lustre/mdt/mdt_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2010, 2013, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/mdt/mdt_handler.c
  37  *
  38  * Lustre Metadata Target (mdt) request handler
  39  *
  40  * Author: Peter Braam <braam@clusterfs.com>
  41  * Author: Andreas Dilger <adilger@clusterfs.com>
  42  * Author: Phil Schwan <phil@clusterfs.com>
  43  * Author: Mike Shaver <shaver@clusterfs.com>
  44  * Author: Nikita Danilov <nikita@clusterfs.com>
  45  * Author: Huang Hua <huanghua@clusterfs.com>
  46  * Author: Yury Umanets <umka@clusterfs.com>
  47  */
  48
  49 #define DEBUG_SUBSYSTEM S_MDS
  50
  51 #include <linux/module.h>
  52 /*
  53  * struct OBD_{ALLOC,FREE}*()
  54  */
  55 #include <obd_support.h>
  56 /* struct ptlrpc_request */
  57 #include <lustre_net.h>
  58 /* struct obd_export */
  59 #include <lustre_export.h>
  60 /* struct obd_device */
  61 #include <obd.h>
  62 /* lu2dt_dev() */
  63 #include <dt_object.h>
  64 #include <lustre_mds.h>
  65 #include <lustre_log.h>
  66 #include "mdt_internal.h"
  67 #include <lustre_acl.h>
  68 #include <lustre_param.h>
  69 #include <lustre_quota.h>
  70 #include <lustre_linkea.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 lm,
 168                        const char *name, int namelen)
 169 {
 170         lh->mlh_reg_mode = lm;
 171         lh->mlh_rreg_mode = lm;
 172         lh->mlh_type = MDT_PDO_LOCK;
 173
 174         if (name != NULL && (name[0] != '\0')) {
 175                 LASSERT(namelen > 0);
 176                 lh->mlh_pdo_hash = full_name_hash(name, namelen);
 177                 /* XXX Workaround for LU-2856
 178                  * Zero is a valid return value of full_name_hash, but several
 179                  * users of mlh_pdo_hash assume a non-zero hash value. We
 180                  * therefore map zero onto an arbitrary, but consistent
 181                  * value (1) to avoid problems further down the road. */
 182                 if (unlikely(!lh->mlh_pdo_hash))
 183                         lh->mlh_pdo_hash = 1;
 184         } else {
 185                 LASSERT(namelen == 0);
 186                 lh->mlh_pdo_hash = 0ull;
 187         }
 188 }
 189
 190 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 191                               struct mdt_lock_handle *lh)
 192 {
 193         mdl_mode_t mode;
 194         ENTRY;
 195
 196         /*
 197          * Any dir access needs couple of locks:
 198          *
 199          * 1) on part of dir we gonna take lookup/modify;
 200          *
 201          * 2) on whole dir to protect it from concurrent splitting and/or to
 202          * flush client's cache for readdir().
 203          *
 204          * so, for a given mode and object this routine decides what lock mode
 205          * to use for lock #2:
 206          *
 207          * 1) if caller's gonna lookup in dir then we need to protect dir from
 208          * being splitted only - LCK_CR
 209          *
 210          * 2) if caller's gonna modify dir then we need to protect dir from
 211          * being splitted and to flush cache - LCK_CW
 212          *
 213          * 3) if caller's gonna modify dir and that dir seems ready for
 214          * splitting then we need to protect it from any type of access
 215          * (lookup/modify/split) - LCK_EX --bzzz
 216          */
 217
 218         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 219         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 220
 221         /*
 222          * Ask underlaying level its opinion about preferable PDO lock mode
 223          * having access type passed as regular lock mode:
 224          *
 225          * - MDL_MINMODE means that lower layer does not want to specify lock
 226          * mode;
 227          *
 228          * - MDL_NL means that no PDO lock should be taken. This is used in some
 229          * cases. Say, for non-splittable directories no need to use PDO locks
 230          * at all.
 231          */
 232         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 233                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 234
 235         if (mode != MDL_MINMODE) {
 236                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 237         } else {
 238                 /*
 239                  * Lower layer does not want to specify locking mode. We do it
 240                  * our selves. No special protection is needed, just flush
 241                  * client's cache on modification and allow concurrent
 242                  * mondification.
 243                  */
 244                 switch (lh->mlh_reg_mode) {
 245                 case LCK_EX:
 246                         lh->mlh_pdo_mode = LCK_EX;
 247                         break;
 248                 case LCK_PR:
 249                         lh->mlh_pdo_mode = LCK_CR;
 250                         break;
 251                 case LCK_PW:
 252                         lh->mlh_pdo_mode = LCK_CW;
 253                         break;
 254                 default:
 255                         CERROR("Not expected lock type (0x%x)\n",
 256                                (int)lh->mlh_reg_mode);
 257                         LBUG();
 258                 }
 259         }
 260
 261         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 262         EXIT;
 263 }
 264
 265 int mdt_getstatus(struct tgt_session_info *tsi)
 266 {
 267         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
 268         struct mdt_device       *mdt = info->mti_mdt;
 269         struct mdt_body         *repbody;
 270         int                      rc;
 271
 272         ENTRY;
 273
 274         rc = mdt_check_ucred(info);
 275         if (rc)
 276                 GOTO(out, rc = err_serious(rc));
 277
 278         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
 279                 GOTO(out, rc = err_serious(-ENOMEM));
 280
 281         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 282         repbody->fid1 = mdt->mdt_md_root_fid;
 283         repbody->valid |= OBD_MD_FLID;
 284
 285         if (tsi->tsi_tgt->lut_mds_capa &&
 286             exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) {
 287                 struct mdt_object       *root;
 288                 struct lustre_capa      *capa;
 289
 290                 root = mdt_object_find(info->mti_env, mdt, &repbody->fid1);
 291                 if (IS_ERR(root))
 292                         GOTO(out, rc = PTR_ERR(root));
 293
 294                 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
 295                 LASSERT(capa);
 296                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 297                 rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa,
 298                                  0);
 299                 mdt_object_put(info->mti_env, root);
 300                 if (rc == 0)
 301                         repbody->valid |= OBD_MD_FLMDSCAPA;
 302         }
 303         EXIT;
 304 out:
 305         mdt_thread_info_fini(info);
 306         return rc;
 307 }
 308
 309 int mdt_statfs(struct tgt_session_info *tsi)
 310 {
 311         struct ptlrpc_request           *req = tgt_ses_req(tsi);
 312         struct mdt_thread_info          *info = tsi2mdt_info(tsi);
 313         struct md_device                *next = info->mti_mdt->mdt_child;
 314         struct ptlrpc_service_part      *svcpt;
 315         struct obd_statfs               *osfs;
 316         int                             rc;
 317
 318         ENTRY;
 319
 320         svcpt = req->rq_rqbd->rqbd_svcpt;
 321
 322         /* This will trigger a watchdog timeout */
 323         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 324                          (MDT_SERVICE_WATCHDOG_FACTOR *
 325                           at_get(&svcpt->scp_at_estimate)) + 1);
 326
 327         rc = mdt_check_ucred(info);
 328         if (rc)
 329                 GOTO(out, rc = err_serious(rc));
 330
 331         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
 332                 GOTO(out, rc = err_serious(-ENOMEM));
 333
 334         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 335         if (!osfs)
 336                 GOTO(out, rc = -EPROTO);
 337
 338         /** statfs information are cached in the mdt_device */
 339         if (cfs_time_before_64(info->mti_mdt->mdt_osfs_age,
 340                                cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS))) {
 341                 /** statfs data is too old, get up-to-date one */
 342                 rc = next->md_ops->mdo_statfs(info->mti_env, next, osfs);
 343                 if (rc)
 344                         GOTO(out, rc);
 345                 spin_lock(&info->mti_mdt->mdt_osfs_lock);
 346                 info->mti_mdt->mdt_osfs = *osfs;
 347                 info->mti_mdt->mdt_osfs_age = cfs_time_current_64();
 348                 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
 349         } else {
 350                 /** use cached statfs data */
 351                 spin_lock(&info->mti_mdt->mdt_osfs_lock);
 352                 *osfs = info->mti_mdt->mdt_osfs;
 353                 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
 354         }
 355
 356         if (rc == 0)
 357                 mdt_counter_incr(req, LPROC_MDT_STATFS);
 358 out:
 359         mdt_thread_info_fini(info);
 360         RETURN(rc);
 361 }
 362
 363 /**
 364  * Pack SOM attributes into the reply.
 365  * Call under a DLM UPDATE lock.
 366  */
 367 static void mdt_pack_size2body(struct mdt_thread_info *info,
 368                                struct mdt_object *mo)
 369 {
 370         struct mdt_body *b;
 371         struct md_attr *ma = &info->mti_attr;
 372
 373         LASSERT(ma->ma_attr.la_valid & LA_MODE);
 374         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 375
 376         /* Check if Size-on-MDS is supported, if this is a regular file,
 377          * if SOM is enabled on the object and if SOM cache exists and valid.
 378          * Otherwise do not pack Size-on-MDS attributes to the reply. */
 379         if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) ||
 380             !S_ISREG(ma->ma_attr.la_mode) ||
 381             !mdt_object_is_som_enabled(mo) ||
 382             !(ma->ma_valid & MA_SOM))
 383                 return;
 384
 385         b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 386         b->size = ma->ma_som->msd_size;
 387         b->blocks = ma->ma_som->msd_blocks;
 388 }
 389
 390 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 391                         const struct lu_attr *attr, const struct lu_fid *fid)
 392 {
 393         struct md_attr *ma = &info->mti_attr;
 394
 395         LASSERT(ma->ma_valid & MA_INODE);
 396
 397         b->atime      = attr->la_atime;
 398         b->mtime      = attr->la_mtime;
 399         b->ctime      = attr->la_ctime;
 400         b->mode       = attr->la_mode;
 401         b->size       = attr->la_size;
 402         b->blocks     = attr->la_blocks;
 403         b->uid        = attr->la_uid;
 404         b->gid        = attr->la_gid;
 405         b->flags      = attr->la_flags;
 406         b->nlink      = attr->la_nlink;
 407         b->rdev       = attr->la_rdev;
 408
 409         /*XXX should pack the reply body according to lu_valid*/
 410         b->valid |= OBD_MD_FLCTIME | OBD_MD_FLUID   |
 411                     OBD_MD_FLGID   | OBD_MD_FLTYPE  |
 412                     OBD_MD_FLMODE  | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
 413                     OBD_MD_FLATIME | OBD_MD_FLMTIME ;
 414
 415         if (!S_ISREG(attr->la_mode)) {
 416                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 417         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
 418                 /* means no objects are allocated on osts. */
 419                 LASSERT(!(ma->ma_valid & MA_LOV));
 420                 /* just ignore blocks occupied by extend attributes on MDS */
 421                 b->blocks = 0;
 422                 /* if no object is allocated on osts, the size on mds is valid. b=22272 */
 423                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 424         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL &&
 425                    ma->ma_lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) {
 426                 /* A released file stores its size on MDS. */
 427                 /* But return 1 block for released file, unless tools like tar
 428                  * will consider it fully sparse. (LU-3864)
 429                  */
 430                 if (unlikely(b->size == 0))
 431                         b->blocks = 0;
 432                 else
 433                         b->blocks = 1;
 434                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 435         }
 436
 437         if (fid) {
 438                 b->fid1 = *fid;
 439                 b->valid |= OBD_MD_FLID;
 440                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n",
 441                                 PFID(fid), b->nlink, b->mode, b->size);
 442         }
 443
 444         if (info)
 445                 mdt_body_reverse_idmap(info, b);
 446
 447         if (fid != NULL && (b->valid & OBD_MD_FLSIZE))
 448                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 449                        PFID(fid), (unsigned long long)b->size);
 450 }
 451
 452 static inline int mdt_body_has_lov(const struct lu_attr *la,
 453                                    const struct mdt_body *body)
 454 {
 455         return ((S_ISREG(la->la_mode) && (body->valid & OBD_MD_FLEASIZE)) ||
 456                 (S_ISDIR(la->la_mode) && (body->valid & OBD_MD_FLDIREA )) );
 457 }
 458
 459 void mdt_client_compatibility(struct mdt_thread_info *info)
 460 {
 461         struct mdt_body       *body;
 462         struct ptlrpc_request *req = mdt_info_req(info);
 463         struct obd_export     *exp = req->rq_export;
 464         struct md_attr        *ma = &info->mti_attr;
 465         struct lu_attr        *la = &ma->ma_attr;
 466         ENTRY;
 467
 468         if (exp_connect_layout(exp))
 469                 /* the client can deal with 16-bit lmm_stripe_count */
 470                 RETURN_EXIT;
 471
 472         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 473
 474         if (!mdt_body_has_lov(la, body))
 475                 RETURN_EXIT;
 476
 477         /* now we have a reply with a lov for a client not compatible with the
 478          * layout lock so we have to clean the layout generation number */
 479         if (S_ISREG(la->la_mode))
 480                 ma->ma_lmm->lmm_layout_gen = 0;
 481         EXIT;
 482 }
 483
 484 static int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
 485                              char *name)
 486 {
 487         const struct lu_env *env = info->mti_env;
 488         int rc;
 489         ENTRY;
 490
 491         LASSERT(info->mti_big_lmm_used == 0);
 492         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
 493         if (rc < 0)
 494                 RETURN(rc);
 495
 496         /* big_lmm may need to be grown */
 497         if (info->mti_big_lmmsize < rc) {
 498                 int size = size_roundup_power2(rc);
 499
 500                 if (info->mti_big_lmmsize > 0) {
 501                         /* free old buffer */
 502                         LASSERT(info->mti_big_lmm);
 503                         OBD_FREE_LARGE(info->mti_big_lmm,
 504                                        info->mti_big_lmmsize);
 505                         info->mti_big_lmm = NULL;
 506                         info->mti_big_lmmsize = 0;
 507                 }
 508
 509                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
 510                 if (info->mti_big_lmm == NULL)
 511                         RETURN(-ENOMEM);
 512                 info->mti_big_lmmsize = size;
 513         }
 514         LASSERT(info->mti_big_lmmsize >= rc);
 515
 516         info->mti_buf.lb_buf = info->mti_big_lmm;
 517         info->mti_buf.lb_len = info->mti_big_lmmsize;
 518         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
 519
 520         RETURN(rc);
 521 }
 522
 523 int mdt_attr_get_lov(struct mdt_thread_info *info,
 524                      struct mdt_object *o, struct md_attr *ma)
 525 {
 526         struct md_object *next = mdt_object_child(o);
 527         struct lu_buf    *buf = &info->mti_buf;
 528         int rc;
 529
 530         buf->lb_buf = ma->ma_lmm;
 531         buf->lb_len = ma->ma_lmm_size;
 532         rc = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_LOV);
 533         if (rc > 0) {
 534                 ma->ma_lmm_size = rc;
 535                 ma->ma_valid |= MA_LOV;
 536                 rc = 0;
 537         } else if (rc == -ENODATA) {
 538                 /* no LOV EA */
 539                 rc = 0;
 540         } else if (rc == -ERANGE) {
 541                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LOV);
 542                 if (rc > 0) {
 543                         info->mti_big_lmm_used = 1;
 544                         ma->ma_valid |= MA_LOV;
 545                         ma->ma_lmm = info->mti_big_lmm;
 546                         ma->ma_lmm_size = rc;
 547                         /* update mdt_max_mdsize so all clients
 548                          * will be aware about that */
 549                         if (info->mti_mdt->mdt_max_mdsize < rc)
 550                                 info->mti_mdt->mdt_max_mdsize = rc;
 551                         rc = 0;
 552                 }
 553         }
 554
 555         return rc;
 556 }
 557
 558 int mdt_attr_get_pfid(struct mdt_thread_info *info,
 559                       struct mdt_object *o, struct lu_fid *pfid)
 560 {
 561         struct lu_buf           *buf = &info->mti_buf;
 562         struct link_ea_header   *leh;
 563         struct link_ea_entry    *lee;
 564         int                      rc;
 565         ENTRY;
 566
 567         buf->lb_buf = info->mti_big_lmm;
 568         buf->lb_len = info->mti_big_lmmsize;
 569         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
 570                           buf, XATTR_NAME_LINK);
 571         /* ignore errors, MA_PFID won't be set and it is
 572          * up to the caller to treat this as an error */
 573         if (rc == -ERANGE || buf->lb_len == 0) {
 574                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
 575                 buf->lb_buf = info->mti_big_lmm;
 576                 buf->lb_len = info->mti_big_lmmsize;
 577         }
 578
 579         if (rc < 0)
 580                 RETURN(rc);
 581         if (rc < sizeof(*leh)) {
 582                 CERROR("short LinkEA on "DFID": rc = %d\n",
 583                        PFID(mdt_object_fid(o)), rc);
 584                 RETURN(-ENODATA);
 585         }
 586
 587         leh = (struct link_ea_header *) buf->lb_buf;
 588         lee = (struct link_ea_entry *)(leh + 1);
 589         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
 590                 leh->leh_magic = LINK_EA_MAGIC;
 591                 leh->leh_reccount = __swab32(leh->leh_reccount);
 592                 leh->leh_len = __swab64(leh->leh_len);
 593         }
 594         if (leh->leh_magic != LINK_EA_MAGIC)
 595                 RETURN(-EINVAL);
 596         if (leh->leh_reccount == 0)
 597                 RETURN(-ENODATA);
 598
 599         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
 600         fid_be_to_cpu(pfid, pfid);
 601
 602         RETURN(0);
 603 }
 604
 605 int mdt_attr_get_complex(struct mdt_thread_info *info,
 606                          struct mdt_object *o, struct md_attr *ma)
 607 {
 608         const struct lu_env *env = info->mti_env;
 609         struct md_object    *next = mdt_object_child(o);
 610         struct lu_buf       *buf = &info->mti_buf;
 611         u32                  mode = lu_object_attr(&next->mo_lu);
 612         int                  need = ma->ma_need;
 613         int                  rc = 0, rc2;
 614         ENTRY;
 615
 616         ma->ma_valid = 0;
 617
 618         if (need & MA_INODE) {
 619                 ma->ma_need = MA_INODE;
 620                 rc = mo_attr_get(env, next, ma);
 621                 if (rc)
 622                         GOTO(out, rc);
 623                 ma->ma_valid |= MA_INODE;
 624         }
 625
 626         if (need & MA_PFID) {
 627                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
 628                 if (rc == 0)
 629                         ma->ma_valid |= MA_PFID;
 630                 /* ignore this error, parent fid is not mandatory */
 631                 rc = 0;
 632         }
 633
 634         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
 635                 rc = mdt_attr_get_lov(info, o, ma);
 636                 if (rc)
 637                         GOTO(out, rc);
 638         }
 639
 640         if (need & MA_LMV && S_ISDIR(mode)) {
 641                 buf->lb_buf = ma->ma_lmv;
 642                 buf->lb_len = ma->ma_lmv_size;
 643                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_LMV);
 644                 if (rc2 > 0) {
 645                         ma->ma_lmv_size = rc2;
 646                         ma->ma_valid |= MA_LMV;
 647                 } else if (rc2 == -ENODATA) {
 648                         /* no LMV EA */
 649                         ma->ma_lmv_size = 0;
 650                 } else
 651                         GOTO(out, rc = rc2);
 652         }
 653
 654         if (need & MA_SOM && S_ISREG(mode)) {
 655                 buf->lb_buf = info->mti_xattr_buf;
 656                 buf->lb_len = sizeof(info->mti_xattr_buf);
 657                 CLASSERT(sizeof(struct som_attrs) <=
 658                          sizeof(info->mti_xattr_buf));
 659                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_SOM);
 660                 rc2 = lustre_buf2som(info->mti_xattr_buf, rc2, ma->ma_som);
 661                 if (rc2 == 0)
 662                         ma->ma_valid |= MA_SOM;
 663                 else if (rc2 < 0 && rc2 != -ENODATA)
 664                         GOTO(out, rc = rc2);
 665         }
 666
 667         if (need & MA_HSM && S_ISREG(mode)) {
 668                 buf->lb_buf = info->mti_xattr_buf;
 669                 buf->lb_len = sizeof(info->mti_xattr_buf);
 670                 CLASSERT(sizeof(struct hsm_attrs) <=
 671                          sizeof(info->mti_xattr_buf));
 672                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
 673                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
 674                 if (rc2 == 0)
 675                         ma->ma_valid |= MA_HSM;
 676                 else if (rc2 < 0 && rc2 != -ENODATA)
 677                         GOTO(out, rc = rc2);
 678         }
 679
 680 #ifdef CONFIG_FS_POSIX_ACL
 681         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
 682                 buf->lb_buf = ma->ma_acl;
 683                 buf->lb_len = ma->ma_acl_size;
 684                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
 685                 if (rc2 > 0) {
 686                         ma->ma_acl_size = rc2;
 687                         ma->ma_valid |= MA_ACL_DEF;
 688                 } else if (rc2 == -ENODATA) {
 689                         /* no ACLs */
 690                         ma->ma_acl_size = 0;
 691                 } else
 692                         GOTO(out, rc = rc2);
 693         }
 694 #endif
 695 out:
 696         ma->ma_need = need;
 697         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
 698                rc, ma->ma_valid, ma->ma_lmm);
 699         RETURN(rc);
 700 }
 701
 702 static int mdt_getattr_internal(struct mdt_thread_info *info,
 703                                 struct mdt_object *o, int ma_need)
 704 {
 705         struct md_object        *next = mdt_object_child(o);
 706         const struct mdt_body   *reqbody = info->mti_body;
 707         struct ptlrpc_request   *req = mdt_info_req(info);
 708         struct md_attr          *ma = &info->mti_attr;
 709         struct lu_attr          *la = &ma->ma_attr;
 710         struct req_capsule      *pill = info->mti_pill;
 711         const struct lu_env     *env = info->mti_env;
 712         struct mdt_body         *repbody;
 713         struct lu_buf           *buffer = &info->mti_buf;
 714         int                     rc;
 715         int                     is_root;
 716         ENTRY;
 717
 718         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
 719                 RETURN(err_serious(-ENOMEM));
 720
 721         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 722
 723         ma->ma_valid = 0;
 724
 725         if (mdt_object_remote(o)) {
 726                 /* This object is located on remote node.*/
 727                 /* Return -EIO for old client */
 728                 if (!mdt_is_dne_client(req->rq_export))
 729                         GOTO(out, rc = -EIO);
 730
 731                 repbody->fid1 = *mdt_object_fid(o);
 732                 repbody->valid = OBD_MD_FLID | OBD_MD_MDS;
 733                 GOTO(out, rc = 0);
 734         }
 735
 736         buffer->lb_len = reqbody->eadatasize;
 737         if (buffer->lb_len > 0)
 738                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
 739         else
 740                 buffer->lb_buf = NULL;
 741
 742         /* If it is dir object and client require MEA, then we got MEA */
 743         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 744             reqbody->valid & OBD_MD_MEA) {
 745                 /* Assumption: MDT_MD size is enough for lmv size. */
 746                 ma->ma_lmv = buffer->lb_buf;
 747                 ma->ma_lmv_size = buffer->lb_len;
 748                 ma->ma_need = MA_LMV | MA_INODE;
 749         } else {
 750                 ma->ma_lmm = buffer->lb_buf;
 751                 ma->ma_lmm_size = buffer->lb_len;
 752                 ma->ma_need = MA_LOV | MA_INODE | MA_HSM;
 753         }
 754
 755         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 756             reqbody->valid & OBD_MD_FLDIREA  &&
 757             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
 758                 /* get default stripe info for this dir. */
 759                 ma->ma_need |= MA_LOV_DEF;
 760         }
 761         ma->ma_need |= ma_need;
 762         if (ma->ma_need & MA_SOM)
 763                 ma->ma_som = &info->mti_u.som.data;
 764
 765         rc = mdt_attr_get_complex(info, o, ma);
 766         if (unlikely(rc)) {
 767                 CERROR("%s: getattr error for "DFID": rc = %d\n",
 768                        mdt_obd_name(info->mti_mdt),
 769                        PFID(mdt_object_fid(o)), rc);
 770                 RETURN(rc);
 771         }
 772
 773         /* if file is released, check if a restore is running */
 774         if ((ma->ma_valid & MA_HSM) && (ma->ma_hsm.mh_flags & HS_RELEASED) &&
 775             mdt_hsm_restore_is_running(info, mdt_object_fid(o))) {
 776                 repbody->t_state = MS_RESTORE;
 777                 repbody->valid |= OBD_MD_TSTATE;
 778         }
 779
 780         is_root = lu_fid_eq(mdt_object_fid(o), &info->mti_mdt->mdt_md_root_fid);
 781
 782         /* the Lustre protocol supposes to return default striping
 783          * on the user-visible root if explicitly requested */
