_valid & MA_LOV) == 0 && S_ISDIR(la->la_mode) &&
 879             (ma->ma_need & MA_LOV_DEF && is_root) && ma->ma_need & MA_LOV) {
 880                 struct lu_fid      rootfid;
 881                 struct mdt_object *root;
 882                 struct mdt_device *mdt = info->mti_mdt;
 883
 884                 rc = dt_root_get(env, mdt->mdt_bottom, &rootfid);
 885                 if (rc)
 886                         RETURN(rc);
 887                 root = mdt_object_find(env, mdt, &rootfid);
 888                 if (IS_ERR(root))
 889                         RETURN(PTR_ERR(root));
 890                 rc = mdt_stripe_get(info, root, ma, XATTR_NAME_LOV);
 891                 mdt_object_put(info->mti_env, root);
 892                 if (unlikely(rc)) {
 893                         CERROR("%s: getattr error for "DFID": rc = %d\n",
 894                                mdt_obd_name(info->mti_mdt),
 895                                PFID(mdt_object_fid(o)), rc);
 896                         RETURN(rc);
 897                 }
 898         }
 899
 900         if (likely(ma->ma_valid & MA_INODE))
 901                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
 902         else
 903                 RETURN(-EFAULT);
 904
 905         if (mdt_body_has_lov(la, reqbody)) {
 906                 if (ma->ma_valid & MA_LOV) {
 907                         LASSERT(ma->ma_lmm_size);
 908                         repbody->mbo_eadatasize = ma->ma_lmm_size;
 909                         if (S_ISDIR(la->la_mode))
 910                                 repbody->mbo_valid |= OBD_MD_FLDIREA;
 911                         else
 912                                 repbody->mbo_valid |= OBD_MD_FLEASIZE;
 913                         mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
 914                 }
 915                 if (ma->ma_valid & MA_LMV) {
 916                         /* Return -ENOTSUPP for old client */
 917                         if (!mdt_is_striped_client(req->rq_export))
 918                                 RETURN(-ENOTSUPP);
 919
 920                         LASSERT(S_ISDIR(la->la_mode));
 921                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
 922                         repbody->mbo_eadatasize = ma->ma_lmv_size;
 923                         repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
 924                 }
 925                 if (ma->ma_valid & MA_LMV_DEF) {
 926                         /* Return -ENOTSUPP for old client */
 927                         if (!mdt_is_striped_client(req->rq_export))
 928                                 RETURN(-ENOTSUPP);
 929                         LASSERT(S_ISDIR(la->la_mode));
 930                         repbody->mbo_eadatasize = ma->ma_lmv_size;
 931                         repbody->mbo_valid |= (OBD_MD_FLDIREA |
 932                                                OBD_MD_DEFAULT_MEA);
 933                 }
 934         } else if (S_ISLNK(la->la_mode) &&
 935                    reqbody->mbo_valid & OBD_MD_LINKNAME) {
 936                 buffer->lb_buf = ma->ma_lmm;
 937                 /* eadatasize from client includes NULL-terminator, so
 938                  * there is no need to read it */
 939                 buffer->lb_len = reqbody->mbo_eadatasize - 1;
 940                 rc = mo_readlink(env, next, buffer);
 941                 if (unlikely(rc <= 0)) {
 942                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
 943                                mdt_obd_name(info->mti_mdt),
 944                                PFID(mdt_object_fid(o)), rc);
 945                         rc = -EFAULT;
 946                 } else {
 947                         int print_limit = min_t(int, PAGE_CACHE_SIZE - 128, rc);
 948
 949                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
 950                                 rc -= 2;
 951                         repbody->mbo_valid |= OBD_MD_LINKNAME;
 952                         /* we need to report back size with NULL-terminator
 953                          * because client expects that */
 954                         repbody->mbo_eadatasize = rc + 1;
 955                         if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
 956                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
 957                                        "on "DFID ", expected %d\n",
 958                                        mdt_obd_name(info->mti_mdt),
 959                                        rc, PFID(mdt_object_fid(o)),
 960                                        reqbody->mbo_eadatasize - 1);
 961                         /* NULL terminate */
 962                         ((char *)ma->ma_lmm)[rc] = 0;
 963
 964                         /* If the total CDEBUG() size is larger than a page, it
 965                          * will print a warning to the console, avoid this by
 966                          * printing just the last part of the symlink. */
 967                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
 968                                print_limit < rc ? "..." : "", print_limit,
 969                                (char *)ma->ma_lmm + rc - print_limit, rc);
 970                         rc = 0;
 971                 }
 972         }
 973
 974         if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
 975                 repbody->mbo_max_cookiesize = 0;
 976                 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
 977                 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
 978                 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
 979                        "MAX_COOKIE to : %d:%d\n", repbody->mbo_max_mdsize,
 980                        repbody->mbo_max_cookiesize);
 981         }
 982
 983         if (exp_connect_rmtclient(info->mti_exp) &&
 984             reqbody->mbo_valid & OBD_MD_FLRMTPERM) {
 985                 void *buf = req_capsule_server_get(pill, &RMF_ACL);
 986
 987                 /* mdt_getattr_lock only */
 988                 rc = mdt_pack_remote_perm(info, o, buf);
 989                 if (rc) {
 990                         repbody->mbo_valid &= ~OBD_MD_FLRMTPERM;
 991                         repbody->mbo_aclsize = 0;
 992                         RETURN(rc);
 993                 } else {
 994                         repbody->mbo_valid |= OBD_MD_FLRMTPERM;
 995                         repbody->mbo_aclsize = sizeof(struct mdt_remote_perm);
 996                 }
 997         }
 998 #ifdef CONFIG_FS_POSIX_ACL
 999         else if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1000                  (reqbody->mbo_valid & OBD_MD_FLACL)) {
1001                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
1002                 buffer->lb_len = req_capsule_get_size(pill,
1003                                                       &RMF_ACL, RCL_SERVER);
1004                 if (buffer->lb_len > 0) {
1005                         rc = mo_xattr_get(env, next, buffer,
1006                                           XATTR_NAME_ACL_ACCESS);
1007                         if (rc < 0) {
1008                                 if (rc == -ENODATA) {
1009                                         repbody->mbo_aclsize = 0;
1010                                         repbody->mbo_valid |= OBD_MD_FLACL;
1011                                         rc = 0;
1012                                 } else if (rc == -EOPNOTSUPP) {
1013                                         rc = 0;
1014                                 } else {
1015                                         CERROR("%s: unable to read "DFID
1016                                                " ACL: rc = %d\n",
1017                                                mdt_obd_name(info->mti_mdt),
1018                                                PFID(mdt_object_fid(o)), rc);
1019                                 }
1020                         } else {
1021                                 repbody->mbo_aclsize = rc;
1022                                 repbody->mbo_valid |= OBD_MD_FLACL;
1023                                 rc = 0;
1024                         }
1025                 }
1026         }
1027 #endif
1028
1029         if (reqbody->mbo_valid & OBD_MD_FLMDSCAPA &&
1030             info->mti_mdt->mdt_lut.lut_mds_capa &&
1031             exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) {
1032                 struct lustre_capa *capa;
1033
1034                 capa = req_capsule_server_get(pill, &RMF_CAPA1);
1035                 LASSERT(capa);
1036                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
1037                 rc = mo_capa_get(env, next, capa, 0);
1038                 if (rc)
1039                         RETURN(rc);
1040                 repbody->mbo_valid |= OBD_MD_FLMDSCAPA;
1041         }
1042
1043 out:
1044         if (rc == 0)
1045                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
1046
1047         RETURN(rc);
1048 }
1049
1050 static int mdt_renew_capa(struct mdt_thread_info *info)
1051 {
1052         struct mdt_object  *obj = info->mti_object;
1053         struct mdt_body    *body;
1054         struct lustre_capa *capa, *c;
1055         int rc;
1056         ENTRY;
1057
1058         /* if object doesn't exist, or server has disabled capability,
1059          * return directly, client will find body->valid OBD_MD_FLOSSCAPA
1060          * flag not set.
1061          */
1062         if (!obj || !info->mti_mdt->mdt_lut.lut_oss_capa ||
1063             !(exp_connect_flags(info->mti_exp) & OBD_CONNECT_OSS_CAPA))
1064                 RETURN(0);
1065
1066         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1067         LASSERT(body != NULL);
1068
1069         c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
1070         LASSERT(c);
1071
1072         capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
1073         LASSERT(capa);
1074
1075         *capa = *c;
1076         rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
1077         if (rc == 0)
1078                 body->mbo_valid |= OBD_MD_FLOSSCAPA;
1079         RETURN(rc);
1080 }
1081
1082 static int mdt_getattr(struct tgt_session_info *tsi)
1083 {
1084         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1085         struct mdt_object       *obj = info->mti_object;
1086         struct req_capsule      *pill = info->mti_pill;
1087         struct mdt_body         *reqbody;
1088         struct mdt_body         *repbody;
1089         int rc, rc2;
1090         ENTRY;
1091
1092         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1093         LASSERT(reqbody);
1094
1095         if (reqbody->mbo_valid & OBD_MD_FLOSSCAPA) {
1096                 rc = req_capsule_server_pack(pill);
1097                 if (unlikely(rc))
1098                         RETURN(err_serious(rc));
1099                 rc = mdt_renew_capa(info);
1100                 GOTO(out_shrink, rc);
1101         }
1102
1103         LASSERT(obj != NULL);
1104         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1105
1106         /* Unlike intent case where we need to pre-fill out buffers early on
1107          * in intent policy for ldlm reasons, here we can have a much better
1108          * guess at EA size by just reading it from disk.
1109          * Exceptions are readdir and (missing) directory striping */
1110         /* Readlink */
1111         if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1112                 /* No easy way to know how long is the symlink, but it cannot
1113                  * be more than PATH_MAX, so we allocate +1 */
1114                 rc = PATH_MAX + 1;
1115
1116         /* A special case for fs ROOT: getattr there might fetch
1117          * default EA for entire fs, not just for this dir!
1118          */
1119         } else if (lu_fid_eq(mdt_object_fid(obj),
1120                              &info->mti_mdt->mdt_md_root_fid) &&
1121                    (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1122                    (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1123                                                                  MDS_GETATTR)) {
1124                 /* Should the default strping be bigger, mdt_fix_reply
1125                  * will reallocate */
1126                 rc = DEF_REP_MD_SIZE;
1127         } else {
1128                 /* Read the actual EA size from disk */
1129                 rc = mdt_attr_get_eabuf_size(info, obj);
1130         }
1131
1132         if (rc < 0)
1133                 GOTO(out_shrink, rc);
1134
1135         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1136
1137         rc = req_capsule_server_pack(pill);
1138         if (unlikely(rc != 0))
1139                 GOTO(out, rc = err_serious(rc));
1140
1141         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1142         LASSERT(repbody != NULL);
1143         repbody->mbo_eadatasize = 0;
1144         repbody->mbo_aclsize = 0;
1145
1146         if (reqbody->mbo_valid & OBD_MD_FLRMTPERM)
1147                 rc = mdt_init_ucred(info, reqbody);
1148         else
1149                 rc = mdt_check_ucred(info);
1150         if (unlikely(rc))
1151                 GOTO(out_shrink, rc);
1152
1153         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1154
1155         /*
1156          * Don't check capability at all, because rename might getattr for
1157          * remote obj, and at that time no capability is available.
1158          */
1159         mdt_set_capainfo(info, 1, &reqbody->mbo_fid1, BYPASS_CAPA);
1160         rc = mdt_getattr_internal(info, obj, 0);
1161         if (reqbody->mbo_valid & OBD_MD_FLRMTPERM)
1162                 mdt_exit_ucred(info);
1163         EXIT;
1164 out_shrink:
1165         mdt_client_compatibility(info);
1166         rc2 = mdt_fix_reply(info);
1167         if (rc == 0)
1168                 rc = rc2;
1169 out:
1170         mdt_thread_info_fini(info);
1171         return rc;
1172 }
1173
1174 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1175 {
1176         struct mdt_thread_info  *info;
1177         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1178         struct obd_export       *exp = req->rq_export;
1179         struct mdt_object       *o1, *o2, *o;
1180         struct mdt_lock_handle  *lh1, *lh2;
1181         struct mdc_swap_layouts *msl;
1182         int                      rc;
1183         ENTRY;
1184
1185         /* client does not support layout lock, so layout swaping
1186          * is disabled.
1187          * FIXME: there is a problem for old clients which don't support
1188          * layout lock yet. If those clients have already opened the file
1189          * they won't be notified at all so that old layout may still be
1190          * used to do IO. This can be fixed after file release is landed by
1191          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1192         if (!exp_connect_layout(exp))
1193                 RETURN(-EOPNOTSUPP);
1194
1195         info = tsi2mdt_info(tsi);
1196
1197         if (info->mti_dlm_req != NULL)
1198                 ldlm_request_cancel(req, info->mti_dlm_req, 0);
1199
1200         if (req_capsule_get_size(info->mti_pill, &RMF_CAPA1, RCL_CLIENT))
1201                 mdt_set_capainfo(info, 0, &info->mti_body->mbo_fid1,
1202                                  req_capsule_client_get(info->mti_pill,
1203                                                         &RMF_CAPA1));
1204
1205         if (req_capsule_get_size(info->mti_pill, &RMF_CAPA2, RCL_CLIENT))
1206                 mdt_set_capainfo(info, 1, &info->mti_body->mbo_fid2,
1207                                  req_capsule_client_get(info->mti_pill,
1208                                                         &RMF_CAPA2));
1209
1210         o1 = info->mti_object;
1211         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1212                                 &info->mti_body->mbo_fid2);
1213         if (IS_ERR(o))
1214                 GOTO(out, rc = PTR_ERR(o));
1215
1216         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1217                 GOTO(put, rc = -ENOENT);
1218
1219         rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1220         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1221                 GOTO(put, rc);
1222
1223         if (rc < 0)
1224                 swap(o1, o2);
1225
1226         /* permission check. Make sure the calling process having permission
1227          * to write both files. */
1228         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1229                                 MAY_WRITE);
1230         if (rc < 0)
1231                 GOTO(put, rc);
1232
1233         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1234                                 MAY_WRITE);
1235         if (rc < 0)
1236                 GOTO(put, rc);
1237
1238         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1239         if (msl == NULL)
1240                 GOTO(put, rc = -EPROTO);
1241
1242         lh1 = &info->mti_lh[MDT_LH_NEW];
1243         mdt_lock_reg_init(lh1, LCK_EX);
1244         lh2 = &info->mti_lh[MDT_LH_OLD];
1245         mdt_lock_reg_init(lh2, LCK_EX);
1246
1247         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1248                              MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK);
1249         if (rc < 0)
1250                 GOTO(put, rc);
1251
1252         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1253                              MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK);
1254         if (rc < 0)
1255                 GOTO(unlock1, rc);
1256
1257         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1258                              mdt_object_child(o2), msl->msl_flags);
1259         GOTO(unlock2, rc);
1260 unlock2:
1261         mdt_object_unlock(info, o2, lh2, rc);
1262 unlock1:
1263         mdt_object_unlock(info, o1, lh1, rc);
1264 put:
1265         mdt_object_put(info->mti_env, o);
1266 out:
1267         mdt_thread_info_fini(info);
1268         RETURN(rc);
1269 }
1270
1271 static int mdt_raw_lookup(struct mdt_thread_info *info,
1272                           struct mdt_object *parent,
1273                           const struct lu_name *lname,
1274                           struct ldlm_reply *ldlm_rep)
1275 {
1276         struct md_object *next = mdt_object_child(info->mti_object);
1277         const struct mdt_body *reqbody = info->mti_body;
1278         struct lu_fid *child_fid = &info->mti_tmp_fid1;
1279         struct mdt_body *repbody;
1280         int rc;
1281         ENTRY;
1282
1283         if (reqbody->mbo_valid != OBD_MD_FLID)
1284                 RETURN(0);
1285
1286         LASSERT(!info->mti_cross_ref);
1287
1288         /* Only got the fid of this obj by name */
1289         fid_zero(child_fid);
1290         rc = mdo_lookup(info->mti_env, next, lname, child_fid,
1291                         &info->mti_spec);
1292 #if 0
1293         /* XXX is raw_lookup possible as intent operation? */
1294         if (rc != 0) {
1295                 if (rc == -ENOENT)
1296                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1297                 RETURN(rc);
1298         } else
1299                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1300
1301         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1302 #endif
1303         if (rc == 0) {
1304                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1305                 repbody->mbo_fid1 = *child_fid;
1306                 repbody->mbo_valid = OBD_MD_FLID;
1307         }
1308         RETURN(1);
1309 }
1310
1311 /*
1312  * UPDATE lock should be taken against parent, and be release before exit;
1313  * child_bits lock should be taken against child, and be returned back:
1314  *            (1)normal request should release the child lock;
1315  *            (2)intent request will grant the lock to client.
1316  */
1317 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1318                                  struct mdt_lock_handle *lhc,
1319                                  __u64 child_bits,
1320                                  struct ldlm_reply *ldlm_rep)
1321 {
1322         struct ptlrpc_request  *req       = mdt_info_req(info);
1323         struct mdt_body        *reqbody   = NULL;
1324         struct mdt_object      *parent    = info->mti_object;
1325         struct mdt_object      *child;
1326         struct md_object       *next      = mdt_object_child(parent);
1327         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1328         struct lu_name         *lname     = NULL;
1329         struct mdt_lock_handle *lhp       = NULL;
1330         struct ldlm_lock       *lock;
1331         bool                    is_resent;
1332         bool                    try_layout;
1333         int                     ma_need = 0;
1334         int                     rc;
1335         ENTRY;
1336
1337         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1338         LASSERT(ergo(is_resent,
1339                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1340
1341         LASSERT(parent != NULL);
1342
1343         if (info->mti_cross_ref) {
1344                 /* Only getattr on the child. Parent is on another node. */
1345                 mdt_set_disposition(info, ldlm_rep,
1346                                     DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1347                 child = parent;
1348                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1349                        "ldlm_rep = %p\n",
1350                        PFID(mdt_object_fid(child)), ldlm_rep);
1351
1352                 rc = mdt_check_resent_lock(info, child, lhc);
1353                 if (rc < 0) {
1354                         RETURN(-EPROTO);
1355                 } else if (rc > 0) {
1356                         mdt_lock_handle_init(lhc);
1357                         mdt_lock_reg_init(lhc, LCK_PR);
1358
1359                         /*
1360                          * Object's name is on another MDS, no lookup or layout
1361                          * lock is needed here but update lock is.
1362                          */
1363                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
1364                                         MDS_INODELOCK_LAYOUT);
1365                         child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1366
1367                         rc = mdt_object_lock(info, child, lhc, child_bits,
1368                                              MDT_LOCAL_LOCK);
1369                         if (rc < 0)
1370                                 RETURN(rc);
1371                 }
1372
1373                 /* Finally, we can get attr for child. */
1374                 if (!mdt_object_exists(child)) {
1375                         LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1376                                         &child->mot_obj,
1377                                         "remote object doesn't exist.\n");
1378                         mdt_object_unlock(info, child, lhc, 1);
1379                         RETURN(-ENOENT);
1380                 }
1381
1382                 mdt_set_capainfo(info, 0, mdt_object_fid(child), BYPASS_CAPA);
1383                 rc = mdt_getattr_internal(info, child, 0);
1384                 if (unlikely(rc != 0))
1385                         mdt_object_unlock(info, child, lhc, 1);
1386
1387                 RETURN(rc);
1388         }
1389
1390         lname = &info->mti_name;
1391         mdt_name_unpack(info->mti_pill, &RMF_NAME, lname, MNF_FIX_ANON);
1392
1393         if (lu_name_is_valid(lname)) {
1394                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1395                        "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1396                        PNAME(lname), ldlm_rep);
1397         } else {
1398                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1399                 if (unlikely(reqbody == NULL))
1400                         RETURN(err_serious(-EPROTO));
1401
1402                 *child_fid = reqbody->mbo_fid2;
1403
1404                 if (unlikely(!fid_is_sane(child_fid)))
1405                         RETURN(err_serious(-EINVAL));
1406
1407                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1408                        "ldlm_rep = %p\n",
1409                        PFID(mdt_object_fid(parent)),
1410                        PFID(&reqbody->mbo_fid2), ldlm_rep);
1411         }
1412
1413         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1414
1415         if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1416                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1417                                 &parent->mot_obj,
1418                                 "Parent doesn't exist!\n");
1419                 RETURN(-ESTALE);
1420         }
1421
1422         if (mdt_object_remote(parent)) {
1423                 CERROR("%s: parent "DFID" is on remote target\n",
1424                        mdt_obd_name(info->mti_mdt),
1425                        PFID(mdt_object_fid(parent)));
1426                 RETURN(-EIO);
1427         }
1428
1429         if (lu_name_is_valid(lname)) {
1430                 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1431                 if (rc != 0) {
1432                         if (rc > 0)
1433                                 rc = 0;
1434                         RETURN(rc);
1435                 }
1436
1437                 /* step 1: lock parent only if parent is a directory */
1438                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1439                         lhp = &info->mti_lh[MDT_LH_PARENT];
1440                         mdt_lock_pdo_init(lhp, LCK_PR, lname);
1441                         rc = mdt_object_lock(info, parent, lhp,
1442                                              MDS_INODELOCK_UPDATE,
1443                                              MDT_LOCAL_LOCK);
1444                         if (unlikely(rc != 0))
1445                                 RETURN(rc);
1446                 }
1447
1448                 /* step 2: lookup child's fid by name */
1449                 fid_zero(child_fid);
1450                 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
1451                                 &info->mti_spec);
1452                 if (rc == -ENOENT)
1453                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1454
1455                 if (rc != 0)
1456                         GOTO(out_parent, rc);
1457         }
1458
1459         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1460
1461         /*
1462          *step 3: find the child object by fid & lock it.
