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_mdt.h>
  66 #include <lustre_log.h>
  67 #include "mdt_internal.h"
  68 #include <lustre_acl.h>
  69 #include <lustre_param.h>
  70 #include <lustre_quota.h>
  71 #include <lustre_linkea.h>
  72 #include <lustre_lfsck.h>
  73
  74 mdl_mode_t mdt_mdl_lock_modes[] = {
  75         [LCK_MINMODE] = MDL_MINMODE,
  76         [LCK_EX]      = MDL_EX,
  77         [LCK_PW]      = MDL_PW,
  78         [LCK_PR]      = MDL_PR,
  79         [LCK_CW]      = MDL_CW,
  80         [LCK_CR]      = MDL_CR,
  81         [LCK_NL]      = MDL_NL,
  82         [LCK_GROUP]   = MDL_GROUP
  83 };
  84
  85 ldlm_mode_t mdt_dlm_lock_modes[] = {
  86         [MDL_MINMODE] = LCK_MINMODE,
  87         [MDL_EX]      = LCK_EX,
  88         [MDL_PW]      = LCK_PW,
  89         [MDL_PR]      = LCK_PR,
  90         [MDL_CW]      = LCK_CW,
  91         [MDL_CR]      = LCK_CR,
  92         [MDL_NL]      = LCK_NL,
  93         [MDL_GROUP]   = LCK_GROUP
  94 };
  95
  96 static struct mdt_device *mdt_dev(struct lu_device *d);
  97 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
  98
  99 static const struct lu_object_operations mdt_obj_ops;
 100
 101 /* Slab for MDT object allocation */
 102 static struct kmem_cache *mdt_object_kmem;
 103
 104 static struct lu_kmem_descr mdt_caches[] = {
 105         {
 106                 .ckd_cache = &mdt_object_kmem,
 107                 .ckd_name  = "mdt_obj",
 108                 .ckd_size  = sizeof(struct mdt_object)
 109         },
 110         {
 111                 .ckd_cache = NULL
 112         }
 113 };
 114
 115 int mdt_get_disposition(struct ldlm_reply *rep, int flag)
 116 {
 117         if (!rep)
 118                 return 0;
 119         return (rep->lock_policy_res1 & flag);
 120 }
 121
 122 void mdt_clear_disposition(struct mdt_thread_info *info,
 123                            struct ldlm_reply *rep, int flag)
 124 {
 125         if (info)
 126                 info->mti_opdata &= ~flag;
 127         if (rep)
 128                 rep->lock_policy_res1 &= ~flag;
 129 }
 130
 131 void mdt_set_disposition(struct mdt_thread_info *info,
 132                          struct ldlm_reply *rep, int flag)
 133 {
 134         if (info)
 135                 info->mti_opdata |= flag;
 136         if (rep)
 137                 rep->lock_policy_res1 |= flag;
 138 }
 139
 140 void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
 141 {
 142         lh->mlh_pdo_hash = 0;
 143         lh->mlh_reg_mode = lm;
 144         lh->mlh_rreg_mode = lm;
 145         lh->mlh_type = MDT_REG_LOCK;
 146 }
 147
 148 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lm,
 149                        const char *name, int namelen)
 150 {
 151         lh->mlh_reg_mode = lm;
 152         lh->mlh_rreg_mode = lm;
 153         lh->mlh_type = MDT_PDO_LOCK;
 154
 155         if (name != NULL && (name[0] != '\0')) {
 156                 LASSERT(namelen > 0);
 157                 lh->mlh_pdo_hash = full_name_hash(name, namelen);
 158                 /* XXX Workaround for LU-2856
 159                  * Zero is a valid return value of full_name_hash, but several
 160                  * users of mlh_pdo_hash assume a non-zero hash value. We
 161                  * therefore map zero onto an arbitrary, but consistent
 162                  * value (1) to avoid problems further down the road. */
 163                 if (unlikely(!lh->mlh_pdo_hash))
 164                         lh->mlh_pdo_hash = 1;
 165         } else {
 166                 LASSERT(namelen == 0);
 167                 lh->mlh_pdo_hash = 0ull;
 168         }
 169 }
 170
 171 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 172                               struct mdt_lock_handle *lh)
 173 {
 174         mdl_mode_t mode;
 175         ENTRY;
 176
 177         /*
 178          * Any dir access needs couple of locks:
 179          *
 180          * 1) on part of dir we gonna take lookup/modify;
 181          *
 182          * 2) on whole dir to protect it from concurrent splitting and/or to
 183          * flush client's cache for readdir().
 184          *
 185          * so, for a given mode and object this routine decides what lock mode
 186          * to use for lock #2:
 187          *
 188          * 1) if caller's gonna lookup in dir then we need to protect dir from
 189          * being splitted only - LCK_CR
 190          *
 191          * 2) if caller's gonna modify dir then we need to protect dir from
 192          * being splitted and to flush cache - LCK_CW
 193          *
 194          * 3) if caller's gonna modify dir and that dir seems ready for
 195          * splitting then we need to protect it from any type of access
 196          * (lookup/modify/split) - LCK_EX --bzzz
 197          */
 198
 199         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 200         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 201
 202         /*
 203          * Ask underlaying level its opinion about preferable PDO lock mode
 204          * having access type passed as regular lock mode:
 205          *
 206          * - MDL_MINMODE means that lower layer does not want to specify lock
 207          * mode;
 208          *
 209          * - MDL_NL means that no PDO lock should be taken. This is used in some
 210          * cases. Say, for non-splittable directories no need to use PDO locks
 211          * at all.
 212          */
 213         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 214                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 215
 216         if (mode != MDL_MINMODE) {
 217                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 218         } else {
 219                 /*
 220                  * Lower layer does not want to specify locking mode. We do it
 221                  * our selves. No special protection is needed, just flush
 222                  * client's cache on modification and allow concurrent
 223                  * mondification.