 784         if ((ma->ma_valid & MA_LOV) == 0 && S_ISDIR(la->la_mode) &&
 785             (ma->ma_need & MA_LOV_DEF && is_root) && ma->ma_need & MA_LOV) {
 786                 struct lu_fid      rootfid;
 787                 struct mdt_object *root;
 788                 struct mdt_device *mdt = info->mti_mdt;
 789
 790                 rc = dt_root_get(env, mdt->mdt_bottom, &rootfid);
 791                 if (rc)
 792                         RETURN(rc);
 793                 root = mdt_object_find(env, mdt, &rootfid);
 794                 if (IS_ERR(root))
 795                         RETURN(PTR_ERR(root));
 796                 rc = mdt_attr_get_lov(info, root, ma);
 797                 mdt_object_put(info->mti_env, root);
 798                 if (unlikely(rc)) {
 799                         CERROR("%s: getattr error for "DFID": rc = %d\n",
 800                                mdt_obd_name(info->mti_mdt),
 801                                PFID(mdt_object_fid(o)), rc);
 802                         RETURN(rc);
 803                 }
 804         }
 805
 806         if (likely(ma->ma_valid & MA_INODE))
 807                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
 808         else
 809                 RETURN(-EFAULT);
 810
 811         if (mdt_body_has_lov(la, reqbody)) {
 812                 if (ma->ma_valid & MA_LOV) {
 813                         LASSERT(ma->ma_lmm_size);
 814                         mdt_dump_lmm(D_INFO, ma->ma_lmm);
 815                         repbody->eadatasize = ma->ma_lmm_size;
 816                         if (S_ISDIR(la->la_mode))
 817                                 repbody->valid |= OBD_MD_FLDIREA;
 818                         else
 819                                 repbody->valid |= OBD_MD_FLEASIZE;
 820                 }
 821                 if (ma->ma_valid & MA_LMV) {
 822                         LASSERT(S_ISDIR(la->la_mode));
 823                         repbody->eadatasize = ma->ma_lmv_size;
 824                         repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
 825                 }
 826         } else if (S_ISLNK(la->la_mode) &&
 827                    reqbody->valid & OBD_MD_LINKNAME) {
 828                 buffer->lb_buf = ma->ma_lmm;
 829                 /* eadatasize from client includes NULL-terminator, so
 830                  * there is no need to read it */
 831                 buffer->lb_len = reqbody->eadatasize - 1;
 832                 rc = mo_readlink(env, next, buffer);
 833                 if (unlikely(rc <= 0)) {
 834                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
 835                                mdt_obd_name(info->mti_mdt),
 836                                PFID(mdt_object_fid(o)), rc);
 837                         rc = -EFAULT;
 838                 } else {
 839                         int print_limit = min_t(int, PAGE_CACHE_SIZE - 128, rc);
 840
 841                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
 842                                 rc -= 2;
 843                         repbody->valid |= OBD_MD_LINKNAME;
 844                         /* we need to report back size with NULL-terminator
 845                          * because client expects that */
 846                         repbody->eadatasize = rc + 1;
 847                         if (repbody->eadatasize != reqbody->eadatasize)
 848                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
 849                                        "on "DFID ", expected %d\n",
 850                                        mdt_obd_name(info->mti_mdt),
 851                                        rc, PFID(mdt_object_fid(o)),
 852                                        reqbody->eadatasize - 1);
 853                         /* NULL terminate */
 854                         ((char *)ma->ma_lmm)[rc] = 0;
 855
 856                         /* If the total CDEBUG() size is larger than a page, it
 857                          * will print a warning to the console, avoid this by
 858                          * printing just the last part of the symlink. */
 859                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
 860                                print_limit < rc ? "..." : "", print_limit,
 861                                (char *)ma->ma_lmm + rc - print_limit, rc);
 862                         rc = 0;
 863                 }
 864         }
 865
 866         if (reqbody->valid & OBD_MD_FLMODEASIZE) {
 867                 repbody->max_cookiesize = 0;
 868                 repbody->max_mdsize = info->mti_mdt->mdt_max_mdsize;
 869                 repbody->valid |= OBD_MD_FLMODEASIZE;
 870                 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
 871                        "MAX_COOKIE to : %d:%d\n", repbody->max_mdsize,
 872                        repbody->max_cookiesize);
 873         }
 874
 875         if (exp_connect_rmtclient(info->mti_exp) &&
 876             reqbody->valid & OBD_MD_FLRMTPERM) {
 877                 void *buf = req_capsule_server_get(pill, &RMF_ACL);
 878
 879                 /* mdt_getattr_lock only */
 880                 rc = mdt_pack_remote_perm(info, o, buf);
 881                 if (rc) {
 882                         repbody->valid &= ~OBD_MD_FLRMTPERM;
 883                         repbody->aclsize = 0;
 884                         RETURN(rc);
 885                 } else {
 886                         repbody->valid |= OBD_MD_FLRMTPERM;
 887                         repbody->aclsize = sizeof(struct mdt_remote_perm);
 888                 }
 889         }
 890 #ifdef CONFIG_FS_POSIX_ACL
 891         else if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
 892                  (reqbody->valid & OBD_MD_FLACL)) {
 893                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
 894                 buffer->lb_len = req_capsule_get_size(pill,
 895                                                       &RMF_ACL, RCL_SERVER);
 896                 if (buffer->lb_len > 0) {
 897                         rc = mo_xattr_get(env, next, buffer,
 898                                           XATTR_NAME_ACL_ACCESS);
 899                         if (rc < 0) {
 900                                 if (rc == -ENODATA) {
 901                                         repbody->aclsize = 0;
 902                                         repbody->valid |= OBD_MD_FLACL;
 903                                         rc = 0;
 904                                 } else if (rc == -EOPNOTSUPP) {
 905                                         rc = 0;
 906                                 } else {
 907                                         CERROR("%s: unable to read "DFID
 908                                                " ACL: rc = %d\n",
 909                                                mdt_obd_name(info->mti_mdt),
 910                                                PFID(mdt_object_fid(o)), rc);
 911                                 }
 912                         } else {
 913                                 repbody->aclsize = rc;
 914                                 repbody->valid |= OBD_MD_FLACL;
 915                                 rc = 0;
 916                         }
 917                 }
 918         }
 919 #endif
 920
 921         if (reqbody->valid & OBD_MD_FLMDSCAPA &&
 922             info->mti_mdt->mdt_lut.lut_mds_capa &&
 923             exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) {
 924                 struct lustre_capa *capa;
 925
 926                 capa = req_capsule_server_get(pill, &RMF_CAPA1);
 927                 LASSERT(capa);
 928                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 929                 rc = mo_capa_get(env, next, capa, 0);
 930                 if (rc)
 931                         RETURN(rc);
 932                 repbody->valid |= OBD_MD_FLMDSCAPA;
 933         }
 934
 935 out:
 936         if (rc == 0)
 937                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
 938
 939         RETURN(rc);
 940 }
 941
 942 static int mdt_renew_capa(struct mdt_thread_info *info)
 943 {
 944         struct mdt_object  *obj = info->mti_object;
 945         struct mdt_body    *body;
 946         struct lustre_capa *capa, *c;
 947         int rc;
 948         ENTRY;
 949
 950         /* if object doesn't exist, or server has disabled capability,
 951          * return directly, client will find body->valid OBD_MD_FLOSSCAPA
 952          * flag not set.
 953          */
 954         if (!obj || !info->mti_mdt->mdt_lut.lut_oss_capa ||
 955             !(exp_connect_flags(info->mti_exp) & OBD_CONNECT_OSS_CAPA))
 956                 RETURN(0);
 957
 958         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 959         LASSERT(body != NULL);
 960
 961         c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
 962         LASSERT(c);
 963
 964         capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
 965         LASSERT(capa);
 966
 967         *capa = *c;
 968         rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
 969         if (rc == 0)
 970                 body->valid |= OBD_MD_FLOSSCAPA;
 971         RETURN(rc);
 972 }
 973
 974 int mdt_getattr(struct tgt_session_info *tsi)
 975 {
 976         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
 977         struct mdt_object       *obj = info->mti_object;
 978         struct req_capsule      *pill = info->mti_pill;
 979         struct mdt_body         *reqbody;
 980         struct mdt_body         *repbody;
 981         mode_t                   mode;
 982         int rc, rc2;
 983         ENTRY;
 984
 985         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
 986         LASSERT(reqbody);
 987
 988         if (reqbody->valid & OBD_MD_FLOSSCAPA) {
 989                 rc = req_capsule_server_pack(pill);
 990                 if (unlikely(rc))
 991                         RETURN(err_serious(rc));
 992                 rc = mdt_renew_capa(info);
 993                 GOTO(out_shrink, rc);
 994         }
 995
 996         LASSERT(obj != NULL);
 997         LASSERT(lu_object_assert_exists(&obj->mot_obj));
 998
 999         mode = lu_object_attr(&obj->mot_obj);
1000
1001         /* old clients may not report needed easize, use max value then */
1002         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1003                              reqbody->eadatasize == 0 ?
1004                              info->mti_mdt->mdt_max_mdsize :
1005                              reqbody->eadatasize);
1006
1007         rc = req_capsule_server_pack(pill);
1008         if (unlikely(rc != 0))
1009                 GOTO(out, rc = err_serious(rc));
1010
1011         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1012         LASSERT(repbody != NULL);
1013         repbody->eadatasize = 0;
1014         repbody->aclsize = 0;
1015
1016         if (reqbody->valid & OBD_MD_FLRMTPERM)
1017                 rc = mdt_init_ucred(info, reqbody);
1018         else
1019                 rc = mdt_check_ucred(info);
1020         if (unlikely(rc))
1021                 GOTO(out_shrink, rc);
1022
1023         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1024
1025         /*
1026          * Don't check capability at all, because rename might getattr for
1027          * remote obj, and at that time no capability is available.
1028          */
1029         mdt_set_capainfo(info, 1, &reqbody->fid1, BYPASS_CAPA);
1030         rc = mdt_getattr_internal(info, obj, 0);
1031         if (reqbody->valid & OBD_MD_FLRMTPERM)
1032                 mdt_exit_ucred(info);
1033         EXIT;
1034 out_shrink:
1035         mdt_client_compatibility(info);
1036         rc2 = mdt_fix_reply(info);
1037         if (rc == 0)
1038                 rc = rc2;
1039 out:
1040         mdt_thread_info_fini(info);
1041         return rc;
1042 }
1043
1044 int mdt_is_subdir(struct tgt_session_info *tsi)
1045 {
1046         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1047         struct mdt_object     *o = info->mti_object;
1048         struct req_capsule    *pill = info->mti_pill;
1049         const struct mdt_body *body = info->mti_body;
1050         struct mdt_body       *repbody;
1051         int                    rc;
1052         ENTRY;
1053
1054         LASSERT(o != NULL);
1055
1056         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1057
1058         /*
1059          * We save last checked parent fid to @repbody->fid1 for remote
1060          * directory case.
1061          */
1062         LASSERT(fid_is_sane(&body->fid2));
1063         LASSERT(mdt_object_exists(o) && !mdt_object_remote(o));
1064         rc = mdo_is_subdir(info->mti_env, mdt_object_child(o),
1065                            &body->fid2, &repbody->fid1);
1066         if (rc == 0 || rc == -EREMOTE)
1067                 repbody->valid |= OBD_MD_FLID;
1068
1069         mdt_thread_info_fini(info);
1070         RETURN(rc);
1071 }
1072
1073 int mdt_swap_layouts(struct tgt_session_info *tsi)
1074 {
1075         struct mdt_thread_info  *info;
1076         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1077         struct obd_export       *exp = req->rq_export;
1078         struct mdt_object       *o1, *o2, *o;
1079         struct mdt_lock_handle  *lh1, *lh2;
1080         struct mdc_swap_layouts *msl;
1081         int                      rc;
1082         ENTRY;
1083
1084         /* client does not support layout lock, so layout swaping
1085          * is disabled.
1086          * FIXME: there is a problem for old clients which don't support
1087          * layout lock yet. If those clients have already opened the file
1088          * they won't be notified at all so that old layout may still be
1089          * used to do IO. This can be fixed after file release is landed by
1090          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1091         if (!exp_connect_layout(exp))
1092                 RETURN(-EOPNOTSUPP);
1093
1094         info = tsi2mdt_info(tsi);
1095         if (req_capsule_get_size(info->mti_pill, &RMF_CAPA1, RCL_CLIENT))
1096                 mdt_set_capainfo(info, 0, &info->mti_body->fid1,
1097                                  req_capsule_client_get(info->mti_pill,
1098                                                         &RMF_CAPA1));
1099
1100         if (req_capsule_get_size(info->mti_pill, &RMF_CAPA2, RCL_CLIENT))
1101                 mdt_set_capainfo(info, 1, &info->mti_body->fid2,
1102                                  req_capsule_client_get(info->mti_pill,
1103                                                         &RMF_CAPA2));
1104
1105         o1 = info->mti_object;
1106         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1107                                 &info->mti_body->fid2);
1108         if (IS_ERR(o))
1109                 GOTO(out, rc = PTR_ERR(o));
1110
1111         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1112                 GOTO(put, rc = -ENOENT);
1113
1114         rc = lu_fid_cmp(&info->mti_body->fid1, &info->mti_body->fid2);
1115         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1116                 GOTO(put, rc);
1117
1118         if (rc < 0)
1119                 swap(o1, o2);
1120
1121         /* permission check. Make sure the calling process having permission
1122          * to write both files. */
1123         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1124                                 MAY_WRITE);
1125         if (rc < 0)
1126                 GOTO(put, rc);
1127
1128         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1129                                 MAY_WRITE);
1130         if (rc < 0)
1131                 GOTO(put, rc);
1132
1133         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1134         if (msl == NULL)
1135                 GOTO(put, rc = -EPROTO);
1136
1137         lh1 = &info->mti_lh[MDT_LH_NEW];
1138         mdt_lock_reg_init(lh1, LCK_EX);
1139         lh2 = &info->mti_lh[MDT_LH_OLD];
1140         mdt_lock_reg_init(lh2, LCK_EX);
1141
1142         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1143                              MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK);
1144         if (rc < 0)
1145                 GOTO(put, rc);
1146
1147         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1148                              MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK);
1149         if (rc < 0)
1150                 GOTO(unlock1, rc);
1151
1152         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1153                              mdt_object_child(o2), msl->msl_flags);
1154         GOTO(unlock2, rc);
1155 unlock2:
1156         mdt_object_unlock(info, o2, lh2, rc);
1157 unlock1:
1158         mdt_object_unlock(info, o1, lh1, rc);
1159 put:
1160         mdt_object_put(info->mti_env, o);
1161 out:
1162         mdt_thread_info_fini(info);
1163         RETURN(rc);
1164 }
1165
1166 static int mdt_raw_lookup(struct mdt_thread_info *info,
1167                           struct mdt_object *parent,
1168                           const struct lu_name *lname,
1169                           struct ldlm_reply *ldlm_rep)
1170 {
1171         struct md_object *next = mdt_object_child(info->mti_object);
1172         const struct mdt_body *reqbody = info->mti_body;
1173         struct lu_fid *child_fid = &info->mti_tmp_fid1;
1174         struct mdt_body *repbody;
1175         int rc;
1176         ENTRY;
1177
1178         if (reqbody->valid != OBD_MD_FLID)
1179                 RETURN(0);
1180
1181         LASSERT(!info->mti_cross_ref);
1182
1183         /* Only got the fid of this obj by name */
1184         fid_zero(child_fid);
1185         rc = mdo_lookup(info->mti_env, next, lname, child_fid,
1186                         &info->mti_spec);
1187 #if 0
1188         /* XXX is raw_lookup possible as intent operation? */
1189         if (rc != 0) {
1190                 if (rc == -ENOENT)
1191                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1192                 RETURN(rc);
1193         } else
1194                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1195
1196         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1197 #endif
1198         if (rc == 0) {
1199                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1200                 repbody->fid1 = *child_fid;
1201                 repbody->valid = OBD_MD_FLID;
1202         }
1203         RETURN(1);
1204 }
1205
1206 /*
1207  * UPDATE lock should be taken against parent, and be release before exit;
1208  * child_bits lock should be taken against child, and be returned back:
1209  *            (1)normal request should release the child lock;
1210  *            (2)intent request will grant the lock to client.
1211  */
1212 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1213                                  struct mdt_lock_handle *lhc,
1214                                  __u64 child_bits,
1215                                  struct ldlm_reply *ldlm_rep)
1216 {
1217         struct ptlrpc_request  *req       = mdt_info_req(info);
1218         struct mdt_body        *reqbody   = NULL;
1219         struct mdt_object      *parent    = info->mti_object;
1220         struct mdt_object      *child;
1221         struct md_object       *next      = mdt_object_child(parent);
1222         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1223         struct lu_name         *lname     = NULL;
1224         const char             *name      = NULL;
1225         int                     namelen   = 0;
1226         struct mdt_lock_handle *lhp       = NULL;
1227         struct ldlm_lock       *lock;
1228         struct ldlm_res_id     *res_id;
1229         int                     is_resent;
1230         int                     ma_need = 0;
1231         int                     rc;
1232
1233         ENTRY;
1234
1235         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1236         LASSERT(ergo(is_resent,
1237                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1238
1239         LASSERT(parent != NULL);
1240         name = req_capsule_client_get(info->mti_pill, &RMF_NAME);
1241         if (name == NULL)
1242                 RETURN(err_serious(-EFAULT));
1243
1244         namelen = req_capsule_get_size(info->mti_pill, &RMF_NAME,
1245                                        RCL_CLIENT) - 1;
1246         if (!info->mti_cross_ref) {
1247                 /*
1248                  * XXX: Check for "namelen == 0" is for getattr by fid
1249                  * (OBD_CONNECT_ATTRFID), otherwise do not allow empty name,
1250                  * that is the name must contain at least one character and
1251                  * the terminating '\0'
1252                  */
1253                 if (namelen == 0) {
1254                         reqbody = req_capsule_client_get(info->mti_pill,
1255                                                          &RMF_MDT_BODY);
1256                         if (unlikely(reqbody == NULL))
1257                                 RETURN(err_serious(-EFAULT));
1258
1259                         if (unlikely(!fid_is_sane(&reqbody->fid2)))
1260                                 RETURN(err_serious(-EINVAL));
1261
1262                         name = NULL;
1263                         CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1264                                "ldlm_rep = %p\n",
1265                                PFID(mdt_object_fid(parent)),
1266                                PFID(&reqbody->fid2), ldlm_rep);
1267                 } else {
1268                         lname = mdt_name(info->mti_env, (char *)name, namelen);
1269                         CDEBUG(D_INODE, "getattr with lock for "DFID"/%s, "
1270                                "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1271                                name, ldlm_rep);
1272                 }
1273         }
1274         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1275
1276         if (unlikely(!mdt_object_exists(parent)) && lname) {
1277                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1278                                 &parent->mot_obj,
1279                                 "Parent doesn't exist!\n");
1280                 RETURN(-ESTALE);
1281         } else if (!info->mti_cross_ref) {
1282                 LASSERTF(!mdt_object_remote(parent),
1283                          "Parent "DFID" is on remote server\n",
1284                          PFID(mdt_object_fid(parent)));
1285         }
1286         if (lname) {
1287                 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1288                 if (rc != 0) {
1289                         if (rc > 0)
1290                                 rc = 0;
1291                         RETURN(rc);
1292                 }
1293         }
1294
1295         if (info->mti_cross_ref) {
1296                 /* Only getattr on the child. Parent is on another node. */
1297                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1298                 child = parent;
1299                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1300                        "ldlm_rep=%p\n", PFID(mdt_object_fid(child)), ldlm_rep);
1301
1302                 if (is_resent) {
1303                         /* Do not take lock for resent case. */
1304                         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1305                         LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
1306                                  lhc->mlh_reg_lh.cookie);
1307                         LASSERT(fid_res_name_eq(mdt_object_fid(child),
1308                                                 &lock->l_resource->lr_name));
1309                         LDLM_LOCK_PUT(lock);
1310                         rc = 0;
1311                 } else {
1312                         mdt_lock_handle_init(lhc);
1313                         mdt_lock_reg_init(lhc, LCK_PR);
1314
1315                         /*
1316                          * Object's name is on another MDS, no lookup or layout
1317                          * lock is needed here but update lock is.
1318                          */
1319                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
1320                                         MDS_INODELOCK_LAYOUT);
1321                         child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1322
1323                         rc = mdt_object_lock(info, child, lhc, child_bits,
1324                                              MDT_LOCAL_LOCK);
1325                 }
1326                 if (rc == 0) {
1327                         /* Finally, we can get attr for child. */
1328                         if (!mdt_object_exists(child)) {
1329                                 LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1330                                                 &child->mot_obj,
1331                                              "remote object doesn't exist.\n");
1332                                 mdt_object_unlock(info, child, lhc, 1);
1333                                 RETURN(-ENOENT);
1334                         }
1335
1336                         mdt_set_capainfo(info, 0, mdt_object_fid(child),
1337                                          BYPASS_CAPA);
1338                         rc = mdt_getattr_internal(info, child, 0);
1339                         if (unlikely(rc != 0))
1340                                 mdt_object_unlock(info, child, lhc, 1);
1341                 }
1342                 RETURN(rc);
1343         }
1344
1345         if (lname) {
1346                 /* step 1: lock parent only if parent is a directory */
1347                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1348                         lhp = &info->mti_lh[MDT_LH_PARENT];
1349                         mdt_lock_pdo_init(lhp, LCK_PR, name, namelen);
1350                         rc = mdt_object_lock(info, parent, lhp,
1351                                              MDS_INODELOCK_UPDATE,
1352                                              MDT_LOCAL_LOCK);
1353                         if (unlikely(rc != 0))
1354                                 RETURN(rc);
1355                 }
1356
1357                 /* step 2: lookup child's fid by name */
1358                 fid_zero(child_fid);
1359                 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
1360                                 &info->mti_spec);
1361
1362                 if (rc != 0) {
1363                         if (rc == -ENOENT)
1364                                 mdt_set_disposition(info, ldlm_rep,
1365                                                     DISP_LOOKUP_NEG);
1366                         GOTO(out_parent, rc);
1367                 } else
1368                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1369         } else {
1370                 *child_fid = reqbody->fid2;
1371                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1372         }
1373
1374         /*
1375          *step 3: find the child object by fid & lock it.
1376          *        regardless if it is local or remote.
1377          *
1378          *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1379          *      set parent dir fid the same as child fid in getattr by fid case
1380          *      we should not lu_object_find() the object again, could lead
1381          *      to hung if there is a concurrent unlink destroyed the object.
