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