 224                  */
 225                 switch (lh->mlh_reg_mode) {
 226                 case LCK_EX:
 227                         lh->mlh_pdo_mode = LCK_EX;
 228                         break;
 229                 case LCK_PR:
 230                         lh->mlh_pdo_mode = LCK_CR;
 231                         break;
 232                 case LCK_PW:
 233                         lh->mlh_pdo_mode = LCK_CW;
 234                         break;
 235                 default:
 236                         CERROR("Not expected lock type (0x%x)\n",
 237                                (int)lh->mlh_reg_mode);
 238                         LBUG();
 239                 }
 240         }
 241
 242         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 243         EXIT;
 244 }
 245
 246 int mdt_getstatus(struct mdt_thread_info *info)
 247 {
 248         struct mdt_device       *mdt  = info->mti_mdt;
 249         struct mdt_body         *repbody;
 250         int                     rc;
 251         ENTRY;
 252
 253         rc = mdt_check_ucred(info);
 254         if (rc)
 255                 RETURN(err_serious(rc));
 256
 257         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
 258                 RETURN(err_serious(-ENOMEM));
 259
 260         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 261         repbody->fid1 = mdt->mdt_md_root_fid;
 262         repbody->valid |= OBD_MD_FLID;
 263
 264         if (mdt->mdt_opts.mo_mds_capa &&
 265             exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) {
 266                 struct mdt_object  *root;
 267                 struct lustre_capa *capa;
 268
 269                 root = mdt_object_find(info->mti_env, mdt, &repbody->fid1);
 270                 if (IS_ERR(root))
 271                         RETURN(PTR_ERR(root));
 272
 273                 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
 274                 LASSERT(capa);
 275                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 276                 rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa,
 277                                  0);
 278                 mdt_object_put(info->mti_env, root);
 279                 if (rc == 0)
 280                         repbody->valid |= OBD_MD_FLMDSCAPA;
 281         }
 282
 283         RETURN(rc);
 284 }
 285
 286 int mdt_statfs(struct mdt_thread_info *info)
 287 {
 288         struct ptlrpc_request           *req = mdt_info_req(info);
 289         struct md_device                *next = info->mti_mdt->mdt_child;
 290         struct ptlrpc_service_part      *svcpt;
 291         struct obd_statfs               *osfs;
 292         int                             rc;
 293
 294         ENTRY;
 295
 296         svcpt = info->mti_pill->rc_req->rq_rqbd->rqbd_svcpt;
 297
 298         /* This will trigger a watchdog timeout */
 299         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 300                          (MDT_SERVICE_WATCHDOG_FACTOR *
 301                           at_get(&svcpt->scp_at_estimate)) + 1);
 302
 303         rc = mdt_check_ucred(info);
 304         if (rc)
 305                 RETURN(err_serious(rc));
 306
 307         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
 308                 RETURN(err_serious(-ENOMEM));
 309
 310         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 311         if (!osfs)
 312                 RETURN(-EPROTO);
 313
 314         /** statfs information are cached in the mdt_device */
 315         if (cfs_time_before_64(info->mti_mdt->mdt_osfs_age,
 316                                cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS))) {
 317                 /** statfs data is too old, get up-to-date one */
 318                 rc = next->md_ops->mdo_statfs(info->mti_env, next, osfs);
 319                 if (rc)
 320                         RETURN(rc);
 321                 spin_lock(&info->mti_mdt->mdt_osfs_lock);
 322                 info->mti_mdt->mdt_osfs = *osfs;
 323                 info->mti_mdt->mdt_osfs_age = cfs_time_current_64();
 324                 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
 325         } else {
 326                 /** use cached statfs data */
 327                 spin_lock(&info->mti_mdt->mdt_osfs_lock);
 328                 *osfs = info->mti_mdt->mdt_osfs;
 329                 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
 330         }
 331
 332         if (rc == 0)
 333                 mdt_counter_incr(req, LPROC_MDT_STATFS);
 334
 335         RETURN(rc);
 336 }
 337
 338 /**
 339  * Pack SOM attributes into the reply.
 340  * Call under a DLM UPDATE lock.
 341  */
 342 static void mdt_pack_size2body(struct mdt_thread_info *info,
 343                                struct mdt_object *mo)
 344 {
 345         struct mdt_body *b;
 346         struct md_attr *ma = &info->mti_attr;
 347
 348         LASSERT(ma->ma_attr.la_valid & LA_MODE);
 349         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 350
 351         /* Check if Size-on-MDS is supported, if this is a regular file,
 352          * if SOM is enabled on the object and if SOM cache exists and valid.
 353          * Otherwise do not pack Size-on-MDS attributes to the reply. */
 354         if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) ||
 355             !S_ISREG(ma->ma_attr.la_mode) ||
 356             !mdt_object_is_som_enabled(mo) ||
 357             !(ma->ma_valid & MA_SOM))
 358                 return;
 359
 360         b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 361         b->size = ma->ma_som->msd_size;
 362         b->blocks = ma->ma_som->msd_blocks;
 363 }
 364
 365 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 366                         const struct lu_attr *attr, const struct lu_fid *fid)
 367 {
 368         struct md_attr *ma = &info->mti_attr;
 369
 370         LASSERT(ma->ma_valid & MA_INODE);
 371
 372         b->atime      = attr->la_atime;
 373         b->mtime      = attr->la_mtime;
 374         b->ctime      = attr->la_ctime;
 375         b->mode       = attr->la_mode;
 376         b->size       = attr->la_size;
 377         b->blocks     = attr->la_blocks;
 378         b->uid        = attr->la_uid;
 379         b->gid        = attr->la_gid;
 380         b->flags      = attr->la_flags;
 381         b->nlink      = attr->la_nlink;
 382         b->rdev       = attr->la_rdev;
 383
 384         /*XXX should pack the reply body according to lu_valid*/
 385         b->valid |= OBD_MD_FLCTIME | OBD_MD_FLUID   |
 386                     OBD_MD_FLGID   | OBD_MD_FLTYPE  |
 387                     OBD_MD_FLMODE  | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
 388                     OBD_MD_FLATIME | OBD_MD_FLMTIME ;
 389
 390         if (!S_ISREG(attr->la_mode)) {
 391                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 392         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