1382          */
1383         if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1384                 mdt_object_get(info->mti_env, parent);
1385                 child = parent;
1386         } else {
1387                 child = mdt_object_find(info->mti_env, info->mti_mdt,
1388                                         child_fid);
1389         }
1390
1391         if (unlikely(IS_ERR(child)))
1392                 GOTO(out_parent, rc = PTR_ERR(child));
1393         if (is_resent) {
1394                 /* Do not take lock for resent case. */
1395                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1396                 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
1397                          lhc->mlh_reg_lh.cookie);
1398
1399                 res_id = &lock->l_resource->lr_name;
1400                 if (!fid_res_name_eq(mdt_object_fid(child),
1401                                      &lock->l_resource->lr_name)) {
1402                         LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
1403                                                  &lock->l_resource->lr_name),
1404                                  "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1405                                  PLDLMRES(lock->l_resource),
1406                                  PFID(mdt_object_fid(parent)));
1407                         CWARN("Although resent, but still not get child lock"
1408                               "parent:"DFID" child:"DFID"\n",
1409                               PFID(mdt_object_fid(parent)),
1410                               PFID(mdt_object_fid(child)));
1411                         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
1412                         LDLM_LOCK_PUT(lock);
1413                         GOTO(relock, 0);
1414                 }
1415                 LDLM_LOCK_PUT(lock);
1416                 rc = 0;
1417         } else {
1418                 bool try_layout = false;
1419
1420 relock:
1421                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout*2);
1422                 mdt_lock_handle_init(lhc);
1423                 mdt_lock_reg_init(lhc, LCK_PR);
1424
1425                 if (!mdt_object_exists(child)) {
1426                         LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1427                                         &child->mot_obj,
1428                                         "Object doesn't exist!\n");
1429                         GOTO(out_child, rc = -ENOENT);
1430                 }
1431
1432                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1433                       mdt_object_exists(child) && !mdt_object_remote(child)) {
1434                         struct md_attr *ma = &info->mti_attr;
1435
1436                         ma->ma_valid = 0;
1437                         ma->ma_need = MA_INODE;
1438                         rc = mdt_attr_get_complex(info, child, ma);
1439                         if (unlikely(rc != 0))
1440                                 GOTO(out_child, rc);
1441
1442                         /* If the file has not been changed for some time, we
1443                          * return not only a LOOKUP lock, but also an UPDATE
1444                          * lock and this might save us RPC on later STAT. For
1445                          * directories, it also let negative dentry cache start
1446                          * working for this dir. */
1447                         if (ma->ma_valid & MA_INODE &&
1448                             ma->ma_attr.la_valid & LA_CTIME &&
1449                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1450                                 ma->ma_attr.la_ctime < cfs_time_current_sec())
1451                                 child_bits |= MDS_INODELOCK_UPDATE;
1452                 }
1453
1454                 /* layout lock must be granted in a best-effort way
1455                  * for IT operations */
1456                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1457                 if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1458                     exp_connect_layout(info->mti_exp) &&
1459                     S_ISREG(lu_object_attr(&child->mot_obj)) &&
1460                     !mdt_object_remote(child) && ldlm_rep != NULL) {
1461                         /* try to grant layout lock for regular file. */
1462                         try_layout = true;
1463                 }
1464
1465                 rc = 0;
1466                 if (try_layout) {
1467                         child_bits |= MDS_INODELOCK_LAYOUT;
1468                         /* try layout lock, it may fail to be granted due to
1469                          * contention at LOOKUP or UPDATE */
1470                         if (!mdt_object_lock_try(info, child, lhc, child_bits,
1471                                                  MDT_CROSS_LOCK)) {
1472                                 child_bits &= ~MDS_INODELOCK_LAYOUT;
1473                                 LASSERT(child_bits != 0);
1474                                 rc = mdt_object_lock(info, child, lhc,
1475                                                 child_bits, MDT_CROSS_LOCK);
1476                         } else {
1477                                 ma_need |= MA_LOV;
1478                         }
1479                 } else {
1480                         rc = mdt_object_lock(info, child, lhc, child_bits,
1481                                                 MDT_CROSS_LOCK);
1482                 }
1483                 if (unlikely(rc != 0))
1484                         GOTO(out_child, rc);
1485         }
1486
1487         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1488         /* Get MA_SOM attributes if update lock is given. */
1489         if (lock &&
1490             lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_UPDATE &&
1491             S_ISREG(lu_object_attr(&mdt_object_child(child)->mo_lu)))
1492                 ma_need |= MA_SOM;
1493
1494         /* finally, we can get attr for child. */
1495         mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
1496         rc = mdt_getattr_internal(info, child, ma_need);
1497         if (unlikely(rc != 0)) {
1498                 mdt_object_unlock(info, child, lhc, 1);
1499         } else if (lock) {
1500                 /* Debugging code. */
1501                 res_id = &lock->l_resource->lr_name;
1502                 LDLM_DEBUG(lock, "Returning lock to client");
1503                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1504                                          &lock->l_resource->lr_name),
1505                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1506                          PLDLMRES(lock->l_resource),
1507                          PFID(mdt_object_fid(child)));
1508                 if (mdt_object_exists(child) && !mdt_object_remote(child))
1509                         mdt_pack_size2body(info, child);
1510         }
1511         if (lock)
1512                 LDLM_LOCK_PUT(lock);
1513
1514         EXIT;
1515 out_child:
1516         mdt_object_put(info->mti_env, child);
1517 out_parent:
1518         if (lhp)
1519                 mdt_object_unlock(info, parent, lhp, 1);
1520         return rc;
1521 }
1522
1523 /* normal handler: should release the child lock */
1524 int mdt_getattr_name(struct tgt_session_info *tsi)
1525 {
1526         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1527         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1528         struct mdt_body        *reqbody;
1529         struct mdt_body        *repbody;
1530         int rc, rc2;
1531         ENTRY;
1532
1533         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1534         LASSERT(reqbody != NULL);
1535         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1536         LASSERT(repbody != NULL);
1537
1538         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1539         repbody->eadatasize = 0;
1540         repbody->aclsize = 0;
1541
1542         rc = mdt_init_ucred(info, reqbody);
1543         if (unlikely(rc))
1544                 GOTO(out_shrink, rc);
1545
1546         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1547         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1548                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1549                 lhc->mlh_reg_lh.cookie = 0;
1550         }
1551         mdt_exit_ucred(info);
1552         EXIT;
1553 out_shrink:
1554         mdt_client_compatibility(info);
1555         rc2 = mdt_fix_reply(info);
1556         if (rc == 0)
1557                 rc = rc2;
1558         mdt_thread_info_fini(info);
1559         return rc;
1560 }
1561
1562 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1563                          void *karg, void *uarg);
1564
1565 int mdt_set_info(struct tgt_session_info *tsi)
1566 {
1567         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1568         char                    *key;
1569         void                    *val;
1570         int                      keylen, vallen, rc = 0;
1571
1572         ENTRY;
1573
1574         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
1575         if (key == NULL) {
1576                 DEBUG_REQ(D_HA, req, "no set_info key");
1577                 RETURN(err_serious(-EFAULT));
1578         }
1579
1580         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
1581                                       RCL_CLIENT);
1582
1583         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
1584         if (val == NULL) {
1585                 DEBUG_REQ(D_HA, req, "no set_info val");
1586                 RETURN(err_serious(-EFAULT));
1587         }
1588
1589         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
1590                                       RCL_CLIENT);
1591
1592         /* Swab any part of val you need to here */
1593         if (KEY_IS(KEY_READ_ONLY)) {
1594                 spin_lock(&req->rq_export->exp_lock);
1595                 if (*(__u32 *)val)
1596                         *exp_connect_flags_ptr(req->rq_export) |=
1597                                 OBD_CONNECT_RDONLY;
1598                 else
1599                         *exp_connect_flags_ptr(req->rq_export) &=
1600                                 ~OBD_CONNECT_RDONLY;
1601                 spin_unlock(&req->rq_export->exp_lock);
1602         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1603                 struct changelog_setinfo *cs = val;
1604
1605                 if (vallen != sizeof(*cs)) {
1606                         CERROR("%s: bad changelog_clear setinfo size %d\n",
1607                                tgt_name(tsi->tsi_tgt), vallen);
1608                         RETURN(-EINVAL);
1609                 }
1610                 if (ptlrpc_req_need_swab(req)) {
1611                         __swab64s(&cs->cs_recno);
1612                         __swab32s(&cs->cs_id);
1613                 }
1614
1615                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
1616                                    vallen, val, NULL);
1617         } else {
1618                 RETURN(-EINVAL);
1619         }
1620         RETURN(rc);
1621 }
1622
1623 int mdt_readpage(struct tgt_session_info *tsi)
1624 {
1625         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
1626         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
1627         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
1628         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
1629         struct mdt_body         *repbody;
1630         int                      rc;
1631         int                      i;
1632
1633         ENTRY;
1634
1635         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1636                 RETURN(err_serious(-ENOMEM));
1637
1638         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
1639         if (repbody == NULL || reqbody == NULL)
1640                 RETURN(err_serious(-EFAULT));
1641
1642         /*
1643          * prepare @rdpg before calling lower layers and transfer itself. Here
1644          * reqbody->size contains offset of where to start to read and
1645          * reqbody->nlink contains number bytes to read.
1646          */
1647         rdpg->rp_hash = reqbody->size;
1648         if (rdpg->rp_hash != reqbody->size) {
1649                 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1650                        rdpg->rp_hash, reqbody->size);
1651                 RETURN(-EFAULT);
1652         }
1653
1654         rdpg->rp_attrs = reqbody->mode;
1655         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
1656                 rdpg->rp_attrs |= LUDA_64BITHASH;
1657         rdpg->rp_count  = min_t(unsigned int, reqbody->nlink,
1658                                 exp_max_brw_size(tsi->tsi_exp));
1659         rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE - 1) >>
1660                           PAGE_CACHE_SHIFT;
1661         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1662         if (rdpg->rp_pages == NULL)
1663                 RETURN(-ENOMEM);
1664
1665         for (i = 0; i < rdpg->rp_npages; ++i) {
1666                 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1667                 if (rdpg->rp_pages[i] == NULL)
1668                         GOTO(free_rdpg, rc = -ENOMEM);
1669         }
1670
1671         /* call lower layers to fill allocated pages with directory data */
1672         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
1673         if (rc < 0)
1674                 GOTO(free_rdpg, rc);
1675
1676         /* send pages to client */
1677         rc = tgt_sendpage(tsi, rdpg, rc);
1678
1679         EXIT;
1680 free_rdpg:
1681
1682         for (i = 0; i < rdpg->rp_npages; i++)
1683                 if (rdpg->rp_pages[i] != NULL)
1684                         __free_page(rdpg->rp_pages[i]);
1685         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1686
1687         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1688                 RETURN(0);
1689
1690         return rc;
1691 }
1692
1693 static int mdt_reint_internal(struct mdt_thread_info *info,
1694                               struct mdt_lock_handle *lhc,
1695                               __u32 op)
1696 {
1697         struct req_capsule      *pill = info->mti_pill;
1698         struct mdt_body         *repbody;
1699         int                      rc = 0, rc2;
1700
1701         ENTRY;
1702
1703         rc = mdt_reint_unpack(info, op);
1704         if (rc != 0) {
1705                 CERROR("Can't unpack reint, rc %d\n", rc);
1706                 RETURN(err_serious(rc));
1707         }
1708
1709         /* for replay (no_create) lmm is not needed, client has it already */
1710         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1711                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1712                                      info->mti_rr.rr_eadatalen);
1713
1714         /* llog cookies are always 0, the field is kept for compatibility */
1715         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1716                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
1717
1718         rc = req_capsule_server_pack(pill);
1719         if (rc != 0) {
1720                 CERROR("Can't pack response, rc %d\n", rc);
1721                 RETURN(err_serious(rc));
1722         }
1723
1724         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1725                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1726                 LASSERT(repbody);
1727                 repbody->eadatasize = 0;
1728                 repbody->aclsize = 0;
1729         }
1730
1731         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1732
1733         /* for replay no cookkie / lmm need, because client have this already */
1734         if (info->mti_spec.no_create)
1735                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1736                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1737
1738         rc = mdt_init_ucred_reint(info);
1739         if (rc)
1740                 GOTO(out_shrink, rc);
1741
1742         rc = mdt_fix_attr_ucred(info, op);
1743         if (rc != 0)
1744                 GOTO(out_ucred, rc = err_serious(rc));
1745
1746         if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1747                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1748                 GOTO(out_ucred, rc);
1749         }
1750         rc = mdt_reint_rec(info, lhc);
1751         EXIT;
1752 out_ucred:
1753         mdt_exit_ucred(info);
1754 out_shrink:
1755         mdt_client_compatibility(info);
1756         rc2 = mdt_fix_reply(info);
1757         if (rc == 0)
1758                 rc = rc2;
1759         return rc;
1760 }
1761
1762 static long mdt_reint_opcode(struct ptlrpc_request *req,
1763                              const struct req_format **fmt)
1764 {
1765         struct mdt_device       *mdt;
1766         struct mdt_rec_reint    *rec;
1767         long                     opc;
1768
1769         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
1770         if (rec != NULL) {
1771                 opc = rec->rr_opcode;
1772                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
1773                 if (opc < REINT_MAX && fmt[opc] != NULL)
1774                         req_capsule_extend(&req->rq_pill, fmt[opc]);
1775                 else {
1776                         mdt = mdt_exp2dev(req->rq_export);
1777                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
1778                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
1779                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
1780                         opc = err_serious(-EFAULT);
1781                 }
1782         } else {
1783                 opc = err_serious(-EFAULT);
1784         }
1785         return opc;
1786 }
1787
1788 int mdt_reint(struct tgt_session_info *tsi)
1789 {
1790         long opc;
1791         int  rc;
1792         static const struct req_format *reint_fmts[REINT_MAX] = {
1793                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
1794                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
1795                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
1796                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
1797                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
1798                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
1799                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
1800                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK
1801         };
1802
1803         ENTRY;
1804
1805         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
1806         if (opc >= 0) {
1807                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1808                 /*
1809                  * No lock possible here from client to pass it to reint code
1810                  * path.
1811                  */
1812                 rc = mdt_reint_internal(info, NULL, opc);
1813                 mdt_thread_info_fini(info);
1814         } else {
1815                 rc = opc;
1816         }
1817
1818         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1819         RETURN(rc);
1820 }
1821
1822 /* this should sync the whole device */
1823 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1824 {
1825         struct dt_device *dt = mdt->mdt_bottom;
1826         int rc;
1827         ENTRY;
1828
1829         rc = dt->dd_ops->dt_sync(env, dt);
1830         RETURN(rc);
1831 }
1832
1833 /* this should sync this object */
1834 static int mdt_object_sync(struct mdt_thread_info *info)
1835 {
1836         struct md_object *next;
1837         int rc;
1838         ENTRY;
1839
1840         if (!mdt_object_exists(info->mti_object)) {
1841                 CWARN("Non existing object  "DFID"!\n",
1842                       PFID(mdt_object_fid(info->mti_object)));
1843                 RETURN(-ESTALE);
1844         }
1845         next = mdt_object_child(info->mti_object);
1846         rc = mo_object_sync(info->mti_env, next);
1847
1848         RETURN(rc);
1849 }
1850
1851 int mdt_sync(struct tgt_session_info *tsi)
1852 {
1853         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1854         struct req_capsule      *pill = tsi->tsi_pill;
1855         struct mdt_body         *body;
1856         int                      rc;
1857
1858         ENTRY;
1859
1860         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1861                 RETURN(err_serious(-ENOMEM));
1862
1863         if (fid_seq(&tsi->tsi_mdt_body->fid1) == 0) {
1864                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
1865         } else {
1866                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1867
1868                 /* sync an object */
1869                 rc = mdt_object_sync(info);
1870                 if (rc == 0) {
1871                         const struct lu_fid *fid;
1872                         struct lu_attr *la = &info->mti_attr.ma_attr;
1873
1874                         info->mti_attr.ma_need = MA_INODE;
1875                         info->mti_attr.ma_valid = 0;
1876                         rc = mdt_attr_get_complex(info, info->mti_object,
1877                                                   &info->mti_attr);
1878                         if (rc == 0) {
1879                                 body = req_capsule_server_get(pill,
1880                                                               &RMF_MDT_BODY);
1881                                 fid = mdt_object_fid(info->mti_object);
1882                                 mdt_pack_attr2body(info, body, la, fid);
1883                         }
1884                 }
1885                 mdt_thread_info_fini(info);
1886         }
1887         if (rc == 0)
1888                 mdt_counter_incr(req, LPROC_MDT_SYNC);
1889
1890         RETURN(rc);
1891 }
1892
1893 /*
1894  * Handle quota control requests to consult current usage/limit, but also
1895  * to configure quota enforcement
1896  */
1897 int mdt_quotactl(struct tgt_session_info *tsi)
1898 {
1899         struct obd_export       *exp  = tsi->tsi_exp;
1900         struct req_capsule      *pill = tsi->tsi_pill;
1901         struct obd_quotactl     *oqctl, *repoqc;
1902         int                      id, rc;
1903         struct mdt_device       *mdt = mdt_exp2dev(exp);
1904         struct lu_device        *qmt = mdt->mdt_qmt_dev;
1905         ENTRY;
1906
1907         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1908         if (oqctl == NULL)
1909                 RETURN(err_serious(-EPROTO));
1910
1911         rc = req_capsule_server_pack(pill);
1912         if (rc)
1913                 RETURN(err_serious(rc));
1914
1915         switch (oqctl->qc_cmd) {
1916         case Q_QUOTACHECK:
1917         case LUSTRE_Q_INVALIDATE:
1918         case LUSTRE_Q_FINVALIDATE:
1919         case Q_QUOTAON:
1920         case Q_QUOTAOFF:
1921         case Q_INITQUOTA:
1922                 /* deprecated, not used any more */
1923                 RETURN(-EOPNOTSUPP);
1924                 /* master quotactl */
1925         case Q_GETINFO:
1926         case Q_SETINFO:
1927         case Q_SETQUOTA:
1928         case Q_GETQUOTA:
1929                 if (qmt == NULL)
1930                         RETURN(-EOPNOTSUPP);
1931                 /* slave quotactl */
1932         case Q_GETOINFO:
1933         case Q_GETOQUOTA:
1934                 break;
1935         default:
1936                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
1937                 RETURN(-EFAULT);
1938         }
1939
1940         /* map uid/gid for remote client */
1941         id = oqctl->qc_id;
1942         if (exp_connect_rmtclient(exp)) {
1943                 struct lustre_idmap_table *idmap;
1944
1945                 idmap = exp->exp_mdt_data.med_idmap;
1946
1947                 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
1948                              oqctl->qc_cmd != Q_GETINFO))
1949                         RETURN(-EPERM);
1950
1951                 if (oqctl->qc_type == USRQUOTA)
1952                         id = lustre_idmap_lookup_uid(NULL, idmap, 0,
1953                                                      oqctl->qc_id);
1954                 else if (oqctl->qc_type == GRPQUOTA)
1955                         id = lustre_idmap_lookup_gid(NULL, idmap, 0,
1956                                                      oqctl->qc_id);
1957                 else
1958                         RETURN(-EINVAL);
1959
1960                 if (id == CFS_IDMAP_NOTFOUND) {
1961                         CDEBUG(D_QUOTA, "no mapping for id %u\n", oqctl->qc_id);
1962                         RETURN(-EACCES);
1963                 }
1964         }
1965
1966         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
1967         if (repoqc == NULL)
1968                 RETURN(err_serious(-EFAULT));
1969
1970         if (oqctl->qc_id != id)
1971                 swap(oqctl->qc_id, id);
1972
1973         switch (oqctl->qc_cmd) {
1974
1975         case Q_GETINFO:
1976         case Q_SETINFO:
1977         case Q_SETQUOTA:
1978         case Q_GETQUOTA:
1979                 /* forward quotactl request to QMT */
1980                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
1981                 break;
1982
1983         case Q_GETOINFO:
1984         case Q_GETOQUOTA:
1985                 /* slave quotactl */
1986                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
1987                                    oqctl);
1988                 break;
1989
1990         default:
1991                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
1992                 RETURN(-EFAULT);
1993         }
1994
1995         if (oqctl->qc_id != id)
1996                 swap(oqctl->qc_id, id);
1997
1998         *repoqc = *oqctl;
1999         RETURN(rc);
2000 }
2001
2002 /** clone llog ctxt from child (mdd)
2003  * This allows remote llog (replicator) access.
2004  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2005  * context was originally set up, or we can handle them directly.
2006  * I choose the latter, but that means I need any llog
2007  * contexts set up by child to be accessable by the mdt.  So we clone the
2008  * context into our context list here.
2009  */
2010 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2011                                int idx)
2012 {
2013         struct md_device  *next = mdt->mdt_child;
2014         struct llog_ctxt *ctxt;
2015         int rc;
2016
2017         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2018                 return 0;
2019
2020         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2021         if (rc || ctxt == NULL) {
2022                 return 0;
2023         }
2024
2025         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2026         if (rc)
2027                 CERROR("Can't set mdt ctxt %d\n", rc);
2028
2029         return rc;
2030 }
2031
2032 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2033                                  struct mdt_device *mdt, int idx)
2034 {
2035         struct llog_ctxt *ctxt;
2036
2037         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2038         if (ctxt == NULL)
2039                 return 0;
2040         /* Put once for the get we just did, and once for the clone */
2041         llog_ctxt_put(ctxt);
2042         llog_ctxt_put(ctxt);
2043         return 0;
2044 }
2045
2046 /*
2047  * sec context handlers
2048  */
2049 int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2050 {
2051         int rc;
2052
2053         rc = mdt_handle_idmap(tsi);
2054         if (unlikely(rc)) {
2055                 struct ptlrpc_request   *req = tgt_ses_req(tsi);
2056                 __u32                    opc;
2057
2058                 opc = lustre_msg_get_opc(req->rq_reqmsg);
2059                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2060                         sptlrpc_svc_ctx_invalidate(req);
2061         }
2062
2063         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2064
2065         return rc;
2066 }
2067
2068 /*
2069  * quota request handlers
2070  */
2071 int mdt_quota_dqacq(struct tgt_session_info *tsi)
2072 {
2073         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
2074         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2075         int                      rc;
2076         ENTRY;
2077
2078         if (qmt == NULL)
2079                 RETURN(err_serious(-EOPNOTSUPP));
2080
2081         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2082         RETURN(rc);
2083 }
2084
2085 struct mdt_object *mdt_object_new(const struct lu_env *env,
2086                                   struct mdt_device *d,
2087                                   const struct lu_fid *f)
2088 {
2089         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2090         struct lu_object *o;
2091         struct mdt_object *m;
2092         ENTRY;
2093
2094         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2095         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2096         if (unlikely(IS_ERR(o)))
2097                 m = (struct mdt_object *)o;
2098         else
2099                 m = mdt_obj(o);
2100         RETURN(m);
2101 }
2102
2103 struct mdt_object *mdt_object_find(const struct lu_env *env,
2104                                    struct mdt_device *d,
2105                                    const struct lu_fid *f)
2106 {
2107         struct lu_object *o;
2108         struct mdt_object *m;
2109         ENTRY;
2110
2111         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2112         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2113         if (unlikely(IS_ERR(o)))
2114                 m = (struct mdt_object *)o;
2115         else
2116                 m = mdt_obj(o);
2117
2118         RETURN(m);
2119 }
2120
2121 /**
2122  * Asyncronous commit for mdt device.
2123  *
2124  * Pass asynchonous commit call down the MDS stack.
2125  *
2126  * \param env environment
2127  * \param mdt the mdt device
2128  */
2129 static void mdt_device_commit_async(const struct lu_env *env,
2130                                     struct mdt_device *mdt)
2131 {
2132         struct dt_device *dt = mdt->mdt_bottom;
2133         int rc;
2134
2135         rc = dt->dd_ops->dt_commit_async(env, dt);
2136         if (unlikely(rc != 0))
2137                 CWARN("async commit start failed with rc = %d", rc);
2138 }
2139
2140 /**
2141  * Mark the lock as "synchonous".
2142  *
2143  * Mark the lock to deffer transaction commit to the unlock time.
2144  *
2145  * \param lock the lock to mark as "synchonous"
2146  *
2147  * \see mdt_is_lock_sync
2148  * \see mdt_save_lock
2149  */
2150 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2151 {
2152         lock->l_ast_data = (void*)1;
2153 }
2154
2155 /**
2156  * Check whehter the lock "synchonous" or not.
2157  *
2158  * \param lock the lock to check
2159  * \retval 1 the lock is "synchonous"
2160  * \retval 0 the lock isn't "synchronous"
2161  *
2162  * \see mdt_set_lock_sync
2163  * \see mdt_save_lock
2164  */
2165 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2166 {
2167         return lock->l_ast_data != NULL;
2168 }
2169
2170 /**
2171  * Blocking AST for mdt locks.
2172  *
2173  * Starts transaction commit if in case of COS lock conflict or
2174  * deffers such a commit to the mdt_save_lock.