1463          *        regardless if it is local or remote.
1464          *
1465          *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1466          *      set parent dir fid the same as child fid in getattr by fid case
1467          *      we should not lu_object_find() the object again, could lead
1468          *      to hung if there is a concurrent unlink destroyed the object.
1469          */
1470         if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1471                 mdt_object_get(info->mti_env, parent);
1472                 child = parent;
1473         } else {
1474                 child = mdt_object_find(info->mti_env, info->mti_mdt,
1475                                         child_fid);
1476         }
1477
1478         if (unlikely(IS_ERR(child)))
1479                 GOTO(out_parent, rc = PTR_ERR(child));
1480
1481         rc = mdt_check_resent_lock(info, child, lhc);
1482         if (rc < 0) {
1483                 GOTO(out_child, rc);
1484         } else if (rc > 0) {
1485                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout*2);
1486                 mdt_lock_handle_init(lhc);
1487                 mdt_lock_reg_init(lhc, LCK_PR);
1488                 try_layout = false;
1489
1490                 if (!mdt_object_exists(child)) {
1491                         LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1492                                         &child->mot_obj,
1493                                         "Object doesn't exist!\n");
1494                         GOTO(out_child, rc = -ENOENT);
1495                 }
1496
1497                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1498                       mdt_object_exists(child) && !mdt_object_remote(child)) {
1499                         struct md_attr *ma = &info->mti_attr;
1500
1501                         ma->ma_valid = 0;
1502                         ma->ma_need = MA_INODE;
1503                         rc = mdt_attr_get_complex(info, child, ma);
1504                         if (unlikely(rc != 0))
1505                                 GOTO(out_child, rc);
1506
1507                         /* If the file has not been changed for some time, we
1508                          * return not only a LOOKUP lock, but also an UPDATE
1509                          * lock and this might save us RPC on later STAT. For
1510                          * directories, it also let negative dentry cache start
1511                          * working for this dir. */
1512                         if (ma->ma_valid & MA_INODE &&
1513                             ma->ma_attr.la_valid & LA_CTIME &&
1514                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1515                                 ma->ma_attr.la_ctime < cfs_time_current_sec())
1516                                 child_bits |= MDS_INODELOCK_UPDATE;
1517                 }
1518
1519                 /* layout lock must be granted in a best-effort way
1520                  * for IT operations */
1521                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1522                 if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1523                     exp_connect_layout(info->mti_exp) &&
1524                     S_ISREG(lu_object_attr(&child->mot_obj)) &&
1525                     !mdt_object_remote(child) && ldlm_rep != NULL) {
1526                         /* try to grant layout lock for regular file. */
1527                         try_layout = true;
1528                 }
1529
1530                 rc = 0;
1531                 if (try_layout) {
1532                         child_bits |= MDS_INODELOCK_LAYOUT;
1533                         /* try layout lock, it may fail to be granted due to
1534                          * contention at LOOKUP or UPDATE */
1535                         if (!mdt_object_lock_try(info, child, lhc, child_bits,
1536                                                  MDT_CROSS_LOCK)) {
1537                                 child_bits &= ~MDS_INODELOCK_LAYOUT;
1538                                 LASSERT(child_bits != 0);
1539                                 rc = mdt_object_lock(info, child, lhc,
1540                                                 child_bits, MDT_CROSS_LOCK);
1541                         } else {
1542                                 ma_need |= MA_LOV;
1543                         }
1544                 } else {
1545                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1546                          * client will enqueue the lock to the remote MDT */
1547                         if (mdt_object_remote(child))
1548                                 child_bits &= ~MDS_INODELOCK_UPDATE;
1549                         rc = mdt_object_lock(info, child, lhc, child_bits,
1550                                                 MDT_CROSS_LOCK);
1551                 }
1552                 if (unlikely(rc != 0))
1553                         GOTO(out_child, rc);
1554         }
1555
1556         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1557         /* Get MA_SOM attributes if update lock is given. */
1558         if (lock &&
1559             lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_UPDATE &&
1560             S_ISREG(lu_object_attr(&mdt_object_child(child)->mo_lu)))
1561                 ma_need |= MA_SOM;
1562
1563         /* finally, we can get attr for child. */
1564         mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
1565         rc = mdt_getattr_internal(info, child, ma_need);
1566         if (unlikely(rc != 0)) {
1567                 mdt_object_unlock(info, child, lhc, 1);
1568         } else if (lock) {
1569                 /* Debugging code. */
1570                 LDLM_DEBUG(lock, "Returning lock to client");
1571                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1572                                          &lock->l_resource->lr_name),
1573                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1574                          PLDLMRES(lock->l_resource),
1575                          PFID(mdt_object_fid(child)));
1576                 if (mdt_object_exists(child) && !mdt_object_remote(child))
1577                         mdt_pack_size2body(info, child);
1578         }
1579         if (lock)
1580                 LDLM_LOCK_PUT(lock);
1581
1582         EXIT;
1583 out_child:
1584         mdt_object_put(info->mti_env, child);
1585 out_parent:
1586         if (lhp)
1587                 mdt_object_unlock(info, parent, lhp, 1);
1588         return rc;
1589 }
1590
1591 /* normal handler: should release the child lock */
1592 static int mdt_getattr_name(struct tgt_session_info *tsi)
1593 {
1594         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1595         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1596         struct mdt_body        *reqbody;
1597         struct mdt_body        *repbody;
1598         int rc, rc2;
1599         ENTRY;
1600
1601         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1602         LASSERT(reqbody != NULL);
1603         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1604         LASSERT(repbody != NULL);
1605
1606         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1607         repbody->mbo_eadatasize = 0;
1608         repbody->mbo_aclsize = 0;
1609
1610         rc = mdt_init_ucred(info, reqbody);
1611         if (unlikely(rc))
1612                 GOTO(out_shrink, rc);
1613
1614         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1615         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1616                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1617                 lhc->mlh_reg_lh.cookie = 0;
1618         }
1619         mdt_exit_ucred(info);
1620         EXIT;
1621 out_shrink:
1622         mdt_client_compatibility(info);
1623         rc2 = mdt_fix_reply(info);
1624         if (rc == 0)
1625                 rc = rc2;
1626         mdt_thread_info_fini(info);
1627         return rc;
1628 }
1629
1630 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1631                          void *karg, void *uarg);
1632
1633 static int mdt_set_info(struct tgt_session_info *tsi)
1634 {
1635         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1636         char                    *key;
1637         void                    *val;
1638         int                      keylen, vallen, rc = 0;
1639
1640         ENTRY;
1641
1642         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
1643         if (key == NULL) {
1644                 DEBUG_REQ(D_HA, req, "no set_info key");
1645                 RETURN(err_serious(-EFAULT));
1646         }
1647
1648         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
1649                                       RCL_CLIENT);
1650
1651         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
1652         if (val == NULL) {
1653                 DEBUG_REQ(D_HA, req, "no set_info val");
1654                 RETURN(err_serious(-EFAULT));
1655         }
1656
1657         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
1658                                       RCL_CLIENT);
1659
1660         /* Swab any part of val you need to here */
1661         if (KEY_IS(KEY_READ_ONLY)) {
1662                 spin_lock(&req->rq_export->exp_lock);
1663                 if (*(__u32 *)val)
1664                         *exp_connect_flags_ptr(req->rq_export) |=
1665                                 OBD_CONNECT_RDONLY;
1666                 else
1667                         *exp_connect_flags_ptr(req->rq_export) &=
1668                                 ~OBD_CONNECT_RDONLY;
1669                 spin_unlock(&req->rq_export->exp_lock);
1670         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1671                 struct changelog_setinfo *cs = val;
1672
1673                 if (vallen != sizeof(*cs)) {
1674                         CERROR("%s: bad changelog_clear setinfo size %d\n",
1675                                tgt_name(tsi->tsi_tgt), vallen);
1676                         RETURN(-EINVAL);
1677                 }
1678                 if (ptlrpc_req_need_swab(req)) {
1679                         __swab64s(&cs->cs_recno);
1680                         __swab32s(&cs->cs_id);
1681                 }
1682
1683                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
1684                                    vallen, val, NULL);
1685         } else {
1686                 RETURN(-EINVAL);
1687         }
1688         RETURN(rc);
1689 }
1690
1691 static int mdt_readpage(struct tgt_session_info *tsi)
1692 {
1693         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
1694         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
1695         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
1696         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
1697         struct mdt_body         *repbody;
1698         int                      rc;
1699         int                      i;
1700
1701         ENTRY;
1702
1703         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1704                 RETURN(err_serious(-ENOMEM));
1705
1706         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
1707         if (repbody == NULL || reqbody == NULL)
1708                 RETURN(err_serious(-EFAULT));
1709
1710         /*
1711          * prepare @rdpg before calling lower layers and transfer itself. Here
1712          * reqbody->size contains offset of where to start to read and
1713          * reqbody->nlink contains number bytes to read.
1714          */
1715         rdpg->rp_hash = reqbody->mbo_size;
1716         if (rdpg->rp_hash != reqbody->mbo_size) {
1717                 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1718                        rdpg->rp_hash, reqbody->mbo_size);
1719                 RETURN(-EFAULT);
1720         }
1721
1722         rdpg->rp_attrs = reqbody->mbo_mode;
1723         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
1724                 rdpg->rp_attrs |= LUDA_64BITHASH;
1725         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
1726                                 exp_max_brw_size(tsi->tsi_exp));
1727         rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE - 1) >>
1728                           PAGE_CACHE_SHIFT;
1729         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1730         if (rdpg->rp_pages == NULL)
1731                 RETURN(-ENOMEM);
1732
1733         for (i = 0; i < rdpg->rp_npages; ++i) {
1734                 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1735                 if (rdpg->rp_pages[i] == NULL)
1736                         GOTO(free_rdpg, rc = -ENOMEM);
1737         }
1738
1739         /* call lower layers to fill allocated pages with directory data */
1740         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
1741         if (rc < 0)
1742                 GOTO(free_rdpg, rc);
1743
1744         /* send pages to client */
1745         rc = tgt_sendpage(tsi, rdpg, rc);
1746
1747         EXIT;
1748 free_rdpg:
1749
1750         for (i = 0; i < rdpg->rp_npages; i++)
1751                 if (rdpg->rp_pages[i] != NULL)
1752                         __free_page(rdpg->rp_pages[i]);
1753         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1754
1755         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1756                 RETURN(0);
1757
1758         return rc;
1759 }
1760
1761 static int mdt_reint_internal(struct mdt_thread_info *info,
1762                               struct mdt_lock_handle *lhc,
1763                               __u32 op)
1764 {
1765         struct req_capsule      *pill = info->mti_pill;
1766         struct mdt_body         *repbody;
1767         int                      rc = 0, rc2;
1768
1769         ENTRY;
1770
1771         rc = mdt_reint_unpack(info, op);
1772         if (rc != 0) {
1773                 CERROR("Can't unpack reint, rc %d\n", rc);
1774                 RETURN(err_serious(rc));
1775         }
1776
1777         /* for replay (no_create) lmm is not needed, client has it already */
1778         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1779                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1780                                      DEF_REP_MD_SIZE);
1781
1782         /* llog cookies are always 0, the field is kept for compatibility */
1783         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1784                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
1785
1786         rc = req_capsule_server_pack(pill);
1787         if (rc != 0) {
1788                 CERROR("Can't pack response, rc %d\n", rc);
1789                 RETURN(err_serious(rc));
1790         }
1791
1792         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1793                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1794                 LASSERT(repbody);
1795                 repbody->mbo_eadatasize = 0;
1796                 repbody->mbo_aclsize = 0;
1797         }
1798
1799         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1800
1801         /* for replay no cookkie / lmm need, because client have this already */
1802         if (info->mti_spec.no_create)
1803                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1804                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1805
1806         rc = mdt_init_ucred_reint(info);
1807         if (rc)
1808                 GOTO(out_shrink, rc);
1809
1810         rc = mdt_fix_attr_ucred(info, op);
1811         if (rc != 0)
1812                 GOTO(out_ucred, rc = err_serious(rc));
1813
1814         if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1815                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
1816                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1817                 GOTO(out_ucred, rc);
1818         }
1819         rc = mdt_reint_rec(info, lhc);
1820         EXIT;
1821 out_ucred:
1822         mdt_exit_ucred(info);
1823 out_shrink:
1824         mdt_client_compatibility(info);
1825         rc2 = mdt_fix_reply(info);
1826         if (rc == 0)
1827                 rc = rc2;
1828         return rc;
1829 }
1830
1831 static long mdt_reint_opcode(struct ptlrpc_request *req,
1832                              const struct req_format **fmt)
1833 {
1834         struct mdt_device       *mdt;
1835         struct mdt_rec_reint    *rec;
1836         long                     opc;
1837
1838         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
1839         if (rec != NULL) {
1840                 opc = rec->rr_opcode;
1841                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
1842                 if (opc < REINT_MAX && fmt[opc] != NULL)
1843                         req_capsule_extend(&req->rq_pill, fmt[opc]);
1844                 else {
1845                         mdt = mdt_exp2dev(req->rq_export);
1846                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
1847                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
1848                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
1849                         opc = err_serious(-EFAULT);
1850                 }
1851         } else {
1852                 opc = err_serious(-EFAULT);
1853         }
1854         return opc;
1855 }
1856
1857 static int mdt_reint(struct tgt_session_info *tsi)
1858 {
1859         long opc;
1860         int  rc;
1861         static const struct req_format *reint_fmts[REINT_MAX] = {
1862                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
1863                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
1864                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
1865                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
1866                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
1867                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
1868                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
1869                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
1870                 [REINT_MIGRATE]  = &RQF_MDS_REINT_RENAME
1871         };
1872
1873         ENTRY;
1874
1875         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
1876         if (opc >= 0) {
1877                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1878                 /*
1879                  * No lock possible here from client to pass it to reint code
1880                  * path.
1881                  */
1882                 rc = mdt_reint_internal(info, NULL, opc);
1883                 mdt_thread_info_fini(info);
1884         } else {
1885                 rc = opc;
1886         }
1887
1888         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1889         RETURN(rc);
1890 }
1891
1892 /* this should sync the whole device */
1893 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1894 {
1895         struct dt_device *dt = mdt->mdt_bottom;
1896         int rc;
1897         ENTRY;
1898
1899         rc = dt->dd_ops->dt_sync(env, dt);
1900         RETURN(rc);
1901 }
1902
1903 /* this should sync this object */
1904 static int mdt_object_sync(struct mdt_thread_info *info)
1905 {
1906         struct md_object *next;
1907         int rc;
1908         ENTRY;
1909
1910         if (!mdt_object_exists(info->mti_object)) {
1911                 CWARN("%s: non existing object "DFID": rc = %d\n",
1912                       mdt_obd_name(info->mti_mdt),
1913                       PFID(mdt_object_fid(info->mti_object)), -ESTALE);
1914                 RETURN(-ESTALE);
1915         }
1916         next = mdt_object_child(info->mti_object);
1917         rc = mo_object_sync(info->mti_env, next);
1918
1919         RETURN(rc);
1920 }
1921
1922 static int mdt_sync(struct tgt_session_info *tsi)
1923 {
1924         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1925         struct req_capsule      *pill = tsi->tsi_pill;
1926         struct mdt_body         *body;
1927         int                      rc;
1928
1929         ENTRY;
1930
1931         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1932                 RETURN(err_serious(-ENOMEM));
1933
1934         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
1935                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
1936         } else {
1937                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1938
1939                 /* sync an object */
1940                 rc = mdt_object_sync(info);
1941                 if (rc == 0) {
1942                         const struct lu_fid *fid;
1943                         struct lu_attr *la = &info->mti_attr.ma_attr;
1944
1945                         info->mti_attr.ma_need = MA_INODE;
1946                         info->mti_attr.ma_valid = 0;
1947                         rc = mdt_attr_get_complex(info, info->mti_object,
1948                                                   &info->mti_attr);
1949                         if (rc == 0) {
1950                                 body = req_capsule_server_get(pill,
1951                                                               &RMF_MDT_BODY);
1952                                 fid = mdt_object_fid(info->mti_object);
1953                                 mdt_pack_attr2body(info, body, la, fid);
1954                         }
1955                 }
1956                 mdt_thread_info_fini(info);
1957         }
1958         if (rc == 0)
1959                 mdt_counter_incr(req, LPROC_MDT_SYNC);
1960
1961         RETURN(rc);
1962 }
1963
1964 /*
1965  * Handle quota control requests to consult current usage/limit, but also
1966  * to configure quota enforcement
1967  */
1968 static int mdt_quotactl(struct tgt_session_info *tsi)
1969 {
1970         struct obd_export       *exp  = tsi->tsi_exp;
1971         struct req_capsule      *pill = tsi->tsi_pill;
1972         struct obd_quotactl     *oqctl, *repoqc;
1973         int                      id, rc;
1974         struct mdt_device       *mdt = mdt_exp2dev(exp);
1975         struct lu_device        *qmt = mdt->mdt_qmt_dev;
1976         ENTRY;
1977
1978         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1979         if (oqctl == NULL)
1980                 RETURN(err_serious(-EPROTO));
1981
1982         rc = req_capsule_server_pack(pill);
1983         if (rc)
1984                 RETURN(err_serious(rc));
1985
1986         switch (oqctl->qc_cmd) {
1987         case Q_QUOTACHECK:
1988         case LUSTRE_Q_INVALIDATE:
1989         case LUSTRE_Q_FINVALIDATE:
1990         case Q_QUOTAON:
1991         case Q_QUOTAOFF:
1992         case Q_INITQUOTA:
1993                 /* deprecated, not used any more */
1994                 RETURN(-EOPNOTSUPP);
1995                 /* master quotactl */
1996         case Q_GETINFO:
1997         case Q_SETINFO:
1998         case Q_SETQUOTA:
1999         case Q_GETQUOTA:
2000                 if (qmt == NULL)
2001                         RETURN(-EOPNOTSUPP);
2002                 /* slave quotactl */
2003         case Q_GETOINFO:
2004         case Q_GETOQUOTA:
2005                 break;
2006         default:
2007                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2008                 RETURN(-EFAULT);
2009         }
2010
2011         /* map uid/gid for remote client */
2012         id = oqctl->qc_id;
2013         if (exp_connect_rmtclient(exp)) {
2014                 struct lustre_idmap_table *idmap;
2015
2016                 idmap = exp->exp_mdt_data.med_idmap;
2017
2018                 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
2019                              oqctl->qc_cmd != Q_GETINFO))
2020                         RETURN(-EPERM);
2021
2022                 if (oqctl->qc_type == USRQUOTA)
2023                         id = lustre_idmap_lookup_uid(NULL, idmap, 0,
2024                                                      oqctl->qc_id);
2025                 else if (oqctl->qc_type == GRPQUOTA)
2026                         id = lustre_idmap_lookup_gid(NULL, idmap, 0,
2027                                                      oqctl->qc_id);
2028                 else
2029                         RETURN(-EINVAL);
2030
2031                 if (id == CFS_IDMAP_NOTFOUND) {
2032                         CDEBUG(D_QUOTA, "no mapping for id %u\n", oqctl->qc_id);
2033                         RETURN(-EACCES);
2034                 }
2035         }
2036
2037         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2038         if (repoqc == NULL)
2039                 RETURN(err_serious(-EFAULT));
2040
2041         if (oqctl->qc_id != id)
2042                 swap(oqctl->qc_id, id);
2043
2044         switch (oqctl->qc_cmd) {
2045
2046         case Q_GETINFO:
2047         case Q_SETINFO:
2048         case Q_SETQUOTA:
2049         case Q_GETQUOTA:
2050                 /* forward quotactl request to QMT */
2051                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2052                 break;
2053
2054         case Q_GETOINFO:
2055         case Q_GETOQUOTA:
2056                 /* slave quotactl */
2057                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2058                                    oqctl);
2059                 break;
2060
2061         default:
2062                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2063                 RETURN(-EFAULT);
2064         }
2065
2066         if (oqctl->qc_id != id)
2067                 swap(oqctl->qc_id, id);
2068
2069         *repoqc = *oqctl;
2070         RETURN(rc);
2071 }
2072
2073 /** clone llog ctxt from child (mdd)
2074  * This allows remote llog (replicator) access.
2075  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2076  * context was originally set up, or we can handle them directly.
2077  * I choose the latter, but that means I need any llog
2078  * contexts set up by child to be accessable by the mdt.  So we clone the
2079  * context into our context list here.