 393                 /* means no objects are allocated on osts. */
 394                 LASSERT(!(ma->ma_valid & MA_LOV));
 395                 /* just ignore blocks occupied by extend attributes on MDS */
 396                 b->blocks = 0;
 397                 /* if no object is allocated on osts, the size on mds is valid. b=22272 */
 398                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 399         }
 400
 401         if (fid) {
 402                 b->fid1 = *fid;
 403                 b->valid |= OBD_MD_FLID;
 404                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n",
 405                                 PFID(fid), b->nlink, b->mode, b->size);
 406         }
 407
 408         if (info)
 409                 mdt_body_reverse_idmap(info, b);
 410
 411         if (b->valid & OBD_MD_FLSIZE)
 412                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 413                        PFID(fid), (unsigned long long)b->size);
 414 }
 415
 416 static inline int mdt_body_has_lov(const struct lu_attr *la,
 417                                    const struct mdt_body *body)
 418 {
 419         return ((S_ISREG(la->la_mode) && (body->valid & OBD_MD_FLEASIZE)) ||
 420                 (S_ISDIR(la->la_mode) && (body->valid & OBD_MD_FLDIREA )) );
 421 }
 422
 423 void mdt_client_compatibility(struct mdt_thread_info *info)
 424 {
 425         struct mdt_body       *body;
 426         struct ptlrpc_request *req = mdt_info_req(info);
 427         struct obd_export     *exp = req->rq_export;
 428         struct md_attr        *ma = &info->mti_attr;
 429         struct lu_attr        *la = &ma->ma_attr;
 430         ENTRY;
 431
 432         if (exp_connect_layout(exp))
 433                 /* the client can deal with 16-bit lmm_stripe_count */
 434                 RETURN_EXIT;
 435
 436         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 437
 438         if (!mdt_body_has_lov(la, body))
 439                 RETURN_EXIT;
 440
 441         /* now we have a reply with a lov for a client not compatible with the
 442          * layout lock so we have to clean the layout generation number */
 443         if (S_ISREG(la->la_mode))
 444                 ma->ma_lmm->lmm_layout_gen = 0;
 445         EXIT;
 446 }
 447
 448 static int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
 449                              char *name)
 450 {
 451         const struct lu_env *env = info->mti_env;
 452         int rc;
 453         ENTRY;
 454
 455         LASSERT(info->mti_big_lmm_used == 0);
 456         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
 457         if (rc < 0)
 458                 RETURN(rc);
 459
 460         /* big_lmm may need to be grown */
 461         if (info->mti_big_lmmsize < rc) {
 462                 int size = size_roundup_power2(rc);
 463
 464                 if (info->mti_big_lmmsize > 0) {
 465                         /* free old buffer */
 466                         LASSERT(info->mti_big_lmm);
 467                         OBD_FREE_LARGE(info->mti_big_lmm,
 468                                        info->mti_big_lmmsize);
 469                         info->mti_big_lmm = NULL;
 470                         info->mti_big_lmmsize = 0;
 471                 }
 472
 473                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
 474                 if (info->mti_big_lmm == NULL)
 475                         RETURN(-ENOMEM);
 476                 info->mti_big_lmmsize = size;
 477         }
 478         LASSERT(info->mti_big_lmmsize >= rc);
 479
 480         info->mti_buf.lb_buf = info->mti_big_lmm;
 481         info->mti_buf.lb_len = info->mti_big_lmmsize;
 482         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
 483
 484         RETURN(rc);
 485 }
 486
 487 int mdt_attr_get_lov(struct mdt_thread_info *info,
 488                      struct mdt_object *o, struct md_attr *ma)
 489 {
 490         struct md_object *next = mdt_object_child(o);
 491         struct lu_buf    *buf = &info->mti_buf;
 492         int rc;
 493
 494         buf->lb_buf = ma->ma_lmm;
 495         buf->lb_len = ma->ma_lmm_size;
 496         rc = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_LOV);
 497         if (rc > 0) {
 498                 ma->ma_lmm_size = rc;
 499                 ma->ma_valid |= MA_LOV;
 500                 rc = 0;
 501         } else if (rc == -ENODATA) {
 502                 /* no LOV EA */
 503                 rc = 0;
 504         } else if (rc == -ERANGE) {
 505                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LOV);
 506                 if (rc > 0) {
 507                         info->mti_big_lmm_used = 1;
 508                         ma->ma_valid |= MA_LOV;
 509                         ma->ma_lmm = info->mti_big_lmm;
 510                         ma->ma_lmm_size = rc;
 511                         /* update mdt_max_mdsize so all clients
 512                          * will be aware about that */
 513                         if (info->mti_mdt->mdt_max_mdsize < rc)
 514                                 info->mti_mdt->mdt_max_mdsize = rc;
 515                         rc = 0;
 516                 }
 517         }
 518
 519         return rc;
 520 }
 521
 522 int mdt_attr_get_pfid(struct mdt_thread_info *info,
 523                       struct mdt_object *o, struct lu_fid *pfid)
 524 {
 525         struct lu_buf           *buf = &info->mti_buf;
 526         struct link_ea_header   *leh;
 527         struct link_ea_entry    *lee;
 528         int                      rc;
 529         ENTRY;
 530
 531         buf->lb_buf = info->mti_big_lmm;
 532         buf->lb_len = info->mti_big_lmmsize;
 533         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
 534                           buf, XATTR_NAME_LINK);
 535         /* ignore errors, MA_PFID won't be set and it is
 536          * up to the caller to treat this as an error */
 537         if (rc == -ERANGE || buf->lb_len == 0) {
 538                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
 539                 buf->lb_buf = info->mti_big_lmm;
 540                 buf->lb_len = info->mti_big_lmmsize;
 541         }
 542
 543         if (rc < 0)
 544                 RETURN(rc);
 545         if (rc < sizeof(*leh)) {
 546                 CERROR("short LinkEA on "DFID": rc = %d\n",
 547                        PFID(mdt_object_fid(o)), rc);
 548                 RETURN(-ENODATA);
 549         }
 550
 551         leh = (struct link_ea_header *) buf->lb_buf;
 552         lee = (struct link_ea_entry *)(leh + 1);
 553         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
 554                 leh->leh_magic = LINK_EA_MAGIC;
 555                 leh->leh_reccount = __swab32(leh->leh_reccount);
 556                 leh->leh_len = __swab64(leh->leh_len);
 557         }
 558         if (leh->leh_magic != LINK_EA_MAGIC)
 559                 RETURN(-EINVAL);
 560         if (leh->leh_reccount == 0)
 561                 RETURN(-ENODATA);