2175  *
2176  * \param lock the lock which blocks a request or cancelling lock
2177  * \param desc unused
2178  * \param data unused
2179  * \param flag indicates whether this cancelling or blocking callback
2180  * \retval 0
2181  * \see ldlm_blocking_ast_nocheck
2182  */
2183 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2184                      void *data, int flag)
2185 {
2186         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2187         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2188         int rc;
2189         ENTRY;
2190
2191         if (flag == LDLM_CB_CANCELING)
2192                 RETURN(0);
2193         lock_res_and_lock(lock);
2194         if (lock->l_blocking_ast != mdt_blocking_ast) {
2195                 unlock_res_and_lock(lock);
2196                 RETURN(0);
2197         }
2198         if (mdt_cos_is_enabled(mdt) &&
2199             lock->l_req_mode & (LCK_PW | LCK_EX) &&
2200             lock->l_blocking_lock != NULL &&
2201             lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
2202                 mdt_set_lock_sync(lock);
2203         }
2204         rc = ldlm_blocking_ast_nocheck(lock);
2205
2206         /* There is no lock conflict if l_blocking_lock == NULL,
2207          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2208          * when the last reference to a local lock was released */
2209         if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2210                 struct lu_env env;
2211
2212                 rc = lu_env_init(&env, LCT_LOCAL);
2213                 if (unlikely(rc != 0))
2214                         CWARN("lu_env initialization failed with rc = %d,"
2215                               "cannot start asynchronous commit\n", rc);
2216                 else
2217                         mdt_device_commit_async(&env, mdt);
2218                 lu_env_fini(&env);
2219         }
2220         RETURN(rc);
2221 }
2222
2223 int mdt_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2224                         void *data, int flag)
2225 {
2226         struct lustre_handle lockh;
2227         int               rc;
2228
2229         switch (flag) {
2230         case LDLM_CB_BLOCKING:
2231                 ldlm_lock2handle(lock, &lockh);
2232                 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
2233                 if (rc < 0) {
2234                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2235                         RETURN(rc);
2236                 }
2237                 break;
2238         case LDLM_CB_CANCELING:
2239                 LDLM_DEBUG(lock, "Revoke remote lock\n");
2240                 break;
2241         default:
2242                 LBUG();
2243         }
2244         RETURN(0);
2245 }
2246
2247 int mdt_remote_object_lock(struct mdt_thread_info *mti,
2248                            struct mdt_object *o, struct lustre_handle *lh,
2249                            ldlm_mode_t mode, __u64 ibits)
2250 {
2251         struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
2252         ldlm_policy_data_t *policy = &mti->mti_policy;
2253         int rc = 0;
2254         ENTRY;
2255
2256         LASSERT(mdt_object_remote(o));
2257
2258         LASSERT(ibits & MDS_INODELOCK_UPDATE);
2259
2260         memset(einfo, 0, sizeof(*einfo));
2261         einfo->ei_type = LDLM_IBITS;
2262         einfo->ei_mode = mode;
2263         einfo->ei_cb_bl = mdt_md_blocking_ast;
2264         einfo->ei_cb_cp = ldlm_completion_ast;
2265
2266         memset(policy, 0, sizeof(*policy));
2267         policy->l_inodebits.bits = ibits;
2268
2269         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2270                             policy);
2271         RETURN(rc);
2272 }
2273
2274 static int mdt_object_lock0(struct mdt_thread_info *info, struct mdt_object *o,
2275                             struct mdt_lock_handle *lh, __u64 ibits,
2276                             bool nonblock, int locality)
2277 {
2278         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2279         ldlm_policy_data_t *policy = &info->mti_policy;
2280         struct ldlm_res_id *res_id = &info->mti_res_id;
2281         __u64 dlmflags;
2282         int rc;
2283         ENTRY;
2284
2285         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2286         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2287         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2288         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2289
2290         if (mdt_object_remote(o)) {
2291                 if (locality == MDT_CROSS_LOCK) {
2292                         ibits &= ~(MDS_INODELOCK_UPDATE | MDS_INODELOCK_PERM |
2293                                    MDS_INODELOCK_LAYOUT);
2294                         ibits |= MDS_INODELOCK_LOOKUP;
2295                 } else {
2296                         LASSERTF(!(ibits &
2297                                  (MDS_INODELOCK_UPDATE | MDS_INODELOCK_PERM |
2298                                   MDS_INODELOCK_LAYOUT)),
2299                                 "%s: wrong bit "LPX64" for remote obj "DFID"\n",
2300                                 mdt_obd_name(info->mti_mdt), ibits,
2301                                 PFID(mdt_object_fid(o)));
2302                         LASSERT(ibits & MDS_INODELOCK_LOOKUP);
2303                 }
2304                 /* No PDO lock on remote object */
2305                 LASSERT(lh->mlh_type != MDT_PDO_LOCK);
2306         }
2307
2308         if (lh->mlh_type == MDT_PDO_LOCK) {
2309                 /* check for exists after object is locked */
2310                 if (mdt_object_exists(o) == 0) {
2311                         /* Non-existent object shouldn't have PDO lock */
2312                         RETURN(-ESTALE);
2313                 } else {
2314                         /* Non-dir object shouldn't have PDO lock */
2315                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2316                                 RETURN(-ENOTDIR);
2317                 }
2318         }
2319
2320         memset(policy, 0, sizeof(*policy));
2321         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2322
2323         dlmflags = LDLM_FL_ATOMIC_CB;
2324         if (nonblock)
2325                 dlmflags |= LDLM_FL_BLOCK_NOWAIT;
2326
2327         /*
2328          * Take PDO lock on whole directory and build correct @res_id for lock
2329          * on part of directory.
2330          */
2331         if (lh->mlh_pdo_hash != 0) {
2332                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2333                 mdt_lock_pdo_mode(info, o, lh);
2334                 if (lh->mlh_pdo_mode != LCK_NL) {
2335                         /*
2336                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2337                          * is never going to be sent to client and we do not
2338                          * want it slowed down due to possible cancels.
2339                          */
2340                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2341                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2342                                           policy, res_id, dlmflags,
2343                                           &info->mti_exp->exp_handle.h_cookie);
2344                         if (unlikely(rc))
2345                                 RETURN(rc);
2346                 }
2347
2348                 /*
2349                  * Finish res_id initializing by name hash marking part of
2350                  * directory which is taking modification.
2351                  */
2352                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2353         }
2354
2355         policy->l_inodebits.bits = ibits;
2356
2357         /*
2358          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2359          * going to be sent to client. If it is - mdt_intent_policy() path will
2360          * fix it up and turn FL_LOCAL flag off.
2361          */
2362         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2363                           res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2364                           &info->mti_exp->exp_handle.h_cookie);
2365         if (rc)
2366                 mdt_object_unlock(info, o, lh, 1);
2367         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2368                  lh->mlh_pdo_hash != 0 &&
2369                  (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX)) {
2370                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2371         }
2372
2373         RETURN(rc);
2374 }
2375
2376 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2377                     struct mdt_lock_handle *lh, __u64 ibits, int locality)
2378 {
2379         return mdt_object_lock0(info, o, lh, ibits, false, locality);
2380 }
2381
2382 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
2383                         struct mdt_lock_handle *lh, __u64 ibits, int locality)
2384 {
2385         struct mdt_lock_handle tmp = *lh;
2386         int rc;
2387
2388         rc = mdt_object_lock0(info, o, &tmp, ibits, true, locality);
2389         if (rc == 0)
2390                 *lh = tmp;
2391
2392         return rc == 0;
2393 }
2394
2395 /**
2396  * Save a lock within request object.
2397  *
2398  * Keep the lock referenced until whether client ACK or transaction
2399  * commit happens or release the lock immediately depending on input
2400  * parameters. If COS is ON, a write lock is converted to COS lock
2401  * before saving.
2402  *
2403  * \param info thead info object
2404  * \param h lock handle
2405  * \param mode lock mode
2406  * \param decref force immediate lock releasing
2407  */
2408 static
2409 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2410                    ldlm_mode_t mode, int decref)
2411 {
2412         ENTRY;
2413
2414         if (lustre_handle_is_used(h)) {
2415                 if (decref || !info->mti_has_trans ||
2416                     !(mode & (LCK_PW | LCK_EX))){
2417                         mdt_fid_unlock(h, mode);
2418                 } else {
2419                         struct mdt_device *mdt = info->mti_mdt;
2420                         struct ldlm_lock *lock = ldlm_handle2lock(h);
2421                         struct ptlrpc_request *req = mdt_info_req(info);
2422                         int no_ack = 0;
2423
2424                         LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2425                                  h->cookie);
2426                         /* there is no request if mdt_object_unlock() is called
2427                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
2428                         if (likely(req != NULL)) {
2429                                 CDEBUG(D_HA, "request = %p reply state = %p"
2430                                        " transno = "LPD64"\n", req,
2431                                        req->rq_reply_state, req->rq_transno);
2432                                 if (mdt_cos_is_enabled(mdt)) {
2433                                         no_ack = 1;
2434                                         ldlm_lock_downgrade(lock, LCK_COS);
2435                                         mode = LCK_COS;
2436                                 }
2437                                 ptlrpc_save_lock(req, h, mode, no_ack);
2438                         } else {
2439                                 ldlm_lock_decref(h, mode);
2440                         }
2441                         if (mdt_is_lock_sync(lock)) {
2442                                 CDEBUG(D_HA, "found sync-lock,"
2443                                        " async commit started\n");
2444                                 mdt_device_commit_async(info->mti_env,
2445                                                         mdt);
2446                         }
2447                         LDLM_LOCK_PUT(lock);
2448                 }
2449                 h->cookie = 0ull;
2450         }
2451
2452         EXIT;
2453 }
2454
2455 /**
2456  * Unlock mdt object.
2457  *
2458  * Immeditely release the regular lock and the PDO lock or save the
2459  * lock in reqeuest and keep them referenced until client ACK or
2460  * transaction commit.
2461  *
2462  * \param info thread info object
2463  * \param o mdt object
2464  * \param lh mdt lock handle referencing regular and PDO locks
2465  * \param decref force immediate lock releasing
2466  */
2467 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2468                        struct mdt_lock_handle *lh, int decref)
2469 {
2470         ENTRY;
2471
2472         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2473         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2474
2475         if (lustre_handle_is_used(&lh->mlh_rreg_lh))
2476                 ldlm_lock_decref(&lh->mlh_rreg_lh, lh->mlh_rreg_mode);
2477
2478         EXIT;
2479 }
2480
2481 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2482                                         const struct lu_fid *f,
2483                                         struct mdt_lock_handle *lh,
2484                                         __u64 ibits)
2485 {
2486         struct mdt_object *o;
2487
2488         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2489         if (!IS_ERR(o)) {
2490                 int rc;
2491
2492                 rc = mdt_object_lock(info, o, lh, ibits,
2493                                      MDT_LOCAL_LOCK);
2494                 if (rc != 0) {
2495                         mdt_object_put(info->mti_env, o);
2496                         o = ERR_PTR(rc);
2497                 }
2498         }
2499         return o;
2500 }
2501
2502 void mdt_object_unlock_put(struct mdt_thread_info * info,
2503                            struct mdt_object * o,
2504                            struct mdt_lock_handle *lh,
2505                            int decref)
2506 {
2507         mdt_object_unlock(info, o, lh, decref);
2508         mdt_object_put(info->mti_env, o);
2509 }
2510
2511 /*
2512  * Generic code handling requests that have struct mdt_body passed in:
2513  *
2514  *  - extract mdt_body from request and save it in @info, if present;
2515  *
2516  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
2517  *  @info;
2518  *
2519  *  - if HABEO_CORPUS flag is set for this request type check whether object
2520  *  actually exists on storage (lu_object_exists()).
2521  *
2522  */
2523 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2524 {
2525         const struct mdt_body    *body;
2526         struct mdt_object        *obj;
2527         const struct lu_env      *env;
2528         struct req_capsule       *pill;
2529         int                       rc;
2530         ENTRY;
2531
2532         env = info->mti_env;
2533         pill = info->mti_pill;
2534
2535         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2536         if (body == NULL)
2537                 RETURN(-EFAULT);
2538
2539         if (!(body->valid & OBD_MD_FLID))
2540                 RETURN(0);
2541
2542         if (!fid_is_sane(&body->fid1)) {
2543                 CERROR("Invalid fid: "DFID"\n", PFID(&body->fid1));
2544                 RETURN(-EINVAL);
2545         }
2546
2547         /*
2548          * Do not get size or any capa fields before we check that request
2549          * contains capa actually. There are some requests which do not, for
2550          * instance MDS_IS_SUBDIR.
2551          */
2552         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2553             req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
2554                 mdt_set_capainfo(info, 0, &body->fid1,
2555                                  req_capsule_client_get(pill, &RMF_CAPA1));
2556
2557         obj = mdt_object_find(env, info->mti_mdt, &body->fid1);
2558         if (!IS_ERR(obj)) {
2559                 if ((flags & HABEO_CORPUS) &&
2560                     !mdt_object_exists(obj)) {
2561                         mdt_object_put(env, obj);
2562                         /* for capability renew ENOENT will be handled in
2563                          * mdt_renew_capa */
2564                         if (body->valid & OBD_MD_FLOSSCAPA)
2565                                 rc = 0;
2566                         else
2567                                 rc = -ENOENT;
2568                 } else {
2569                         info->mti_object = obj;
2570                         rc = 0;
2571                 }
2572         } else
2573                 rc = PTR_ERR(obj);
2574
2575         RETURN(rc);
2576 }
2577
2578 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2579 {
2580         struct req_capsule *pill = info->mti_pill;
2581         int rc;
2582         ENTRY;
2583
2584         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2585                 rc = mdt_body_unpack(info, flags);
2586         else
2587                 rc = 0;
2588
2589         if (rc == 0 && (flags & HABEO_REFERO)) {
2590                 /* Pack reply. */
2591                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2592                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2593                                              info->mti_body->eadatasize);
2594                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2595                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
2596                                              RCL_SERVER, 0);
2597
2598                 rc = req_capsule_server_pack(pill);
2599         }
2600         RETURN(rc);
2601 }
2602
2603 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2604 {
2605         struct md_device *next = m->mdt_child;
2606
2607         return next->md_ops->mdo_init_capa_ctxt(env, next,
2608                                                 m->mdt_lut.lut_mds_capa,
2609                                                 m->mdt_capa_timeout,
2610                                                 m->mdt_capa_alg,
2611                                                 m->mdt_capa_keys);
2612 }
2613
2614 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2615 {
2616         lh->mlh_type = MDT_NUL_LOCK;
2617         lh->mlh_reg_lh.cookie = 0ull;
2618         lh->mlh_reg_mode = LCK_MINMODE;
2619         lh->mlh_pdo_lh.cookie = 0ull;
2620         lh->mlh_pdo_mode = LCK_MINMODE;
2621         lh->mlh_rreg_lh.cookie = 0ull;
2622         lh->mlh_rreg_mode = LCK_MINMODE;
2623 }
2624
2625 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2626 {
2627         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2628         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2629 }
2630
2631 /*
2632  * Initialize fields of struct mdt_thread_info. Other fields are left in
2633  * uninitialized state, because it's too expensive to zero out whole
2634  * mdt_thread_info (> 1K) on each request arrival.
2635  */
2636 void mdt_thread_info_init(struct ptlrpc_request *req,
2637                           struct mdt_thread_info *info)
2638 {
2639         int i;
2640
2641         info->mti_pill = &req->rq_pill;
2642
2643         /* lock handle */
2644         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2645                 mdt_lock_handle_init(&info->mti_lh[i]);
2646
2647         /* mdt device: it can be NULL while CONNECT */
2648         if (req->rq_export) {
2649                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2650                 info->mti_exp = req->rq_export;
2651         } else
2652                 info->mti_mdt = NULL;
2653         info->mti_env = req->rq_svc_thread->t_env;
2654         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2655
2656         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2657         info->mti_big_buf = LU_BUF_NULL;
2658         info->mti_body = NULL;
2659         info->mti_object = NULL;
2660         info->mti_dlm_req = NULL;
2661         info->mti_has_trans = 0;
2662         info->mti_cross_ref = 0;
2663         info->mti_opdata = 0;
2664         info->mti_big_lmm_used = 0;
2665
2666         /* To not check for split by default. */
2667         info->mti_spec.no_create = 0;
2668         info->mti_spec.sp_rm_entry = 0;
2669 }
2670
2671 void mdt_thread_info_fini(struct mdt_thread_info *info)
2672 {
2673         int i;
2674
2675         if (info->mti_object != NULL) {
2676                 mdt_object_put(info->mti_env, info->mti_object);
2677                 info->mti_object = NULL;
2678         }
2679
2680         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2681                 mdt_lock_handle_fini(&info->mti_lh[i]);
2682         info->mti_env = NULL;
2683         info->mti_pill = NULL;
2684         info->mti_exp = NULL;
2685
2686         if (unlikely(info->mti_big_buf.lb_buf != NULL))
2687                 lu_buf_free(&info->mti_big_buf);
2688 }
2689
2690 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
2691 {
2692         struct mdt_thread_info  *mti;
2693         struct lustre_capa      *lc;
2694
2695         mti = mdt_th_info(tsi->tsi_env);
2696         LASSERT(mti != NULL);
2697
2698         mdt_thread_info_init(tgt_ses_req(tsi), mti);
2699         if (tsi->tsi_corpus != NULL) {
2700                 struct req_capsule *pill = tsi->tsi_pill;
2701
2702                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
2703                 lu_object_get(tsi->tsi_corpus);
2704
2705                 /*
2706                  * XXX: must be part of tgt_mdt_body_unpack but moved here
2707                  * due to mdt_set_capainfo().
2708                  */
2709                 if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2710                     req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT) > 0) {
2711                         lc = req_capsule_client_get(pill, &RMF_CAPA1);
2712                         mdt_set_capainfo(mti, 0, &tsi->tsi_mdt_body->fid1, lc);
2713                 }
2714         }
2715         mti->mti_body = tsi->tsi_mdt_body;
2716         mti->mti_dlm_req = tsi->tsi_dlm_req;
2717
2718         return mti;
2719 }
2720
2721 int mdt_tgt_connect(struct tgt_session_info *tsi)
2722 {
2723         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2724         int                      rc;
2725
2726         ENTRY;
2727
2728         rc = tgt_connect(tsi);
2729         if (rc != 0)
2730                 RETURN(rc);
2731
2732         rc = mdt_init_idmap(tsi);
2733         if (rc != 0)
2734                 GOTO(err, rc);
2735         RETURN(0);
2736 err:
2737         obd_disconnect(class_export_get(req->rq_export));
2738         return rc;
2739 }
2740
2741 enum mdt_it_code {
2742         MDT_IT_OPEN,
2743         MDT_IT_OCREAT,
2744         MDT_IT_CREATE,
2745         MDT_IT_GETATTR,
2746         MDT_IT_READDIR,
2747         MDT_IT_LOOKUP,
2748         MDT_IT_UNLINK,
2749         MDT_IT_TRUNC,
2750         MDT_IT_GETXATTR,
2751         MDT_IT_LAYOUT,
2752         MDT_IT_QUOTA,
2753         MDT_IT_NR
2754 };
2755
2756 static int mdt_intent_getattr(enum mdt_it_code opcode,
2757                               struct mdt_thread_info *info,
2758                               struct ldlm_lock **,
2759                               __u64);
2760
2761 static int mdt_intent_getxattr(enum mdt_it_code opcode,
2762                                 struct mdt_thread_info *info,
2763                                 struct ldlm_lock **lockp,
2764                                 __u64 flags);
2765
2766 static int mdt_intent_layout(enum mdt_it_code opcode,
2767                              struct mdt_thread_info *info,
2768                              struct ldlm_lock **,
2769                              __u64);
2770 static int mdt_intent_reint(enum mdt_it_code opcode,
2771                             struct mdt_thread_info *info,
2772                             struct ldlm_lock **,
2773                             __u64);
2774
2775 static struct mdt_it_flavor {
2776         const struct req_format *it_fmt;
2777         __u32                    it_flags;
2778         int                    (*it_act)(enum mdt_it_code ,
2779                                          struct mdt_thread_info *,
2780                                          struct ldlm_lock **,
2781                                          __u64);
2782         long                     it_reint;
2783 } mdt_it_flavor[] = {
2784         [MDT_IT_OPEN]     = {
2785                 .it_fmt   = &RQF_LDLM_INTENT,
2786                 /*.it_flags = HABEO_REFERO,*/
2787                 .it_flags = 0,
2788                 .it_act   = mdt_intent_reint,
2789                 .it_reint = REINT_OPEN
2790         },
2791         [MDT_IT_OCREAT]   = {
2792                 .it_fmt   = &RQF_LDLM_INTENT,
2793                 /*
2794                  * OCREAT is not a MUTABOR request as if the file
2795                  * already exists.
2796                  * We do the extra check of OBD_CONNECT_RDONLY in
2797                  * mdt_reint_open() when we really need to create
2798                  * the object.
2799                  */
2800                 .it_flags = 0,
2801                 .it_act   = mdt_intent_reint,
2802                 .it_reint = REINT_OPEN
2803         },
2804         [MDT_IT_CREATE]   = {
2805                 .it_fmt   = &RQF_LDLM_INTENT,
2806                 .it_flags = MUTABOR,
2807                 .it_act   = mdt_intent_reint,
2808                 .it_reint = REINT_CREATE
2809         },
2810         [MDT_IT_GETATTR]  = {
2811                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2812                 .it_flags = HABEO_REFERO,
2813                 .it_act   = mdt_intent_getattr
2814         },
2815         [MDT_IT_READDIR]  = {
2816                 .it_fmt   = NULL,
2817                 .it_flags = 0,
2818                 .it_act   = NULL
2819         },
2820         [MDT_IT_LOOKUP]   = {
2821                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2822                 .it_flags = HABEO_REFERO,
2823                 .it_act   = mdt_intent_getattr
2824         },
2825         [MDT_IT_UNLINK]   = {
2826                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
2827                 .it_flags = MUTABOR,
2828                 .it_act   = NULL,
2829                 .it_reint = REINT_UNLINK
2830         },
2831         [MDT_IT_TRUNC]    = {
2832                 .it_fmt   = NULL,
2833                 .it_flags = MUTABOR,
2834                 .it_act   = NULL
2835         },
2836         [MDT_IT_GETXATTR] = {
2837                 .it_fmt   = &RQF_LDLM_INTENT_GETXATTR,
2838                 .it_flags = HABEO_CORPUS,
2839                 .it_act   = mdt_intent_getxattr
2840         },
2841         [MDT_IT_LAYOUT] = {
2842                 .it_fmt   = &RQF_LDLM_INTENT_LAYOUT,
2843                 .it_flags = 0,
2844                 .it_act   = mdt_intent_layout
2845         }
2846 };
2847
2848 int mdt_intent_lock_replace(struct mdt_thread_info *info,
2849                             struct ldlm_lock **lockp,
2850                             struct ldlm_lock *new_lock,
2851                             struct mdt_lock_handle *lh,
2852                             __u64 flags)
2853 {
2854         struct ptlrpc_request  *req = mdt_info_req(info);
2855         struct ldlm_lock       *lock = *lockp;
2856
2857         /*
2858          * Get new lock only for cases when possible resent did not find any
2859          * lock.
2860          */
2861         if (new_lock == NULL)
2862                 new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
2863
2864         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
2865                 lh->mlh_reg_lh.cookie = 0;
2866                 RETURN(0);
2867         }
2868
2869         LASSERTF(new_lock != NULL,
2870                  "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
2871
2872         /*
2873          * If we've already given this lock to a client once, then we should
2874          * have no readers or writers.  Otherwise, we should have one reader
2875          * _or_ writer ref (which will be zeroed below) before returning the
2876          * lock to a client.
2877          */
2878         if (new_lock->l_export == req->rq_export) {
2879                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
2880         } else {
2881                 LASSERT(new_lock->l_export == NULL);
2882                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
2883         }
2884
2885         *lockp = new_lock;
2886
2887         if (new_lock->l_export == req->rq_export) {
2888                 /*
2889                  * Already gave this to the client, which means that we
2890                  * reconstructed a reply.
2891                  */
2892                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
2893                         MSG_RESENT);
2894                 lh->mlh_reg_lh.cookie = 0;
2895                 RETURN(ELDLM_LOCK_REPLACED);
2896         }
2897
2898         /*
2899          * Fixup the lock to be given to the client.