2080  */
2081 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2082                                int idx)
2083 {
2084         struct md_device  *next = mdt->mdt_child;
2085         struct llog_ctxt *ctxt;
2086         int rc;
2087
2088         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2089                 return 0;
2090
2091         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2092         if (rc || ctxt == NULL) {
2093                 return 0;
2094         }
2095
2096         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2097         if (rc)
2098                 CERROR("Can't set mdt ctxt %d\n", rc);
2099
2100         return rc;
2101 }
2102
2103 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2104                                  struct mdt_device *mdt, int idx)
2105 {
2106         struct llog_ctxt *ctxt;
2107
2108         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2109         if (ctxt == NULL)
2110                 return 0;
2111         /* Put once for the get we just did, and once for the clone */
2112         llog_ctxt_put(ctxt);
2113         llog_ctxt_put(ctxt);
2114         return 0;
2115 }
2116
2117 /*
2118  * sec context handlers
2119  */
2120 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2121 {
2122         int rc;
2123
2124         rc = mdt_handle_idmap(tsi);
2125         if (unlikely(rc)) {
2126                 struct ptlrpc_request   *req = tgt_ses_req(tsi);
2127                 __u32                    opc;
2128
2129                 opc = lustre_msg_get_opc(req->rq_reqmsg);
2130                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2131                         sptlrpc_svc_ctx_invalidate(req);
2132         }
2133
2134         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2135
2136         return rc;
2137 }
2138
2139 /*
2140  * quota request handlers
2141  */
2142 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2143 {
2144         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
2145         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2146         int                      rc;
2147         ENTRY;
2148
2149         if (qmt == NULL)
2150                 RETURN(err_serious(-EOPNOTSUPP));
2151
2152         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2153         RETURN(rc);
2154 }
2155
2156 struct mdt_object *mdt_object_new(const struct lu_env *env,
2157                                   struct mdt_device *d,
2158                                   const struct lu_fid *f)
2159 {
2160         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2161         struct lu_object *o;
2162         struct mdt_object *m;
2163         ENTRY;
2164
2165         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2166         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2167         if (unlikely(IS_ERR(o)))
2168                 m = (struct mdt_object *)o;
2169         else
2170                 m = mdt_obj(o);
2171         RETURN(m);
2172 }
2173
2174 struct mdt_object *mdt_object_find(const struct lu_env *env,
2175                                    struct mdt_device *d,
2176                                    const struct lu_fid *f)
2177 {
2178         struct lu_object *o;
2179         struct mdt_object *m;
2180         ENTRY;
2181
2182         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2183         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2184         if (unlikely(IS_ERR(o)))
2185                 m = (struct mdt_object *)o;
2186         else
2187                 m = mdt_obj(o);
2188
2189         RETURN(m);
2190 }
2191
2192 /**
2193  * Asyncronous commit for mdt device.
2194  *
2195  * Pass asynchonous commit call down the MDS stack.
2196  *
2197  * \param env environment
2198  * \param mdt the mdt device
2199  */
2200 static void mdt_device_commit_async(const struct lu_env *env,
2201                                     struct mdt_device *mdt)
2202 {
2203         struct dt_device *dt = mdt->mdt_bottom;
2204         int rc;
2205
2206         rc = dt->dd_ops->dt_commit_async(env, dt);
2207         if (unlikely(rc != 0))
2208                 CWARN("%s: async commit start failed: rc = %d\n",
2209                       mdt_obd_name(mdt), rc);
2210 }
2211
2212 /**
2213  * Mark the lock as "synchonous".
2214  *
2215  * Mark the lock to deffer transaction commit to the unlock time.
2216  *
2217  * \param lock the lock to mark as "synchonous"
2218  *
2219  * \see mdt_is_lock_sync
2220  * \see mdt_save_lock
2221  */
2222 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2223 {
2224         lock->l_ast_data = (void*)1;
2225 }
2226
2227 /**
2228  * Check whehter the lock "synchonous" or not.
2229  *
2230  * \param lock the lock to check
2231  * \retval 1 the lock is "synchonous"
2232  * \retval 0 the lock isn't "synchronous"
2233  *
2234  * \see mdt_set_lock_sync
2235  * \see mdt_save_lock
2236  */
2237 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2238 {
2239         return lock->l_ast_data != NULL;
2240 }
2241
2242 /**
2243  * Blocking AST for mdt locks.
2244  *
2245  * Starts transaction commit if in case of COS lock conflict or
2246  * deffers such a commit to the mdt_save_lock.
2247  *
2248  * \param lock the lock which blocks a request or cancelling lock
2249  * \param desc unused
2250  * \param data unused
2251  * \param flag indicates whether this cancelling or blocking callback
2252  * \retval 0
2253  * \see ldlm_blocking_ast_nocheck
2254  */
2255 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2256                      void *data, int flag)
2257 {
2258         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2259         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2260         int rc;
2261         ENTRY;
2262
2263         if (flag == LDLM_CB_CANCELING)
2264                 RETURN(0);
2265         lock_res_and_lock(lock);
2266         if (lock->l_blocking_ast != mdt_blocking_ast) {
2267                 unlock_res_and_lock(lock);
2268                 RETURN(0);
2269         }
2270         if (mdt_cos_is_enabled(mdt) &&
2271             lock->l_req_mode & (LCK_PW | LCK_EX) &&
2272             lock->l_blocking_lock != NULL &&
2273             lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
2274                 mdt_set_lock_sync(lock);
2275         }
2276         rc = ldlm_blocking_ast_nocheck(lock);
2277
2278         /* There is no lock conflict if l_blocking_lock == NULL,
2279          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2280          * when the last reference to a local lock was released */
2281         if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2282                 struct lu_env env;
2283
2284                 rc = lu_env_init(&env, LCT_LOCAL);
2285                 if (unlikely(rc != 0))
2286                         CWARN("%s: lu_env initialization failed, cannot "
2287                               "start asynchronous commit: rc = %d\n",
2288                               obd->obd_name, rc);
2289                 else
2290                         mdt_device_commit_async(&env, mdt);
2291                 lu_env_fini(&env);
2292         }
2293         RETURN(rc);
2294 }
2295
2296 /* Used for cross-MDT lock */
2297 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2298                             void *data, int flag)
2299 {
2300         struct lustre_handle lockh;
2301         int               rc;
2302
2303         switch (flag) {
2304         case LDLM_CB_BLOCKING:
2305                 ldlm_lock2handle(lock, &lockh);
2306                 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
2307                 if (rc < 0) {
2308                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2309                         RETURN(rc);
2310                 }
2311                 break;
2312         case LDLM_CB_CANCELING:
2313                 LDLM_DEBUG(lock, "Revoke remote lock\n");
2314                 break;
2315         default:
2316                 LBUG();
2317         }
2318         RETURN(0);
2319 }
2320
2321 int mdt_check_resent_lock(struct mdt_thread_info *info,
2322                           struct mdt_object *mo,
2323                           struct mdt_lock_handle *lhc)
2324 {
2325         /* the lock might already be gotten in ldlm_handle_enqueue() */
2326         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
2327                 struct ptlrpc_request *req = mdt_info_req(info);
2328                 struct ldlm_lock      *lock;
2329
2330                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2331                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
2332                 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
2333                          lhc->mlh_reg_lh.cookie);
2334
2335                 if (!fid_res_name_eq(mdt_object_fid(mo),
2336                                      &lock->l_resource->lr_name)) {
2337                         CWARN("%s: Although resent, but still not "
2338                               "get child lock:"DFID"\n",
2339                               info->mti_exp->exp_obd->obd_name,
2340                               PFID(mdt_object_fid(mo)));
2341                         LDLM_LOCK_PUT(lock);
2342                         RETURN(-EPROTO);
2343                 }
2344                 LDLM_LOCK_PUT(lock);
2345                 return 0;
2346         }
2347         return 1;
2348 }
2349
2350 int mdt_remote_object_lock(struct mdt_thread_info *mti,
2351                            struct mdt_object *o, const struct lu_fid *fid,
2352                            struct lustre_handle *lh, ldlm_mode_t mode,
2353                            __u64 ibits)
2354 {
2355         struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
2356         ldlm_policy_data_t *policy = &mti->mti_policy;
2357         struct ldlm_res_id *res_id = &mti->mti_res_id;
2358         int rc = 0;
2359         ENTRY;
2360
2361         LASSERT(mdt_object_remote(o));
2362
2363         LASSERT(ibits == MDS_INODELOCK_UPDATE);
2364
2365         fid_build_reg_res_name(fid, res_id);
2366
2367         memset(einfo, 0, sizeof(*einfo));
2368         einfo->ei_type = LDLM_IBITS;
2369         einfo->ei_mode = mode;
2370         einfo->ei_cb_bl = mdt_remote_blocking_ast;
2371         einfo->ei_cb_cp = ldlm_completion_ast;
2372         einfo->ei_enq_slave = 0;
2373         einfo->ei_res_id = res_id;
2374
2375         memset(policy, 0, sizeof(*policy));
2376         policy->l_inodebits.bits = ibits;
2377
2378         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2379                             policy);
2380         RETURN(rc);
2381 }
2382
2383 static int mdt_object_local_lock(struct mdt_thread_info *info,
2384                                  struct mdt_object *o,
2385                                  struct mdt_lock_handle *lh, __u64 ibits,
2386                                  bool nonblock, int locality)
2387 {
2388         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2389         ldlm_policy_data_t *policy = &info->mti_policy;
2390         struct ldlm_res_id *res_id = &info->mti_res_id;
2391         __u64 dlmflags;
2392         int rc;
2393         ENTRY;
2394
2395         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2396         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2397         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2398         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2399
2400         /* Only enqueue LOOKUP lock for remote object */
2401         if (mdt_object_remote(o))
2402                 LASSERT(ibits == MDS_INODELOCK_LOOKUP);
2403
2404         if (lh->mlh_type == MDT_PDO_LOCK) {
2405                 /* check for exists after object is locked */
2406                 if (mdt_object_exists(o) == 0) {
2407                         /* Non-existent object shouldn't have PDO lock */
2408                         RETURN(-ESTALE);
2409                 } else {
2410                         /* Non-dir object shouldn't have PDO lock */
2411                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2412                                 RETURN(-ENOTDIR);
2413                 }
2414         }
2415
2416         memset(policy, 0, sizeof(*policy));
2417         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2418
2419         dlmflags = LDLM_FL_ATOMIC_CB;
2420         if (nonblock)
2421                 dlmflags |= LDLM_FL_BLOCK_NOWAIT;
2422
2423         /*
2424          * Take PDO lock on whole directory and build correct @res_id for lock
2425          * on part of directory.
2426          */
2427         if (lh->mlh_pdo_hash != 0) {
2428                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2429                 mdt_lock_pdo_mode(info, o, lh);
2430                 if (lh->mlh_pdo_mode != LCK_NL) {
2431                         /*
2432                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2433                          * is never going to be sent to client and we do not
2434                          * want it slowed down due to possible cancels.
2435                          */
2436                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2437                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2438                                           policy, res_id, dlmflags,
2439                                           info->mti_exp == NULL ? NULL :
2440                                           &info->mti_exp->exp_handle.h_cookie);
2441                         if (unlikely(rc))
2442                                 RETURN(rc);
2443                 }
2444
2445                 /*
2446                  * Finish res_id initializing by name hash marking part of
2447                  * directory which is taking modification.
2448                  */
2449                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2450         }
2451
2452         policy->l_inodebits.bits = ibits;
2453
2454         /*
2455          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2456          * going to be sent to client. If it is - mdt_intent_policy() path will
2457          * fix it up and turn FL_LOCAL flag off.
2458          */
2459         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2460                           res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2461                           info->mti_exp == NULL ? NULL :
2462                           &info->mti_exp->exp_handle.h_cookie);
2463         if (rc)
2464                 mdt_object_unlock(info, o, lh, 1);
2465         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2466                  lh->mlh_pdo_hash != 0 &&
2467                  (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX)) {
2468                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2469         }
2470
2471         RETURN(rc);
2472 }
2473
2474 static int
2475 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
2476                          struct mdt_lock_handle *lh, __u64 ibits,
2477                          bool nonblock, int locality)
2478 {
2479         int rc;
2480         ENTRY;
2481
2482         if (!mdt_object_remote(o))
2483                 return mdt_object_local_lock(info, o, lh, ibits, nonblock,
2484                                              locality);
2485
2486         if (locality == MDT_LOCAL_LOCK) {
2487                 CERROR("%s: try to get local lock for remote object"
2488                        DFID".\n", mdt_obd_name(info->mti_mdt),
2489                        PFID(mdt_object_fid(o)));
2490                 RETURN(-EPROTO);
2491         }
2492
2493         /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
2494         ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
2495                    MDS_INODELOCK_XATTR);
2496         if (ibits & MDS_INODELOCK_UPDATE) {
2497                 /* Sigh, PDO needs to enqueue 2 locks right now, but
2498                  * enqueue RPC can only request 1 lock, to avoid extra
2499                  * RPC, so it will instead enqueue EX lock for remote
2500                  * object anyway XXX*/
2501                 if (lh->mlh_type == MDT_PDO_LOCK &&
2502                     lh->mlh_pdo_hash != 0) {
2503                         CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
2504                                "EX lock.\n", mdt_obd_name(info->mti_mdt),
2505                                PFID(mdt_object_fid(o)));
2506                         lh->mlh_pdo_hash = 0;
2507                         lh->mlh_rreg_mode = LCK_EX;
2508                         lh->mlh_type = MDT_REG_LOCK;
2509                 }
2510                 rc = mdt_remote_object_lock(info, o, mdt_object_fid(o),
2511                                             &lh->mlh_rreg_lh,
2512                                             lh->mlh_rreg_mode,
2513                                             MDS_INODELOCK_UPDATE);
2514                 if (rc != ELDLM_OK)
2515                         RETURN(rc);
2516         }
2517
2518         /* Only enqueue LOOKUP lock for remote object */
2519         if (ibits & MDS_INODELOCK_LOOKUP) {
2520                 rc = mdt_object_local_lock(info, o, lh,
2521                                            MDS_INODELOCK_LOOKUP,
2522                                            nonblock, locality);
2523                 if (rc != ELDLM_OK)
2524                         RETURN(rc);
2525         }
2526
2527         RETURN(0);
2528 }
2529
2530 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2531                     struct mdt_lock_handle *lh, __u64 ibits, int locality)
2532 {
2533         return mdt_object_lock_internal(info, o, lh, ibits, false, locality);
2534 }
2535
2536 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
2537                         struct mdt_lock_handle *lh, __u64 ibits, int locality)
2538 {
2539         struct mdt_lock_handle tmp = *lh;
2540         int rc;
2541
2542         rc = mdt_object_lock_internal(info, o, &tmp, ibits, true, locality);
2543         if (rc == 0)
2544                 *lh = tmp;
2545
2546         return rc == 0;
2547 }
2548
2549 /**
2550  * Save a lock within request object.
2551  *
2552  * Keep the lock referenced until whether client ACK or transaction
2553  * commit happens or release the lock immediately depending on input
2554  * parameters. If COS is ON, a write lock is converted to COS lock
2555  * before saving.
2556  *
2557  * \param info thead info object
2558  * \param h lock handle
2559  * \param mode lock mode
2560  * \param decref force immediate lock releasing
2561  */
2562 static
2563 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2564                    ldlm_mode_t mode, int decref)
2565 {
2566         ENTRY;
2567
2568         if (lustre_handle_is_used(h)) {
2569                 if (decref || !info->mti_has_trans ||
2570                     !(mode & (LCK_PW | LCK_EX))){
2571                         mdt_fid_unlock(h, mode);
2572                 } else {
2573                         struct mdt_device *mdt = info->mti_mdt;
2574                         struct ldlm_lock *lock = ldlm_handle2lock(h);
2575                         struct ptlrpc_request *req = mdt_info_req(info);
2576                         int no_ack = 0;
2577
2578                         LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2579                                  h->cookie);
2580                         /* there is no request if mdt_object_unlock() is called
2581                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
2582                         if (likely(req != NULL)) {
2583                                 CDEBUG(D_HA, "request = %p reply state = %p"
2584                                        " transno = "LPD64"\n", req,
2585                                        req->rq_reply_state, req->rq_transno);
2586                                 if (mdt_cos_is_enabled(mdt)) {
2587                                         no_ack = 1;
2588                                         ldlm_lock_downgrade(lock, LCK_COS);
2589                                         mode = LCK_COS;
2590                                 }
2591                                 ptlrpc_save_lock(req, h, mode, no_ack);
2592                         } else {
2593                                 ldlm_lock_decref(h, mode);
2594                         }
2595                         if (mdt_is_lock_sync(lock)) {
2596                                 CDEBUG(D_HA, "found sync-lock,"
2597                                        " async commit started\n");
2598                                 mdt_device_commit_async(info->mti_env,
2599                                                         mdt);
2600                         }
2601                         LDLM_LOCK_PUT(lock);
2602                 }
2603                 h->cookie = 0ull;
2604         }
2605
2606         EXIT;
2607 }
2608
2609 /**
2610  * Unlock mdt object.
2611  *
2612  * Immeditely release the regular lock and the PDO lock or save the
2613  * lock in reqeuest and keep them referenced until client ACK or
2614  * transaction commit.
2615  *
2616  * \param info thread info object
2617  * \param o mdt object
2618  * \param lh mdt lock handle referencing regular and PDO locks
2619  * \param decref force immediate lock releasing
2620  */
2621 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2622                        struct mdt_lock_handle *lh, int decref)
2623 {
2624         ENTRY;
2625
2626         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2627         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2628
2629         if (lustre_handle_is_used(&lh->mlh_rreg_lh))
2630                 ldlm_lock_decref(&lh->mlh_rreg_lh, lh->mlh_rreg_mode);
2631
2632         EXIT;
2633 }
2634
2635 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2636                                         const struct lu_fid *f,
2637                                         struct mdt_lock_handle *lh,
2638                                         __u64 ibits)
2639 {
2640         struct mdt_object *o;
2641
2642         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2643         if (!IS_ERR(o)) {
2644                 int rc;
2645
2646                 rc = mdt_object_lock(info, o, lh, ibits,
2647                                      MDT_LOCAL_LOCK);
2648                 if (rc != 0) {
2649                         mdt_object_put(info->mti_env, o);
2650                         o = ERR_PTR(rc);
2651                 }
2652         }
2653         return o;
2654 }
2655
2656 void mdt_object_unlock_put(struct mdt_thread_info * info,
2657                            struct mdt_object * o,
2658                            struct mdt_lock_handle *lh,
2659                            int decref)
2660 {
2661         mdt_object_unlock(info, o, lh, decref);
2662         mdt_object_put(info->mti_env, o);
2663 }
2664
2665 /*
2666  * Generic code handling requests that have struct mdt_body passed in:
2667  *
2668  *  - extract mdt_body from request and save it in @info, if present;
2669  *
2670  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
2671  *  @info;
2672  *
2673  *  - if HABEO_CORPUS flag is set for this request type check whether object
2674  *  actually exists on storage (lu_object_exists()).
2675  *
2676  */
2677 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2678 {
2679         const struct mdt_body    *body;
2680         struct mdt_object        *obj;
2681         const struct lu_env      *env;
2682         struct req_capsule       *pill;
2683         int                       rc;
2684         ENTRY;
2685
2686         env = info->mti_env;
2687         pill = info->mti_pill;
2688
2689         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2690         if (body == NULL)
2691                 RETURN(-EFAULT);
2692
2693         if (!(body->mbo_valid & OBD_MD_FLID))
2694                 RETURN(0);
2695
2696         if (!fid_is_sane(&body->mbo_fid1)) {
2697                 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
2698                 RETURN(-EINVAL);
2699         }
2700
2701         /*
2702          * Do not get size or any capa fields before we check that request
2703          * contains capa actually. There are some requests which do not, for
2704          * instance MDS_IS_SUBDIR.
2705          */
2706         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2707             req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
2708                 mdt_set_capainfo(info, 0, &body->mbo_fid1,
2709                                  req_capsule_client_get(pill, &RMF_CAPA1));
2710
2711         obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
2712         if (!IS_ERR(obj)) {
2713                 if ((flags & HABEO_CORPUS) &&
2714                     !mdt_object_exists(obj)) {
2715                         mdt_object_put(env, obj);
2716                         /* for capability renew ENOENT will be handled in
2717                          * mdt_renew_capa */
2718                         if (body->mbo_valid & OBD_MD_FLOSSCAPA)
2719                                 rc = 0;
2720                         else
2721                                 rc = -ENOENT;
2722                 } else {
2723                         info->mti_object = obj;
2724                         rc = 0;
2725                 }
2726         } else
2727                 rc = PTR_ERR(obj);
2728
2729         RETURN(rc);
2730 }
2731
2732 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2733 {
2734         struct req_capsule *pill = info->mti_pill;
2735         int rc;
2736         ENTRY;
2737
2738         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2739                 rc = mdt_body_unpack(info, flags);
2740         else
2741                 rc = 0;
2742
2743         if (rc == 0 && (flags & HABEO_REFERO)) {
2744                 /* Pack reply. */
2745                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2746                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2747                                              DEF_REP_MD_SIZE);
2748                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2749                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
2750                                              RCL_SERVER, 0);
2751
2752                 rc = req_capsule_server_pack(pill);
2753         }
2754         RETURN(rc);
2755 }
2756
2757 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2758 {
2759         struct md_device *next = m->mdt_child;
2760
2761         return next->md_ops->mdo_init_capa_ctxt(env, next,
2762                                                 m->mdt_lut.lut_mds_capa,
2763                                                 m->mdt_capa_timeout,
2764                                                 m->mdt_capa_alg,
2765                                                 m->mdt_capa_keys);
2766 }
2767
2768 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2769 {
2770         lh->mlh_type = MDT_NUL_LOCK;
2771         lh->mlh_reg_lh.cookie = 0ull;
2772         lh->mlh_reg_mode = LCK_MINMODE;
2773         lh->mlh_pdo_lh.cookie = 0ull;
2774         lh->mlh_pdo_mode = LCK_MINMODE;
2775         lh->mlh_rreg_lh.cookie = 0ull;
2776         lh->mlh_rreg_mode = LCK_MINMODE;
2777 }
2778
2779 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2780 {
2781         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2782         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2783 }
2784
2785 /*
2786  * Initialize fields of struct mdt_thread_info. Other fields are left in
2787  * uninitialized state, because it's too expensive to zero out whole
2788  * mdt_thread_info (> 1K) on each request arrival.