 562
 563         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
 564         fid_be_to_cpu(pfid, pfid);
 565
 566         RETURN(0);
 567 }
 568
 569 int mdt_attr_get_complex(struct mdt_thread_info *info,
 570                          struct mdt_object *o, struct md_attr *ma)
 571 {
 572         const struct lu_env *env = info->mti_env;
 573         struct md_object    *next = mdt_object_child(o);
 574         struct lu_buf       *buf = &info->mti_buf;
 575         u32                  mode = lu_object_attr(&next->mo_lu);
 576         int                  need = ma->ma_need;
 577         int                  rc = 0, rc2;
 578         ENTRY;
 579
 580         ma->ma_valid = 0;
 581
 582         if (need & MA_INODE) {
 583                 ma->ma_need = MA_INODE;
 584                 rc = mo_attr_get(env, next, ma);
 585                 if (rc)
 586                         GOTO(out, rc);
 587                 ma->ma_valid |= MA_INODE;
 588         }
 589
 590         if (need & MA_PFID) {
 591                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
 592                 if (rc == 0)
 593                         ma->ma_valid |= MA_PFID;
 594                 /* ignore this error, parent fid is not mandatory */
 595                 rc = 0;
 596         }
 597
 598         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
 599                 rc = mdt_attr_get_lov(info, o, ma);
 600                 if (rc)
 601                         GOTO(out, rc);
 602         }
 603
 604         if (need & MA_LMV && S_ISDIR(mode)) {
 605                 buf->lb_buf = ma->ma_lmv;
 606                 buf->lb_len = ma->ma_lmv_size;
 607                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_LMV);
 608                 if (rc2 > 0) {
 609                         ma->ma_lmv_size = rc2;
 610                         ma->ma_valid |= MA_LMV;
 611                 } else if (rc2 == -ENODATA) {
 612                         /* no LMV EA */
 613                         ma->ma_lmv_size = 0;
 614                 } else
 615                         GOTO(out, rc = rc2);
 616         }
 617
 618         if (need & MA_SOM && S_ISREG(mode)) {
 619                 buf->lb_buf = info->mti_xattr_buf;
 620                 buf->lb_len = sizeof(info->mti_xattr_buf);
 621                 CLASSERT(sizeof(struct som_attrs) <=
 622                          sizeof(info->mti_xattr_buf));
 623                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_SOM);
 624                 rc2 = lustre_buf2som(info->mti_xattr_buf, rc2, ma->ma_som);
 625                 if (rc2 == 0)
 626                         ma->ma_valid |= MA_SOM;
 627                 else if (rc2 < 0 && rc2 != -ENODATA)
 628                         GOTO(out, rc = rc2);
 629         }
 630
 631         if (need & MA_HSM && S_ISREG(mode)) {
 632                 buf->lb_buf = info->mti_xattr_buf;
 633                 buf->lb_len = sizeof(info->mti_xattr_buf);
 634                 CLASSERT(sizeof(struct hsm_attrs) <=
 635                          sizeof(info->mti_xattr_buf));
 636                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
 637                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
 638                 if (rc2 == 0)
 639                         ma->ma_valid |= MA_HSM;
 640                 else if (rc2 < 0 && rc2 != -ENODATA)
 641                         GOTO(out, rc = rc2);
 642         }
 643
 644 #ifdef CONFIG_FS_POSIX_ACL
 645         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
 646                 buf->lb_buf = ma->ma_acl;
 647                 buf->lb_len = ma->ma_acl_size;
 648                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
 649                 if (rc2 > 0) {
 650                         ma->ma_acl_size = rc2;
 651                         ma->ma_valid |= MA_ACL_DEF;
 652                 } else if (rc2 == -ENODATA) {
 653                         /* no ACLs */
 654                         ma->ma_acl_size = 0;
 655                 } else
 656                         GOTO(out, rc = rc2);
 657         }
 658 #endif
 659 out:
 660         ma->ma_need = need;
 661         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
 662                rc, ma->ma_valid, ma->ma_lmm);
 663         RETURN(rc);
 664 }
 665
 666 static int mdt_getattr_internal(struct mdt_thread_info *info,
 667                                 struct mdt_object *o, int ma_need)
 668 {
 669         struct md_object        *next = mdt_object_child(o);
 670         const struct mdt_body   *reqbody = info->mti_body;
 671         struct ptlrpc_request   *req = mdt_info_req(info);
 672         struct md_attr          *ma = &info->mti_attr;
 673         struct lu_attr          *la = &ma->ma_attr;
 674         struct req_capsule      *pill = info->mti_pill;
 675         const struct lu_env     *env = info->mti_env;
 676         struct mdt_body         *repbody;
 677         struct lu_buf           *buffer = &info->mti_buf;
 678         int                     rc;
 679         int                     is_root;
 680         ENTRY;
 681
 682         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
 683                 RETURN(err_serious(-ENOMEM));
 684
 685         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 686
 687         ma->ma_valid = 0;
 688
 689         if (mdt_object_remote(o)) {
 690                 /* This object is located on remote node.*/
 691                 /* Return -EIO for old client */
 692                 if (!mdt_is_dne_client(req->rq_export))
 693                         GOTO(out, rc = -EIO);
 694
 695                 repbody->fid1 = *mdt_object_fid(o);
 696                 repbody->valid = OBD_MD_FLID | OBD_MD_MDS;
 697                 GOTO(out, rc = 0);
 698         }
 699
 700         buffer->lb_len = reqbody->eadatasize;
 701         if (buffer->lb_len > 0)
 702                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
 703         else
 704                 buffer->lb_buf = NULL;
 705
 706         /* If it is dir object and client require MEA, then we got MEA */
 707         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 708             reqbody->valid & OBD_MD_MEA) {
 709                 /* Assumption: MDT_MD size is enough for lmv size. */
 710                 ma->ma_lmv = buffer->lb_buf;
 711                 ma->ma_lmv_size = buffer->lb_len;
 712                 ma->ma_need = MA_LMV | MA_INODE;
 713         } else {
 714                 ma->ma_lmm = buffer->lb_buf;
 715                 ma->ma_lmm_size = buffer->lb_len;
 716                 ma->ma_need = MA_LOV | MA_INODE | MA_HSM;
 717         }
 718
 719         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 720             reqbody->valid & OBD_MD_FLDIREA  &&
 721             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
 722                 /* get default stripe info for this dir. */
 723                 ma->ma_need |= MA_LOV_DEF;