2900          */
2901         lock_res_and_lock(new_lock);
2902         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
2903          * possible blocking AST. */
2904         while (new_lock->l_readers > 0) {
2905                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
2906                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
2907                 new_lock->l_readers--;
2908         }
2909         while (new_lock->l_writers > 0) {
2910                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
2911                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
2912                 new_lock->l_writers--;
2913         }
2914
2915         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
2916         new_lock->l_blocking_ast = lock->l_blocking_ast;
2917         new_lock->l_completion_ast = lock->l_completion_ast;
2918         new_lock->l_remote_handle = lock->l_remote_handle;
2919         new_lock->l_flags &= ~LDLM_FL_LOCAL;
2920
2921         unlock_res_and_lock(new_lock);
2922
2923         cfs_hash_add(new_lock->l_export->exp_lock_hash,
2924                      &new_lock->l_remote_handle,
2925                      &new_lock->l_exp_hash);
2926
2927         LDLM_LOCK_RELEASE(new_lock);
2928         lh->mlh_reg_lh.cookie = 0;
2929
2930         RETURN(ELDLM_LOCK_REPLACED);
2931 }
2932
2933 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
2934                                     struct ldlm_lock *new_lock,
2935                                     struct ldlm_lock **old_lock,
2936                                     struct mdt_lock_handle *lh)
2937 {
2938         struct ptlrpc_request  *req = mdt_info_req(info);
2939         struct obd_export      *exp = req->rq_export;
2940         struct lustre_handle    remote_hdl;
2941         struct ldlm_request    *dlmreq;
2942         struct ldlm_lock       *lock;
2943
2944         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
2945                 return;
2946
2947         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
2948         remote_hdl = dlmreq->lock_handle[0];
2949
2950         /* In the function below, .hs_keycmp resolves to
2951          * ldlm_export_lock_keycmp() */
2952         /* coverity[overrun-buffer-val] */
2953         lock = cfs_hash_lookup(exp->exp_lock_hash, &remote_hdl);
2954         if (lock) {
2955                 if (lock != new_lock) {
2956                         lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie;
2957                         lh->mlh_reg_mode = lock->l_granted_mode;
2958
2959                         LDLM_DEBUG(lock, "Restoring lock cookie");
2960                         DEBUG_REQ(D_DLMTRACE, req,
2961                                   "restoring lock cookie "LPX64,
2962                                   lh->mlh_reg_lh.cookie);
2963                         if (old_lock)
2964                                 *old_lock = LDLM_LOCK_GET(lock);
2965                         cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
2966                         return;
2967                 }
2968
2969                 cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
2970         }
2971
2972         /*
2973          * If the xid matches, then we know this is a resent request, and allow
2974          * it. (It's probably an OPEN, for which we don't send a lock.
2975          */
2976         if (req_xid_is_last(req))
2977                 return;
2978
2979         /*
2980          * This remote handle isn't enqueued, so we never received or processed
2981          * this request.  Clear MSG_RESENT, because it can be handled like any
2982          * normal request now.
2983          */
2984         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
2985
2986         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
2987                   remote_hdl.cookie);
2988 }
2989
2990 static int mdt_intent_getxattr(enum mdt_it_code opcode,
2991                                 struct mdt_thread_info *info,
2992                                 struct ldlm_lock **lockp,
2993                                 __u64 flags)
2994 {
2995         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
2996         struct ldlm_reply      *ldlm_rep = NULL;
2997         int rc, grc;
2998
2999         /*
3000          * Initialize lhc->mlh_reg_lh either from a previously granted lock
3001          * (for the resend case) or a new lock. Below we will use it to
3002          * replace the original lock.
3003          */
3004         mdt_intent_fixup_resent(info, *lockp, NULL, lhc);
3005         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3006                 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3007                 rc = mdt_object_lock(info, info->mti_object, lhc,
3008                                         MDS_INODELOCK_XATTR,
3009                                         MDT_LOCAL_LOCK);
3010                 if (rc)
3011                         return rc;
3012         }
3013
3014         grc = mdt_getxattr(info);
3015
3016         rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3017
3018         if (mdt_info_req(info)->rq_repmsg != NULL)
3019                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3020         if (ldlm_rep == NULL)
3021                 RETURN(err_serious(-EFAULT));
3022
3023         ldlm_rep->lock_policy_res2 = grc;
3024
3025         return rc;
3026 }
3027
3028 static int mdt_intent_getattr(enum mdt_it_code opcode,
3029                               struct mdt_thread_info *info,
3030                               struct ldlm_lock **lockp,
3031                               __u64 flags)
3032 {
3033         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3034         struct ldlm_lock       *new_lock = NULL;
3035         __u64                   child_bits;
3036         struct ldlm_reply      *ldlm_rep;
3037         struct ptlrpc_request  *req;
3038         struct mdt_body        *reqbody;
3039         struct mdt_body        *repbody;
3040         int                     rc, rc2;
3041         ENTRY;
3042
3043         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3044         LASSERT(reqbody);
3045
3046         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3047         LASSERT(repbody);
3048
3049         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
3050         repbody->eadatasize = 0;
3051         repbody->aclsize = 0;
3052
3053         switch (opcode) {
3054         case MDT_IT_LOOKUP:
3055                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3056                 break;
3057         case MDT_IT_GETATTR:
3058                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3059                              MDS_INODELOCK_PERM;
3060                 break;
3061         default:
3062                 CERROR("Unsupported intent (%d)\n", opcode);
3063                 GOTO(out_shrink, rc = -EINVAL);
3064         }
3065
3066         rc = mdt_init_ucred(info, reqbody);
3067         if (rc)
3068                 GOTO(out_shrink, rc);
3069
3070         req = info->mti_pill->rc_req;
3071         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3072         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3073
3074         /* Get lock from request for possible resent case. */
3075         mdt_intent_fixup_resent(info, *lockp, &new_lock, lhc);
3076
3077         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3078         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3079
3080         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3081                 ldlm_rep->lock_policy_res2 = 0;
3082         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3083             ldlm_rep->lock_policy_res2) {
3084                 lhc->mlh_reg_lh.cookie = 0ull;
3085                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3086         }
3087
3088         rc = mdt_intent_lock_replace(info, lockp, new_lock, lhc, flags);
3089         EXIT;
3090 out_ucred:
3091         mdt_exit_ucred(info);
3092 out_shrink:
3093         mdt_client_compatibility(info);
3094         rc2 = mdt_fix_reply(info);
3095         if (rc == 0)
3096                 rc = rc2;
3097         return rc;
3098 }
3099
3100 static int mdt_intent_layout(enum mdt_it_code opcode,
3101                              struct mdt_thread_info *info,
3102                              struct ldlm_lock **lockp,
3103                              __u64 flags)
3104 {
3105         struct layout_intent *layout;
3106         struct lu_fid *fid;
3107         struct mdt_object *obj = NULL;
3108         struct md_object *child = NULL;
3109         int rc;
3110         ENTRY;
3111
3112         if (opcode != MDT_IT_LAYOUT) {
3113                 CERROR("%s: Unknown intent (%d)\n", mdt_obd_name(info->mti_mdt),
3114                         opcode);
3115                 RETURN(-EINVAL);
3116         }
3117
3118         fid = &info->mti_tmp_fid2;
3119         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3120
3121         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3122         if (IS_ERR(obj))
3123                 RETURN(PTR_ERR(obj));
3124
3125         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3126                 child = mdt_object_child(obj);
3127
3128                 /* get the length of lsm */
3129                 rc = mo_xattr_get(info->mti_env, child, &LU_BUF_NULL,
3130                                   XATTR_NAME_LOV);
3131
3132                 if (rc > info->mti_mdt->mdt_max_mdsize)
3133                         info->mti_mdt->mdt_max_mdsize = rc;
3134         }
3135
3136         mdt_object_put(info->mti_env, obj);
3137
3138         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3139         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
3140                         ldlm_lvbo_size(*lockp));
3141         rc = req_capsule_server_pack(info->mti_pill);
3142         if (rc != 0)
3143                 RETURN(-EINVAL);
3144
3145         layout = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3146         LASSERT(layout != NULL);
3147         if (layout->li_opc == LAYOUT_INTENT_ACCESS)
3148                 /* return to normal ldlm handling */
3149                 RETURN(0);
3150
3151         CERROR("%s: Unsupported layout intent (%d)\n",
3152                 mdt_obd_name(info->mti_mdt), layout->li_opc);
3153         RETURN(-EINVAL);
3154 }
3155
3156 static int mdt_intent_reint(enum mdt_it_code opcode,
3157                             struct mdt_thread_info *info,
3158                             struct ldlm_lock **lockp,
3159                             __u64 flags)
3160 {
3161         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3162         struct ldlm_reply      *rep = NULL;
3163         long                    opc;
3164         int                     rc;
3165
3166         static const struct req_format *intent_fmts[REINT_MAX] = {
3167                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3168                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3169         };
3170
3171         ENTRY;
3172
3173         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
3174         if (opc < 0)
3175                 RETURN(opc);
3176
3177         if (mdt_it_flavor[opcode].it_reint != opc) {
3178                 CERROR("Reint code %ld doesn't match intent: %d\n",
3179                        opc, opcode);
3180                 RETURN(err_serious(-EPROTO));
3181         }
3182
3183         /* Get lock from request for possible resent case. */
3184         mdt_intent_fixup_resent(info, *lockp, NULL, lhc);
3185
3186         rc = mdt_reint_internal(info, lhc, opc);
3187
3188         /* Check whether the reply has been packed successfully. */
3189         if (mdt_info_req(info)->rq_repmsg != NULL)
3190                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3191         if (rep == NULL)
3192                 RETURN(err_serious(-EFAULT));
3193
3194         /* MDC expects this in any case */
3195         if (rc != 0)
3196                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3197
3198         /* the open lock or the lock for cross-ref object should be
3199          * returned to the client */
3200         if (rc == -EREMOTE || mdt_get_disposition(rep, DISP_OPEN_LOCK)) {
3201                 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
3202                 rep->lock_policy_res2 = 0;
3203                 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3204                 RETURN(rc);
3205         }
3206
3207         rep->lock_policy_res2 = clear_serious(rc);
3208
3209         if (rep->lock_policy_res2 == -ENOENT &&
3210             mdt_get_disposition(rep, DISP_LOOKUP_NEG))
3211                 rep->lock_policy_res2 = 0;
3212
3213         if (rc == -ENOTCONN || rc == -ENODEV ||
3214             rc == -EOVERFLOW) { /**< if VBR failure then return error */
3215                 /*
3216                  * If it is the disconnect error (ENODEV & ENOCONN), the error
3217                  * will be returned by rq_status, and client at ptlrpc layer
3218                  * will detect this, then disconnect, reconnect the import
3219                  * immediately, instead of impacting the following the rpc.
3220                  */
3221                 lhc->mlh_reg_lh.cookie = 0ull;
3222                 RETURN(rc);
3223         } else {
3224                 /*
3225                  * For other cases, the error will be returned by intent.
3226                  * and client will retrieve the result from intent.
3227                  */
3228                  /*
3229                   * FIXME: when open lock is finished, that should be
3230                   * checked here.
3231                   */
3232                 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3233                         LASSERTF(rc == 0, "Error occurred but lock handle "
3234                                  "is still in use, rc = %d\n", rc);
3235                         rep->lock_policy_res2 = 0;
3236                         rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3237                         RETURN(rc);
3238                 } else {
3239                         lhc->mlh_reg_lh.cookie = 0ull;
3240                         RETURN(ELDLM_LOCK_ABORTED);
3241                 }
3242         }
3243 }
3244
3245 static int mdt_intent_code(long itcode)
3246 {
3247         int rc;
3248
3249         switch(itcode) {
3250         case IT_OPEN:
3251                 rc = MDT_IT_OPEN;
3252                 break;
3253         case IT_OPEN|IT_CREAT:
3254                 rc = MDT_IT_OCREAT;
3255                 break;
3256         case IT_CREAT:
3257                 rc = MDT_IT_CREATE;
3258                 break;
3259         case IT_READDIR:
3260                 rc = MDT_IT_READDIR;
3261                 break;
3262         case IT_GETATTR:
3263                 rc = MDT_IT_GETATTR;
3264                 break;
3265         case IT_LOOKUP:
3266                 rc = MDT_IT_LOOKUP;
3267                 break;
3268         case IT_UNLINK:
3269                 rc = MDT_IT_UNLINK;
3270                 break;
3271         case IT_TRUNC:
3272                 rc = MDT_IT_TRUNC;
3273                 break;
3274         case IT_GETXATTR:
3275                 rc = MDT_IT_GETXATTR;
3276                 break;
3277         case IT_LAYOUT:
3278                 rc = MDT_IT_LAYOUT;
3279                 break;
3280         case IT_QUOTA_DQACQ:
3281         case IT_QUOTA_CONN:
3282                 rc = MDT_IT_QUOTA;
3283                 break;
3284         default:
3285                 CERROR("Unknown intent opcode: %ld\n", itcode);
3286                 rc = -EINVAL;
3287                 break;
3288         }
3289         return rc;
3290 }
3291
3292 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
3293                           struct ldlm_lock **lockp, __u64 flags)
3294 {
3295         struct req_capsule   *pill;
3296         struct mdt_it_flavor *flv;
3297         int opc;
3298         int rc;
3299         ENTRY;
3300
3301         opc = mdt_intent_code(itopc);
3302         if (opc < 0)
3303                 RETURN(-EINVAL);
3304
3305         pill = info->mti_pill;
3306
3307         if (opc == MDT_IT_QUOTA) {
3308                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
3309
3310                 if (qmt == NULL)
3311                         RETURN(-EOPNOTSUPP);
3312
3313                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
3314                 /* pass the request to quota master */
3315                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
3316                                                  mdt_info_req(info), lockp,
3317                                                  flags);
3318                 RETURN(rc);
3319         }
3320
3321         flv  = &mdt_it_flavor[opc];
3322         if (flv->it_fmt != NULL)
3323                 req_capsule_extend(pill, flv->it_fmt);
3324
3325         rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3326         if (rc == 0) {
3327                 struct ptlrpc_request *req = mdt_info_req(info);
3328                 if (flv->it_flags & MUTABOR &&
3329                     exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
3330                         RETURN(-EROFS);
3331         }
3332         if (rc == 0 && flv->it_act != NULL) {
3333                 struct ldlm_reply *rep;
3334
3335                 /* execute policy */
3336                 rc = flv->it_act(opc, info, lockp, flags);
3337
3338                 /* Check whether the reply has been packed successfully. */
3339                 if (mdt_info_req(info)->rq_repmsg != NULL) {
3340                         rep = req_capsule_server_get(info->mti_pill,
3341                                                      &RMF_DLM_REP);
3342                         rep->lock_policy_res2 =
3343                                 ptlrpc_status_hton(rep->lock_policy_res2);
3344                 }
3345         } else {
3346                 rc = -EPROTO;
3347         }
3348         RETURN(rc);
3349 }
3350
3351 static int mdt_intent_policy(struct ldlm_namespace *ns,
3352                              struct ldlm_lock **lockp, void *req_cookie,
3353                              ldlm_mode_t mode, __u64 flags, void *data)
3354 {
3355         struct tgt_session_info *tsi;
3356         struct mdt_thread_info  *info;
3357         struct ptlrpc_request   *req  =  req_cookie;
3358         struct ldlm_intent      *it;
3359         struct req_capsule      *pill;
3360         int rc;
3361
3362         ENTRY;
3363
3364         LASSERT(req != NULL);
3365
3366         tsi = tgt_ses_info(req->rq_svc_thread->t_env);
3367
3368         info = tsi2mdt_info(tsi);
3369         LASSERT(info != NULL);
3370         pill = info->mti_pill;
3371         LASSERT(pill->rc_req == req);
3372
3373         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3374                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
3375                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3376                 if (it != NULL) {
3377                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
3378                         if (rc == 0)
3379                                 rc = ELDLM_OK;
3380
3381                         /* Lock without inodebits makes no sense and will oops
3382                          * later in ldlm. Let's check it now to see if we have
3383                          * ibits corrupted somewhere in mdt_intent_opc().
3384                          * The case for client miss to set ibits has been
3385                          * processed by others. */
3386                         LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
3387                                         lr_type == LDLM_IBITS,
3388                                      info->mti_dlm_req->lock_desc.\
3389                                         l_policy_data.l_inodebits.bits != 0));
3390                 } else
3391                         rc = err_serious(-EFAULT);
3392         } else {
3393                 /* No intent was provided */
3394                 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3395                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
3396                 rc = req_capsule_server_pack(pill);
3397                 if (rc)
3398                         rc = err_serious(rc);
3399         }
3400         mdt_thread_info_fini(info);
3401         RETURN(rc);
3402 }
3403
3404 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
3405 {
3406         struct seq_server_site  *ss = mdt_seq_site(mdt);
3407
3408         if (ss->ss_node_id == 0)
3409                 return;
3410
3411         if (ss->ss_client_seq != NULL) {
3412                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3413                 ss->ss_client_seq->lcs_exp = NULL;
3414         }
3415
3416         if (ss->ss_server_fld != NULL) {
3417                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
3418                 ss->ss_server_fld->lsf_control_exp = NULL;
3419         }
3420 }
3421
3422 static void mdt_seq_fini_cli(struct mdt_device *mdt)
3423 {
3424         struct seq_server_site *ss = mdt_seq_site(mdt);
3425
3426         if (ss == NULL)
3427                 return;
3428
3429         if (ss->ss_server_seq == NULL)
3430                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
3431
3432         return;
3433 }
3434
3435 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
3436 {
3437         mdt_seq_fini_cli(mdt);
3438         mdt_deregister_seq_exp(mdt);
3439
3440         return seq_site_fini(env, mdt_seq_site(mdt));
3441 }
3442
3443 /**
3444  * It will retrieve its FLDB entries from MDT0, and it only happens
3445  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
3446  * and it needs to refresh FLDB from the MDT0.
3447  **/
3448 static int mdt_register_lwp_callback(void *data)
3449 {
3450         struct lu_env           env;
3451         struct mdt_device       *mdt = data;
3452         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
3453         int                     rc;
3454         ENTRY;
3455
3456         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
3457
3458         if (!likely(fld->lsf_new))
3459                 RETURN(0);
3460
3461         rc = lu_env_init(&env, LCT_MD_THREAD);
3462         if (rc) {
3463                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
3464                 RETURN(rc);
3465         }
3466
3467         rc = fld_update_from_controller(&env, fld);
3468         if (rc != 0) {
3469                 CERROR("%s: cannot update controller: rc = %d\n",
3470                        mdt_obd_name(mdt), rc);
3471                 GOTO(out, rc);
3472         }
3473 out:
3474         lu_env_fini(&env);
3475         RETURN(rc);
3476 }
3477
3478 static int mdt_register_seq_exp(struct mdt_device *mdt)
3479 {
3480         struct seq_server_site  *ss = mdt_seq_site(mdt);
3481         char                    *lwp_name = NULL;
3482         int                     rc;
3483
3484         if (ss->ss_node_id == 0)
3485                 return 0;
3486
3487         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
3488         if (lwp_name == NULL)
3489                 GOTO(out_free, rc = -ENOMEM);
3490
3491         rc = tgt_name2lwpname(mdt_obd_name(mdt), lwp_name);
3492         if (rc != 0)
3493                 GOTO(out_free, rc);
3494
3495         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
3496                                       NULL, NULL);
3497         if (rc != 0)
3498                 GOTO(out_free, rc);
3499
3500         rc = lustre_register_lwp_item(lwp_name,
3501                                       &ss->ss_server_fld->lsf_control_exp,
3502                                       mdt_register_lwp_callback, mdt);
3503         if (rc != 0) {
3504                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3505                 ss->ss_client_seq->lcs_exp = NULL;
3506                 GOTO(out_free, rc);
3507         }
3508 out_free:
3509         if (lwp_name != NULL)
3510                 OBD_FREE(lwp_name, MAX_OBD_NAME);
3511
3512         return rc;
3513 }
3514
3515 /*
3516  * Init client sequence manager which is used by local MDS to talk to sequence
3517  * controller on remote node.
3518  */
3519 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
3520 {
3521         struct seq_server_site  *ss = mdt_seq_site(mdt);
3522         int                     rc;
3523         char                    *prefix;
3524         ENTRY;
3525
3526         /* check if this is adding the first MDC and controller is not yet
3527          * initialized. */
3528         OBD_ALLOC_PTR(ss->ss_client_seq);
3529         if (ss->ss_client_seq == NULL)
3530                 RETURN(-ENOMEM);
3531
3532         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3533         if (prefix == NULL) {
3534                 OBD_FREE_PTR(ss->ss_client_seq);
3535                 ss->ss_client_seq = NULL;
3536                 RETURN(-ENOMEM);
3537         }
3538
3539         /* Note: seq_client_fini will be called in seq_site_fini */
3540         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
3541         rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
3542                              prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
3543                                                             NULL);
3544         OBD_FREE(prefix, MAX_OBD_NAME + 5);
3545         if (rc != 0) {
3546                 OBD_FREE_PTR(ss->ss_client_seq);
3547                 ss->ss_client_seq = NULL;
3548                 RETURN(rc);
3549         }
3550
3551         rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
3552
3553         RETURN(rc);
3554 }
3555
3556 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
3557 {
3558         struct seq_server_site  *ss;
3559         int                     rc;
3560         ENTRY;
3561
3562         ss = mdt_seq_site(mdt);
3563         /* init sequence controller server(MDT0) */
3564         if (ss->ss_node_id == 0) {
3565                 OBD_ALLOC_PTR(ss->ss_control_seq);
3566                 if (ss->ss_control_seq == NULL)
3567                         RETURN(-ENOMEM);
3568
3569                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
3570                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
3571                                      ss);
3572                 if (rc)
3573                         GOTO(out_seq_fini, rc);
3574         }
3575
3576         /* Init normal sequence server */
3577         OBD_ALLOC_PTR(ss->ss_server_seq);
3578         if (ss->ss_server_seq == NULL)
3579                 GOTO(out_seq_fini, rc = -ENOMEM);
3580
3581         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
3582                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
3583         if (rc)
3584                 GOTO(out_seq_fini, rc);
3585
3586         /* init seq client for seq server to talk to seq controller(MDT0) */
3587         rc = mdt_seq_init_cli(env, mdt);
3588         if (rc != 0)
3589                 GOTO(out_seq_fini, rc);
3590
3591         if (ss->ss_node_id != 0)
3592                 /* register controler export through lwp */
3593                 rc = mdt_register_seq_exp(mdt);
3594
3595         EXIT;
3596 out_seq_fini:
3597         if (rc)
3598                 mdt_seq_fini(env, mdt);
3599
3600         return rc;
3601 }
3602
3603 /*
3604  * FLD wrappers
3605  */
3606 static int mdt_fld_fini(const struct lu_env *env,
3607                         struct mdt_device *m)
3608 {
3609         struct seq_server_site *ss = mdt_seq_site(m);
3610         ENTRY;
3611
3612         if (ss && ss->ss_server_fld) {
3613                 fld_server_fini(env, ss->ss_server_fld);
3614                 OBD_FREE_PTR(ss->ss_server_fld);
3615                 ss->ss_server_fld = NULL;
3616         }
3617
3618         RETURN(0);
3619 }
3620
3621 static int mdt_fld_init(const struct lu_env *env,
3622                         const char *uuid,
3623                         struct mdt_device *m)
3624 {
3625         struct seq_server_site *ss;
3626         int rc;
3627         ENTRY;
3628
3629         ss = mdt_seq_site(m);
3630
3631         OBD_ALLOC_PTR(ss->ss_server_fld);
3632         if (ss->ss_server_fld == NULL)
3633                 RETURN(rc = -ENOMEM);
3634
3635         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
3636                              LU_SEQ_RANGE_MDT);
3637         if (rc) {
3638                 OBD_FREE_PTR(ss->ss_server_fld);
3639                 ss->ss_server_fld = NULL;
3640                 RETURN(rc);
3641         }
3642
3643         RETURN(0);
3644 }
3645
3646 static void mdt_stack_pre_fini(const struct lu_env *env,
3647                            struct mdt_device *m, struct lu_device *top)
3648 {
3649         struct obd_device       *obd;
3650         struct lustre_cfg_bufs  *bufs;
3651         struct lustre_cfg       *lcfg;
3652         struct mdt_thread_info  *info;
3653         ENTRY;
3654
3655         LASSERT(top);
3656
3657         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3658         LASSERT(info != NULL);
3659
3660         bufs = &info->mti_u.bufs;
3661
3662         LASSERT(m->mdt_child_exp);
3663         LASSERT(m->mdt_child_exp->exp_obd);
3664         obd = m->mdt_child_exp->exp_obd;
3665
3666         /* process cleanup, pass mdt obd name to get obd umount flags */
3667         /* XXX: this is needed because all layers are referenced by
3668          * objects (some of them are pinned by osd, for example *
3669          * the proper solution should be a model where object used
3670          * by osd only doesn't have mdt/mdd slices -bzzz */
3671         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
3672         lustre_cfg_bufs_set_string(bufs, 1, NULL);
3673         lcfg = lustre_cfg_new(LCFG_PRE_CLEANUP, bufs);
3674         if (!lcfg) {
3675                 CERROR("%s:Cannot alloc lcfg!\n", mdt_obd_name(m));
3676                 return;
3677         }
3678         top->ld_ops->ldo_process_config(env, top, lcfg);
3679         lustre_cfg_free(lcfg);
3680         EXIT;
3681 }
3682
3683 static void mdt_stack_fini(const struct lu_env *env,
3684                            struct mdt_device *m, struct lu_device *top)
3685 {
3686         struct obd_device       *obd = mdt2obd_dev(m);
3687         struct lustre_cfg_bufs  *bufs;
3688         struct lustre_cfg       *lcfg;
3689         struct mdt_thread_info  *info;
3690         char                     flags[3] = "";
3691         ENTRY;
3692
3693         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3694         LASSERT(info != NULL);
3695
3696         lu_dev_del_linkage(top->ld_site, top);
3697
3698         lu_site_purge(env, top->ld_site, -1);
3699
3700         bufs = &info->mti_u.bufs;
3701         /* process cleanup, pass mdt obd name to get obd umount flags */
3702         /* another purpose is to let all layers to release their objects */
3703         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
3704         if (obd->obd_force)
3705                 strcat(flags, "F");
3706         if (obd->obd_fail)
3707                 strcat(flags, "A");
3708         lustre_cfg_bufs_set_string(bufs, 1, flags);
3709         lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
3710         if (!lcfg) {
3711                 CERROR("Cannot alloc lcfg!\n");
3712                 return;
3713         }
3714         LASSERT(top);
3715         top->ld_ops->ldo_process_config(env, top, lcfg);
3716         lustre_cfg_free(lcfg);
3717
3718         lu_site_purge(env, top->ld_site, -1);
3719
3720         m->mdt_child = NULL;
3721         m->mdt_bottom = NULL;
3722
3723         obd_disconnect(m->mdt_child_exp);
3724         m->mdt_child_exp = NULL;
3725
3726         obd_disconnect(m->mdt_bottom_exp);
3727         m->mdt_child_exp = NULL;
3728 }
3729
3730 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
3731                                const char *next, struct obd_export **exp)
3732 {
3733         struct obd_connect_data *data = NULL;
3734         struct obd_device       *obd;
3735         int                      rc;
3736         ENTRY;
3737
3738         OBD_ALLOC_PTR(data);
3739         if (data == NULL)
3740                 GOTO(out, rc = -ENOMEM);
3741
3742         obd = class_name2obd(next);
3743         if (obd == NULL) {
3744                 CERROR("%s: can't locate next device: %s\n",
3745                        mdt_obd_name(m), next);
3746                 GOTO(out, rc = -ENOTCONN);
3747         }
3748
3749         data->ocd_connect_flags = OBD_CONNECT_VERSION;
3750         data->ocd_version = LUSTRE_VERSION_CODE;
3751
3752         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
3753         if (rc) {
3754                 CERROR("%s: cannot connect to next dev %s (%d)\n",
3755                        mdt_obd_name(m), next, rc);
3756                 GOTO(out, rc);
3757         }
3758
3759 out:
3760         if (data)
3761                 OBD_FREE_PTR(data);
3762         RETURN(rc);
3763 }
3764
3765 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
3766                           struct lustre_cfg *cfg)
3767 {
3768         char                   *dev = lustre_cfg_string(cfg, 0);
3769         int                     rc, name_size, uuid_size;
3770         char                   *name, *uuid, *p;
3771         struct lustre_cfg_bufs *bufs;
3772         struct lustre_cfg      *lcfg;
3773         struct obd_device      *obd;
3774         struct lustre_profile  *lprof;
3775         struct lu_site         *site;
3776         ENTRY;
3777
3778         /* in 1.8 we had the only device in the stack - MDS.