2789  */
2790 void mdt_thread_info_init(struct ptlrpc_request *req,
2791                           struct mdt_thread_info *info)
2792 {
2793         int i;
2794
2795         info->mti_pill = &req->rq_pill;
2796
2797         /* lock handle */
2798         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2799                 mdt_lock_handle_init(&info->mti_lh[i]);
2800
2801         /* mdt device: it can be NULL while CONNECT */
2802         if (req->rq_export) {
2803                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2804                 info->mti_exp = req->rq_export;
2805         } else
2806                 info->mti_mdt = NULL;
2807         info->mti_env = req->rq_svc_thread->t_env;
2808         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2809
2810         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2811         info->mti_big_buf = LU_BUF_NULL;
2812         info->mti_body = NULL;
2813         info->mti_object = NULL;
2814         info->mti_dlm_req = NULL;
2815         info->mti_has_trans = 0;
2816         info->mti_cross_ref = 0;
2817         info->mti_opdata = 0;
2818         info->mti_big_lmm_used = 0;
2819
2820         info->mti_spec.no_create = 0;
2821         info->mti_spec.sp_rm_entry = 0;
2822
2823         info->mti_spec.u.sp_ea.eadata = NULL;
2824         info->mti_spec.u.sp_ea.eadatalen = 0;
2825 }
2826
2827 void mdt_thread_info_fini(struct mdt_thread_info *info)
2828 {
2829         int i;
2830
2831         if (info->mti_object != NULL) {
2832                 mdt_object_put(info->mti_env, info->mti_object);
2833                 info->mti_object = NULL;
2834         }
2835
2836         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2837                 mdt_lock_handle_fini(&info->mti_lh[i]);
2838         info->mti_env = NULL;
2839         info->mti_pill = NULL;
2840         info->mti_exp = NULL;
2841
2842         if (unlikely(info->mti_big_buf.lb_buf != NULL))
2843                 lu_buf_free(&info->mti_big_buf);
2844 }
2845
2846 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
2847 {
2848         struct mdt_thread_info  *mti;
2849         struct lustre_capa      *lc;
2850
2851         mti = mdt_th_info(tsi->tsi_env);
2852         LASSERT(mti != NULL);
2853
2854         mdt_thread_info_init(tgt_ses_req(tsi), mti);
2855         if (tsi->tsi_corpus != NULL) {
2856                 struct req_capsule *pill = tsi->tsi_pill;
2857
2858                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
2859                 lu_object_get(tsi->tsi_corpus);
2860
2861                 /*
2862                  * XXX: must be part of tgt_mdt_body_unpack but moved here
2863                  * due to mdt_set_capainfo().
2864                  */
2865                 if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2866                     req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT) > 0) {
2867                         lc = req_capsule_client_get(pill, &RMF_CAPA1);
2868                         mdt_set_capainfo(mti, 0, &tsi->tsi_mdt_body->mbo_fid1,
2869                                          lc);
2870                 }
2871         }
2872         mti->mti_body = tsi->tsi_mdt_body;
2873         mti->mti_dlm_req = tsi->tsi_dlm_req;
2874
2875         return mti;
2876 }
2877
2878 static int mdt_tgt_connect(struct tgt_session_info *tsi)
2879 {
2880         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2881         int                      rc;
2882
2883         ENTRY;
2884
2885         if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
2886             cfs_fail_val ==
2887             tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id)
2888                 schedule_timeout_and_set_state(TASK_UNINTERRUPTIBLE, HZ * 3);
2889
2890         rc = tgt_connect(tsi);
2891         if (rc != 0)
2892                 RETURN(rc);
2893
2894         rc = mdt_init_idmap(tsi);
2895         if (rc != 0)
2896                 GOTO(err, rc);
2897         RETURN(0);
2898 err:
2899         obd_disconnect(class_export_get(req->rq_export));
2900         return rc;
2901 }
2902
2903 enum mdt_it_code {
2904         MDT_IT_OPEN,
2905         MDT_IT_OCREAT,
2906         MDT_IT_CREATE,
2907         MDT_IT_GETATTR,
2908         MDT_IT_READDIR,
2909         MDT_IT_LOOKUP,
2910         MDT_IT_UNLINK,
2911         MDT_IT_TRUNC,
2912         MDT_IT_GETXATTR,
2913         MDT_IT_LAYOUT,
2914         MDT_IT_QUOTA,
2915         MDT_IT_NR
2916 };
2917
2918 static int mdt_intent_getattr(enum mdt_it_code opcode,
2919                               struct mdt_thread_info *info,
2920                               struct ldlm_lock **,
2921                               __u64);
2922
2923 static int mdt_intent_getxattr(enum mdt_it_code opcode,
2924                                 struct mdt_thread_info *info,
2925                                 struct ldlm_lock **lockp,
2926                                 __u64 flags);
2927
2928 static int mdt_intent_layout(enum mdt_it_code opcode,
2929                              struct mdt_thread_info *info,
2930                              struct ldlm_lock **,
2931                              __u64);
2932 static int mdt_intent_reint(enum mdt_it_code opcode,
2933                             struct mdt_thread_info *info,
2934                             struct ldlm_lock **,
2935                             __u64);
2936
2937 static struct mdt_it_flavor {
2938         const struct req_format *it_fmt;
2939         __u32                    it_flags;
2940         int                    (*it_act)(enum mdt_it_code ,
2941                                          struct mdt_thread_info *,
2942                                          struct ldlm_lock **,
2943                                          __u64);
2944         long                     it_reint;
2945 } mdt_it_flavor[] = {
2946         [MDT_IT_OPEN]     = {
2947                 .it_fmt   = &RQF_LDLM_INTENT,
2948                 /*.it_flags = HABEO_REFERO,*/
2949                 .it_flags = 0,
2950                 .it_act   = mdt_intent_reint,
2951                 .it_reint = REINT_OPEN
2952         },
2953         [MDT_IT_OCREAT]   = {
2954                 .it_fmt   = &RQF_LDLM_INTENT,
2955                 /*
2956                  * OCREAT is not a MUTABOR request as if the file
2957                  * already exists.
2958                  * We do the extra check of OBD_CONNECT_RDONLY in
2959                  * mdt_reint_open() when we really need to create
2960                  * the object.
2961                  */
2962                 .it_flags = 0,
2963                 .it_act   = mdt_intent_reint,
2964                 .it_reint = REINT_OPEN
2965         },
2966         [MDT_IT_CREATE]   = {
2967                 .it_fmt   = &RQF_LDLM_INTENT,
2968                 .it_flags = MUTABOR,
2969                 .it_act   = mdt_intent_reint,
2970                 .it_reint = REINT_CREATE
2971         },
2972         [MDT_IT_GETATTR]  = {
2973                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2974                 .it_flags = HABEO_REFERO,
2975                 .it_act   = mdt_intent_getattr
2976         },
2977         [MDT_IT_READDIR]  = {
2978                 .it_fmt   = NULL,
2979                 .it_flags = 0,
2980                 .it_act   = NULL
2981         },
2982         [MDT_IT_LOOKUP]   = {
2983                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2984                 .it_flags = HABEO_REFERO,
2985                 .it_act   = mdt_intent_getattr
2986         },
2987         [MDT_IT_UNLINK]   = {
2988                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
2989                 .it_flags = MUTABOR,
2990                 .it_act   = NULL,
2991                 .it_reint = REINT_UNLINK
2992         },
2993         [MDT_IT_TRUNC]    = {
2994                 .it_fmt   = NULL,
2995                 .it_flags = MUTABOR,
2996                 .it_act   = NULL
2997         },
2998         [MDT_IT_GETXATTR] = {
2999                 .it_fmt   = &RQF_LDLM_INTENT_GETXATTR,
3000                 .it_flags = HABEO_CORPUS,
3001                 .it_act   = mdt_intent_getxattr
3002         },
3003         [MDT_IT_LAYOUT] = {
3004                 .it_fmt   = &RQF_LDLM_INTENT_LAYOUT,
3005                 .it_flags = 0,
3006                 .it_act   = mdt_intent_layout
3007         }
3008 };
3009
3010 static int
3011 mdt_intent_lock_replace(struct mdt_thread_info *info,
3012                         struct ldlm_lock **lockp,
3013                         struct mdt_lock_handle *lh,
3014                         __u64 flags)
3015 {
3016         struct ptlrpc_request  *req = mdt_info_req(info);
3017         struct ldlm_lock       *lock = *lockp;
3018         struct ldlm_lock       *new_lock;
3019
3020         /* If possible resent found a lock, @lh is set to its handle */
3021         new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3022
3023         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3024                 lh->mlh_reg_lh.cookie = 0;
3025                 RETURN(0);
3026         }
3027
3028         LASSERTF(new_lock != NULL,
3029                  "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
3030
3031         /*
3032          * If we've already given this lock to a client once, then we should
3033          * have no readers or writers.  Otherwise, we should have one reader
3034          * _or_ writer ref (which will be zeroed below) before returning the
3035          * lock to a client.
3036          */
3037         if (new_lock->l_export == req->rq_export) {
3038                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3039         } else {
3040                 LASSERT(new_lock->l_export == NULL);
3041                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3042         }
3043
3044         *lockp = new_lock;
3045
3046         if (new_lock->l_export == req->rq_export) {
3047                 /*
3048                  * Already gave this to the client, which means that we
3049                  * reconstructed a reply.
3050                  */
3051                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3052                         MSG_RESENT);
3053
3054                 LDLM_LOCK_RELEASE(new_lock);
3055                 lh->mlh_reg_lh.cookie = 0;
3056                 RETURN(ELDLM_LOCK_REPLACED);
3057         }
3058
3059         /*
3060          * Fixup the lock to be given to the client.
3061          */
3062         lock_res_and_lock(new_lock);
3063         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3064          * possible blocking AST. */
3065         while (new_lock->l_readers > 0) {
3066                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3067                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3068                 new_lock->l_readers--;
3069         }
3070         while (new_lock->l_writers > 0) {
3071                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3072                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3073                 new_lock->l_writers--;
3074         }
3075
3076         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3077         new_lock->l_blocking_ast = lock->l_blocking_ast;
3078         new_lock->l_completion_ast = lock->l_completion_ast;
3079         new_lock->l_remote_handle = lock->l_remote_handle;
3080         new_lock->l_flags &= ~LDLM_FL_LOCAL;
3081
3082         unlock_res_and_lock(new_lock);
3083
3084         cfs_hash_add(new_lock->l_export->exp_lock_hash,
3085                      &new_lock->l_remote_handle,
3086                      &new_lock->l_exp_hash);
3087
3088         LDLM_LOCK_RELEASE(new_lock);
3089         lh->mlh_reg_lh.cookie = 0;
3090
3091         RETURN(ELDLM_LOCK_REPLACED);
3092 }
3093
3094 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3095                                     struct ldlm_lock *new_lock,
3096                                     struct mdt_lock_handle *lh,
3097                                     __u64 flags)
3098 {
3099         struct ptlrpc_request  *req = mdt_info_req(info);
3100         struct ldlm_request    *dlmreq;
3101
3102         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3103                 return;
3104
3105         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3106
3107         /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
3108          * lock was found by ldlm_handle_enqueue(); if so @lh must be
3109          * initialized. */
3110         if (flags & LDLM_FL_RESENT) {
3111                 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
3112                 lh->mlh_reg_mode = new_lock->l_granted_mode;
3113
3114                 LDLM_DEBUG(new_lock, "Restoring lock cookie");
3115                 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie "LPX64,
3116                           lh->mlh_reg_lh.cookie);
3117                 return;
3118         }
3119
3120         /*
3121          * If the xid matches, then we know this is a resent request, and allow
3122          * it. (It's probably an OPEN, for which we don't send a lock.
3123          */
3124         if (req_xid_is_last(req))
3125                 return;
3126
3127         /*
3128          * This remote handle isn't enqueued, so we never received or processed
3129          * this request.  Clear MSG_RESENT, because it can be handled like any
3130          * normal request now.
3131          */
3132         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3133
3134         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
3135                   dlmreq->lock_handle[0].cookie);
3136 }
3137
3138 static int mdt_intent_getxattr(enum mdt_it_code opcode,
3139                                 struct mdt_thread_info *info,
3140                                 struct ldlm_lock **lockp,
3141                                 __u64 flags)
3142 {
3143         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3144         struct ldlm_reply      *ldlm_rep = NULL;
3145         int rc, grc;
3146
3147         /*
3148          * Initialize lhc->mlh_reg_lh either from a previously granted lock
3149          * (for the resend case) or a new lock. Below we will use it to
3150          * replace the original lock.
3151          */
3152         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3153         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3154                 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3155                 rc = mdt_object_lock(info, info->mti_object, lhc,
3156                                         MDS_INODELOCK_XATTR,
3157                                         MDT_LOCAL_LOCK);
3158                 if (rc)
3159                         return rc;
3160         }
3161
3162         grc = mdt_getxattr(info);
3163
3164         rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3165
3166         if (mdt_info_req(info)->rq_repmsg != NULL)
3167                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3168         if (ldlm_rep == NULL)
3169                 RETURN(err_serious(-EFAULT));
3170
3171         ldlm_rep->lock_policy_res2 = grc;
3172
3173         return rc;
3174 }
3175
3176 static int mdt_intent_getattr(enum mdt_it_code opcode,
3177                               struct mdt_thread_info *info,
3178                               struct ldlm_lock **lockp,
3179                               __u64 flags)
3180 {
3181         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3182         __u64                   child_bits;
3183         struct ldlm_reply      *ldlm_rep;
3184         struct mdt_body        *reqbody;
3185         struct mdt_body        *repbody;
3186         int                     rc, rc2;
3187         ENTRY;
3188
3189         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3190         LASSERT(reqbody);
3191
3192         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3193         LASSERT(repbody);
3194
3195         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
3196         repbody->mbo_eadatasize = 0;
3197         repbody->mbo_aclsize = 0;
3198
3199         switch (opcode) {
3200         case MDT_IT_LOOKUP:
3201                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3202                 break;
3203         case MDT_IT_GETATTR:
3204                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3205                              MDS_INODELOCK_PERM;
3206                 break;
3207         default:
3208                 CERROR("Unsupported intent (%d)\n", opcode);
3209                 GOTO(out_shrink, rc = -EINVAL);
3210         }
3211
3212         rc = mdt_init_ucred(info, reqbody);
3213         if (rc)
3214                 GOTO(out_shrink, rc);
3215
3216         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3217         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3218
3219         /* Get lock from request for possible resent case. */
3220         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3221
3222         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3223         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3224
3225         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3226                 ldlm_rep->lock_policy_res2 = 0;
3227         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3228             ldlm_rep->lock_policy_res2) {
3229                 lhc->mlh_reg_lh.cookie = 0ull;
3230                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3231         }
3232
3233         rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3234         EXIT;
3235 out_ucred:
3236         mdt_exit_ucred(info);
3237 out_shrink:
3238         mdt_client_compatibility(info);
3239         rc2 = mdt_fix_reply(info);
3240         if (rc == 0)
3241                 rc = rc2;
3242         return rc;
3243 }
3244
3245 static int mdt_intent_layout(enum mdt_it_code opcode,
3246                              struct mdt_thread_info *info,
3247                              struct ldlm_lock **lockp,
3248                              __u64 flags)
3249 {
3250         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_LAYOUT];
3251         struct layout_intent *layout;
3252         struct lu_fid *fid;
3253         struct mdt_object *obj = NULL;
3254         int rc = 0;
3255         ENTRY;
3256
3257         if (opcode != MDT_IT_LAYOUT) {
3258                 CERROR("%s: Unknown intent (%d)\n", mdt_obd_name(info->mti_mdt),
3259                         opcode);
3260                 RETURN(-EINVAL);
3261         }
3262
3263         fid = &info->mti_tmp_fid2;
3264         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3265
3266         /* Get lock from request for possible resent case. */
3267         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3268
3269         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3270         if (IS_ERR(obj))
3271                 RETURN(PTR_ERR(obj));
3272
3273         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3274                 /* get the length of lsm */
3275                 rc = mdt_attr_get_eabuf_size(info, obj);
3276                 if (rc < 0) {
3277                         mdt_object_put(info->mti_env, obj);
3278                         RETURN(rc);
3279                 }
3280
3281                 if (rc > info->mti_mdt->mdt_max_mdsize)
3282                         info->mti_mdt->mdt_max_mdsize = rc;
3283         }
3284
3285         mdt_object_put(info->mti_env, obj);
3286
3287         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3288         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER, rc);
3289         rc = req_capsule_server_pack(info->mti_pill);
3290         if (rc != 0)
3291                 RETURN(-EINVAL);
3292
3293         if (lustre_handle_is_used(&lhc->mlh_reg_lh))
3294                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3295
3296         layout = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3297         LASSERT(layout != NULL);
3298         if (layout->li_opc == LAYOUT_INTENT_ACCESS)
3299                 /* return to normal/resent ldlm handling */
3300                 RETURN(rc);
3301
3302         CERROR("%s: Unsupported layout intent (%d)\n",
3303                 mdt_obd_name(info->mti_mdt), layout->li_opc);
3304         RETURN(-EINVAL);
3305 }
3306
3307 static int mdt_intent_reint(enum mdt_it_code opcode,
3308                             struct mdt_thread_info *info,
3309                             struct ldlm_lock **lockp,
3310                             __u64 flags)
3311 {
3312         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3313         struct ldlm_reply      *rep = NULL;
3314         long                    opc;
3315         int                     rc;
3316
3317         static const struct req_format *intent_fmts[REINT_MAX] = {
3318                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3319                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3320         };
3321
3322         ENTRY;
3323
3324         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
3325         if (opc < 0)
3326                 RETURN(opc);
3327
3328         if (mdt_it_flavor[opcode].it_reint != opc) {
3329                 CERROR("Reint code %ld doesn't match intent: %d\n",
3330                        opc, opcode);
3331                 RETURN(err_serious(-EPROTO));
3332         }
3333
3334         /* Get lock from request for possible resent case. */
3335         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3336
3337         rc = mdt_reint_internal(info, lhc, opc);
3338
3339         /* Check whether the reply has been packed successfully. */
3340         if (mdt_info_req(info)->rq_repmsg != NULL)
3341                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3342         if (rep == NULL)
3343                 RETURN(err_serious(-EFAULT));
3344
3345         /* MDC expects this in any case */
3346         if (rc != 0)
3347                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3348
3349         /* the open lock or the lock for cross-ref object should be
3350          * returned to the client */
3351         if (rc == -EREMOTE || mdt_get_disposition(rep, DISP_OPEN_LOCK)) {
3352                 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
3353                 rep->lock_policy_res2 = 0;
3354                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3355                 RETURN(rc);
3356         }
3357
3358         rep->lock_policy_res2 = clear_serious(rc);
3359
3360         if (rep->lock_policy_res2 == -ENOENT &&
3361             mdt_get_disposition(rep, DISP_LOOKUP_NEG))
3362                 rep->lock_policy_res2 = 0;
3363
3364         if (rc == -ENOTCONN || rc == -ENODEV ||
3365             rc == -EOVERFLOW) { /**< if VBR failure then return error */
3366                 /*
3367                  * If it is the disconnect error (ENODEV & ENOCONN), the error
3368                  * will be returned by rq_status, and client at ptlrpc layer
3369                  * will detect this, then disconnect, reconnect the import
3370                  * immediately, instead of impacting the following the rpc.
3371                  */
3372                 lhc->mlh_reg_lh.cookie = 0ull;
3373                 RETURN(rc);
3374         } else {
3375                 /*
3376                  * For other cases, the error will be returned by intent.
3377                  * and client will retrieve the result from intent.
3378                  */
3379                  /*
3380                   * FIXME: when open lock is finished, that should be
3381                   * checked here.