 724         }
 725         ma->ma_need |= ma_need;
 726         if (ma->ma_need & MA_SOM)
 727                 ma->ma_som = &info->mti_u.som.data;
 728
 729         rc = mdt_attr_get_complex(info, o, ma);
 730         if (unlikely(rc)) {
 731                 CERROR("%s: getattr error for "DFID": rc = %d\n",
 732                        mdt_obd_name(info->mti_mdt),
 733                        PFID(mdt_object_fid(o)), rc);
 734                 RETURN(rc);
 735         }
 736
 737         is_root = lu_fid_eq(mdt_object_fid(o), &info->mti_mdt->mdt_md_root_fid);
 738
 739         /* the Lustre protocol supposes to return default striping
 740          * on the user-visible root if explicitly requested */
 741         if ((ma->ma_valid & MA_LOV) == 0 && S_ISDIR(la->la_mode) &&
 742             (ma->ma_need & MA_LOV_DEF && is_root) && (ma->ma_need & MA_LOV)) {
 743                 struct lu_fid      rootfid;
 744                 struct mdt_object *root;
 745                 struct mdt_device *mdt = info->mti_mdt;
 746
 747                 rc = dt_root_get(env, mdt->mdt_bottom, &rootfid);
 748                 if (rc)
 749                         RETURN(rc);
 750                 root = mdt_object_find(env, mdt, &rootfid);
 751                 if (IS_ERR(root))
 752                         RETURN(PTR_ERR(root));
 753                 rc = mdt_attr_get_lov(info, root, ma);
 754                 mdt_object_put(info->mti_env, root);
 755                 if (unlikely(rc)) {
 756                         CERROR("%s: getattr error for "DFID": rc = %d\n",
 757                                mdt_obd_name(info->mti_mdt),
 758                                PFID(mdt_object_fid(o)), rc);
 759                         RETURN(rc);
 760                 }
 761         }
 762
 763         if (likely(ma->ma_valid & MA_INODE))
 764                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
 765         else
 766                 RETURN(-EFAULT);
 767
 768         if (mdt_body_has_lov(la, reqbody)) {
 769                 if (ma->ma_valid & MA_LOV) {
 770                         LASSERT(ma->ma_lmm_size);
 771                         mdt_dump_lmm(D_INFO, ma->ma_lmm);
 772                         repbody->eadatasize = ma->ma_lmm_size;
 773                         if (S_ISDIR(la->la_mode))
 774                                 repbody->valid |= OBD_MD_FLDIREA;
 775                         else
 776                                 repbody->valid |= OBD_MD_FLEASIZE;
 777                 }
 778                 if (ma->ma_valid & MA_LMV) {
 779                         LASSERT(S_ISDIR(la->la_mode));
 780                         repbody->eadatasize = ma->ma_lmv_size;
 781                         repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
 782                 }
 783         } else if (S_ISLNK(la->la_mode) &&
 784                    reqbody->valid & OBD_MD_LINKNAME) {
 785                 buffer->lb_buf = ma->ma_lmm;
 786                 /* eadatasize from client includes NULL-terminator, so
 787                  * there is no need to read it */
 788                 buffer->lb_len = reqbody->eadatasize - 1;
 789                 rc = mo_readlink(env, next, buffer);
 790                 if (unlikely(rc <= 0)) {
 791                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
 792                                mdt_obd_name(info->mti_mdt),
 793                                PFID(mdt_object_fid(o)), rc);
 794                         rc = -EFAULT;
 795                 } else {
 796                         int print_limit = min_t(int, PAGE_CACHE_SIZE - 128, rc);
 797
 798                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
 799                                 rc -= 2;
 800                         repbody->valid |= OBD_MD_LINKNAME;
 801                         /* we need to report back size with NULL-terminator
 802                          * because client expects that */
 803                         repbody->eadatasize = rc + 1;
 804                         if (repbody->eadatasize != reqbody->eadatasize)
 805                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
 806                                        "on "DFID ", expected %d\n",
 807                                        mdt_obd_name(info->mti_mdt),
 808                                        rc, PFID(mdt_object_fid(o)),
 809                                        reqbody->eadatasize - 1);
 810                         /* NULL terminate */
 811                         ((char *)ma->ma_lmm)[rc] = 0;
 812
 813                         /* If the total CDEBUG() size is larger than a page, it
 814                          * will print a warning to the console, avoid this by
 815                          * printing just the last part of the symlink. */
 816                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
 817                                print_limit < rc ? "..." : "", print_limit,
 818                                (char *)ma->ma_lmm + rc - print_limit, rc);
 819                         rc = 0;
 820                 }
 821         }
 822
 823         if (reqbody->valid & OBD_MD_FLMODEASIZE) {
 824                 repbody->max_cookiesize = 0;
 825                 repbody->max_mdsize = info->mti_mdt->mdt_max_mdsize;
 826                 repbody->valid |= OBD_MD_FLMODEASIZE;
 827                 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
 828                        "MAX_COOKIE to : %d:%d\n", repbody->max_mdsize,
 829                        repbody->max_cookiesize);
 830         }
 831
 832         if (exp_connect_rmtclient(info->mti_exp) &&
 833             reqbody->valid & OBD_MD_FLRMTPERM) {
 834                 void *buf = req_capsule_server_get(pill, &RMF_ACL);
 835
 836                 /* mdt_getattr_lock only */
 837                 rc = mdt_pack_remote_perm(info, o, buf);
 838                 if (rc) {
 839                         repbody->valid &= ~OBD_MD_FLRMTPERM;
 840                         repbody->aclsize = 0;
 841                         RETURN(rc);
 842                 } else {
 843                         repbody->valid |= OBD_MD_FLRMTPERM;
 844                         repbody->aclsize = sizeof(struct mdt_remote_perm);
 845                 }
 846         }
 847 #ifdef CONFIG_FS_POSIX_ACL
 848         else if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
 849                  (reqbody->valid & OBD_MD_FLACL)) {
 850                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
 851                 buffer->lb_len = req_capsule_get_size(pill,
 852                                                       &RMF_ACL, RCL_SERVER);
 853                 if (buffer->lb_len > 0) {
 854                         rc = mo_xattr_get(env, next, buffer,
 855                                           XATTR_NAME_ACL_ACCESS);
 856                         if (rc < 0) {
 857                                 if (rc == -ENODATA) {