3779          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
3780          * in 2.3 OSD is instantiated by obd_mount.c, so we need
3781          * to generate names and setup MDT, MDD. MDT will be using
3782          * generated name to connect to MDD. for MDD the next device
3783          * will be LOD with name taken from so called "profile" which
3784          * is generated by mount_option line
3785          *
3786          * 1.8 MGS generates config. commands like this:
3787          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
3788          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
3789          * 2.0 MGS generates config. commands like this:
3790          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
3791          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
3792          *                    3:lustre-MDT0000-mdtlov  4:f
3793          *
3794          * we generate MDD name from MDT one, just replacing T with D
3795          *
3796          * after all the preparations, the logical equivalent will be
3797          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
3798          *                    3:lustre-MDT0000-mdtlov  4:f
3799          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
3800          *                    3:lustre-MDD0000  4:f
3801          *
3802          *  notice we build the stack from down to top: MDD first, then MDT */
3803
3804         name_size = MAX_OBD_NAME;
3805         uuid_size = MAX_OBD_NAME;
3806
3807         OBD_ALLOC(name, name_size);
3808         OBD_ALLOC(uuid, uuid_size);
3809         if (name == NULL || uuid == NULL)
3810                 GOTO(cleanup_mem, rc = -ENOMEM);
3811
3812         OBD_ALLOC_PTR(bufs);
3813         if (!bufs)
3814                 GOTO(cleanup_mem, rc = -ENOMEM);
3815
3816         strcpy(name, dev);
3817         p = strstr(name, "-MDT");
3818         if (p == NULL)
3819                 GOTO(free_bufs, rc = -ENOMEM);
3820         p[3] = 'D';
3821
3822         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
3823
3824         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
3825         if (lprof == NULL || lprof->lp_dt == NULL) {
3826                 CERROR("can't find the profile: %s\n",
3827                        lustre_cfg_string(cfg, 0));
3828                 GOTO(free_bufs, rc = -EINVAL);
3829         }
3830
3831         lustre_cfg_bufs_reset(bufs, name);
3832         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
3833         lustre_cfg_bufs_set_string(bufs, 2, uuid);
3834         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
3835
3836         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
3837         if (!lcfg)
3838                 GOTO(free_bufs, rc = -ENOMEM);
3839
3840         rc = class_attach(lcfg);
3841         if (rc)
3842                 GOTO(lcfg_cleanup, rc);
3843
3844         obd = class_name2obd(name);
3845         if (!obd) {
3846                 CERROR("Can not find obd %s (%s in config)\n",
3847                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
3848                 GOTO(class_detach, rc = -EINVAL);
3849         }
3850
3851         lustre_cfg_free(lcfg);
3852
3853         lustre_cfg_bufs_reset(bufs, name);
3854         lustre_cfg_bufs_set_string(bufs, 1, uuid);
3855         lustre_cfg_bufs_set_string(bufs, 2, dev);
3856         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
3857
3858         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
3859
3860         rc = class_setup(obd, lcfg);
3861         if (rc)
3862                 GOTO(class_detach, rc);
3863
3864         /* connect to MDD we just setup */
3865         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
3866         if (rc)
3867                 GOTO(class_detach, rc);
3868
3869         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
3870         LASSERT(site);
3871         LASSERT(mdt_lu_site(mdt) == NULL);
3872         mdt->mdt_lu_dev.ld_site = site;
3873         site->ls_top_dev = &mdt->mdt_lu_dev;
3874         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
3875
3876
3877         /* now connect to bottom OSD */
3878         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
3879         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
3880         if (rc)
3881                 GOTO(class_detach, rc);
3882         mdt->mdt_bottom =
3883                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
3884
3885
3886         rc = lu_env_refill((struct lu_env *)env);
3887         if (rc != 0)
3888                 CERROR("Failure to refill session: '%d'\n", rc);
3889
3890         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
3891
3892         EXIT;
3893 class_detach:
3894         if (rc)
3895                 class_detach(obd, lcfg);
3896 lcfg_cleanup:
3897         lustre_cfg_free(lcfg);
3898 free_bufs:
3899         OBD_FREE_PTR(bufs);
3900 cleanup_mem:
3901         if (name)
3902                 OBD_FREE(name, name_size);
3903         if (uuid)
3904                 OBD_FREE(uuid, uuid_size);
3905         RETURN(rc);
3906 }
3907
3908 /* setup quota master target on MDT0 */
3909 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
3910                           struct lustre_cfg *cfg)
3911 {
3912         struct obd_device       *obd;
3913         char                    *dev = lustre_cfg_string(cfg, 0);
3914         char                    *qmtname, *uuid, *p;
3915         struct lustre_cfg_bufs  *bufs;
3916         struct lustre_cfg       *lcfg;
3917         struct lustre_profile   *lprof;
3918         struct obd_connect_data *data;
3919         int                      rc;
3920         ENTRY;
3921
3922         LASSERT(mdt->mdt_qmt_exp == NULL);
3923         LASSERT(mdt->mdt_qmt_dev == NULL);
3924
3925         /* quota master is on MDT0 only for now */
3926         if (mdt->mdt_seq_site.ss_node_id != 0)
3927                 RETURN(0);
3928
3929         /* MGS generates config commands which look as follows:
3930          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
3931          *                    3:lustre-MDT0000-mdtlov  4:f
3932          *
3933          * We generate the QMT name from the MDT one, just replacing MD with QM
3934          * after all the preparations, the logical equivalent will be:
3935          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
3936          *                    3:lustre-MDT0000-osd  4:f */
3937         OBD_ALLOC(qmtname, MAX_OBD_NAME);
3938         OBD_ALLOC(uuid, UUID_MAX);
3939         OBD_ALLOC_PTR(bufs);
3940         OBD_ALLOC_PTR(data);
3941         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
3942                 GOTO(cleanup_mem, rc = -ENOMEM);
3943
3944         strcpy(qmtname, dev);
3945         p = strstr(qmtname, "-MDT");
3946         if (p == NULL)
3947                 GOTO(cleanup_mem, rc = -ENOMEM);
3948         /* replace MD with QM */
3949         p[1] = 'Q';
3950         p[2] = 'M';
3951
3952         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
3953
3954         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
3955         if (lprof == NULL || lprof->lp_dt == NULL) {
3956                 CERROR("can't find profile for %s\n",
3957                        lustre_cfg_string(cfg, 0));
3958                 GOTO(cleanup_mem, rc = -EINVAL);
3959         }
3960
3961         lustre_cfg_bufs_reset(bufs, qmtname);
3962         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
3963         lustre_cfg_bufs_set_string(bufs, 2, uuid);
3964         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
3965
3966         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
3967         if (!lcfg)
3968                 GOTO(cleanup_mem, rc = -ENOMEM);
3969
3970         rc = class_attach(lcfg);
3971         if (rc)
3972                 GOTO(lcfg_cleanup, rc);
3973
3974         obd = class_name2obd(qmtname);
3975         if (!obd) {
3976                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
3977                        lustre_cfg_string(cfg, 0));
3978                 GOTO(class_detach, rc = -EINVAL);
3979         }
3980
3981         lustre_cfg_free(lcfg);
3982
3983         lustre_cfg_bufs_reset(bufs, qmtname);
3984         lustre_cfg_bufs_set_string(bufs, 1, uuid);
3985         lustre_cfg_bufs_set_string(bufs, 2, dev);
3986
3987         /* for quota, the next device should be the OSD device */
3988         lustre_cfg_bufs_set_string(bufs, 3,
3989                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
3990
3991         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
3992
3993         rc = class_setup(obd, lcfg);
3994         if (rc)
3995                 GOTO(class_detach, rc);
3996
3997         mdt->mdt_qmt_dev = obd->obd_lu_dev;
3998
3999         /* configure local quota objects */
4000         rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
4001                                                    &mdt->mdt_lu_dev,
4002                                                    mdt->mdt_qmt_dev);
4003         if (rc)
4004                 GOTO(class_cleanup, rc);
4005
4006         /* connect to quota master target */
4007         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4008         data->ocd_version = LUSTRE_VERSION_CODE;
4009         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
4010                          data, NULL);
4011         if (rc) {
4012                 CERROR("cannot connect to quota master device %s (%d)\n",
4013                        qmtname, rc);
4014                 GOTO(class_cleanup, rc);
4015         }
4016
4017         EXIT;
4018 class_cleanup:
4019         if (rc) {
4020                 class_manual_cleanup(obd);
4021                 mdt->mdt_qmt_dev = NULL;
4022         }
4023 class_detach:
4024         if (rc)
4025                 class_detach(obd, lcfg);
4026 lcfg_cleanup:
4027         lustre_cfg_free(lcfg);
4028 cleanup_mem:
4029         if (bufs)
4030                 OBD_FREE_PTR(bufs);
4031         if (qmtname)
4032                 OBD_FREE(qmtname, MAX_OBD_NAME);
4033         if (uuid)
4034                 OBD_FREE(uuid, UUID_MAX);
4035         if (data)
4036                 OBD_FREE_PTR(data);
4037         return rc;
4038 }
4039
4040 /* Shutdown quota master target associated with mdt */
4041 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
4042 {
4043         ENTRY;
4044
4045         if (mdt->mdt_qmt_exp == NULL)
4046                 RETURN_EXIT;
4047         LASSERT(mdt->mdt_qmt_dev != NULL);
4048
4049         /* the qmt automatically shuts down when the mdt disconnects */
4050         obd_disconnect(mdt->mdt_qmt_exp);
4051         mdt->mdt_qmt_exp = NULL;
4052         mdt->mdt_qmt_dev = NULL;
4053         EXIT;
4054 }
4055
4056 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
4057  * for now. This will be removed along with converting rest of MDT code
4058  * to use tgt_session_info */
4059 int mdt_tgt_getxattr(struct tgt_session_info *tsi)
4060 {
4061         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
4062         int                      rc;
4063
4064         rc = mdt_getxattr(info);
4065
4066         mdt_thread_info_fini(info);
4067         return rc;
4068 }
4069
4070 static struct tgt_handler mdt_tgt_handlers[] = {
4071 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4072                 0,                      MDS_CONNECT,    mdt_tgt_connect,
4073                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
4074 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4075                 0,                      MDS_DISCONNECT, tgt_disconnect,
4076                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
4077 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4078                 HABEO_REFERO,           MDS_SET_INFO,   mdt_set_info,
4079                 &RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION),
4080 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
4081 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_GETSTATUS,  mdt_getstatus),
4082 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETATTR,    mdt_getattr),
4083 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_GETATTR_NAME,
4084                                                         mdt_getattr_name),
4085 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETXATTR,   mdt_tgt_getxattr),
4086 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_STATFS,     mdt_statfs),
4087 TGT_MDT_HDL(0           | MUTABOR,      MDS_REINT,      mdt_reint),
4088 TGT_MDT_HDL(HABEO_CORPUS,               MDS_CLOSE,      mdt_close),
4089 TGT_MDT_HDL(HABEO_CORPUS,               MDS_DONE_WRITING,
4090                                                         mdt_done_writing),
4091 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_READPAGE,   mdt_readpage),
4092 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_SYNC,       mdt_sync),
4093 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_IS_SUBDIR,  mdt_is_subdir),
4094 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
4095 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_PROGRESS,
4096                                                         mdt_hsm_progress),
4097 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_REGISTER,
4098                                                         mdt_hsm_ct_register),
4099 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_UNREGISTER,
4100                                                         mdt_hsm_ct_unregister),
4101 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_STATE_GET,
4102                                                         mdt_hsm_state_get),
4103 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_STATE_SET,
4104                                                         mdt_hsm_state_set),
4105 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_ACTION, mdt_hsm_action),
4106 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_REQUEST,
4107                                                         mdt_hsm_request),
4108 TGT_MDT_HDL(HABEO_CORPUS|HABEO_REFERO | MUTABOR, MDS_SWAP_LAYOUTS,
4109                                                         mdt_swap_layouts)
4110 };
4111
4112 static struct tgt_handler mdt_sec_ctx_ops[] = {
4113 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
4114 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
4115 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
4116 };
4117
4118 static struct tgt_handler mdt_quota_ops[] = {
4119 TGT_QUOTA_HDL(HABEO_REFERO,             QUOTA_DQACQ,      mdt_quota_dqacq),
4120 };
4121
4122 static struct tgt_opc_slice mdt_common_slice[] = {
4123         {
4124                 .tos_opc_start  = MDS_FIRST_OPC,
4125                 .tos_opc_end    = MDS_LAST_OPC,
4126                 .tos_hs         = mdt_tgt_handlers
4127         },
4128         {
4129                 .tos_opc_start  = OBD_FIRST_OPC,
4130                 .tos_opc_end    = OBD_LAST_OPC,
4131                 .tos_hs         = tgt_obd_handlers
4132         },
4133         {
4134                 .tos_opc_start  = LDLM_FIRST_OPC,
4135                 .tos_opc_end    = LDLM_LAST_OPC,
4136                 .tos_hs         = tgt_dlm_handlers
4137         },
4138         {
4139                 .tos_opc_start  = SEC_FIRST_OPC,
4140                 .tos_opc_end    = SEC_LAST_OPC,
4141                 .tos_hs         = mdt_sec_ctx_ops
4142         },
4143         {
4144                 .tos_opc_start  = UPDATE_OBJ,
4145                 .tos_opc_end    = UPDATE_LAST_OPC,
4146                 .tos_hs         = tgt_out_handlers
4147         },
4148         {
4149                 .tos_opc_start  = FLD_FIRST_OPC,
4150                 .tos_opc_end    = FLD_LAST_OPC,
4151                 .tos_hs         = fld_handlers
4152         },
4153         {
4154                 .tos_opc_start  = SEQ_FIRST_OPC,
4155                 .tos_opc_end    = SEQ_LAST_OPC,
4156                 .tos_hs         = seq_handlers
4157         },
4158         {
4159                 .tos_opc_start  = QUOTA_DQACQ,
4160                 .tos_opc_end    = QUOTA_LAST_OPC,
4161                 .tos_hs         = mdt_quota_ops
4162         },
4163         {
4164                 .tos_opc_start  = LLOG_FIRST_OPC,
4165                 .tos_opc_end    = LLOG_LAST_OPC,
4166                 .tos_hs         = tgt_llog_handlers
4167         },
4168
4169         {
4170                 .tos_hs         = NULL
4171         }
4172 };
4173
4174 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4175 {
4176         struct md_device  *next = m->mdt_child;
4177         struct lu_device  *d    = &m->mdt_lu_dev;
4178         struct obd_device *obd = mdt2obd_dev(m);
4179         ENTRY;
4180
4181         target_recovery_fini(obd);
4182
4183         ping_evictor_stop();
4184
4185         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
4186
4187         if (m->mdt_opts.mo_coordinator)
4188                 mdt_hsm_cdt_stop(m);
4189
4190         mdt_hsm_cdt_fini(m);
4191
4192         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
4193         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4194         obd_exports_barrier(obd);
4195         obd_zombie_barrier();
4196
4197         mdt_procfs_fini(m);
4198
4199         tgt_fini(env, &m->mdt_lut);
4200         mdt_fs_cleanup(env, m);
4201         upcall_cache_cleanup(m->mdt_identity_cache);
4202         m->mdt_identity_cache = NULL;
4203
4204         if (m->mdt_namespace != NULL) {
4205                 ldlm_namespace_free(m->mdt_namespace, NULL,
4206                                     d->ld_obd->obd_force);
4207                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4208         }
4209
4210         mdt_quota_fini(env, m);
4211
4212         cfs_free_nidlist(&m->mdt_nosquash_nids);
4213         if (m->mdt_nosquash_str) {
4214                 OBD_FREE(m->mdt_nosquash_str, m->mdt_nosquash_strlen);
4215                 m->mdt_nosquash_str = NULL;
4216                 m->mdt_nosquash_strlen = 0;
4217         }
4218
4219         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_PAUSE_LFSCK, 0, NULL);
4220
4221         mdt_seq_fini(env, m);
4222         mdt_fld_fini(env, m);
4223
4224         next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL);
4225         cfs_timer_disarm(&m->mdt_ck_timer);
4226         mdt_ck_thread_stop(m);
4227
4228         /*
4229          * Finish the stack
4230          */
4231         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4232
4233         LASSERT(cfs_atomic_read(&d->ld_ref) == 0);
4234
4235         server_put_mount(mdt_obd_name(m));
4236
4237         EXIT;
4238 }
4239
4240 int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4241
4242 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4243                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
4244 {
4245         struct mdt_thread_info    *info;
4246         struct obd_device         *obd;
4247         const char                *dev = lustre_cfg_string(cfg, 0);
4248         const char                *num = lustre_cfg_string(cfg, 2);
4249         struct lustre_mount_info  *lmi = NULL;
4250         struct lustre_sb_info     *lsi;
4251         struct lu_site            *s;
4252         struct seq_server_site    *ss_site;
4253         const char                *identity_upcall = "NONE";
4254         struct md_device          *next;
4255         int                        rc;
4256         long                       node_id;
4257         mntopt_t                   mntopts;
4258         ENTRY;
4259
4260         lu_device_init(&m->mdt_lu_dev, ldt);
4261         /*
4262          * Environment (env) might be missing mdt_thread_key values at that
4263          * point, if device is allocated when mdt_thread_key is in QUIESCENT
4264          * mode.
4265          *
4266          * Usually device allocation path doesn't use module key values, but
4267          * mdt has to do a lot of work here, so allocate key value.
4268          */
4269         rc = lu_env_refill((struct lu_env *)env);
4270         if (rc != 0)
4271                 RETURN(rc);
4272
4273         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4274         LASSERT(info != NULL);
4275
4276         obd = class_name2obd(dev);
4277         LASSERT(obd != NULL);
4278
4279         m->mdt_max_mdsize = MAX_MD_SIZE; /* 4 stripes */
4280
4281         m->mdt_som_conf = 0;
4282
4283         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4284
4285         /* default is coordinator off, it is started through conf_param
4286          * or /proc */
4287         m->mdt_opts.mo_coordinator = 0;
4288
4289         lmi = server_get_mount(dev);
4290         if (lmi == NULL) {
4291                 CERROR("Cannot get mount info for %s!\n", dev);
4292                 RETURN(-EFAULT);
4293         } else {
4294                 lsi = s2lsi(lmi->lmi_sb);
4295                 /* CMD is supported only in IAM mode */
4296                 LASSERT(num);
4297                 node_id = simple_strtol(num, NULL, 10);
4298                 obd->u.obt.obt_magic = OBT_MAGIC;
4299         }
4300
4301         spin_lock_init(&m->mdt_ioepoch_lock);
4302         m->mdt_capa_timeout = CAPA_TIMEOUT;
4303         m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1;
4304         m->mdt_ck_timeout = CAPA_KEY_TIMEOUT;
4305         m->mdt_squash_uid = 0;
4306         m->mdt_squash_gid = 0;
4307         CFS_INIT_LIST_HEAD(&m->mdt_nosquash_nids);
4308         m->mdt_nosquash_str = NULL;
4309         m->mdt_nosquash_strlen = 0;
4310         init_rwsem(&m->mdt_squash_sem);
4311         spin_lock_init(&m->mdt_osfs_lock);
4312         m->mdt_osfs_age = cfs_time_shift_64(-1000);
4313         m->mdt_enable_remote_dir = 0;
4314         m->mdt_enable_remote_dir_gid = 0;
4315
4316         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
4317         m->mdt_lu_dev.ld_obd = obd;
4318         /* Set this lu_device to obd for error handling purposes. */
4319         obd->obd_lu_dev = &m->mdt_lu_dev;
4320
4321         /* init the stack */
4322         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
4323         if (rc) {
4324                 CERROR("%s: Can't init device stack, rc %d\n",
4325                        mdt_obd_name(m), rc);
4326                 GOTO(err_lmi, rc);
4327         }
4328
4329         s = mdt_lu_site(m);
4330         ss_site = mdt_seq_site(m);
4331         s->ld_seq_site = ss_site;
4332         ss_site->ss_lu = s;
4333
4334         /* set server index */
4335         ss_site->ss_node_id = node_id;
4336
4337         /* failover is the default
4338          * FIXME: we do not failout mds0/mgs, which may cause some problems.