3382                   */
3383                 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3384                         LASSERTF(rc == 0, "Error occurred but lock handle "
3385                                  "is still in use, rc = %d\n", rc);
3386                         rep->lock_policy_res2 = 0;
3387                         rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3388                         RETURN(rc);
3389                 } else {
3390                         lhc->mlh_reg_lh.cookie = 0ull;
3391                         RETURN(ELDLM_LOCK_ABORTED);
3392                 }
3393         }
3394 }
3395
3396 static int mdt_intent_code(long itcode)
3397 {
3398         int rc;
3399
3400         switch(itcode) {
3401         case IT_OPEN:
3402                 rc = MDT_IT_OPEN;
3403                 break;
3404         case IT_OPEN|IT_CREAT:
3405                 rc = MDT_IT_OCREAT;
3406                 break;
3407         case IT_CREAT:
3408                 rc = MDT_IT_CREATE;
3409                 break;
3410         case IT_READDIR:
3411                 rc = MDT_IT_READDIR;
3412                 break;
3413         case IT_GETATTR:
3414                 rc = MDT_IT_GETATTR;
3415                 break;
3416         case IT_LOOKUP:
3417                 rc = MDT_IT_LOOKUP;
3418                 break;
3419         case IT_UNLINK:
3420                 rc = MDT_IT_UNLINK;
3421                 break;
3422         case IT_TRUNC:
3423                 rc = MDT_IT_TRUNC;
3424                 break;
3425         case IT_GETXATTR:
3426                 rc = MDT_IT_GETXATTR;
3427                 break;
3428         case IT_LAYOUT:
3429                 rc = MDT_IT_LAYOUT;
3430                 break;
3431         case IT_QUOTA_DQACQ:
3432         case IT_QUOTA_CONN:
3433                 rc = MDT_IT_QUOTA;
3434                 break;
3435         default:
3436                 CERROR("Unknown intent opcode: %ld\n", itcode);
3437                 rc = -EINVAL;
3438                 break;
3439         }
3440         return rc;
3441 }
3442
3443 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
3444                           struct ldlm_lock **lockp, __u64 flags)
3445 {
3446         struct req_capsule   *pill;
3447         struct mdt_it_flavor *flv;
3448         int opc;
3449         int rc;
3450         ENTRY;
3451
3452         opc = mdt_intent_code(itopc);
3453         if (opc < 0)
3454                 RETURN(-EINVAL);
3455
3456         pill = info->mti_pill;
3457
3458         if (opc == MDT_IT_QUOTA) {
3459                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
3460
3461                 if (qmt == NULL)
3462                         RETURN(-EOPNOTSUPP);
3463
3464                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
3465                 /* pass the request to quota master */
3466                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
3467                                                  mdt_info_req(info), lockp,
3468                                                  flags);
3469                 RETURN(rc);
3470         }
3471
3472         flv  = &mdt_it_flavor[opc];
3473         if (flv->it_fmt != NULL)
3474                 req_capsule_extend(pill, flv->it_fmt);
3475
3476         rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3477         if (rc == 0) {
3478                 struct ptlrpc_request *req = mdt_info_req(info);
3479                 if (flv->it_flags & MUTABOR &&
3480                     exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
3481                         RETURN(-EROFS);
3482         }
3483         if (rc == 0 && flv->it_act != NULL) {
3484                 struct ldlm_reply *rep;
3485
3486                 /* execute policy */
3487                 rc = flv->it_act(opc, info, lockp, flags);
3488
3489                 /* Check whether the reply has been packed successfully. */
3490                 if (mdt_info_req(info)->rq_repmsg != NULL) {
3491                         rep = req_capsule_server_get(info->mti_pill,
3492                                                      &RMF_DLM_REP);
3493                         rep->lock_policy_res2 =
3494                                 ptlrpc_status_hton(rep->lock_policy_res2);
3495                 }
3496         } else {
3497                 rc = -EPROTO;
3498         }
3499         RETURN(rc);
3500 }
3501
3502 static int mdt_intent_policy(struct ldlm_namespace *ns,
3503                              struct ldlm_lock **lockp, void *req_cookie,
3504                              ldlm_mode_t mode, __u64 flags, void *data)
3505 {
3506         struct tgt_session_info *tsi;
3507         struct mdt_thread_info  *info;
3508         struct ptlrpc_request   *req  =  req_cookie;
3509         struct ldlm_intent      *it;
3510         struct req_capsule      *pill;
3511         int rc;
3512
3513         ENTRY;
3514
3515         LASSERT(req != NULL);
3516
3517         tsi = tgt_ses_info(req->rq_svc_thread->t_env);
3518
3519         info = tsi2mdt_info(tsi);
3520         LASSERT(info != NULL);
3521         pill = info->mti_pill;
3522         LASSERT(pill->rc_req == req);
3523
3524         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3525                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
3526                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3527                 if (it != NULL) {
3528                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
3529                         if (rc == 0)
3530                                 rc = ELDLM_OK;
3531
3532                         /* Lock without inodebits makes no sense and will oops
3533                          * later in ldlm. Let's check it now to see if we have
3534                          * ibits corrupted somewhere in mdt_intent_opc().
3535                          * The case for client miss to set ibits has been
3536                          * processed by others. */
3537                         LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
3538                                         lr_type == LDLM_IBITS,
3539                                      info->mti_dlm_req->lock_desc.\
3540                                         l_policy_data.l_inodebits.bits != 0));
3541                 } else
3542                         rc = err_serious(-EFAULT);
3543         } else {
3544                 /* No intent was provided */
3545                 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3546                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
3547                 rc = req_capsule_server_pack(pill);
3548                 if (rc)
3549                         rc = err_serious(rc);
3550         }
3551         mdt_thread_info_fini(info);
3552         RETURN(rc);
3553 }
3554
3555 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
3556 {
3557         struct seq_server_site  *ss = mdt_seq_site(mdt);
3558
3559         if (ss->ss_node_id == 0)
3560                 return;
3561
3562         if (ss->ss_client_seq != NULL) {
3563                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3564                 ss->ss_client_seq->lcs_exp = NULL;
3565         }
3566
3567         if (ss->ss_server_fld != NULL) {
3568                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
3569                 ss->ss_server_fld->lsf_control_exp = NULL;
3570         }
3571 }
3572
3573 static void mdt_seq_fini_cli(struct mdt_device *mdt)
3574 {
3575         struct seq_server_site *ss = mdt_seq_site(mdt);
3576
3577         if (ss == NULL)
3578                 return;
3579
3580         if (ss->ss_server_seq != NULL)
3581                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
3582 }
3583
3584 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
3585 {
3586         mdt_seq_fini_cli(mdt);
3587         mdt_deregister_seq_exp(mdt);
3588
3589         return seq_site_fini(env, mdt_seq_site(mdt));
3590 }
3591
3592 /**
3593  * It will retrieve its FLDB entries from MDT0, and it only happens
3594  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
3595  * and it needs to refresh FLDB from the MDT0.
3596  **/
3597 static int mdt_register_lwp_callback(void *data)
3598 {
3599         struct lu_env           env;
3600         struct mdt_device       *mdt = data;
3601         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
3602         int                     rc;
3603         ENTRY;
3604
3605         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
3606
3607         if (!likely(fld->lsf_new))
3608                 RETURN(0);
3609
3610         rc = lu_env_init(&env, LCT_MD_THREAD);
3611         if (rc) {
3612                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
3613                 RETURN(rc);
3614         }
3615
3616         rc = fld_update_from_controller(&env, fld);
3617         if (rc != 0) {
3618                 CERROR("%s: cannot update controller: rc = %d\n",
3619                        mdt_obd_name(mdt), rc);
3620                 GOTO(out, rc);
3621         }
3622 out:
3623         lu_env_fini(&env);
3624         RETURN(rc);
3625 }
3626
3627 static int mdt_register_seq_exp(struct mdt_device *mdt)
3628 {
3629         struct seq_server_site  *ss = mdt_seq_site(mdt);
3630         char                    *lwp_name = NULL;
3631         int                     rc;
3632
3633         if (ss->ss_node_id == 0)
3634                 return 0;
3635
3636         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
3637         if (lwp_name == NULL)
3638                 GOTO(out_free, rc = -ENOMEM);
3639
3640         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
3641         if (rc != 0)
3642                 GOTO(out_free, rc);
3643
3644         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
3645                                       NULL, NULL);
3646         if (rc != 0)
3647                 GOTO(out_free, rc);
3648
3649         rc = lustre_register_lwp_item(lwp_name,
3650                                       &ss->ss_server_fld->lsf_control_exp,
3651                                       mdt_register_lwp_callback, mdt);
3652         if (rc != 0) {
3653                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3654                 ss->ss_client_seq->lcs_exp = NULL;
3655                 GOTO(out_free, rc);
3656         }
3657 out_free:
3658         if (lwp_name != NULL)
3659                 OBD_FREE(lwp_name, MAX_OBD_NAME);
3660
3661         return rc;
3662 }
3663
3664 /*
3665  * Init client sequence manager which is used by local MDS to talk to sequence
3666  * controller on remote node.
3667  */
3668 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
3669 {
3670         struct seq_server_site  *ss = mdt_seq_site(mdt);
3671         int                     rc;
3672         char                    *prefix;
3673         ENTRY;
3674
3675         /* check if this is adding the first MDC and controller is not yet
3676          * initialized. */
3677         OBD_ALLOC_PTR(ss->ss_client_seq);
3678         if (ss->ss_client_seq == NULL)
3679                 RETURN(-ENOMEM);
3680
3681         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3682         if (prefix == NULL) {
3683                 OBD_FREE_PTR(ss->ss_client_seq);
3684                 ss->ss_client_seq = NULL;
3685                 RETURN(-ENOMEM);
3686         }
3687
3688         /* Note: seq_client_fini will be called in seq_site_fini */
3689         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
3690         rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
3691                              prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
3692                                                             NULL);
3693         OBD_FREE(prefix, MAX_OBD_NAME + 5);
3694         if (rc != 0) {
3695                 OBD_FREE_PTR(ss->ss_client_seq);
3696                 ss->ss_client_seq = NULL;
3697                 RETURN(rc);
3698         }
3699
3700         rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
3701
3702         RETURN(rc);
3703 }
3704
3705 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
3706 {
3707         struct seq_server_site  *ss;
3708         int                     rc;
3709         ENTRY;
3710
3711         ss = mdt_seq_site(mdt);
3712         /* init sequence controller server(MDT0) */
3713         if (ss->ss_node_id == 0) {
3714                 OBD_ALLOC_PTR(ss->ss_control_seq);
3715                 if (ss->ss_control_seq == NULL)
3716                         RETURN(-ENOMEM);
3717
3718                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
3719                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
3720                                      ss);
3721                 if (rc)
3722                         GOTO(out_seq_fini, rc);
3723         }
3724
3725         /* Init normal sequence server */
3726         OBD_ALLOC_PTR(ss->ss_server_seq);
3727         if (ss->ss_server_seq == NULL)
3728                 GOTO(out_seq_fini, rc = -ENOMEM);
3729
3730         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
3731                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
3732         if (rc)
3733                 GOTO(out_seq_fini, rc);
3734
3735         /* init seq client for seq server to talk to seq controller(MDT0) */
3736         rc = mdt_seq_init_cli(env, mdt);
3737         if (rc != 0)
3738                 GOTO(out_seq_fini, rc);
3739
3740         if (ss->ss_node_id != 0)
3741                 /* register controler export through lwp */
3742                 rc = mdt_register_seq_exp(mdt);
3743
3744         EXIT;
3745 out_seq_fini:
3746         if (rc)
3747                 mdt_seq_fini(env, mdt);
3748
3749         return rc;
3750 }
3751
3752 /*
3753  * FLD wrappers
3754  */
3755 static int mdt_fld_fini(const struct lu_env *env,
3756                         struct mdt_device *m)
3757 {
3758         struct seq_server_site *ss = mdt_seq_site(m);
3759         ENTRY;
3760
3761         if (ss && ss->ss_server_fld) {
3762                 fld_server_fini(env, ss->ss_server_fld);
3763                 OBD_FREE_PTR(ss->ss_server_fld);
3764                 ss->ss_server_fld = NULL;
3765         }
3766
3767         RETURN(0);
3768 }
3769
3770 static int mdt_fld_init(const struct lu_env *env,
3771                         const char *uuid,
3772                         struct mdt_device *m)
3773 {
3774         struct seq_server_site *ss;
3775         int rc;
3776         ENTRY;
3777
3778         ss = mdt_seq_site(m);
3779
3780         OBD_ALLOC_PTR(ss->ss_server_fld);
3781         if (ss->ss_server_fld == NULL)
3782                 RETURN(rc = -ENOMEM);
3783
3784         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
3785                              LU_SEQ_RANGE_MDT);
3786         if (rc) {
3787                 OBD_FREE_PTR(ss->ss_server_fld);
3788                 ss->ss_server_fld = NULL;
3789                 RETURN(rc);
3790         }
3791
3792         RETURN(0);
3793 }
3794
3795 static void mdt_stack_pre_fini(const struct lu_env *env,
3796                            struct mdt_device *m, struct lu_device *top)
3797 {
3798         struct lustre_cfg_bufs  *bufs;
3799         struct lustre_cfg       *lcfg;
3800         struct mdt_thread_info  *info;
3801         ENTRY;
3802
3803         LASSERT(top);
3804
3805         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3806         LASSERT(info != NULL);
3807
3808         bufs = &info->mti_u.bufs;
3809
3810         LASSERT(m->mdt_child_exp);
3811         LASSERT(m->mdt_child_exp->exp_obd);
3812
3813         /* process cleanup, pass mdt obd name to get obd umount flags */
3814         /* XXX: this is needed because all layers are referenced by
3815          * objects (some of them are pinned by osd, for example *
3816          * the proper solution should be a model where object used
3817          * by osd only doesn't have mdt/mdd slices -bzzz */
3818         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
3819         lustre_cfg_bufs_set_string(bufs, 1, NULL);
3820         lcfg = lustre_cfg_new(LCFG_PRE_CLEANUP, bufs);
3821         if (!lcfg) {
3822                 CERROR("%s: cannot alloc lcfg\n", mdt_obd_name(m));
3823                 return;
3824         }
3825         top->ld_ops->ldo_process_config(env, top, lcfg);
3826         lustre_cfg_free(lcfg);
3827         EXIT;
3828 }
3829
3830 static void mdt_stack_fini(const struct lu_env *env,
3831                            struct mdt_device *m, struct lu_device *top)
3832 {
3833         struct obd_device       *obd = mdt2obd_dev(m);
3834         struct lustre_cfg_bufs  *bufs;
3835         struct lustre_cfg       *lcfg;
3836         struct mdt_thread_info  *info;
3837         char                     flags[3] = "";
3838         ENTRY;
3839
3840         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3841         LASSERT(info != NULL);
3842
3843         lu_dev_del_linkage(top->ld_site, top);
3844
3845         lu_site_purge(env, top->ld_site, -1);
3846
3847         bufs = &info->mti_u.bufs;
3848         /* process cleanup, pass mdt obd name to get obd umount flags */
3849         /* another purpose is to let all layers to release their objects */
3850         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
3851         if (obd->obd_force)
3852                 strcat(flags, "F");
3853         if (obd->obd_fail)
3854                 strcat(flags, "A");
3855         lustre_cfg_bufs_set_string(bufs, 1, flags);
3856         lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
3857         if (!lcfg) {
3858                 CERROR("Cannot alloc lcfg!\n");
3859                 return;
3860         }
3861         LASSERT(top);
3862         top->ld_ops->ldo_process_config(env, top, lcfg);
3863         lustre_cfg_free(lcfg);
3864
3865         lu_site_purge(env, top->ld_site, -1);
3866
3867         m->mdt_child = NULL;
3868         m->mdt_bottom = NULL;
3869
3870         obd_disconnect(m->mdt_child_exp);
3871         m->mdt_child_exp = NULL;
3872
3873         obd_disconnect(m->mdt_bottom_exp);
3874         m->mdt_child_exp = NULL;
3875 }
3876
3877 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
3878                                const char *next, struct obd_export **exp)
3879 {
3880         struct obd_connect_data *data = NULL;
3881         struct obd_device       *obd;
3882         int                      rc;
3883         ENTRY;
3884
3885         OBD_ALLOC_PTR(data);
3886         if (data == NULL)
3887                 GOTO(out, rc = -ENOMEM);
3888
3889         obd = class_name2obd(next);
3890         if (obd == NULL) {
3891                 CERROR("%s: can't locate next device: %s\n",
3892                        mdt_obd_name(m), next);
3893                 GOTO(out, rc = -ENOTCONN);
3894         }
3895
3896         data->ocd_connect_flags = OBD_CONNECT_VERSION;
3897         data->ocd_version = LUSTRE_VERSION_CODE;
3898
3899         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
3900         if (rc) {
3901                 CERROR("%s: cannot connect to next dev %s (%d)\n",
3902                        mdt_obd_name(m), next, rc);
3903                 GOTO(out, rc);
3904         }
3905
3906 out:
3907         if (data)
3908                 OBD_FREE_PTR(data);
3909         RETURN(rc);
3910 }
3911
3912 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
3913                           struct lustre_cfg *cfg)
3914 {
3915         char                   *dev = lustre_cfg_string(cfg, 0);
3916         int                     rc, name_size, uuid_size;
3917         char                   *name, *uuid, *p;
3918         struct lustre_cfg_bufs *bufs;
3919         struct lustre_cfg      *lcfg;
3920         struct obd_device      *obd;
3921         struct lustre_profile  *lprof;
3922         struct lu_site         *site;
3923         ENTRY;
3924
3925         /* in 1.8 we had the only device in the stack - MDS.