 858                                         repbody->aclsize = 0;
 859                                         repbody->valid |= OBD_MD_FLACL;
 860                                         rc = 0;
 861                                 } else if (rc == -EOPNOTSUPP) {
 862                                         rc = 0;
 863                                 } else {
 864                                         CERROR("%s: unable to read "DFID
 865                                                " ACL: rc = %d\n",
 866                                                mdt_obd_name(info->mti_mdt),
 867                                                PFID(mdt_object_fid(o)), rc);
 868                                 }
 869                         } else {
 870                                 repbody->aclsize = rc;
 871                                 repbody->valid |= OBD_MD_FLACL;
 872                                 rc = 0;
 873                         }
 874                 }
 875         }
 876 #endif
 877
 878         if (reqbody->valid & OBD_MD_FLMDSCAPA &&
 879             info->mti_mdt->mdt_opts.mo_mds_capa &&
 880             exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) {
 881                 struct lustre_capa *capa;
 882
 883                 capa = req_capsule_server_get(pill, &RMF_CAPA1);
 884                 LASSERT(capa);
 885                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 886                 rc = mo_capa_get(env, next, capa, 0);
 887                 if (rc)
 888                         RETURN(rc);
 889                 repbody->valid |= OBD_MD_FLMDSCAPA;
 890         }
 891
 892 out:
 893         if (rc == 0)
 894                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
 895
 896         RETURN(rc);
 897 }
 898
 899 static int mdt_renew_capa(struct mdt_thread_info *info)
 900 {
 901         struct mdt_object  *obj = info->mti_object;
 902         struct mdt_body    *body;
 903         struct lustre_capa *capa, *c;
 904         int rc;
 905         ENTRY;
 906
 907         /* if object doesn't exist, or server has disabled capability,
 908          * return directly, client will find body->valid OBD_MD_FLOSSCAPA
 909          * flag not set.
 910          */
 911         if (!obj || !info->mti_mdt->mdt_opts.mo_oss_capa ||
 912             !(exp_connect_flags(info->mti_exp) & OBD_CONNECT_OSS_CAPA))
 913                 RETURN(0);
 914
 915         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 916         LASSERT(body != NULL);
 917
 918         c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
 919         LASSERT(c);
 920
 921         capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
 922         LASSERT(capa);
 923
 924         *capa = *c;
 925         rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
 926         if (rc == 0)
 927                 body->valid |= OBD_MD_FLOSSCAPA;
 928         RETURN(rc);
 929 }
 930
 931 int mdt_getattr(struct mdt_thread_info *info)
 932 {
 933         struct mdt_object       *obj = info->mti_object;
 934         struct req_capsule      *pill = info->mti_pill;
 935         struct mdt_body         *reqbody;
 936         struct mdt_body         *repbody;
 937         mode_t                   mode;
 938         int rc, rc2;
 939         ENTRY;
 940
 941         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
 942         LASSERT(reqbody);
 943
 944         if (reqbody->valid & OBD_MD_FLOSSCAPA) {
 945                 rc = req_capsule_server_pack(pill);
 946                 if (unlikely(rc))
 947                         RETURN(err_serious(rc));
 948                 rc = mdt_renew_capa(info);
 949                 GOTO(out_shrink, rc);
 950         }
 951
 952         LASSERT(obj != NULL);
 953         LASSERT(lu_object_assert_exists(&obj->mot_obj.mo_lu));
 954
 955         mode = lu_object_attr(&obj->mot_obj.mo_lu);
 956
 957         /* old clients may not report needed easize, use max value then */
 958         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
 959                              reqbody->eadatasize == 0 ?
 960                              info->mti_mdt->mdt_max_mdsize :
 961                              reqbody->eadatasize);
 962
 963         rc = req_capsule_server_pack(pill);
 964         if (unlikely(rc != 0))
 965                 RETURN(err_serious(rc));
 966
 967         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 968         LASSERT(repbody != NULL);
 969         repbody->eadatasize = 0;
 970         repbody->aclsize = 0;
 971
 972         if (reqbody->valid & OBD_MD_FLRMTPERM)
 973                 rc = mdt_init_ucred(info, reqbody);
 974         else
 975                 rc = mdt_check_ucred(info);
 976         if (unlikely(rc))
 977                 GOTO(out_shrink, rc);
 978
 979         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
 980
 981         /*
 982          * Don't check capability at all, because rename might getattr for
 983          * remote obj, and at that time no capability is available.
 984          */
 985         mdt_set_capainfo(info, 1, &reqbody->fid1, BYPASS_CAPA);
 986         rc = mdt_getattr_internal(info, obj, 0);
 987         if (reqbody->valid & OBD_MD_FLRMTPERM)
 988                 mdt_exit_ucred(info);
 989         EXIT;
 990 out_shrink:
 991         mdt_client_compatibility(info);
 992         rc2 = mdt_fix_reply(info);
 993         if (rc == 0)
 994                 rc = rc2;
 995         return rc;
 996 }
 997
 998 int mdt_is_subdir(struct mdt_thread_info *info)
 999 {
1000         struct mdt_object     *o = info->mti_object;
1001         struct req_capsule    *pill = info->mti_pill;
1002         const struct mdt_body *body = info->mti_body;
1003         struct mdt_body       *repbody;
1004         int                    rc;
1005         ENTRY;
1006
1007         LASSERT(o != NULL);
1008
1009         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1010
1011         /*
1012          * We save last checked parent fid to @repbody->fid1 for remote
1013          * directory case.
1014          */
1015         LASSERT(fid_is_sane(&body->fid2));
1016         LASSERT(mdt_object_exists(o) && !mdt_object_remote(o));
1017         rc = mdo_is_subdir(info->mti_env, mdt_object_child(o),
1018                            &body->fid2, &repbody->fid1);
1019         if (rc == 0 || rc == -EREMOTE)
1020                 repbody->valid |= OBD_MD_FLID;
1021
1022         RETURN(rc);
1023 }
1024
1025 int mdt_swap_layouts(struct mdt_thread_info *info)
1026 {
1027         struct ptlrpc_request   *req = mdt_info_req(info);
1028         struct obd_export       *exp = req->rq_export;
1029         struct mdt_object       *o1, *o2, *o;
1030         struct mdt_lock_handle  *lh1, *lh2;
1031         struct mdc_swap_layouts *msl;
1032         int                      rc;
1033         ENTRY;
1034
1035         /* client does not support layout lock, so layout swaping
1036          * is disabled.