4339          * assumed whose ss_node_id == 0 XXX
4340          * */
4341         obd->obd_replayable = 1;
4342         /* No connection accepted until configurations will finish */
4343         obd->obd_no_conn = 1;
4344
4345         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4346                 char *str = lustre_cfg_string(cfg, 4);
4347                 if (strchr(str, 'n')) {
4348                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
4349                         obd->obd_replayable = 0;
4350                 }
4351         }
4352
4353         rc = mdt_fld_init(env, mdt_obd_name(m), m);
4354         if (rc)
4355                 GOTO(err_fini_stack, rc);
4356
4357         rc = mdt_seq_init(env, m);
4358         if (rc)
4359                 GOTO(err_fini_fld, rc);
4360
4361         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
4362                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
4363         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4364                                               LDLM_NAMESPACE_SERVER,
4365                                               LDLM_NAMESPACE_GREEDY,
4366                                               LDLM_NS_TYPE_MDT);
4367         if (m->mdt_namespace == NULL)
4368                 GOTO(err_fini_seq, rc = -ENOMEM);
4369
4370         m->mdt_namespace->ns_lvbp = m;
4371         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
4372
4373         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4374         /* set obd_namespace for compatibility with old code */
4375         obd->obd_namespace = m->mdt_namespace;
4376
4377         cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m);
4378
4379         rc = mdt_hsm_cdt_init(m);
4380         if (rc != 0) {
4381                 CERROR("%s: error initializing coordinator, rc %d\n",
4382                        mdt_obd_name(m), rc);
4383                 GOTO(err_free_ns, rc);
4384         }
4385
4386         rc = mdt_ck_thread_start(m);
4387         if (rc)
4388                 GOTO(err_free_hsm, rc);
4389
4390         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
4391                       OBD_FAIL_MDS_ALL_REQUEST_NET,
4392                       OBD_FAIL_MDS_ALL_REPLY_NET);
4393         if (rc)
4394                 GOTO(err_capa, rc);
4395
4396         rc = mdt_fs_setup(env, m, obd, lsi);
4397         if (rc)
4398                 GOTO(err_tgt, rc);
4399
4400         tgt_adapt_sptlrpc_conf(&m->mdt_lut, 1);
4401
4402         next = m->mdt_child;
4403         rc = next->md_ops->mdo_iocontrol(env, next, OBD_IOC_GET_MNTOPT, 0,
4404                                          &mntopts);
4405         if (rc)
4406                 GOTO(err_fs_cleanup, rc);
4407
4408         if (mntopts & MNTOPT_USERXATTR)
4409                 m->mdt_opts.mo_user_xattr = 1;
4410         else
4411                 m->mdt_opts.mo_user_xattr = 0;
4412
4413         rc = next->md_ops->mdo_maxeasize_get(env, next, &m->mdt_max_ea_size);
4414         if (rc)
4415                 GOTO(err_fs_cleanup, rc);
4416
4417         if (mntopts & MNTOPT_ACL)
4418                 m->mdt_opts.mo_acl = 1;
4419         else
4420                 m->mdt_opts.mo_acl = 0;
4421
4422         /* XXX: to support suppgid for ACL, we enable identity_upcall
4423          * by default, otherwise, maybe got unexpected -EACCESS. */
4424         if (m->mdt_opts.mo_acl)
4425                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4426
4427         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
4428                                                 identity_upcall,
4429                                                 &mdt_identity_upcall_cache_ops);
4430         if (IS_ERR(m->mdt_identity_cache)) {
4431                 rc = PTR_ERR(m->mdt_identity_cache);
4432                 m->mdt_identity_cache = NULL;
4433                 GOTO(err_fs_cleanup, rc);
4434         }
4435
4436         rc = mdt_procfs_init(m, dev);
4437         if (rc) {
4438                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4439                 GOTO(err_recovery, rc);
4440         }
4441
4442         rc = mdt_quota_init(env, m, cfg);
4443         if (rc)
4444                 GOTO(err_procfs, rc);
4445
4446         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
4447         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
4448                            "mdt_ldlm_client", m->mdt_ldlm_client);
4449
4450         ping_evictor_start();
4451
4452         /* recovery will be started upon mdt_prepare()
4453          * when the whole stack is complete and ready
4454          * to serve the requests */
4455
4456         mdt_init_capa_ctxt(env, m);
4457
4458         /* Reduce the initial timeout on an MDS because it doesn't need such
4459          * a long timeout as an OST does. Adaptive timeouts will adjust this
4460          * value appropriately. */
4461         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4462                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4463
4464         RETURN(0);
4465 err_procfs:
4466         mdt_procfs_fini(m);
4467 err_recovery:
4468         target_recovery_fini(obd);
4469         upcall_cache_cleanup(m->mdt_identity_cache);
4470         m->mdt_identity_cache = NULL;
4471 err_fs_cleanup:
4472         mdt_fs_cleanup(env, m);
4473 err_tgt:
4474         tgt_fini(env, &m->mdt_lut);
4475 err_capa:
4476         cfs_timer_disarm(&m->mdt_ck_timer);
4477         mdt_ck_thread_stop(m);
4478 err_free_hsm:
4479         mdt_hsm_cdt_fini(m);
4480 err_free_ns:
4481         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4482         obd->obd_namespace = m->mdt_namespace = NULL;
4483 err_fini_seq:
4484         mdt_seq_fini(env, m);
4485 err_fini_fld:
4486         mdt_fld_fini(env, m);
4487 err_fini_stack:
4488         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4489 err_lmi:
4490         if (lmi)
4491                 server_put_mount(dev);
4492         return(rc);
4493 }
4494
4495 /* For interoperability, the left element is old parameter, the right one
4496  * is the new version of the parameter, if some parameter is deprecated,
4497  * the new version should be set as NULL. */
4498 static struct cfg_interop_param mdt_interop_param[] = {
4499         { "mdt.group_upcall",   NULL },
4500         { "mdt.quota_type",     NULL },
4501         { "mdd.quota_type",     NULL },
4502         { "mdt.rootsquash",     "mdt.root_squash" },
4503         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
4504         { NULL }
4505 };
4506
4507 /* used by MGS to process specific configurations */
4508 static int mdt_process_config(const struct lu_env *env,
4509                               struct lu_device *d, struct lustre_cfg *cfg)
4510 {
4511         struct mdt_device *m = mdt_dev(d);
4512         struct md_device *md_next = m->mdt_child;
4513         struct lu_device *next = md2lu_dev(md_next);
4514         int rc;
4515         ENTRY;
4516
4517         switch (cfg->lcfg_command) {
4518         case LCFG_PARAM: {
4519                 struct lprocfs_static_vars  lvars;
4520                 struct obd_device          *obd = d->ld_obd;
4521
4522                 /* For interoperability */
4523                 struct cfg_interop_param   *ptr = NULL;
4524                 struct lustre_cfg          *old_cfg = NULL;
4525                 char                       *param = NULL;
4526
4527                 param = lustre_cfg_string(cfg, 1);
4528                 if (param == NULL) {
4529                         CERROR("param is empty\n");
4530                         rc = -EINVAL;
4531                         break;
4532                 }
4533
4534                 ptr = class_find_old_param(param, mdt_interop_param);
4535                 if (ptr != NULL) {
4536                         if (ptr->new_param == NULL) {
4537                                 rc = 0;
4538                                 CWARN("For interoperability, skip this %s."
4539                                       " It is obsolete.\n", ptr->old_param);
4540                                 break;
4541                         }
4542
4543                         CWARN("Found old param %s, changed it to %s.\n",
4544                               ptr->old_param, ptr->new_param);
4545
4546                         old_cfg = cfg;
4547                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
4548                         if (IS_ERR(cfg)) {
4549                                 rc = PTR_ERR(cfg);
4550                                 break;
4551                         }
4552                 }
4553
4554                 lprocfs_mdt_init_vars(&lvars);
4555                 rc = class_process_proc_param(PARAM_MDT, lvars.obd_vars,
4556                                               cfg, obd);
4557                 if (rc > 0 || rc == -ENOSYS) {
4558                         /* is it an HSM var ? */
4559                         rc = class_process_proc_param(PARAM_HSM,
4560                                                       hsm_cdt_get_proc_vars(),
4561                                                       cfg, obd);
4562                         if (rc > 0 || rc == -ENOSYS)
4563                                 /* we don't understand; pass it on */
4564                                 rc = next->ld_ops->ldo_process_config(env, next,
4565                                                                       cfg);
4566                 }
4567
4568                 if (old_cfg != NULL)
4569                         lustre_cfg_free(cfg);
4570
4571                 break;
4572         }
4573         default:
4574                 /* others are passed further */
4575                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4576                 break;
4577         }
4578         RETURN(rc);
4579 }
4580
4581 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4582                                           const struct lu_object_header *hdr,
4583                                           struct lu_device *d)
4584 {
4585         struct mdt_object *mo;
4586
4587         ENTRY;
4588
4589         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, __GFP_IO);
4590         if (mo != NULL) {
4591                 struct lu_object *o;
4592                 struct lu_object_header *h;
4593
4594                 o = &mo->mot_obj;
4595                 h = &mo->mot_header;
4596                 lu_object_header_init(h);
4597                 lu_object_init(o, h, d);
4598                 lu_object_add_top(h, o);
4599                 o->lo_ops = &mdt_obj_ops;
4600                 mutex_init(&mo->mot_ioepoch_mutex);
4601                 mutex_init(&mo->mot_lov_mutex);
4602                 init_rwsem(&mo->mot_open_sem);
4603                 RETURN(o);
4604         }
4605         RETURN(NULL);
4606 }
4607
4608 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
4609                            const struct lu_object_conf *unused)
4610 {
4611         struct mdt_device *d = mdt_dev(o->lo_dev);
4612         struct lu_device  *under;
4613         struct lu_object  *below;
4614         int                rc = 0;
4615         ENTRY;
4616
4617         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4618                PFID(lu_object_fid(o)));
4619
4620         under = &d->mdt_child->md_lu_dev;
4621         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4622         if (below != NULL) {
4623                 lu_object_add(o, below);
4624         } else
4625                 rc = -ENOMEM;
4626
4627         RETURN(rc);
4628 }
4629
4630 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4631 {
4632         struct mdt_object *mo = mdt_obj(o);
4633         struct lu_object_header *h;
4634         ENTRY;
4635
4636         h = o->lo_header;
4637         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4638                PFID(lu_object_fid(o)));
4639
4640         LASSERT(atomic_read(&mo->mot_open_count) == 0);
4641         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
4642
4643         lu_object_fini(o);
4644         lu_object_header_fini(h);
4645         OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
4646
4647         EXIT;
4648 }
4649
4650 static int mdt_object_print(const struct lu_env *env, void *cookie,
4651                             lu_printer_t p, const struct lu_object *o)
4652 {
4653         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
4654         return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p(ioepoch="LPU64" "
4655                     "flags="LPX64", epochcount=%d, writecount=%d)",
4656                     mdto, mdto->mot_ioepoch, mdto->mot_flags,
4657                     mdto->mot_ioepoch_count, mdto->mot_writecount);
4658 }
4659
4660 static int mdt_prepare(const struct lu_env *env,
4661                 struct lu_device *pdev,
4662                 struct lu_device *cdev)
4663 {
4664         struct mdt_device *mdt = mdt_dev(cdev);
4665         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
4666         struct obd_device *obd = cdev->ld_obd;
4667         struct lfsck_start_param lsp;
4668         int rc;
4669
4670         ENTRY;
4671
4672         LASSERT(obd);
4673
4674         rc = next->ld_ops->ldo_prepare(env, cdev, next);
4675         if (rc)
4676                 RETURN(rc);
4677
4678         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
4679         if (rc)
4680                 RETURN(rc);
4681
4682         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
4683         if (rc)
4684                 RETURN(rc);
4685
4686         lsp.lsp_start = NULL;
4687         lsp.lsp_namespace = mdt->mdt_namespace;
4688         rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
4689                                                    OBD_IOC_START_LFSCK,
4690                                                    0, &lsp);
4691         if (rc != 0) {
4692                 CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
4693                       mdt_obd_name(mdt), rc);
4694                 rc = 0;
4695         }
4696
4697         if (mdt->mdt_seq_site.ss_node_id == 0) {
4698                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
4699                                                          &mdt->mdt_md_root_fid);
4700                 if (rc)
4701                         RETURN(rc);
4702         }
4703
4704         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
4705         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
4706         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
4707         LASSERT(obd->obd_no_conn);
4708         spin_lock(&obd->obd_dev_lock);
4709         obd->obd_no_conn = 0;
4710         spin_unlock(&obd->obd_dev_lock);
4711
4712         if (obd->obd_recovering == 0)
4713                 mdt_postrecov(env, mdt);
4714
4715         RETURN(rc);
4716 }
4717
4718 const struct lu_device_operations mdt_lu_ops = {
4719         .ldo_object_alloc   = mdt_object_alloc,
4720         .ldo_process_config = mdt_process_config,
4721         .ldo_prepare        = mdt_prepare,
4722 };
4723
4724 static const struct lu_object_operations mdt_obj_ops = {
4725         .loo_object_init    = mdt_object_init,
4726         .loo_object_free    = mdt_object_free,
4727         .loo_object_print   = mdt_object_print
4728 };
4729
4730 static int mdt_obd_set_info_async(const struct lu_env *env,
4731                                   struct obd_export *exp,
4732                                   __u32 keylen, void *key,
4733                                   __u32 vallen, void *val,
4734                                   struct ptlrpc_request_set *set)
4735 {
4736         int rc;
4737
4738         ENTRY;
4739
4740         if (KEY_IS(KEY_SPTLRPC_CONF)) {
4741                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp), 0);
4742                 RETURN(rc);
4743         }
4744
4745         RETURN(0);
4746 }
4747
4748 /**
4749  * Match client and server connection feature flags.
4750  *
4751  * Compute the compatibility flags for a connection request based on
4752  * features mutually supported by client and server.
4753  *
4754  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
4755  * must not be updated here, otherwise a partially initialized value may
4756  * be exposed. After the connection request is successfully processed,
4757  * the top-level MDT connect request handler atomically updates the export
4758  * connect flags from the obd_connect_data::ocd_connect_flags field of the
4759  * reply. \see mdt_connect().
4760  *
4761  * \param exp   the obd_export associated with this client/target pair
4762  * \param mdt   the target device for the connection
4763  * \param data  stores data for this connect request
4764  *
4765  * \retval 0       success
4766  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
4767  * \retval -EBADE  client and server feature requirements are incompatible
4768  */
4769 static int mdt_connect_internal(struct obd_export *exp,
4770                                 struct mdt_device *mdt,
4771                                 struct obd_connect_data *data)
4772 {
4773         LASSERT(data != NULL);
4774
4775         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
4776         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
4777
4778         /* If no known bits (which should not happen, probably,
4779            as everybody should support LOOKUP and UPDATE bits at least)
4780            revert to compat mode with plain locks. */
4781         if (!data->ocd_ibits_known &&
4782             data->ocd_connect_flags & OBD_CONNECT_IBITS)
4783                 data->ocd_connect_flags &= ~OBD_CONNECT_IBITS;
4784
4785         if (!mdt->mdt_opts.mo_acl)
4786                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
4787
4788         if (!mdt->mdt_opts.mo_user_xattr)
4789                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
4790
4791         if (!mdt->mdt_som_conf)
4792                 data->ocd_connect_flags &= ~OBD_CONNECT_SOM;
4793
4794         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
4795                 data->ocd_brw_size = min(data->ocd_brw_size,
4796                                          (__u32)MD_MAX_BRW_SIZE);
4797                 if (data->ocd_brw_size == 0) {
4798                         CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
4799                                " ocd_version: %x ocd_grant: %d "
4800                                "ocd_index: %u ocd_brw_size is "
4801                                "unexpectedly zero, network data "
4802                                "corruption? Refusing connection of this"
4803                                " client\n",
4804                                mdt_obd_name(mdt),
4805                                exp->exp_client_uuid.uuid,
4806                                exp, data->ocd_connect_flags, data->ocd_version,
4807                                data->ocd_grant, data->ocd_index);
4808                         return -EPROTO;
4809                 }
4810         }
4811
4812         /* NB: Disregard the rule against updating
4813          * exp_connect_data.ocd_connect_flags in this case, since
4814          * tgt_client_new() needs to know if this is a lightweight
4815          * connection, and it is safe to expose this flag before
4816          * connection processing completes. */
4817         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
4818                 spin_lock(&exp->exp_lock);
4819                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
4820                 spin_unlock(&exp->exp_lock);
4821         }
4822
4823         data->ocd_version = LUSTRE_VERSION_CODE;
4824
4825         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
4826                 CWARN("%s: MDS requires FID support, but client not\n",
4827                       mdt_obd_name(mdt));
4828                 return -EBADE;
4829         }
4830
4831         if (mdt->mdt_som_conf &&
4832             !(data->ocd_connect_flags & (OBD_CONNECT_LIGHTWEIGHT |
4833                                          OBD_CONNECT_MDS_MDS |
4834                                          OBD_CONNECT_SOM))) {
4835                 CWARN("%s: MDS has SOM enabled, but client does not support "
4836                       "it\n", mdt_obd_name(mdt));
4837                 return -EBADE;
4838         }
4839
4840         if (OCD_HAS_FLAG(data, PINGLESS)) {
4841                 if (ptlrpc_pinger_suppress_pings()) {
4842                         spin_lock(&exp->exp_obd->obd_dev_lock);
4843                         list_del_init(&exp->exp_obd_chain_timed);
4844                         spin_unlock(&exp->exp_obd->obd_dev_lock);
4845                 } else {
4846                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
4847                 }
4848         }
4849
4850         data->ocd_max_easize = mdt->mdt_max_ea_size;
4851
4852         return 0;
4853 }
4854
4855 /* mds_connect copy */
4856 static int mdt_obd_connect(const struct lu_env *env,
4857                            struct obd_export **exp, struct obd_device *obd,
4858                            struct obd_uuid *cluuid,
4859                            struct obd_connect_data *data,
4860                            void *localdata)
4861 {
4862         struct obd_export      *lexp;
4863         struct lustre_handle    conn = { 0 };
4864         struct mdt_device      *mdt;
4865         int                     rc;
4866         ENTRY;
4867
4868         LASSERT(env != NULL);
4869         if (!exp || !obd || !cluuid)
4870                 RETURN(-EINVAL);
4871
4872         mdt = mdt_dev(obd->obd_lu_dev);
4873
4874         /*
4875          * first, check whether the stack is ready to handle requests
4876          * XXX: probably not very appropriate method is used now
4877          *      at some point we should find a better one
4878          */
4879         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) && data != NULL &&
4880             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
4881                 rc = obd_get_info(env, mdt->mdt_child_exp,
4882                                   sizeof(KEY_OSP_CONNECTED),
4883                                   KEY_OSP_CONNECTED, NULL, NULL, NULL);
4884                 if (rc)
4885                         RETURN(-EAGAIN);
4886                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
4887         }
4888
4889         rc = class_connect(&conn, obd, cluuid);
4890         if (rc)
4891                 RETURN(rc);
4892
4893         lexp = class_conn2export(&conn);
4894         LASSERT(lexp != NULL);
4895
4896         rc = mdt_connect_internal(lexp, mdt, data);
4897         if (rc == 0) {
4898                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
4899
4900                 LASSERT(lcd);
4901                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
4902                 rc = tgt_client_new(env, lexp);
4903                 if (rc == 0)
4904                         mdt_export_stats_init(obd, lexp, localdata);
4905         }
4906
4907         if (rc != 0) {
4908                 class_disconnect(lexp);
4909                 *exp = NULL;
4910         } else {
4911                 *exp = lexp;
4912         }
4913
4914         RETURN(rc);
4915 }
4916
4917 static int mdt_obd_reconnect(const struct lu_env *env,
4918                              struct obd_export *exp, struct obd_device *obd,
4919                              struct obd_uuid *cluuid,
4920                              struct obd_connect_data *data,
4921                              void *localdata)
4922 {
4923         int                     rc;
4924         ENTRY;
4925
4926         if (exp == NULL || obd == NULL || cluuid == NULL)
4927                 RETURN(-EINVAL);
4928
4929         rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
4930         if (rc == 0)
4931                 mdt_export_stats_init(obd, exp, localdata);
4932
4933         RETURN(rc);
4934 }
4935
4936 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
4937                                    struct mdt_thread_info *info,
4938                                    struct mdt_file_data *mfd)
4939 {
4940         struct lu_context ses;
4941         int rc;
4942         ENTRY;
4943
4944         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
4945         if (rc)
4946                 RETURN(rc);
4947
4948         env->le_ses = &ses;
4949         lu_context_enter(&ses);
4950
4951         mdt_ucred(info)->uc_valid = UCRED_OLD;
4952         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
4953
4954         lu_context_exit(&ses);
4955         lu_context_fini(&ses);
4956         env->le_ses = NULL;
4957
4958         RETURN(rc);
4959 }
4960
4961 static int mdt_export_cleanup(struct obd_export *exp)
4962 {
4963         struct mdt_export_data *med = &exp->exp_mdt_data;
4964         struct obd_device      *obd = exp->exp_obd;
4965         struct mdt_device      *mdt;
4966         struct mdt_thread_info *info;
4967         struct lu_env           env;
4968         CFS_LIST_HEAD(closing_list);
4969         struct mdt_file_data *mfd, *n;
4970         int rc = 0;
4971         ENTRY;
4972
4973         spin_lock(&med->med_open_lock);
4974         while (!cfs_list_empty(&med->med_open_head)) {
4975                 cfs_list_t *tmp = med->med_open_head.next;
4976                 mfd = cfs_list_entry(tmp, struct mdt_file_data, mfd_list);
4977
4978                 /* Remove mfd handle so it can't be found again.
4979                  * We are consuming the mfd_list reference here. */
4980                 class_handle_unhash(&mfd->mfd_handle);
4981                 cfs_list_move_tail(&mfd->mfd_list, &closing_list);
4982         }
4983         spin_unlock(&med->med_open_lock);
4984         mdt = mdt_dev(obd->obd_lu_dev);
4985         LASSERT(mdt != NULL);
4986
4987         rc = lu_env_init(&env, LCT_MD_THREAD);
4988         if (rc)
4989                 RETURN(rc);
4990
4991         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
4992         LASSERT(info != NULL);
4993         memset(info, 0, sizeof *info);
4994         info->mti_env = &env;
4995         info->mti_mdt = mdt;
4996         info->mti_exp = exp;
4997
4998         if (!cfs_list_empty(&closing_list)) {
4999                 struct md_attr *ma = &info->mti_attr;
5000
5001                 /* Close any open files (which may also cause orphan unlinking). */
5002                 cfs_list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5003                         cfs_list_del_init(&mfd->mfd_list);
5004                         ma->ma_need = ma->ma_valid = 0;
5005
5006                         /* This file is being closed due to an eviction, it
5007                          * could have been modified and now dirty regarding to
5008                          * HSM archive, check this!
5009                          * The logic here is to mark a file dirty if there's a
5010                          * chance it was dirtied before the client was evicted,
5011                          * so that we don't have to wait for a release attempt
5012                          * before finding out the file was actually dirty and
5013                          * fail the release. Aggressively marking it dirty here
5014                          * will cause the policy engine to attempt to
5015                          * re-archive it; when rearchiving, we can compare the
5016                          * current version to the HSM data_version and make the
5017                          * archive request into a noop if it's not actually
5018                          * dirty.
5019                          */
5020                         if (mfd->mfd_mode & (FMODE_WRITE|MDS_FMODE_TRUNC))
5021                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
5022
5023                         /* Don't unlink orphan on failover umount, LU-184 */
5024                         if (exp->exp_flags & OBD_OPT_FAILOVER) {
5025                                 ma->ma_valid = MA_FLAGS;
5026                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5027                         }
5028                         mdt_mfd_close(info, mfd);
5029                 }
5030         }
5031         info->mti_mdt = NULL;
5032         /* cleanup client slot early */
5033         /* Do not erase record for recoverable client. */
5034         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5035                 tgt_client_del(&env, exp);
5036         lu_env_fini(&env);
5037
5038         RETURN(rc);
5039 }
5040
5041 static int mdt_obd_disconnect(struct obd_export *exp)
5042 {
5043         int rc;
5044         ENTRY;
5045
5046         LASSERT(exp);
5047         class_export_get(exp);
5048
5049         rc = server_disconnect_export(exp);
5050         if (rc != 0)
5051                 CDEBUG(D_IOCTL, "server disconnect error: %d\n", rc);
5052
5053         rc = mdt_export_cleanup(exp);
5054         class_export_put(exp);
5055         RETURN(rc);
5056 }
5057
5058 /* FIXME: Can we avoid using these two interfaces? */
5059 static int mdt_init_export(struct obd_export *exp)
5060 {
5061         struct mdt_export_data *med = &exp->exp_mdt_data;
5062         int                     rc;
5063         ENTRY;
5064
5065         CFS_INIT_LIST_HEAD(&med->med_open_head);
5066         spin_lock_init(&med->med_open_lock);
5067         mutex_init(&med->med_idmap_mutex);
5068         med->med_idmap = NULL;
5069         spin_lock(&exp->exp_lock);
5070         exp->exp_connecting = 1;
5071         spin_unlock(&exp->exp_lock);
5072
5073         /* self-export doesn't need client data and ldlm initialization */
5074         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5075                                      &exp->exp_client_uuid)))
5076                 RETURN(0);
5077
5078         rc = tgt_client_alloc(exp);
5079         if (rc)
5080                 GOTO(err, rc);
5081
5082         rc = ldlm_init_export(exp);
5083         if (rc)
5084                 GOTO(err_free, rc);
5085
5086         RETURN(rc);
5087
5088 err_free:
5089         tgt_client_free(exp);
5090 err:
5091         CERROR("%s: Failed to initialize export: rc = %d\n",
5092                exp->exp_obd->obd_name, rc);
5093         return rc;
5094 }
5095
5096 static int mdt_destroy_export(struct obd_export *exp)
5097 {
5098         ENTRY;
5099
5100         if (exp_connect_rmtclient(exp))
5101                 mdt_cleanup_idmap(&exp->exp_mdt_data);
5102
5103         target_destroy_export(exp);
5104         /* destroy can be called from failed obd_setup, so
5105          * checking uuid is safer than obd_self_export */
5106         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5107                                      &exp->exp_client_uuid)))
5108                 RETURN(0);
5109
5110         ldlm_destroy_export(exp);
5111         tgt_client_free(exp);
5112
5113         LASSERT(cfs_list_empty(&exp->exp_outstanding_replies));
5114         LASSERT(cfs_list_empty(&exp->exp_mdt_data.med_open_head));
5115
5116         RETURN(0);
5117 }
5118
5119 /** The maximum depth that fid2path() will search.