3926          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
3927          * in 2.3 OSD is instantiated by obd_mount.c, so we need
3928          * to generate names and setup MDT, MDD. MDT will be using
3929          * generated name to connect to MDD. for MDD the next device
3930          * will be LOD with name taken from so called "profile" which
3931          * is generated by mount_option line
3932          *
3933          * 1.8 MGS generates config. commands like this:
3934          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
3935          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
3936          * 2.0 MGS generates config. commands like this:
3937          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
3938          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
3939          *                    3:lustre-MDT0000-mdtlov  4:f
3940          *
3941          * we generate MDD name from MDT one, just replacing T with D
3942          *
3943          * after all the preparations, the logical equivalent will be
3944          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
3945          *                    3:lustre-MDT0000-mdtlov  4:f
3946          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
3947          *                    3:lustre-MDD0000  4:f
3948          *
3949          *  notice we build the stack from down to top: MDD first, then MDT */
3950
3951         name_size = MAX_OBD_NAME;
3952         uuid_size = MAX_OBD_NAME;
3953
3954         OBD_ALLOC(name, name_size);
3955         OBD_ALLOC(uuid, uuid_size);
3956         if (name == NULL || uuid == NULL)
3957                 GOTO(cleanup_mem, rc = -ENOMEM);
3958
3959         OBD_ALLOC_PTR(bufs);
3960         if (!bufs)
3961                 GOTO(cleanup_mem, rc = -ENOMEM);
3962
3963         strcpy(name, dev);
3964         p = strstr(name, "-MDT");
3965         if (p == NULL)
3966                 GOTO(free_bufs, rc = -ENOMEM);
3967         p[3] = 'D';
3968
3969         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
3970
3971         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
3972         if (lprof == NULL || lprof->lp_dt == NULL) {
3973                 CERROR("can't find the profile: %s\n",
3974                        lustre_cfg_string(cfg, 0));
3975                 GOTO(free_bufs, rc = -EINVAL);
3976         }
3977
3978         lustre_cfg_bufs_reset(bufs, name);
3979         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
3980         lustre_cfg_bufs_set_string(bufs, 2, uuid);
3981         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
3982
3983         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
3984         if (!lcfg)
3985                 GOTO(free_bufs, rc = -ENOMEM);
3986
3987         rc = class_attach(lcfg);
3988         if (rc)
3989                 GOTO(lcfg_cleanup, rc);
3990
3991         obd = class_name2obd(name);
3992         if (!obd) {
3993                 CERROR("Can not find obd %s (%s in config)\n",
3994                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
3995                 GOTO(class_detach, rc = -EINVAL);
3996         }
3997
3998         lustre_cfg_free(lcfg);
3999
4000         lustre_cfg_bufs_reset(bufs, name);
4001         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4002         lustre_cfg_bufs_set_string(bufs, 2, dev);
4003         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4004
4005         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4006
4007         rc = class_setup(obd, lcfg);
4008         if (rc)
4009                 GOTO(class_detach, rc);
4010
4011         /* connect to MDD we just setup */
4012         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
4013         if (rc)
4014                 GOTO(class_detach, rc);
4015
4016         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
4017         LASSERT(site);
4018         LASSERT(mdt_lu_site(mdt) == NULL);
4019         mdt->mdt_lu_dev.ld_site = site;
4020         site->ls_top_dev = &mdt->mdt_lu_dev;
4021         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
4022
4023         /* now connect to bottom OSD */
4024         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
4025         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
4026         if (rc)
4027                 GOTO(class_detach, rc);
4028         mdt->mdt_bottom =
4029                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
4030
4031         rc = lu_env_refill((struct lu_env *)env);
4032         if (rc != 0)
4033                 CERROR("Failure to refill session: '%d'\n", rc);
4034
4035         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
4036
4037         EXIT;
4038 class_detach:
4039         if (rc)
4040                 class_detach(obd, lcfg);
4041 lcfg_cleanup:
4042         lustre_cfg_free(lcfg);
4043 free_bufs:
4044         OBD_FREE_PTR(bufs);
4045 cleanup_mem:
4046         if (name)
4047                 OBD_FREE(name, name_size);
4048         if (uuid)
4049                 OBD_FREE(uuid, uuid_size);
4050         RETURN(rc);
4051 }
4052
4053 /* setup quota master target on MDT0 */
4054 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
4055                           struct lustre_cfg *cfg)
4056 {
4057         struct obd_device       *obd;
4058         char                    *dev = lustre_cfg_string(cfg, 0);
4059         char                    *qmtname, *uuid, *p;
4060         struct lustre_cfg_bufs  *bufs;
4061         struct lustre_cfg       *lcfg;
4062         struct lustre_profile   *lprof;
4063         struct obd_connect_data *data;
4064         int                      rc;
4065         ENTRY;
4066
4067         LASSERT(mdt->mdt_qmt_exp == NULL);
4068         LASSERT(mdt->mdt_qmt_dev == NULL);
4069
4070         /* quota master is on MDT0 only for now */
4071         if (mdt->mdt_seq_site.ss_node_id != 0)
4072                 RETURN(0);
4073
4074         /* MGS generates config commands which look as follows:
4075          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4076          *                    3:lustre-MDT0000-mdtlov  4:f
4077          *
4078          * We generate the QMT name from the MDT one, just replacing MD with QM
4079          * after all the preparations, the logical equivalent will be:
4080          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
4081          *                    3:lustre-MDT0000-osd  4:f */
4082         OBD_ALLOC(qmtname, MAX_OBD_NAME);
4083         OBD_ALLOC(uuid, UUID_MAX);
4084         OBD_ALLOC_PTR(bufs);
4085         OBD_ALLOC_PTR(data);
4086         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
4087                 GOTO(cleanup_mem, rc = -ENOMEM);
4088
4089         strcpy(qmtname, dev);
4090         p = strstr(qmtname, "-MDT");
4091         if (p == NULL)
4092                 GOTO(cleanup_mem, rc = -ENOMEM);
4093         /* replace MD with QM */
4094         p[1] = 'Q';
4095         p[2] = 'M';
4096
4097         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
4098
4099         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4100         if (lprof == NULL || lprof->lp_dt == NULL) {
4101                 CERROR("can't find profile for %s\n",
4102                        lustre_cfg_string(cfg, 0));
4103                 GOTO(cleanup_mem, rc = -EINVAL);
4104         }
4105
4106         lustre_cfg_bufs_reset(bufs, qmtname);
4107         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
4108         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4109         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4110
4111         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4112         if (!lcfg)
4113                 GOTO(cleanup_mem, rc = -ENOMEM);
4114
4115         rc = class_attach(lcfg);
4116         if (rc)
4117                 GOTO(lcfg_cleanup, rc);
4118
4119         obd = class_name2obd(qmtname);
4120         if (!obd) {
4121                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
4122                        lustre_cfg_string(cfg, 0));
4123                 GOTO(class_detach, rc = -EINVAL);
4124         }
4125
4126         lustre_cfg_free(lcfg);
4127
4128         lustre_cfg_bufs_reset(bufs, qmtname);
4129         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4130         lustre_cfg_bufs_set_string(bufs, 2, dev);
4131
4132         /* for quota, the next device should be the OSD device */
4133         lustre_cfg_bufs_set_string(bufs, 3,
4134                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
4135
4136         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4137
4138         rc = class_setup(obd, lcfg);
4139         if (rc)
4140                 GOTO(class_detach, rc);
4141
4142         mdt->mdt_qmt_dev = obd->obd_lu_dev;
4143
4144         /* configure local quota objects */
4145         rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
4146                                                    &mdt->mdt_lu_dev,
4147                                                    mdt->mdt_qmt_dev);
4148         if (rc)
4149                 GOTO(class_cleanup, rc);
4150
4151         /* connect to quota master target */
4152         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4153         data->ocd_version = LUSTRE_VERSION_CODE;
4154         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
4155                          data, NULL);
4156         if (rc) {
4157                 CERROR("cannot connect to quota master device %s (%d)\n",
4158                        qmtname, rc);
4159                 GOTO(class_cleanup, rc);
4160         }
4161
4162         EXIT;
4163 class_cleanup:
4164         if (rc) {
4165                 class_manual_cleanup(obd);
4166                 mdt->mdt_qmt_dev = NULL;
4167         }
4168 class_detach:
4169         if (rc)
4170                 class_detach(obd, lcfg);
4171 lcfg_cleanup:
4172         lustre_cfg_free(lcfg);
4173 cleanup_mem:
4174         if (bufs)
4175                 OBD_FREE_PTR(bufs);
4176         if (qmtname)
4177                 OBD_FREE(qmtname, MAX_OBD_NAME);
4178         if (uuid)
4179                 OBD_FREE(uuid, UUID_MAX);
4180         if (data)
4181                 OBD_FREE_PTR(data);
4182         return rc;
4183 }
4184
4185 /* Shutdown quota master target associated with mdt */
4186 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
4187 {
4188         ENTRY;
4189
4190         if (mdt->mdt_qmt_exp == NULL)
4191                 RETURN_EXIT;
4192         LASSERT(mdt->mdt_qmt_dev != NULL);
4193
4194         /* the qmt automatically shuts down when the mdt disconnects */
4195         obd_disconnect(mdt->mdt_qmt_exp);
4196         mdt->mdt_qmt_exp = NULL;
4197         mdt->mdt_qmt_dev = NULL;
4198         EXIT;
4199 }
4200
4201 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
4202  * for now. This will be removed along with converting rest of MDT code
4203  * to use tgt_session_info */
4204 int mdt_tgt_getxattr(struct tgt_session_info *tsi)
4205 {
4206         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
4207         int                      rc;
4208
4209         rc = mdt_getxattr(info);
4210
4211         mdt_thread_info_fini(info);
4212         return rc;
4213 }
4214
4215 static struct tgt_handler mdt_tgt_handlers[] = {
4216 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4217                 0,                      MDS_CONNECT,    mdt_tgt_connect,
4218                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
4219 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4220                 0,                      MDS_DISCONNECT, tgt_disconnect,
4221                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
4222 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4223                 HABEO_REFERO,           MDS_SET_INFO,   mdt_set_info,
4224                 &RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION),
4225 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
4226 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_GETSTATUS,  mdt_getstatus),
4227 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETATTR,    mdt_getattr),
4228 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_GETATTR_NAME,
4229                                                         mdt_getattr_name),
4230 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETXATTR,   mdt_tgt_getxattr),
4231 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_STATFS,     mdt_statfs),
4232 TGT_MDT_HDL(0           | MUTABOR,      MDS_REINT,      mdt_reint),
4233 TGT_MDT_HDL(HABEO_CORPUS,               MDS_CLOSE,      mdt_close),
4234 TGT_MDT_HDL(HABEO_CORPUS,               MDS_DONE_WRITING,
4235                                                         mdt_done_writing),
4236 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_READPAGE,   mdt_readpage),
4237 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_SYNC,       mdt_sync),
4238 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
4239 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_PROGRESS,
4240                                                         mdt_hsm_progress),
4241 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_REGISTER,
4242                                                         mdt_hsm_ct_register),
4243 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_UNREGISTER,
4244                                                         mdt_hsm_ct_unregister),
4245 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_STATE_GET,
4246                                                         mdt_hsm_state_get),
4247 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_STATE_SET,
4248                                                         mdt_hsm_state_set),
4249 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_ACTION, mdt_hsm_action),
4250 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_REQUEST,
4251                                                         mdt_hsm_request),
4252 TGT_MDT_HDL(HABEO_CLAVIS | HABEO_CORPUS | HABEO_REFERO | MUTABOR,
4253             MDS_SWAP_LAYOUTS,
4254             mdt_swap_layouts),
4255 };
4256
4257 static struct tgt_handler mdt_sec_ctx_ops[] = {
4258 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
4259 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
4260 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
4261 };
4262
4263 static struct tgt_handler mdt_quota_ops[] = {
4264 TGT_QUOTA_HDL(HABEO_REFERO,             QUOTA_DQACQ,      mdt_quota_dqacq),
4265 };
4266
4267 static struct tgt_opc_slice mdt_common_slice[] = {
4268         {
4269                 .tos_opc_start  = MDS_FIRST_OPC,
4270                 .tos_opc_end    = MDS_LAST_OPC,
4271                 .tos_hs         = mdt_tgt_handlers
4272         },
4273         {
4274                 .tos_opc_start  = OBD_FIRST_OPC,
4275                 .tos_opc_end    = OBD_LAST_OPC,
4276                 .tos_hs         = tgt_obd_handlers
4277         },
4278         {
4279                 .tos_opc_start  = LDLM_FIRST_OPC,
4280                 .tos_opc_end    = LDLM_LAST_OPC,
4281                 .tos_hs         = tgt_dlm_handlers
4282         },
4283         {
4284                 .tos_opc_start  = SEC_FIRST_OPC,
4285                 .tos_opc_end    = SEC_LAST_OPC,
4286                 .tos_hs         = mdt_sec_ctx_ops
4287         },
4288         {
4289                 .tos_opc_start  = OUT_UPDATE_FIRST_OPC,
4290                 .tos_opc_end    = OUT_UPDATE_LAST_OPC,
4291                 .tos_hs         = tgt_out_handlers
4292         },
4293         {
4294                 .tos_opc_start  = FLD_FIRST_OPC,
4295                 .tos_opc_end    = FLD_LAST_OPC,
4296                 .tos_hs         = fld_handlers
4297         },
4298         {
4299                 .tos_opc_start  = SEQ_FIRST_OPC,
4300                 .tos_opc_end    = SEQ_LAST_OPC,
4301                 .tos_hs         = seq_handlers
4302         },
4303         {
4304                 .tos_opc_start  = QUOTA_DQACQ,
4305                 .tos_opc_end    = QUOTA_LAST_OPC,
4306                 .tos_hs         = mdt_quota_ops
4307         },
4308         {
4309                 .tos_opc_start  = LLOG_FIRST_OPC,
4310                 .tos_opc_end    = LLOG_LAST_OPC,
4311                 .tos_hs         = tgt_llog_handlers
4312         },
4313         {
4314                 .tos_opc_start  = LFSCK_FIRST_OPC,
4315                 .tos_opc_end    = LFSCK_LAST_OPC,
4316                 .tos_hs         = tgt_lfsck_handlers
4317         },
4318
4319         {
4320                 .tos_hs         = NULL
4321         }
4322 };
4323
4324 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4325 {
4326         struct md_device        *next = m->mdt_child;
4327         struct lu_device        *d    = &m->mdt_lu_dev;
4328         struct obd_device       *obd  = mdt2obd_dev(m);
4329         struct lfsck_stop        stop;
4330         ENTRY;
4331
4332         stop.ls_status = LS_PAUSED;
4333         stop.ls_flags = 0;
4334         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
4335
4336         target_recovery_fini(obd);
4337         ping_evictor_stop();
4338         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
4339
4340         if (m->mdt_opts.mo_coordinator)
4341                 mdt_hsm_cdt_stop(m);
4342
4343         mdt_hsm_cdt_fini(m);
4344
4345         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
4346         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4347         obd_exports_barrier(obd);
4348         obd_zombie_barrier();
4349
4350         mdt_procfs_fini(m);
4351
4352         tgt_fini(env, &m->mdt_lut);
4353         mdt_fs_cleanup(env, m);
4354         upcall_cache_cleanup(m->mdt_identity_cache);
4355         m->mdt_identity_cache = NULL;
4356
4357         if (m->mdt_namespace != NULL) {
4358                 ldlm_namespace_free(m->mdt_namespace, NULL,
4359                                     d->ld_obd->obd_force);
4360                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4361         }
4362
4363         mdt_quota_fini(env, m);
4364
4365         cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
4366
4367         mdt_seq_fini(env, m);
4368         mdt_fld_fini(env, m);
4369
4370         next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL);
4371         cfs_timer_disarm(&m->mdt_ck_timer);
4372         mdt_ck_thread_stop(m);
4373
4374         /*
4375          * Finish the stack
4376          */
4377         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4378
4379         LASSERT(atomic_read(&d->ld_ref) == 0);
4380
4381         server_put_mount(mdt_obd_name(m), true);
4382
4383         EXIT;
4384 }
4385
4386 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4387
4388 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4389                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
4390 {
4391         struct mdt_thread_info    *info;
4392         struct obd_device         *obd;
4393         const char                *dev = lustre_cfg_string(cfg, 0);
4394         const char                *num = lustre_cfg_string(cfg, 2);
4395         struct lustre_mount_info  *lmi = NULL;
4396         struct lustre_sb_info     *lsi;
4397         struct lu_site            *s;
4398         struct seq_server_site    *ss_site;
4399         const char                *identity_upcall = "NONE";
4400         struct md_device          *next;
4401         int                        rc;
4402         long                       node_id;
4403         mntopt_t                   mntopts;
4404         ENTRY;
4405
4406         lu_device_init(&m->mdt_lu_dev, ldt);
4407         /*
4408          * Environment (env) might be missing mdt_thread_key values at that
4409          * point, if device is allocated when mdt_thread_key is in QUIESCENT
4410          * mode.
4411          *
4412          * Usually device allocation path doesn't use module key values, but
4413          * mdt has to do a lot of work here, so allocate key value.
4414          */
4415         rc = lu_env_refill((struct lu_env *)env);
4416         if (rc != 0)
4417                 RETURN(rc);
4418
4419         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4420         LASSERT(info != NULL);
4421
4422         obd = class_name2obd(dev);
4423         LASSERT(obd != NULL);
4424
4425         m->mdt_max_mdsize = MAX_MD_SIZE; /* 4 stripes */
4426
4427         m->mdt_som_conf = 0;
4428
4429         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4430
4431         /* default is coordinator off, it is started through conf_param
4432          * or /proc */
4433         m->mdt_opts.mo_coordinator = 0;
4434
4435         lmi = server_get_mount(dev);
4436         if (lmi == NULL) {
4437                 CERROR("Cannot get mount info for %s!\n", dev);
4438                 RETURN(-EFAULT);
4439         } else {
4440                 lsi = s2lsi(lmi->lmi_sb);
4441                 /* CMD is supported only in IAM mode */
4442                 LASSERT(num);
4443                 node_id = simple_strtol(num, NULL, 10);
4444                 obd->u.obt.obt_magic = OBT_MAGIC;
4445         }
4446
4447         spin_lock_init(&m->mdt_ioepoch_lock);
4448         m->mdt_capa_timeout = CAPA_TIMEOUT;
4449         m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1;
4450         m->mdt_ck_timeout = CAPA_KEY_TIMEOUT;
4451         m->mdt_squash.rsi_uid = 0;
4452         m->mdt_squash.rsi_gid = 0;
4453         INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
4454         init_rwsem(&m->mdt_squash.rsi_sem);
4455         spin_lock_init(&m->mdt_osfs_lock);
4456         m->mdt_osfs_age = cfs_time_shift_64(-1000);
4457         m->mdt_enable_remote_dir = 0;
4458         m->mdt_enable_remote_dir_gid = 0;
4459
4460         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
4461         m->mdt_lu_dev.ld_obd = obd;
4462         /* Set this lu_device to obd for error handling purposes. */
4463         obd->obd_lu_dev = &m->mdt_lu_dev;
4464
4465         /* init the stack */
4466         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
4467         if (rc) {
4468                 CERROR("%s: Can't init device stack, rc %d\n",
4469                        mdt_obd_name(m), rc);
4470                 GOTO(err_lmi, rc);
4471         }
4472
4473         s = mdt_lu_site(m);
4474         ss_site = mdt_seq_site(m);
4475         s->ld_seq_site = ss_site;
4476         ss_site->ss_lu = s;
4477
4478         /* set server index */
4479         ss_site->ss_node_id = node_id;
4480
4481         /* failover is the default
4482          * FIXME: we do not failout mds0/mgs, which may cause some problems.
4483          * assumed whose ss_node_id == 0 XXX
4484          * */
4485         obd->obd_replayable = 1;
4486         /* No connection accepted until configurations will finish */
4487         obd->obd_no_conn = 1;
4488
4489         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4490                 char *str = lustre_cfg_string(cfg, 4);
4491                 if (strchr(str, 'n')) {
4492                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
4493                         obd->obd_replayable = 0;
4494                 }
4495         }
4496
4497         rc = mdt_fld_init(env, mdt_obd_name(m), m);
4498         if (rc)
4499                 GOTO(err_fini_stack, rc);
4500
4501         rc = mdt_seq_init(env, m);
4502         if (rc)
4503                 GOTO(err_fini_fld, rc);
4504
4505         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
4506                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
4507         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4508                                               LDLM_NAMESPACE_SERVER,
4509                                               LDLM_NAMESPACE_GREEDY,
4510                                               LDLM_NS_TYPE_MDT);
4511         if (m->mdt_namespace == NULL)
4512                 GOTO(err_fini_seq, rc = -ENOMEM);
4513
4514         m->mdt_namespace->ns_lvbp = m;
4515         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
4516
4517         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4518         /* set obd_namespace for compatibility with old code */
4519         obd->obd_namespace = m->mdt_namespace;
4520
4521         cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m);
4522
4523         rc = mdt_hsm_cdt_init(m);
4524         if (rc != 0) {
4525                 CERROR("%s: error initializing coordinator, rc %d\n",
4526                        mdt_obd_name(m), rc);
4527                 GOTO(err_free_ns, rc);
4528         }
4529
4530         rc = mdt_ck_thread_start(m);
4531         if (rc)
4532                 GOTO(err_free_hsm, rc);
4533
4534         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
4535                       OBD_FAIL_MDS_ALL_REQUEST_NET,
4536                       OBD_FAIL_MDS_ALL_REPLY_NET);
4537         if (rc)
4538                 GOTO(err_capa, rc);
4539
4540         rc = mdt_fs_setup(env, m, obd, lsi);
4541         if (rc)
4542                 GOTO(err_tgt, rc);
4543
4544         tgt_adapt_sptlrpc_conf(&m->mdt_lut, 1);
4545
4546         next = m->mdt_child;
4547         rc = next->md_ops->mdo_iocontrol(env, next, OBD_IOC_GET_MNTOPT, 0,
4548                                          &mntopts);
4549         if (rc)
4550                 GOTO(err_fs_cleanup, rc);
4551
4552         if (mntopts & MNTOPT_USERXATTR)
4553                 m->mdt_opts.mo_user_xattr = 1;
4554         else
4555                 m->mdt_opts.mo_user_xattr = 0;
4556
4557         rc = next->md_ops->mdo_maxeasize_get(env, next, &m->mdt_max_ea_size);
4558         if (rc)
4559                 GOTO(err_fs_cleanup, rc);
4560
4561         if (mntopts & MNTOPT_ACL)
4562                 m->mdt_opts.mo_acl = 1;
4563         else
4564                 m->mdt_opts.mo_acl = 0;
4565
4566         /* XXX: to support suppgid for ACL, we enable identity_upcall
4567          * by default, otherwise, maybe got unexpected -EACCESS. */
4568         if (m->mdt_opts.mo_acl)
4569                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4570
4571         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
4572                                                 identity_upcall,
4573                                                 &mdt_identity_upcall_cache_ops);
4574         if (IS_ERR(m->mdt_identity_cache)) {
4575                 rc = PTR_ERR(m->mdt_identity_cache);
4576                 m->mdt_identity_cache = NULL;
4577                 GOTO(err_fs_cleanup, rc);
4578         }
4579
4580         rc = mdt_procfs_init(m, dev);
4581         if (rc) {
4582                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4583                 GOTO(err_recovery, rc);
4584         }
4585
4586         rc = mdt_quota_init(env, m, cfg);
4587         if (rc)
4588                 GOTO(err_procfs, rc);
4589
4590         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
4591         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
4592                            "mdt_ldlm_client", m->mdt_ldlm_client);
4593
4594         ping_evictor_start();
4595
4596         /* recovery will be started upon mdt_prepare()
4597          * when the whole stack is complete and ready
4598          * to serve the requests */
4599
4600         mdt_init_capa_ctxt(env, m);
4601
4602         /* Reduce the initial timeout on an MDS because it doesn't need such
4603          * a long timeout as an OST does. Adaptive timeouts will adjust this
4604          * value appropriately. */
4605         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4606                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4607
4608         RETURN(0);
4609 err_procfs:
4610         mdt_procfs_fini(m);
4611 err_recovery:
4612         target_recovery_fini(obd);
4613         upcall_cache_cleanup(m->mdt_identity_cache);
4614         m->mdt_identity_cache = NULL;
4615 err_fs_cleanup:
4616         mdt_fs_cleanup(env, m);
4617 err_tgt:
4618         tgt_fini(env, &m->mdt_lut);
4619 err_capa:
4620         cfs_timer_disarm(&m->mdt_ck_timer);
4621         mdt_ck_thread_stop(m);
4622 err_free_hsm:
4623         mdt_hsm_cdt_fini(m);
4624 err_free_ns:
4625         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4626         obd->obd_namespace = m->mdt_namespace = NULL;
4627 err_fini_seq:
4628         mdt_seq_fini(env, m);
4629 err_fini_fld:
4630         mdt_fld_fini(env, m);
4631 err_fini_stack:
4632         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4633 err_lmi:
4634         if (lmi)
4635                 server_put_mount(dev, true);
4636         return(rc);
4637 }
4638
4639 /* For interoperability, the left element is old parameter, the right one
4640  * is the new version of the parameter, if some parameter is deprecated,
4641  * the new version should be set as NULL. */
4642 static struct cfg_interop_param mdt_interop_param[] = {
4643         { "mdt.group_upcall",   NULL },
4644         { "mdt.quota_type",     NULL },
4645         { "mdd.quota_type",     NULL },
4646         { "mdt.rootsquash",     "mdt.root_squash" },
4647         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
4648         { NULL }
4649 };
4650
4651 /* used by MGS to process specific configurations */
4652 static int mdt_process_config(const struct lu_env *env,
4653                               struct lu_device *d, struct lustre_cfg *cfg)
4654 {
4655         struct mdt_device *m = mdt_dev(d);
4656         struct md_device *md_next = m->mdt_child;
4657         struct lu_device *next = md2lu_dev(md_next);
4658         int rc;
4659         ENTRY;
4660
4661         switch (cfg->lcfg_command) {
4662         case LCFG_PARAM: {
4663                 struct obd_device          *obd = d->ld_obd;
4664
4665                 /* For interoperability */
4666                 struct cfg_interop_param   *ptr = NULL;
4667                 struct lustre_cfg          *old_cfg = NULL;
4668                 char                       *param = NULL;
4669
4670                 param = lustre_cfg_string(cfg, 1);
4671                 if (param == NULL) {
4672                         CERROR("param is empty\n");
4673                         rc = -EINVAL;
4674                         break;
4675                 }
4676
4677                 ptr = class_find_old_param(param, mdt_interop_param);
4678                 if (ptr != NULL) {
4679                         if (ptr->new_param == NULL) {
4680                                 rc = 0;