1037          * FIXME: there is a problem for old clients which don't support
1038          * layout lock yet. If those clients have already opened the file
1039          * they won't be notified at all so that old layout may still be
1040          * used to do IO. This can be fixed after file release is landed by
1041          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1042         if (!exp_connect_layout(exp))
1043                 RETURN(-EOPNOTSUPP);
1044
1045         if (req_capsule_get_size(info->mti_pill, &RMF_CAPA1, RCL_CLIENT))
1046                 mdt_set_capainfo(info, 0, &info->mti_body->fid1,
1047                                  req_capsule_client_get(info->mti_pill,
1048                                                         &RMF_CAPA1));
1049
1050         if (req_capsule_get_size(info->mti_pill, &RMF_CAPA2, RCL_CLIENT))
1051                 mdt_set_capainfo(info, 1, &info->mti_body->fid2,
1052                                  req_capsule_client_get(info->mti_pill,
1053                                                         &RMF_CAPA2));
1054
1055         o1 = info->mti_object;
1056         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1057                                 &info->mti_body->fid2);
1058         if (IS_ERR(o))
1059                 GOTO(out, rc = PTR_ERR(o));
1060
1061         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1062                 GOTO(put, rc = -ENOENT);
1063
1064         rc = lu_fid_cmp(&info->mti_body->fid1, &info->mti_body->fid2);
1065         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1066                 GOTO(put, rc);
1067
1068         if (rc < 0)
1069                 swap(o1, o2);
1070
1071         /* permission check. Make sure the calling process having permission
1072          * to write both files. */
1073         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1074                                 MAY_WRITE);
1075         if (rc < 0)
1076                 GOTO(put, rc);
1077
1078         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1079                                 MAY_WRITE);
1080         if (rc < 0)
1081                 GOTO(put, rc);
1082
1083         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1084         if (msl == NULL)
1085                 GOTO(put, rc = -EPROTO);
1086
1087         lh1 = &info->mti_lh[MDT_LH_NEW];
1088         mdt_lock_reg_init(lh1, LCK_EX);
1089         lh2 = &info->mti_lh[MDT_LH_OLD];
1090         mdt_lock_reg_init(lh2, LCK_EX);
1091
1092         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT,
1093                              MDT_LOCAL_LOCK);
1094         if (rc < 0)
1095                 GOTO(put, rc);
1096
1097         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT,
1098                              MDT_LOCAL_LOCK);
1099         if (rc < 0)
1100                 GOTO(unlock1, rc);
1101
1102         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1103                              mdt_object_child(o2), msl->msl_flags);
1104         GOTO(unlock2, rc);
1105 unlock2:
1106         mdt_object_unlock(info, o2, lh2, rc);
1107 unlock1:
1108         mdt_object_unlock(info, o1, lh1, rc);
1109 put:
1110         mdt_object_put(info->mti_env, o);
1111 out:
1112         RETURN(rc);
1113 }
1114
1115 static int mdt_raw_lookup(struct mdt_thread_info *info,
1116                           struct mdt_object *parent,
1117                           const struct lu_name *lname,
1118                           struct ldlm_reply *ldlm_rep)
1119 {
1120         struct md_object *next = mdt_object_child(info->mti_object);
1121         const struct mdt_body *reqbody = info->mti_body;
1122         struct lu_fid *child_fid = &info->mti_tmp_fid1;
1123         struct mdt_body *repbody;
1124         int rc;
1125         ENTRY;
1126
1127         if (reqbody->valid != OBD_MD_FLID)
1128                 RETURN(0);
1129
1130         LASSERT(!info->mti_cross_ref);
1131
1132         /* Only got the fid of this obj by name */
1133         fid_zero(child_fid);
1134         rc = mdo_lookup(info->mti_env, next, lname, child_fid,
1135                         &info->mti_spec);
1136 #if 0
1137         /* XXX is raw_lookup possible as intent operation? */
1138         if (rc != 0) {
1139                 if (rc == -ENOENT)
1140                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1141                 RETURN(rc);
1142         } else
1143                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1144
1145         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1146 #endif
1147         if (rc == 0) {
1148                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1149                 repbody->fid1 = *child_fid;
1150                 repbody->valid = OBD_MD_FLID;
1151         }
1152         RETURN(1);
1153 }
1154
1155 /*
1156  * UPDATE lock should be taken against parent, and be release before exit;
1157  * child_bits lock should be taken against child, and be returned back:
1158  *            (1)normal request should release the child lock;
1159  *            (2)intent request will grant the lock to client.