5120  * This is limited only because we want to store the fids for
5121  * historical path lookup purposes.
5122  */
5123 #define MAX_PATH_DEPTH 100
5124
5125 /** mdt_path() lookup structure. */
5126 struct path_lookup_info {
5127         __u64                   pli_recno;      /**< history point */
5128         __u64                   pli_currec;     /**< current record */
5129         struct lu_fid           pli_fid;
5130         struct lu_fid           pli_fids[MAX_PATH_DEPTH]; /**< path, in fids */
5131         struct mdt_object       *pli_mdt_obj;
5132         char                    *pli_path;      /**< full path */
5133         int                     pli_pathlen;
5134         int                     pli_linkno;     /**< which hardlink to follow */
5135         int                     pli_fidcount;   /**< number of \a pli_fids */
5136 };
5137
5138 static int mdt_links_read(struct mdt_thread_info *info,
5139                           struct mdt_object *mdt_obj, struct linkea_data *ldata)
5140 {
5141         int rc;
5142
5143         LASSERT(ldata->ld_buf->lb_buf != NULL);
5144
5145         if (!mdt_object_exists(mdt_obj))
5146                 return -ENODATA;
5147
5148         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5149                           ldata->ld_buf, XATTR_NAME_LINK);
5150         if (rc == -ERANGE) {
5151                 /* Buf was too small, figure out what we need. */
5152                 lu_buf_free(ldata->ld_buf);
5153                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5154                                   ldata->ld_buf, XATTR_NAME_LINK);
5155                 if (rc < 0)
5156                         return rc;
5157                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
5158                 if (ldata->ld_buf->lb_buf == NULL)
5159                         return -ENOMEM;
5160                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5161                                   ldata->ld_buf, XATTR_NAME_LINK);
5162         }
5163         if (rc < 0)
5164                 return rc;
5165
5166         return linkea_init(ldata);
5167 }
5168
5169 static int mdt_path_current(struct mdt_thread_info *info,
5170                             struct path_lookup_info *pli)
5171 {
5172         struct mdt_device       *mdt = info->mti_mdt;
5173         struct mdt_object       *mdt_obj;
5174         struct link_ea_header   *leh;
5175         struct link_ea_entry    *lee;
5176         struct lu_name          *tmpname = &info->mti_name;
5177         struct lu_fid           *tmpfid = &info->mti_tmp_fid1;
5178         struct lu_buf           *buf = &info->mti_big_buf;
5179         char                    *ptr;
5180         int                     reclen;
5181         struct linkea_data      ldata = { 0 };
5182         int                     rc = 0;
5183         ENTRY;
5184
5185         /* temp buffer for path element, the buffer will be finally freed
5186          * in mdt_thread_info_fini */
5187         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
5188         if (buf->lb_buf == NULL)
5189                 RETURN(-ENOMEM);
5190
5191         ldata.ld_buf = buf;
5192         ptr = pli->pli_path + pli->pli_pathlen - 1;
5193         *ptr = 0;
5194         --ptr;
5195         pli->pli_fidcount = 0;
5196         pli->pli_fids[0] = *(struct lu_fid *)mdt_object_fid(pli->pli_mdt_obj);
5197
5198         /* root FID only exists on MDT0, and fid2path should also ends at MDT0,
5199          * so checking root_fid can only happen on MDT0. */
5200         while (!lu_fid_eq(&mdt->mdt_md_root_fid,
5201                           &pli->pli_fids[pli->pli_fidcount])) {
5202                 mdt_obj = mdt_object_find(info->mti_env, mdt,
5203                                           &pli->pli_fids[pli->pli_fidcount]);
5204                 if (IS_ERR(mdt_obj))
5205                         GOTO(out, rc = PTR_ERR(mdt_obj));
5206                 if (mdt_object_remote(mdt_obj)) {
5207                         mdt_object_put(info->mti_env, mdt_obj);
5208                         GOTO(remote_out, rc = -EREMOTE);
5209                 }
5210                 if (!mdt_object_exists(mdt_obj)) {
5211                         mdt_object_put(info->mti_env, mdt_obj);
5212                         GOTO(out, rc = -ENOENT);
5213                 }
5214
5215                 rc = mdt_links_read(info, mdt_obj, &ldata);
5216                 mdt_object_put(info->mti_env, mdt_obj);
5217                 if (rc != 0)
5218                         GOTO(out, rc);
5219
5220                 leh = buf->lb_buf;
5221                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
5222                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
5223                 /* If set, use link #linkno for path lookup, otherwise use
5224                    link #0.  Only do this for the final path element. */
5225                 if (pli->pli_fidcount == 0 &&
5226                     pli->pli_linkno < leh->leh_reccount) {
5227                         int count;
5228                         for (count = 0; count < pli->pli_linkno; count++) {
5229                                 lee = (struct link_ea_entry *)
5230                                      ((char *)lee + reclen);
5231                                 linkea_entry_unpack(lee, &reclen, tmpname,
5232                                                     tmpfid);
5233                         }
5234                         if (pli->pli_linkno < leh->leh_reccount - 1)
5235                                 /* indicate to user there are more links */
5236                                 pli->pli_linkno++;
5237                 }
5238
5239                 /* Pack the name in the end of the buffer */
5240                 ptr -= tmpname->ln_namelen;
5241                 if (ptr - 1 <= pli->pli_path)
5242                         GOTO(out, rc = -EOVERFLOW);
5243                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
5244                 *(--ptr) = '/';
5245
5246                 /* Store the parent fid for historic lookup */
5247                 if (++pli->pli_fidcount >= MAX_PATH_DEPTH)
5248                         GOTO(out, rc = -EOVERFLOW);
5249                 pli->pli_fids[pli->pli_fidcount] = *tmpfid;
5250         }
5251
5252 remote_out:
5253         ptr++; /* skip leading / */
5254         memmove(pli->pli_path, ptr, pli->pli_path + pli->pli_pathlen - ptr);
5255
5256         EXIT;
5257 out:
5258         return rc;
5259 }
5260
5261 /* Returns the full path to this fid, as of changelog record recno. */
5262 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
5263                     char *path, int pathlen, __u64 *recno, int *linkno,
5264                     struct lu_fid *fid)
5265 {
5266         struct mdt_device       *mdt = info->mti_mdt;
5267         struct path_lookup_info *pli;
5268         int                     tries = 3;
5269         int                     rc = -EAGAIN;
5270         ENTRY;
5271
5272         if (pathlen < 3)
5273                 RETURN(-EOVERFLOW);
5274
5275         if (lu_fid_eq(&mdt->mdt_md_root_fid, mdt_object_fid(obj))) {
5276                 path[0] = '\0';
5277                 RETURN(0);
5278         }
5279
5280         OBD_ALLOC_PTR(pli);
5281         if (pli == NULL)
5282                 RETURN(-ENOMEM);
5283
5284         pli->pli_mdt_obj = obj;
5285         pli->pli_recno = *recno;
5286         pli->pli_path = path;
5287         pli->pli_pathlen = pathlen;
5288         pli->pli_linkno = *linkno;
5289
5290         /* Retry multiple times in case file is being moved */
5291         while (tries-- && rc == -EAGAIN)
5292                 rc = mdt_path_current(info, pli);
5293
5294         /* return the last resolved fids to the client, so the client will
5295          * build the left path on another MDT for remote object */
5296         *fid = pli->pli_fids[pli->pli_fidcount];
5297
5298         *recno = pli->pli_currec;
5299         /* Return next link index to caller */
5300         *linkno = pli->pli_linkno;
5301
5302         OBD_FREE_PTR(pli);
5303
5304         RETURN(rc);
5305 }
5306
5307 static int mdt_fid2path(struct mdt_thread_info *info,
5308                         struct getinfo_fid2path *fp)
5309 {
5310         struct mdt_device *mdt = info->mti_mdt;
5311         struct mdt_object *obj;
5312         int    rc;
5313         ENTRY;
5314
5315         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
5316                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
5317
5318         if (!fid_is_sane(&fp->gf_fid))
5319                 RETURN(-EINVAL);
5320
5321         if (!fid_is_namespace_visible(&fp->gf_fid)) {
5322                 CWARN("%s: "DFID" is invalid, sequence should be "
5323                       ">= "LPX64"\n", mdt_obd_name(mdt),
5324                       PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
5325                 RETURN(-EINVAL);
5326         }
5327
5328         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
5329         if (obj == NULL || IS_ERR(obj)) {
5330                 CDEBUG(D_IOCTL, "no object "DFID": %ld\n", PFID(&fp->gf_fid),
5331                        PTR_ERR(obj));
5332                 RETURN(-EINVAL);
5333         }
5334
5335         if (mdt_object_remote(obj))
5336                 rc = -EREMOTE;
5337         else if (!mdt_object_exists(obj))
5338                 rc = -ENOENT;
5339         else
5340                 rc = 0;
5341
5342         if (rc < 0) {
5343                 mdt_object_put(info->mti_env, obj);
5344                 CDEBUG(D_IOCTL, "nonlocal object "DFID": %d\n",
5345                        PFID(&fp->gf_fid), rc);
5346                 RETURN(rc);
5347         }
5348
5349         rc = mdt_path(info, obj, fp->gf_path, fp->gf_pathlen, &fp->gf_recno,
5350                       &fp->gf_linkno, &fp->gf_fid);
5351
5352         CDEBUG(D_INFO, "fid "DFID", path %s recno "LPX64" linkno %u\n",
5353                PFID(&fp->gf_fid), fp->gf_path, fp->gf_recno, fp->gf_linkno);
5354
5355         mdt_object_put(info->mti_env, obj);
5356
5357         RETURN(rc);
5358 }
5359
5360 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key,
5361                             void *val, int vallen)
5362 {
5363         struct getinfo_fid2path *fpout, *fpin;
5364         int rc = 0;
5365
5366         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
5367         fpout = val;
5368
5369         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5370                 lustre_swab_fid2path(fpin);
5371
5372         memcpy(fpout, fpin, sizeof(*fpin));
5373         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
5374                 RETURN(-EINVAL);
5375
5376         rc = mdt_fid2path(info, fpout);
5377         RETURN(rc);
5378 }
5379
5380 int mdt_get_info(struct tgt_session_info *tsi)
5381 {
5382         char    *key;
5383         int      keylen;
5384         __u32   *vallen;
5385         void    *valout;
5386         int      rc;
5387
5388         ENTRY;
5389
5390         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
5391         if (key == NULL) {
5392                 CDEBUG(D_IOCTL, "No GETINFO key");
5393                 RETURN(err_serious(-EFAULT));
5394         }
5395         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
5396                                       RCL_CLIENT);
5397
5398         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
5399         if (vallen == NULL) {
5400                 CDEBUG(D_IOCTL, "Unable to get RMF_GETINFO_VALLEN buffer");
5401                 RETURN(err_serious(-EFAULT));
5402         }
5403
5404         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
5405                              *vallen);
5406         rc = req_capsule_server_pack(tsi->tsi_pill);
5407         if (rc)
5408                 RETURN(err_serious(rc));
5409
5410         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
5411         if (valout == NULL) {
5412                 CDEBUG(D_IOCTL, "Unable to get get-info RPC out buffer");
5413                 RETURN(err_serious(-EFAULT));
5414         }
5415
5416         if (KEY_IS(KEY_FID2PATH)) {
5417                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
5418
5419                 rc = mdt_rpc_fid2path(info, key, valout, *vallen);
5420                 mdt_thread_info_fini(info);
5421         } else {
5422                 rc = -EINVAL;
5423         }
5424         RETURN(rc);
5425 }
5426
5427 /* Pass the ioc down */
5428 static int mdt_ioc_child(struct lu_env *env, struct mdt_device *mdt,
5429                          unsigned int cmd, int len, void *data)
5430 {
5431         struct lu_context ioctl_session;
5432         struct md_device *next = mdt->mdt_child;
5433         int rc;
5434         ENTRY;
5435
5436         rc = lu_context_init(&ioctl_session, LCT_SERVER_SESSION);
5437         if (rc)
5438                 RETURN(rc);
5439         ioctl_session.lc_thread = (struct ptlrpc_thread *)current;
5440         lu_context_enter(&ioctl_session);
5441         env->le_ses = &ioctl_session;
5442
5443         LASSERT(next->md_ops->mdo_iocontrol);
5444         rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data);
5445
5446         lu_context_exit(&ioctl_session);
5447         lu_context_fini(&ioctl_session);
5448         RETURN(rc);
5449 }
5450
5451 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
5452 {
5453         struct obd_ioctl_data *data = karg;
5454         struct lu_fid *fid;
5455         __u64 version;
5456         struct mdt_object *obj;
5457         struct mdt_lock_handle  *lh;
5458         int rc;
5459         ENTRY;
5460
5461         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
5462             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
5463                 RETURN(-EINVAL);
5464
5465         fid = (struct lu_fid *)data->ioc_inlbuf1;
5466
5467         if (!fid_is_sane(fid))
5468                 RETURN(-EINVAL);
5469
5470         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
5471
5472         lh = &mti->mti_lh[MDT_LH_PARENT];
5473         mdt_lock_reg_init(lh, LCK_CR);
5474
5475         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
5476         if (IS_ERR(obj))
5477                 RETURN(PTR_ERR(obj));
5478
5479         if (mdt_object_remote(obj)) {
5480                 rc = -EREMOTE;
5481                 /**
5482                  * before calling version get the correct MDS should be
5483                  * fid, this is error to find remote object here
5484                  */
5485                 CERROR("nonlocal object "DFID"\n", PFID(fid));
5486         } else if (!mdt_object_exists(obj)) {
5487                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
5488                 rc = -ENOENT;
5489         } else {
5490                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
5491                *(__u64 *)data->ioc_inlbuf2 = version;
5492                 rc = 0;
5493         }
5494         mdt_object_unlock_put(mti, obj, lh, 1);
5495         RETURN(rc);
5496 }
5497
5498 /* ioctls on obd dev */
5499 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
5500                          void *karg, void *uarg)
5501 {
5502         struct lu_env      env;
5503         struct obd_device *obd = exp->exp_obd;
5504         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
5505         struct dt_device  *dt = mdt->mdt_bottom;
5506         int rc;
5507
5508         ENTRY;
5509         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
5510         rc = lu_env_init(&env, LCT_MD_THREAD);
5511         if (rc)
5512                 RETURN(rc);
5513
5514         switch (cmd) {
5515         case OBD_IOC_SYNC:
5516                 rc = mdt_device_sync(&env, mdt);
5517                 break;
5518         case OBD_IOC_SET_READONLY:
5519                 rc = dt->dd_ops->dt_ro(&env, dt);
5520                 break;
5521         case OBD_IOC_ABORT_RECOVERY:
5522                 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
5523                 target_stop_recovery_thread(obd);
5524                 rc = 0;
5525                 break;
5526         case OBD_IOC_CHANGELOG_REG:
5527         case OBD_IOC_CHANGELOG_DEREG:
5528         case OBD_IOC_CHANGELOG_CLEAR:
5529                 rc = mdt_ioc_child(&env, mdt, cmd, len, karg);
5530                 break;
5531         case OBD_IOC_START_LFSCK: {
5532                 struct md_device *next = mdt->mdt_child;
5533                 struct obd_ioctl_data *data = karg;
5534                 struct lfsck_start_param lsp;
5535
5536                 if (unlikely(data == NULL)) {
5537                         rc = -EINVAL;
5538                         break;
5539                 }
5540
5541                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
5542                 lsp.lsp_namespace = mdt->mdt_namespace;
5543                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
5544                 break;
5545         }
5546         case OBD_IOC_STOP_LFSCK: {
5547                 struct md_device *next = mdt->mdt_child;
5548
5549                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, NULL);
5550                 break;
5551         }
5552         case OBD_IOC_GET_OBJ_VERSION: {
5553                 struct mdt_thread_info *mti;
5554                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5555                 memset(mti, 0, sizeof *mti);
5556                 mti->mti_env = &env;
5557                 mti->mti_mdt = mdt;
5558                 mti->mti_exp = exp;
5559
5560                 rc = mdt_ioc_version_get(mti, karg);
5561                 break;
5562         }
5563         case OBD_IOC_CATLOGLIST:
5564                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg);
5565                 break;
5566         default:
5567                 rc = -EOPNOTSUPP;
5568                 CERROR("%s: Not supported cmd = %d, rc = %d\n",
5569                         mdt_obd_name(mdt), cmd, rc);
5570         }
5571
5572         lu_env_fini(&env);
5573         RETURN(rc);
5574 }
5575
5576 int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
5577 {
5578         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
5579         int rc;
5580         ENTRY;
5581
5582         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
5583         RETURN(rc);
5584 }
5585
5586 int mdt_obd_postrecov(struct obd_device *obd)
5587 {
5588         struct lu_env env;
5589         int rc;
5590
5591         rc = lu_env_init(&env, LCT_MD_THREAD);
5592         if (rc)
5593                 RETURN(rc);
5594         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
5595         lu_env_fini(&env);
5596         return rc;
5597 }
5598
5599 static struct obd_ops mdt_obd_device_ops = {
5600         .o_owner          = THIS_MODULE,
5601         .o_set_info_async = mdt_obd_set_info_async,
5602         .o_connect        = mdt_obd_connect,
5603         .o_reconnect      = mdt_obd_reconnect,
5604         .o_disconnect     = mdt_obd_disconnect,
5605         .o_init_export    = mdt_init_export,
5606         .o_destroy_export = mdt_destroy_export,
5607         .o_iocontrol      = mdt_iocontrol,
5608         .o_postrecov      = mdt_obd_postrecov,
5609 };
5610
5611 static struct lu_device* mdt_device_fini(const struct lu_env *env,
5612                                          struct lu_device *d)
5613 {
5614         struct mdt_device *m = mdt_dev(d);
5615         ENTRY;
5616
5617         mdt_fini(env, m);
5618         RETURN(NULL);
5619 }
5620
5621 static struct lu_device *mdt_device_free(const struct lu_env *env,
5622                                          struct lu_device *d)
5623 {
5624         struct mdt_device *m = mdt_dev(d);
5625         ENTRY;
5626
5627         lu_device_fini(&m->mdt_lu_dev);
5628         OBD_FREE_PTR(m);
5629
5630         RETURN(NULL);
5631 }
5632
5633 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
5634                                           struct lu_device_type *t,
5635                                           struct lustre_cfg *cfg)
5636 {
5637         struct lu_device  *l;
5638         struct mdt_device *m;
5639
5640         OBD_ALLOC_PTR(m);
5641         if (m != NULL) {
5642                 int rc;
5643
5644                 l = &m->mdt_lu_dev;
5645                 rc = mdt_init0(env, m, t, cfg);
5646                 if (rc != 0) {
5647                         mdt_device_free(env, l);
5648                         l = ERR_PTR(rc);
5649                         return l;
5650                 }
5651         } else
5652                 l = ERR_PTR(-ENOMEM);
5653         return l;
5654 }
5655
5656 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
5657 LU_KEY_INIT(mdt, struct mdt_thread_info);
5658
5659 static void mdt_key_fini(const struct lu_context *ctx,
5660                          struct lu_context_key *key, void* data)
5661 {
5662         struct mdt_thread_info *info = data;
5663
5664         if (info->mti_big_lmm) {
5665                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
5666                 info->mti_big_lmm = NULL;
5667                 info->mti_big_lmmsize = 0;
5668         }
5669         OBD_FREE_PTR(info);
5670 }
5671
5672 /* context key: mdt_thread_key */
5673 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
5674
5675 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
5676 {
5677         return lu_ucred(info->mti_env);
5678 }
5679
5680 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
5681 {
5682         return lu_ucred_check(info->mti_env);
5683 }
5684
5685 /**
5686  * Enable/disable COS (Commit On Sharing).
5687  *
5688  * Set/Clear the COS flag in mdt options.
5689  *
5690  * \param mdt mdt device
5691  * \param val 0 disables COS, other values enable COS
5692  */
5693 void mdt_enable_cos(struct mdt_device *mdt, int val)
5694 {
5695         struct lu_env env;
5696         int rc;
5697
5698         mdt->mdt_opts.mo_cos = !!val;
5699         rc = lu_env_init(&env, LCT_LOCAL);
5700         if (unlikely(rc != 0)) {
5701                 CWARN("lu_env initialization failed with rc = %d,"
5702                       "cannot sync\n", rc);
5703                 return;
5704         }
5705         mdt_device_sync(&env, mdt);
5706         lu_env_fini(&env);
5707 }
5708
5709 /**
5710  * Check COS (Commit On Sharing) status.
5711  *
5712  * Return COS flag status.
5713  *
5714  * \param mdt mdt device
5715  */
5716 int mdt_cos_is_enabled(struct mdt_device *mdt)
5717 {
5718         return mdt->mdt_opts.mo_cos != 0;
5719 }
5720
5721 static struct lu_device_type_operations mdt_device_type_ops = {
5722         .ldto_device_alloc = mdt_device_alloc,
5723         .ldto_device_free  = mdt_device_free,
5724         .ldto_device_fini  = mdt_device_fini
5725 };
5726
5727 static struct lu_device_type mdt_device_type = {
5728         .ldt_tags     = LU_DEVICE_MD,
5729         .ldt_name     = LUSTRE_MDT_NAME,
5730         .ldt_ops      = &mdt_device_type_ops,
5731         .ldt_ctx_tags = LCT_MD_THREAD
5732 };
5733
5734 static int __init mdt_mod_init(void)
5735 {
5736         struct lprocfs_static_vars lvars;
5737         int rc;
5738
5739         CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") ==
5740                  FID_NOBRACE_LEN + 1);
5741         CLASSERT(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") ==
5742                  FID_LEN + 1);
5743         rc = lu_kmem_init(mdt_caches);
5744         if (rc)
5745                 return rc;
5746
5747         rc = mds_mod_init();
5748         if (rc)
5749                 GOTO(lu_fini, rc);
5750
5751         lprocfs_mdt_init_vars(&lvars);
5752         rc = class_register_type(&mdt_obd_device_ops, NULL, NULL,
5753 #ifndef HAVE_ONLY_PROCFS_SEQ
5754                                 lvars.module_vars,
5755 #endif
5756                                 LUSTRE_MDT_NAME, &mdt_device_type);
5757         if (rc)
5758                 GOTO(mds_fini, rc);
5759 lu_fini:
5760         if (rc)
5761                 lu_kmem_fini(mdt_caches);
5762 mds_fini:
5763         if (rc)
5764                 mds_mod_exit();
5765         return rc;
5766 }
5767
5768 static void __exit mdt_mod_exit(void)
5769 {
5770         class_unregister_type(LUSTRE_MDT_NAME);
5771         mds_mod_exit();
5772         lu_kmem_fini(mdt_caches);
5773 }
5774
5775 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
5776 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
5777 MODULE_LICENSE("GPL");
5778
5779 cfs_module(mdt, LUSTRE_VERSION_STRING, mdt_mod_init, mdt_mod_exit);