4681                                 CWARN("For interoperability, skip this %s."
4682                                       " It is obsolete.\n", ptr->old_param);
4683                                 break;
4684                         }
4685
4686                         CWARN("Found old param %s, changed it to %s.\n",
4687                               ptr->old_param, ptr->new_param);
4688
4689                         old_cfg = cfg;
4690                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
4691                         if (IS_ERR(cfg)) {
4692                                 rc = PTR_ERR(cfg);
4693                                 break;
4694                         }
4695                 }
4696
4697                 rc = class_process_proc_seq_param(PARAM_MDT, obd->obd_vars,
4698                                                         cfg, obd);
4699                 if (rc > 0 || rc == -ENOSYS) {
4700                         /* is it an HSM var ? */
4701                         rc = class_process_proc_seq_param(PARAM_HSM,
4702                                                         hsm_cdt_get_proc_vars(),
4703                                                         cfg, obd);
4704                         if (rc > 0 || rc == -ENOSYS)
4705                                 /* we don't understand; pass it on */
4706                                 rc = next->ld_ops->ldo_process_config(env, next,
4707                                                                       cfg);
4708                 }
4709
4710                 if (old_cfg != NULL)
4711                         lustre_cfg_free(cfg);
4712
4713                 break;
4714         }
4715         default:
4716                 /* others are passed further */
4717                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4718                 break;
4719         }
4720         RETURN(rc);
4721 }
4722
4723 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4724                                           const struct lu_object_header *hdr,
4725                                           struct lu_device *d)
4726 {
4727         struct mdt_object *mo;
4728
4729         ENTRY;
4730
4731         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
4732         if (mo != NULL) {
4733                 struct lu_object *o;
4734                 struct lu_object_header *h;
4735
4736                 o = &mo->mot_obj;
4737                 h = &mo->mot_header;
4738                 lu_object_header_init(h);
4739                 lu_object_init(o, h, d);
4740                 lu_object_add_top(h, o);
4741                 o->lo_ops = &mdt_obj_ops;
4742                 mutex_init(&mo->mot_ioepoch_mutex);
4743                 mutex_init(&mo->mot_lov_mutex);
4744                 init_rwsem(&mo->mot_open_sem);
4745                 RETURN(o);
4746         }
4747         RETURN(NULL);
4748 }
4749
4750 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
4751                            const struct lu_object_conf *unused)
4752 {
4753         struct mdt_device *d = mdt_dev(o->lo_dev);
4754         struct lu_device  *under;
4755         struct lu_object  *below;
4756         int                rc = 0;
4757         ENTRY;
4758
4759         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4760                PFID(lu_object_fid(o)));
4761
4762         under = &d->mdt_child->md_lu_dev;
4763         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4764         if (below != NULL) {
4765                 lu_object_add(o, below);
4766         } else
4767                 rc = -ENOMEM;
4768
4769         RETURN(rc);
4770 }
4771
4772 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4773 {
4774         struct mdt_object *mo = mdt_obj(o);
4775         struct lu_object_header *h;
4776         ENTRY;
4777
4778         h = o->lo_header;
4779         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4780                PFID(lu_object_fid(o)));
4781
4782         LASSERT(atomic_read(&mo->mot_open_count) == 0);
4783         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
4784
4785         lu_object_fini(o);
4786         lu_object_header_fini(h);
4787         OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
4788
4789         EXIT;
4790 }
4791
4792 static int mdt_object_print(const struct lu_env *env, void *cookie,
4793                             lu_printer_t p, const struct lu_object *o)
4794 {
4795         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
4796         return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p(ioepoch="LPU64" "
4797                     "flags="LPX64", epochcount=%d, writecount=%d)",
4798                     mdto, mdto->mot_ioepoch, mdto->mot_flags,
4799                     mdto->mot_ioepoch_count, mdto->mot_writecount);
4800 }
4801
4802 static int mdt_prepare(const struct lu_env *env,
4803                 struct lu_device *pdev,
4804                 struct lu_device *cdev)
4805 {
4806         struct mdt_device *mdt = mdt_dev(cdev);
4807         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
4808         struct obd_device *obd = cdev->ld_obd;
4809         int rc;
4810
4811         ENTRY;
4812
4813         LASSERT(obd);
4814
4815         rc = next->ld_ops->ldo_prepare(env, cdev, next);
4816         if (rc)
4817                 RETURN(rc);
4818
4819         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
4820         if (rc)
4821                 RETURN(rc);
4822
4823         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
4824         if (rc)
4825                 RETURN(rc);
4826
4827         rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
4828         /* The LFSCK instance is registered just now, so it must be there when
4829          * register the namespace to such instance. */
4830         LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
4831
4832         if (mdt->mdt_seq_site.ss_node_id == 0) {
4833                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
4834                                                          &mdt->mdt_md_root_fid);
4835                 if (rc)
4836                         RETURN(rc);
4837         }
4838
4839         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
4840         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
4841         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
4842         LASSERT(obd->obd_no_conn);
4843         spin_lock(&obd->obd_dev_lock);
4844         obd->obd_no_conn = 0;
4845         spin_unlock(&obd->obd_dev_lock);
4846
4847         if (obd->obd_recovering == 0)
4848                 mdt_postrecov(env, mdt);
4849
4850         RETURN(rc);
4851 }
4852
4853 const struct lu_device_operations mdt_lu_ops = {
4854         .ldo_object_alloc   = mdt_object_alloc,
4855         .ldo_process_config = mdt_process_config,
4856         .ldo_prepare        = mdt_prepare,
4857 };
4858
4859 static const struct lu_object_operations mdt_obj_ops = {
4860         .loo_object_init    = mdt_object_init,
4861         .loo_object_free    = mdt_object_free,
4862         .loo_object_print   = mdt_object_print
4863 };
4864
4865 static int mdt_obd_set_info_async(const struct lu_env *env,
4866                                   struct obd_export *exp,
4867                                   __u32 keylen, void *key,
4868                                   __u32 vallen, void *val,
4869                                   struct ptlrpc_request_set *set)
4870 {
4871         int rc;
4872
4873         ENTRY;
4874
4875         if (KEY_IS(KEY_SPTLRPC_CONF)) {
4876                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp), 0);
4877                 RETURN(rc);
4878         }
4879
4880         RETURN(0);
4881 }
4882
4883 /**
4884  * Match client and server connection feature flags.
4885  *
4886  * Compute the compatibility flags for a connection request based on
4887  * features mutually supported by client and server.
4888  *
4889  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
4890  * must not be updated here, otherwise a partially initialized value may
4891  * be exposed. After the connection request is successfully processed,
4892  * the top-level MDT connect request handler atomically updates the export
4893  * connect flags from the obd_connect_data::ocd_connect_flags field of the
4894  * reply. \see mdt_connect().
4895  *
4896  * \param exp   the obd_export associated with this client/target pair
4897  * \param mdt   the target device for the connection
4898  * \param data  stores data for this connect request
4899  *
4900  * \retval 0       success
4901  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
4902  * \retval -EBADE  client and server feature requirements are incompatible
4903  */
4904 static int mdt_connect_internal(struct obd_export *exp,
4905                                 struct mdt_device *mdt,
4906                                 struct obd_connect_data *data)
4907 {
4908         LASSERT(data != NULL);
4909
4910         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
4911         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
4912
4913         if (!(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
4914             !(data->ocd_connect_flags & OBD_CONNECT_IBITS)) {
4915                 CWARN("%s: client %s does not support ibits lock, either "
4916                       "very old or an invalid client: flags "LPX64"\n",
4917                       mdt_obd_name(mdt), exp->exp_client_uuid.uuid,
4918                       data->ocd_connect_flags);
4919                 return -EBADE;
4920         }
4921
4922         if (!mdt->mdt_opts.mo_acl)
4923                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
4924
4925         if (!mdt->mdt_opts.mo_user_xattr)
4926                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
4927
4928         if (!mdt->mdt_som_conf)
4929                 data->ocd_connect_flags &= ~OBD_CONNECT_SOM;
4930
4931         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
4932                 data->ocd_brw_size = min(data->ocd_brw_size,
4933                                          (__u32)MD_MAX_BRW_SIZE);
4934                 if (data->ocd_brw_size == 0) {
4935                         CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
4936                                " ocd_version: %x ocd_grant: %d "
4937                                "ocd_index: %u ocd_brw_size is "
4938                                "unexpectedly zero, network data "
4939                                "corruption? Refusing connection of this"
4940                                " client\n",
4941                                mdt_obd_name(mdt),
4942                                exp->exp_client_uuid.uuid,
4943                                exp, data->ocd_connect_flags, data->ocd_version,
4944                                data->ocd_grant, data->ocd_index);
4945                         return -EPROTO;
4946                 }
4947         }
4948
4949         /* NB: Disregard the rule against updating
4950          * exp_connect_data.ocd_connect_flags in this case, since
4951          * tgt_client_new() needs to know if this is a lightweight
4952          * connection, and it is safe to expose this flag before
4953          * connection processing completes. */
4954         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
4955                 spin_lock(&exp->exp_lock);
4956                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
4957                 spin_unlock(&exp->exp_lock);
4958         }
4959
4960         data->ocd_version = LUSTRE_VERSION_CODE;
4961
4962         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
4963                 CWARN("%s: MDS requires FID support, but client not\n",
4964                       mdt_obd_name(mdt));
4965                 return -EBADE;
4966         }
4967
4968         if (mdt->mdt_som_conf &&
4969             !(data->ocd_connect_flags & (OBD_CONNECT_LIGHTWEIGHT |
4970                                          OBD_CONNECT_MDS_MDS |
4971                                          OBD_CONNECT_SOM))) {
4972                 CWARN("%s: MDS has SOM enabled, but client does not support "
4973                       "it\n", mdt_obd_name(mdt));
4974                 return -EBADE;
4975         }
4976
4977         if (OCD_HAS_FLAG(data, PINGLESS)) {
4978                 if (ptlrpc_pinger_suppress_pings()) {
4979                         spin_lock(&exp->exp_obd->obd_dev_lock);
4980                         list_del_init(&exp->exp_obd_chain_timed);
4981                         spin_unlock(&exp->exp_obd->obd_dev_lock);
4982                 } else {
4983                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
4984                 }
4985         }
4986
4987         data->ocd_max_easize = mdt->mdt_max_ea_size;
4988
4989         return 0;
4990 }
4991
4992 /* mds_connect copy */
4993 static int mdt_obd_connect(const struct lu_env *env,
4994                            struct obd_export **exp, struct obd_device *obd,
4995                            struct obd_uuid *cluuid,
4996                            struct obd_connect_data *data,
4997                            void *localdata)
4998 {
4999         struct obd_export      *lexp;
5000         struct lustre_handle    conn = { 0 };
5001         struct mdt_device      *mdt;
5002         int                     rc;
5003         ENTRY;
5004
5005         LASSERT(env != NULL);
5006         if (!exp || !obd || !cluuid)
5007                 RETURN(-EINVAL);
5008
5009         mdt = mdt_dev(obd->obd_lu_dev);
5010
5011         /*
5012          * first, check whether the stack is ready to handle requests
5013          * XXX: probably not very appropriate method is used now
5014          *      at some point we should find a better one
5015          */
5016         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) && data != NULL &&
5017             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
5018                 rc = obd_get_info(env, mdt->mdt_child_exp,
5019                                   sizeof(KEY_OSP_CONNECTED),
5020                                   KEY_OSP_CONNECTED, NULL, NULL, NULL);
5021                 if (rc)
5022                         RETURN(-EAGAIN);
5023                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
5024         }
5025
5026         rc = class_connect(&conn, obd, cluuid);
5027         if (rc)
5028                 RETURN(rc);
5029
5030         lexp = class_conn2export(&conn);
5031         LASSERT(lexp != NULL);
5032
5033         rc = mdt_connect_internal(lexp, mdt, data);
5034         if (rc == 0) {
5035                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5036
5037                 LASSERT(lcd);
5038                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5039                 rc = tgt_client_new(env, lexp);
5040                 if (rc == 0)
5041                         mdt_export_stats_init(obd, lexp, localdata);
5042
5043                 /* For phase I, sync for cross-ref operation. */
5044                 lexp->exp_keep_sync = 1;
5045         }
5046
5047         if (rc != 0) {
5048                 class_disconnect(lexp);
5049                 *exp = NULL;
5050         } else {
5051                 *exp = lexp;
5052         }
5053
5054         RETURN(rc);
5055 }
5056
5057 static int mdt_obd_reconnect(const struct lu_env *env,
5058                              struct obd_export *exp, struct obd_device *obd,
5059                              struct obd_uuid *cluuid,
5060                              struct obd_connect_data *data,
5061                              void *localdata)
5062 {
5063         int                     rc;
5064         ENTRY;
5065
5066         if (exp == NULL || obd == NULL || cluuid == NULL)
5067                 RETURN(-EINVAL);
5068
5069         rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
5070         if (rc == 0)
5071                 mdt_export_stats_init(obd, exp, localdata);
5072
5073         RETURN(rc);
5074 }
5075
5076 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
5077                                    struct mdt_thread_info *info,
5078                                    struct mdt_file_data *mfd)
5079 {
5080         struct lu_context ses;
5081         int rc;
5082         ENTRY;
5083
5084         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
5085         if (rc)
5086                 RETURN(rc);
5087
5088         env->le_ses = &ses;
5089         lu_context_enter(&ses);
5090
5091         mdt_ucred(info)->uc_valid = UCRED_OLD;
5092         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
5093
5094         lu_context_exit(&ses);
5095         lu_context_fini(&ses);
5096         env->le_ses = NULL;
5097
5098         RETURN(rc);
5099 }
5100
5101 static int mdt_export_cleanup(struct obd_export *exp)
5102 {
5103         struct mdt_export_data *med = &exp->exp_mdt_data;
5104         struct obd_device      *obd = exp->exp_obd;
5105         struct mdt_device      *mdt;
5106         struct mdt_thread_info *info;
5107         struct lu_env           env;
5108         CFS_LIST_HEAD(closing_list);
5109         struct mdt_file_data *mfd, *n;
5110         int rc = 0;
5111         ENTRY;
5112
5113         spin_lock(&med->med_open_lock);
5114         while (!cfs_list_empty(&med->med_open_head)) {
5115                 cfs_list_t *tmp = med->med_open_head.next;
5116                 mfd = cfs_list_entry(tmp, struct mdt_file_data, mfd_list);
5117
5118                 /* Remove mfd handle so it can't be found again.
5119                  * We are consuming the mfd_list reference here. */
5120                 class_handle_unhash(&mfd->mfd_handle);
5121                 cfs_list_move_tail(&mfd->mfd_list, &closing_list);
5122         }
5123         spin_unlock(&med->med_open_lock);
5124         mdt = mdt_dev(obd->obd_lu_dev);
5125         LASSERT(mdt != NULL);
5126
5127         rc = lu_env_init(&env, LCT_MD_THREAD);
5128         if (rc)
5129                 RETURN(rc);
5130
5131         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5132         LASSERT(info != NULL);
5133         memset(info, 0, sizeof *info);
5134         info->mti_env = &env;
5135         info->mti_mdt = mdt;
5136         info->mti_exp = exp;
5137
5138         if (!cfs_list_empty(&closing_list)) {
5139                 struct md_attr *ma = &info->mti_attr;
5140
5141                 /* Close any open files (which may also cause orphan unlinking). */
5142                 cfs_list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5143                         cfs_list_del_init(&mfd->mfd_list);
5144                         ma->ma_need = ma->ma_valid = 0;
5145
5146                         /* This file is being closed due to an eviction, it
5147                          * could have been modified and now dirty regarding to
5148                          * HSM archive, check this!
5149                          * The logic here is to mark a file dirty if there's a
5150                          * chance it was dirtied before the client was evicted,
5151                          * so that we don't have to wait for a release attempt
5152                          * before finding out the file was actually dirty and
5153                          * fail the release. Aggressively marking it dirty here
5154                          * will cause the policy engine to attempt to
5155                          * re-archive it; when rearchiving, we can compare the
5156                          * current version to the HSM data_version and make the
5157                          * archive request into a noop if it's not actually
5158                          * dirty.
5159                          */
5160                         if (mfd->mfd_mode & (FMODE_WRITE|MDS_FMODE_TRUNC))
5161                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
5162
5163                         /* Don't unlink orphan on failover umount, LU-184 */
5164                         if (exp->exp_flags & OBD_OPT_FAILOVER) {
5165                                 ma->ma_valid = MA_FLAGS;
5166                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5167                         }
5168                         mdt_mfd_close(info, mfd);
5169                 }
5170         }
5171         info->mti_mdt = NULL;
5172         /* cleanup client slot early */
5173         /* Do not erase record for recoverable client. */
5174         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5175                 tgt_client_del(&env, exp);
5176         lu_env_fini(&env);
5177
5178         RETURN(rc);
5179 }
5180
5181 static int mdt_obd_disconnect(struct obd_export *exp)
5182 {
5183         int rc;
5184         ENTRY;
5185
5186         LASSERT(exp);
5187         class_export_get(exp);
5188
5189         rc = server_disconnect_export(exp);
5190         if (rc != 0)
5191                 CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
5192
5193         rc = mdt_export_cleanup(exp);
5194         class_export_put(exp);
5195         RETURN(rc);
5196 }
5197
5198 /* FIXME: Can we avoid using these two interfaces? */
5199 static int mdt_init_export(struct obd_export *exp)
5200 {
5201         struct mdt_export_data *med = &exp->exp_mdt_data;
5202         int                     rc;
5203         ENTRY;
5204
5205         CFS_INIT_LIST_HEAD(&med->med_open_head);
5206         spin_lock_init(&med->med_open_lock);
5207         mutex_init(&med->med_idmap_mutex);
5208         med->med_idmap = NULL;
5209         spin_lock(&exp->exp_lock);
5210         exp->exp_connecting = 1;
5211         spin_unlock(&exp->exp_lock);
5212
5213         /* self-export doesn't need client data and ldlm initialization */
5214         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5215                                      &exp->exp_client_uuid)))
5216                 RETURN(0);
5217
5218         rc = tgt_client_alloc(exp);
5219         if (rc)
5220                 GOTO(err, rc);
5221
5222         rc = ldlm_init_export(exp);
5223         if (rc)
5224                 GOTO(err_free, rc);
5225
5226         RETURN(rc);
5227
5228 err_free:
5229         tgt_client_free(exp);
5230 err:
5231         CERROR("%s: Failed to initialize export: rc = %d\n",
5232                exp->exp_obd->obd_name, rc);
5233         return rc;
5234 }
5235
5236 static int mdt_destroy_export(struct obd_export *exp)
5237 {
5238         ENTRY;
5239
5240         if (exp_connect_rmtclient(exp))
5241                 mdt_cleanup_idmap(&exp->exp_mdt_data);
5242
5243         target_destroy_export(exp);
5244         /* destroy can be called from failed obd_setup, so
5245          * checking uuid is safer than obd_self_export */
5246         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5247                                      &exp->exp_client_uuid)))
5248                 RETURN(0);
5249
5250         ldlm_destroy_export(exp);
5251         tgt_client_free(exp);
5252
5253         LASSERT(cfs_list_empty(&exp->exp_outstanding_replies));
5254         LASSERT(cfs_list_empty(&exp->exp_mdt_data.med_open_head));
5255
5256         RETURN(0);
5257 }
5258
5259 /** The maximum depth that fid2path() will search.
5260  * This is limited only because we want to store the fids for
5261  * historical path lookup purposes.
5262  */
5263 #define MAX_PATH_DEPTH 100
5264
5265 /** mdt_path() lookup structure. */
5266 struct path_lookup_info {
5267         __u64                   pli_recno;      /**< history point */
5268         __u64                   pli_currec;     /**< current record */
5269         struct lu_fid           pli_fid;
5270         struct lu_fid           pli_fids[MAX_PATH_DEPTH]; /**< path, in fids */
5271         struct mdt_object       *pli_mdt_obj;
5272         char                    *pli_path;      /**< full path */
5273         int                     pli_pathlen;
5274         int                     pli_linkno;     /**< which hardlink to follow */
5275         int                     pli_fidcount;   /**< number of \a pli_fids */
5276 };
5277
5278 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
5279                    struct linkea_data *ldata)
5280 {
5281         int rc;
5282
5283         LASSERT(ldata->ld_buf->lb_buf != NULL);
5284
5285         if (!mdt_object_exists(mdt_obj))
5286                 return -ENODATA;
5287
5288         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5289                           ldata->ld_buf, XATTR_NAME_LINK);
5290         if (rc == -ERANGE) {
5291                 /* Buf was too small, figure out what we need. */
5292                 lu_buf_free(ldata->ld_buf);
5293                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5294                                   ldata->ld_buf, XATTR_NAME_LINK);
5295                 if (rc < 0)
5296                         return rc;
5297                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
5298                 if (ldata->ld_buf->lb_buf == NULL)
5299                         return -ENOMEM;
5300                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5301                                   ldata->ld_buf, XATTR_NAME_LINK);
5302         }
5303         if (rc < 0)
5304                 return rc;
5305
5306         return linkea_init(ldata);
5307 }
5308
5309 static int mdt_path_current(struct mdt_thread_info *info,
5310                             struct path_lookup_info *pli)
5311 {
5312         struct mdt_device       *mdt = info->mti_mdt;
5313         struct mdt_object       *mdt_obj;
5314         struct link_ea_header   *leh;
5315         struct link_ea_entry    *lee;
5316         struct lu_name          *tmpname = &info->mti_name;
5317         struct lu_fid           *tmpfid = &info->mti_tmp_fid1;
5318         struct lu_buf           *buf = &info->mti_big_buf;
5319         char                    *ptr;
5320         int                     reclen;
5321         struct linkea_data      ldata = { 0 };
5322         int                     rc = 0;
5323         ENTRY;
5324
5325         /* temp buffer for path element, the buffer will be finally freed
5326          * in mdt_thread_info_fini */
5327         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
5328         if (buf->lb_buf == NULL)
5329                 RETURN(-ENOMEM);
5330
5331         ldata.ld_buf = buf;
5332         ptr = pli->pli_path + pli->pli_pathlen - 1;
5333         *ptr = 0;
5334         --ptr;
5335         pli->pli_fidcount = 0;
5336         pli->pli_fids[0] = *(struct lu_fid *)mdt_object_fid(pli->pli_mdt_obj);
5337         *tmpfid = pli->pli_fids[0];
5338         /* root FID only exists on MDT0, and fid2path should also ends at MDT0,
5339          * so checking root_fid can only happen on MDT0. */
5340         while (!lu_fid_eq(&mdt->mdt_md_root_fid,
5341                           &pli->pli_fids[pli->pli_fidcount])) {
5342                 struct lu_buf           lmv_buf;
5343
5344                 mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
5345                 if (IS_ERR(mdt_obj))
5346                         GOTO(out, rc = PTR_ERR(mdt_obj));
5347
5348                 if (!mdt_object_exists(mdt_obj)) {
5349                         mdt_object_put(info->mti_env, mdt_obj);
5350                         GOTO(out, rc = -ENOENT);
5351                 }
5352
5353                 if (mdt_object_remote(mdt_obj)) {
5354                         mdt_object_put(info->mti_env, mdt_obj);
5355                         GOTO(remote_out, rc = -EREMOTE);
5356                 }
5357
5358                 rc = mdt_links_read(info, mdt_obj, &ldata);
5359                 if (rc != 0) {
5360                         mdt_object_put(info->mti_env, mdt_obj);
5361                         GOTO(out, rc);
5362                 }
5363
5364                 leh = buf->lb_buf;
5365                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
5366                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
5367                 /* If set, use link #linkno for path lookup, otherwise use
5368                    link #0.  Only do this for the final path element. */
5369                 if (pli->pli_fidcount == 0 &&
5370                     pli->pli_linkno < leh->leh_reccount) {
5371                         int count;
5372                         for (count = 0; count < pli->pli_linkno; count++) {
5373                                 lee = (struct link_ea_entry *)
5374                                      ((char *)lee + reclen);
5375                                 linkea_entry_unpack(lee, &reclen, tmpname,
5376                                                     tmpfid);
5377                         }
5378                         if (pli->pli_linkno < leh->leh_reccount - 1)
5379                                 /* indicate to user there are more links */
5380                                 pli->pli_linkno++;
5381                 }
5382
5383                 lmv_buf.lb_buf = info->mti_xattr_buf;
5384                 lmv_buf.lb_len = sizeof(info->mti_xattr_buf);
5385                 /* Check if it is slave stripes */
5386                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5387                                   &lmv_buf, XATTR_NAME_LMV);
5388                 mdt_object_put(info->mti_env, mdt_obj);
5389                 if (rc > 0) {
5390                         union lmv_mds_md *lmm = lmv_buf.lb_buf;
5391
5392                         /* For slave stripes, get its master */
5393                         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE) {
5394                                 pli->pli_fids[pli->pli_fidcount] = *tmpfid;
5395                                 continue;
5396                         }
5397                 } else if (rc < 0 && rc != -ENODATA) {
5398                         GOTO(out, rc);
5399                 }
5400
5401                 rc = 0;
5402
5403                 /* Pack the name in the end of the buffer */
5404                 ptr -= tmpname->ln_namelen;
5405                 if (ptr - 1 <= pli->pli_path)
5406                         GOTO(out, rc = -EOVERFLOW);
5407                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
5408                 *(--ptr) = '/';
5409
5410                 /* Store the parent fid for historic lookup */
5411                 if (++pli->pli_fidcount >= MAX_PATH_DEPTH)
5412                         GOTO(out, rc = -EOVERFLOW);
5413                 pli->pli_fids[pli->pli_fidcount] = *tmpfid;
5414         }
5415
5416 remote_out:
5417         ptr++; /* skip leading / */
5418         memmove(pli->pli_path, ptr, pli->pli_path + pli->pli_pathlen - ptr);
5419
5420         EXIT;
5421 out:
5422         return rc;
5423 }
5424
5425 /* Returns the full path to this fid, as of changelog record recno. */
5426 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
5427                     char *path, int pathlen, __u64 *recno, int *linkno,
5428                     struct lu_fid *fid)
5429 {
5430         struct mdt_device       *mdt = info->mti_mdt;
5431         struct path_lookup_info *pli;
5432         int                     tries = 3;
5433         int                     rc = -EAGAIN;
5434         ENTRY;
5435
5436         if (pathlen < 3)
5437                 RETURN(-EOVERFLOW);
5438
5439         if (lu_fid_eq(&mdt->mdt_md_root_fid, mdt_object_fid(obj))) {
5440                 path[0] = '\0';
5441                 RETURN(0);
5442         }
5443
5444         OBD_ALLOC_PTR(pli);
5445         if (pli == NULL)
5446                 RETURN(-ENOMEM);
5447
5448         pli->pli_mdt_obj = obj;
5449         pli->pli_recno = *recno;
5450         pli->pli_path = path;
5451         pli->pli_pathlen = pathlen;
5452         pli->pli_linkno = *linkno;
5453
5454         /* Retry multiple times in case file is being moved */
5455         while (tries-- && rc == -EAGAIN)
5456                 rc = mdt_path_current(info, pli);
5457
5458         /* return the last resolved fids to the client, so the client will
5459          * build the left path on another MDT for remote object */
5460         *fid = pli->pli_fids[pli->pli_fidcount];
5461
5462         *recno = pli->pli_currec;
5463         /* Return next link index to caller */
5464         *linkno = pli->pli_linkno;
5465
5466         OBD_FREE_PTR(pli);
5467
5468         RETURN(rc);
5469 }
5470
5471 static int mdt_fid2path(struct mdt_thread_info *info,
5472                         struct getinfo_fid2path *fp)
5473 {
5474         struct mdt_device *mdt = info->mti_mdt;
5475         struct mdt_object *obj;
5476         int    rc;
5477         ENTRY;
5478
5479         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
5480                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
5481
5482         if (!fid_is_sane(&fp->gf_fid))
5483                 RETURN(-EINVAL);
5484
5485         if (!fid_is_namespace_visible(&fp->gf_fid)) {
5486                 CWARN("%s: "DFID" is invalid, sequence should be "
5487                       ">= "LPX64"\n", mdt_obd_name(mdt),
5488                       PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
5489                 RETURN(-EINVAL);
5490         }
5491
5492         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
5493         if (obj == NULL || IS_ERR(obj)) {
5494                 CDEBUG(D_IOCTL, "no object "DFID": %ld\n", PFID(&fp->gf_fid),
5495                        PTR_ERR(obj));
5496                 RETURN(-EINVAL);
5497         }
5498
5499         if (mdt_object_remote(obj))
5500                 rc = -EREMOTE;
5501         else if (!mdt_object_exists(obj))
5502                 rc = -ENOENT;
5503         else
5504                 rc = 0;
5505
5506         if (rc < 0) {
5507                 mdt_object_put(info->mti_env, obj);
5508                 CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n",
5509                        PFID(&fp->gf_fid), rc);
5510                 RETURN(rc);
5511         }
5512
5513         rc = mdt_path(info, obj, fp->gf_path, fp->gf_pathlen, &fp->gf_recno,
5514                       &fp->gf_linkno, &fp->gf_fid);
5515
5516         CDEBUG(D_INFO, "fid "DFID", path %s recno "LPX64" linkno %u\n",
5517                PFID(&fp->gf_fid), fp->gf_path, fp->gf_recno, fp->gf_linkno);
5518
5519         mdt_object_put(info->mti_env, obj);
5520
5521         RETURN(rc);
5522 }
5523
5524 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key,
5525                             void *val, int vallen)
5526 {
5527         struct getinfo_fid2path *fpout, *fpin;
5528         int rc = 0;
5529
5530         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
5531         fpout = val;
5532
5533         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5534                 lustre_swab_fid2path(fpin);
5535
5536         memcpy(fpout, fpin, sizeof(*fpin));
5537         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
5538                 RETURN(-EINVAL);
5539
5540         rc = mdt_fid2path(info, fpout);
5541         RETURN(rc);
5542 }
5543
5544 int mdt_get_info(struct tgt_session_info *tsi)
5545 {
5546         char    *key;
5547         int      keylen;
5548         __u32   *vallen;
5549         void    *valout;
5550         int      rc;
5551
5552         ENTRY;
5553
5554         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
5555         if (key == NULL) {
5556                 CDEBUG(D_IOCTL, "No GETINFO key\n");
5557                 RETURN(err_serious(-EFAULT));
5558         }
5559         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
5560                                       RCL_CLIENT);
5561
5562         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
5563         if (vallen == NULL) {
5564                 CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n",
5565                                 tgt_name(tsi->tsi_tgt));
5566                 RETURN(err_serious(-EFAULT));
5567         }
5568
5569         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
5570                              *vallen);
5571         rc = req_capsule_server_pack(tsi->tsi_pill);
5572         if (rc)
5573                 RETURN(err_serious(rc));
5574
5575         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
5576         if (valout == NULL) {
5577                 CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n",
5578                                 tgt_name(tsi->tsi_tgt));
5579                 RETURN(err_serious(-EFAULT));
5580         }
5581
5582         if (KEY_IS(KEY_FID2PATH)) {
5583                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
5584
5585                 rc = mdt_rpc_fid2path(info, key, valout, *vallen);
5586                 mdt_thread_info_fini(info);
5587         } else {
5588                 rc = -EINVAL;
5589         }
5590         RETURN(rc);
5591 }
5592
5593 /* Pass the ioc down */
5594 static int mdt_ioc_child(struct lu_env *env, struct mdt_device *mdt,
5595                          unsigned int cmd, int len, void *data)
5596 {
5597         struct lu_context ioctl_session;
5598         struct md_device *next = mdt->mdt_child;
5599         int rc;
5600         ENTRY;
5601
5602         rc = lu_context_init(&ioctl_session, LCT_SERVER_SESSION);
5603         if (rc)
5604                 RETURN(rc);
5605         ioctl_session.lc_thread = (struct ptlrpc_thread *)current;
5606         lu_context_enter(&ioctl_session);
5607         env->le_ses = &ioctl_session;
5608
5609         LASSERT(next->md_ops->mdo_iocontrol);
5610         rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data);
5611
5612         lu_context_exit(&ioctl_session);
5613         lu_context_fini(&ioctl_session);
5614         RETURN(rc);
5615 }
5616
5617 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
5618 {
5619         struct obd_ioctl_data *data = karg;
5620         struct lu_fid *fid;
5621         __u64 version;
5622         struct mdt_object *obj;
5623         struct mdt_lock_handle  *lh;
5624         int rc;
5625         ENTRY;
5626
5627         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
5628             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
5629                 RETURN(-EINVAL);
5630
5631         fid = (struct lu_fid *)data->ioc_inlbuf1;
5632
5633         if (!fid_is_sane(fid))
5634                 RETURN(-EINVAL);
5635
5636         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
5637
5638         lh = &mti->mti_lh[MDT_LH_PARENT];
5639         mdt_lock_reg_init(lh, LCK_CR);
5640
5641         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
5642         if (IS_ERR(obj))
5643                 RETURN(PTR_ERR(obj));
5644
5645         if (mdt_object_remote(obj)) {
5646                 rc = -EREMOTE;
5647                 /**
5648                  * before calling version get the correct MDS should be
5649                  * fid, this is error to find remote object here
5650                  */
5651                 CERROR("nonlocal object "DFID"\n", PFID(fid));
5652         } else if (!mdt_object_exists(obj)) {
5653                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
5654                 rc = -ENOENT;
5655         } else {
5656                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
5657                *(__u64 *)data->ioc_inlbuf2 = version;
5658                 rc = 0;
5659         }
5660         mdt_object_unlock_put(mti, obj, lh, 1);
5661         RETURN(rc);
5662 }
5663
5664 /* ioctls on obd dev */
5665 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
5666                          void *karg, void *uarg)
5667 {
5668         struct lu_env      env;
5669         struct obd_device *obd = exp->exp_obd;
5670         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
5671         struct dt_device  *dt = mdt->mdt_bottom;
5672         int rc;
5673
5674         ENTRY;
5675         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
5676         rc = lu_env_init(&env, LCT_MD_THREAD);
5677         if (rc)
5678                 RETURN(rc);
5679
5680         switch (cmd) {
5681         case OBD_IOC_SYNC:
5682                 rc = mdt_device_sync(&env, mdt);
5683                 break;
5684         case OBD_IOC_SET_READONLY:
5685                 rc = dt->dd_ops->dt_ro(&env, dt);
5686                 break;
5687         case OBD_IOC_ABORT_RECOVERY:
5688                 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
5689                 target_stop_recovery_thread(obd);
5690                 rc = 0;
5691                 break;
5692         case OBD_IOC_CHANGELOG_REG:
5693         case OBD_IOC_CHANGELOG_DEREG:
5694         case OBD_IOC_CHANGELOG_CLEAR:
5695                 rc = mdt_ioc_child(&env, mdt, cmd, len, karg);
5696                 break;
5697         case OBD_IOC_START_LFSCK: {
5698                 struct md_device *next = mdt->mdt_child;
5699                 struct obd_ioctl_data *data = karg;
5700                 struct lfsck_start_param lsp;
5701
5702                 if (unlikely(data == NULL)) {
5703                         rc = -EINVAL;
5704                         break;
5705                 }
5706
5707                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
5708                 lsp.lsp_index_valid = 0;
5709                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
5710                 break;
5711         }
5712         case OBD_IOC_STOP_LFSCK: {
5713                 struct md_device        *next = mdt->mdt_child;
5714                 struct obd_ioctl_data   *data = karg;
5715                 struct lfsck_stop        stop;
5716
5717                 stop.ls_status = LS_STOPPED;
5718                 /* Old lfsck utils may pass NULL @stop. */
5719                 if (data->ioc_inlbuf1 == NULL)
5720                         stop.ls_flags = 0;
5721                 else
5722                         stop.ls_flags =
5723                         ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
5724
5725                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
5726                 break;
5727         }
5728         case OBD_IOC_GET_OBJ_VERSION: {
5729                 struct mdt_thread_info *mti;
5730                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5731                 memset(mti, 0, sizeof *mti);
5732                 mti->mti_env = &env;
5733                 mti->mti_mdt = mdt;
5734                 mti->mti_exp = exp;
5735
5736                 rc = mdt_ioc_version_get(mti, karg);
5737                 break;
5738         }
5739         case OBD_IOC_CATLOGLIST: {
5740                 struct mdt_thread_info *mti;
5741
5742                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5743                 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
5744                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
5745                                        &mti->mti_tmp_fid1);
5746                 break;
5747          }
5748         default:
5749                 rc = -EOPNOTSUPP;
5750                 CERROR("%s: Not supported cmd = %d, rc = %d\n",
5751                         mdt_obd_name(mdt), cmd, rc);
5752         }
5753
5754         lu_env_fini(&env);
5755         RETURN(rc);
5756 }
5757
5758 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
5759 {
5760         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
5761         struct lfsck_start_param lsp;
5762         int rc;
5763         ENTRY;
5764
5765         lsp.lsp_start = NULL;
5766         lsp.lsp_index_valid = 0;
5767         rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
5768                                                    OBD_IOC_START_LFSCK,
5769                                                    0, &lsp);
5770         if (rc != 0 && rc != -EALREADY)
5771                 CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
5772                       mdt_obd_name(mdt), rc);
5773
5774         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
5775         RETURN(rc);
5776 }
5777
5778 static int mdt_obd_postrecov(struct obd_device *obd)
5779 {
5780         struct lu_env env;
5781         int rc;
5782
5783         rc = lu_env_init(&env, LCT_MD_THREAD);
5784         if (rc)
5785                 RETURN(rc);
5786         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
5787         lu_env_fini(&env);
5788         return rc;
5789 }
5790
5791 static struct obd_ops mdt_obd_device_ops = {
5792         .o_owner          = THIS_MODULE,
5793         .o_set_info_async = mdt_obd_set_info_async,
5794         .o_connect        = mdt_obd_connect,
5795         .o_reconnect      = mdt_obd_reconnect,
5796         .o_disconnect     = mdt_obd_disconnect,
5797         .o_init_export    = mdt_init_export,
5798         .o_destroy_export = mdt_destroy_export,
5799         .o_iocontrol      = mdt_iocontrol,
5800         .o_postrecov      = mdt_obd_postrecov,
5801 };
5802
5803 static struct lu_device* mdt_device_fini(const struct lu_env *env,
5804                                          struct lu_device *d)
5805 {
5806         struct mdt_device *m = mdt_dev(d);
5807         ENTRY;
5808
5809         mdt_fini(env, m);
5810         RETURN(NULL);
5811 }
5812
5813 static struct lu_device *mdt_device_free(const struct lu_env *env,
5814                                          struct lu_device *d)
5815 {
5816         struct mdt_device *m = mdt_dev(d);
5817         ENTRY;
5818
5819         lu_device_fini(&m->mdt_lu_dev);
5820         OBD_FREE_PTR(m);
5821
5822         RETURN(NULL);
5823 }
5824
5825 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
5826                                           struct lu_device_type *t,
5827                                           struct lustre_cfg *cfg)
5828 {
5829         struct lu_device  *l;
5830         struct mdt_device *m;
5831
5832         OBD_ALLOC_PTR(m);
5833         if (m != NULL) {
5834                 int rc;
5835
5836                 l = &m->mdt_lu_dev;
5837                 rc = mdt_init0(env, m, t, cfg);
5838                 if (rc != 0) {
5839                         mdt_device_free(env, l);
5840                         l = ERR_PTR(rc);
5841                         return l;
5842                 }
5843         } else
5844                 l = ERR_PTR(-ENOMEM);
5845         return l;
5846 }
5847
5848 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
5849 LU_KEY_INIT(mdt, struct mdt_thread_info);
5850
5851 static void mdt_key_fini(const struct lu_context *ctx,
5852                          struct lu_context_key *key, void* data)
5853 {
5854         struct mdt_thread_info *info = data;
5855
5856         if (info->mti_big_lmm) {
5857                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
5858                 info->mti_big_lmm = NULL;
5859                 info->mti_big_lmmsize = 0;
5860         }
5861         OBD_FREE_PTR(info);
5862 }
5863
5864 /* context key: mdt_thread_key */
5865 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
5866
5867 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
5868 {
5869         return lu_ucred(info->mti_env);
5870 }
5871
5872 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
5873 {
5874         return lu_ucred_check(info->mti_env);
5875 }
5876
5877 /**
5878  * Enable/disable COS (Commit On Sharing).
5879  *
5880  * Set/Clear the COS flag in mdt options.
5881  *
5882  * \param mdt mdt device
5883  * \param val 0 disables COS, other values enable COS
5884  */
5885 void mdt_enable_cos(struct mdt_device *mdt, int val)
5886 {
5887         struct lu_env env;
5888         int rc;
5889
5890         mdt->mdt_opts.mo_cos = !!val;
5891         rc = lu_env_init(&env, LCT_LOCAL);
5892         if (unlikely(rc != 0)) {
5893                 CWARN("%s: lu_env initialization failed, cannot "
5894                       "sync: rc = %d\n", mdt_obd_name(mdt), rc);
5895                 return;
5896         }
5897         mdt_device_sync(&env, mdt);
5898         lu_env_fini(&env);
5899 }
5900
5901 /**
5902  * Check COS (Commit On Sharing) status.
5903  *
5904  * Return COS flag status.
5905  *
5906  * \param mdt mdt device
5907  */
5908 int mdt_cos_is_enabled(struct mdt_device *mdt)
5909 {
5910         return mdt->mdt_opts.mo_cos != 0;
5911 }
5912
5913 static struct lu_device_type_operations mdt_device_type_ops = {
5914         .ldto_device_alloc = mdt_device_alloc,
5915         .ldto_device_free  = mdt_device_free,
5916         .ldto_device_fini  = mdt_device_fini
5917 };
5918
5919 static struct lu_device_type mdt_device_type = {
5920         .ldt_tags     = LU_DEVICE_MD,
5921         .ldt_name     = LUSTRE_MDT_NAME,
5922         .ldt_ops      = &mdt_device_type_ops,
5923         .ldt_ctx_tags = LCT_MD_THREAD
5924 };
5925
5926 static int __init mdt_mod_init(void)
5927 {
5928         int rc;
5929
5930         CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") ==
5931                  FID_NOBRACE_LEN + 1);
5932         CLASSERT(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") ==
5933                  FID_LEN + 1);
5934         rc = lu_kmem_init(mdt_caches);
5935         if (rc)
5936                 return rc;
5937
5938         rc = mds_mod_init();
5939         if (rc)
5940                 GOTO(lu_fini, rc);
5941
5942         rc = class_register_type(&mdt_obd_device_ops, NULL, true, NULL,
5943 #ifndef HAVE_ONLY_PROCFS_SEQ
5944                                  NULL,
5945 #endif
5946                                  LUSTRE_MDT_NAME, &mdt_device_type);
5947         if (rc)
5948                 GOTO(mds_fini, rc);
5949 lu_fini:
5950         if (rc)
5951                 lu_kmem_fini(mdt_caches);
5952 mds_fini:
5953         if (rc)
5954                 mds_mod_exit();
5955         return rc;
5956 }
5957
5958 static void __exit mdt_mod_exit(void)
5959 {
5960         class_unregister_type(LUSTRE_MDT_NAME);
5961         mds_mod_exit();
5962         lu_kmem_fini(mdt_caches);
5963 }
5964
5965 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
5966 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
5967 MODULE_LICENSE("GPL");
5968
5969 cfs_module(mdt, LUSTRE_VERSION_STRING, mdt_mod_init, mdt_mod_exit);