1160  */
1161 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1162                                  struct mdt_lock_handle *lhc,
1163                                  __u64 child_bits,
1164                                  struct ldlm_reply *ldlm_rep)
1165 {
1166         struct ptlrpc_request  *req       = mdt_info_req(info);
1167         struct mdt_body        *reqbody   = NULL;
1168         struct mdt_object      *parent    = info->mti_object;
1169         struct mdt_object      *child;
1170         struct md_object       *next      = mdt_object_child(parent);
1171         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1172         struct lu_name         *lname     = NULL;
1173         const char             *name      = NULL;
1174         int                     namelen   = 0;
1175         struct mdt_lock_handle *lhp       = NULL;
1176         struct ldlm_lock       *lock;
1177         struct ldlm_res_id     *res_id;
1178         int                     is_resent;
1179         int                     ma_need = 0;
1180         int                     rc;
1181
1182         ENTRY;
1183
1184         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1185         LASSERT(ergo(is_resent,
1186                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1187
1188         LASSERT(parent != NULL);
1189         name = req_capsule_client_get(info->mti_pill, &RMF_NAME);
1190         if (name == NULL)
1191                 RETURN(err_serious(-EFAULT));
1192
1193         namelen = req_capsule_get_size(info->mti_pill, &RMF_NAME,
1194                                        RCL_CLIENT) - 1;
1195         if (!info->mti_cross_ref) {
1196                 /*
1197                  * XXX: Check for "namelen == 0" is for getattr by fid
1198                  * (OBD_CONNECT_ATTRFID), otherwise do not allow empty name,
1199                  * that is the name must contain at least one character and
1200                  * the terminating '\0'
1201                  */
1202                 if (namelen == 0) {
1203                         reqbody = req_capsule_client_get(info->mti_pill,
1204                                                          &RMF_MDT_BODY);
1205                         if (unlikely(reqbody == NULL))
1206                                 RETURN(err_serious(-EFAULT));
1207
1208                         if (unlikely(!fid_is_sane(&reqbody->fid2)))
1209                                 RETURN(err_serious(-EINVAL));
1210
1211                         name = NULL;
1212                         CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1213                                "ldlm_rep = %p\n",
1214                                PFID(mdt_object_fid(parent)),
1215                                PFID(&reqbody->fid2), ldlm_rep);
1216                 } else {
1217                         lname = mdt_name(info->mti_env, (char *)name, namelen);
1218                         CDEBUG(D_INODE, "getattr with lock for "DFID"/%s, "
1219                                "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1220                                name, ldlm_rep);
1221                 }
1222         }
1223         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1224
1225         if (unlikely(!mdt_object_exists(parent))) {
1226                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1227                                 &parent->mot_obj.mo_lu,
1228                                 "Parent doesn't exist!\n");
1229                 RETURN(-ESTALE);
1230         } else if (!info->mti_cross_ref) {
1231                 LASSERTF(!mdt_object_remote(parent),
1232                          "Parent "DFID" is on remote server\n",
1233                          PFID(mdt_object_fid(parent)));
1234         }
1235         if (lname) {
1236                 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1237                 if (rc != 0) {
1238                         if (rc > 0)
1239                                 rc = 0;
1240                         RETURN(rc);
1241                 }
1242         }
1243
1244         if (info->mti_cross_ref) {
1245                 /* Only getattr on the child. Parent is on another node. */
1246                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1247                 child = parent;
1248                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1249                        "ldlm_rep=%p\n", PFID(mdt_object_fid(child)), ldlm_rep);
1250
1251                 if (is_resent) {
1252                         /* Do not take lock for resent case. */
1253                         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1254                         LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
1255                                  lhc->mlh_reg_lh.cookie);
1256                         LASSERT(fid_res_name_eq(mdt_object_fid(child),
1257                                                 &lock->l_resource->lr_name));
1258                         LDLM_LOCK_PUT(lock);
1259                         rc = 0;
1260                 } else {
1261                         mdt_lock_handle_init(lhc);
1262                         mdt_lock_reg_init(lhc, LCK_PR);
1263
1264                         /*
1265                          * Object's name is on another MDS, no lookup lock is
1266                          * needed here but update is.
1267                          */
1268                         child_bits &= ~MDS_INODELOCK_LOOKUP;
1269                         child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1270
1271                         rc = mdt_object_lock(info, child, lhc, child_bits,
1272                                              MDT_LOCAL_LOCK);
1273                 }
1274                 if (rc == 0) {
1275                         /* Finally, we can get attr for child. */
1276                         mdt_set_capainfo(info, 0, mdt_object_fid(child),
1277                                          BYPASS_CAPA);
1278                         rc = mdt_getattr_internal(info, child, 0);
1279                         if (unlikely(rc != 0))
1280                                 mdt_object_unlock(info, child, lhc, 1);
1281                 }
1282                 RETURN(rc);
1283         }
1284
1285         if (lname) {
1286                 /* step 1: lock parent only if parent is a directory */
1287                 if (S_ISDIR(lu_object_attr(&parent->mot_obj.mo_lu))) {
1288                         lhp = &info->mti_lh[MDT_LH_PARENT];
1289                         mdt_lock_pdo_init(lhp, LCK_PR, name, namelen);
1290                         rc = mdt_object_lock(info, parent, lhp,
1291                                              MDS_INODELOCK_UPDATE,
1292                                              MDT_LOCAL_LOCK);
1293                         if (unlikely(rc != 0))
1294                                 RETURN(rc);
1295                 }
1296
1297                 /* step 2: lookup child's fid by name */
1298                 fid_zero(child_fid);
1299                 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
1300                                 &info->mti_spec);
1301
1302                 if (rc != 0) {
1303                         if (rc == -ENOENT)
1304                                 mdt_set_disposition(info, ldlm_rep,
1305                                                     DISP_LOOKUP_NEG);
1306                         GOTO(out_parent, rc);
1307                 } else
1308                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1309         } else {
1310                 *child_fid = reqbody->fid2;
1311                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1312         }
1313
1314         /*
1315          *step 3: find the child object by fid & lock it.
1316          *        regardless if it is local or remote.
1317          */
1318         child = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
1319
1320         if (unlikely(IS_ERR(child)))
1321                 GOTO(out_parent, rc = PTR_ERR(child));
1322         if (is_resent) {
1323                 /* Do not take lock for resent case. */
1324                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1325                 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
1326                          lhc->mlh_reg_lh.cookie);
1327
1328                 res_id = &lock->l_resource->lr_name;
1329                 if (!fid_res_name_eq(mdt_object_fid(child),
1330                                     &lock->l_resource->lr_name)) {
1331                          LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
1332                                                  &lock->l_resource->lr_name),
1333                                  "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
1334                                  (unsigned long)res_id->name[0],
1335                                  (unsigned long)res_id->name[1],
1336                                  (unsigned long)res_id->name[2],
1337                                  PFID(mdt_object_fid(parent)));
1338                           CWARN("Although resent, but still not get child lock"
1339                                 "parent:"DFID" child:"DFID"\n",
1340                                 PFID(mdt_object_fid(parent)),
1341                                 PFID(mdt_object_fid(child)));
1342                           lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
1343                           LDLM_LOCK_PUT(lock);
1344                           GOTO(relock, 0);
1345                 }
1346                 LDLM_LOCK_PUT(lock);
1347                 rc = 0;
1348         } else {
1349                 bool try_layout = false;
1350
1351 relock:
1352                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout*2);
1353                 mdt_lock_handle_init(lhc);
1354                 mdt_lock_reg_init(lhc, LCK_PR);
1355
1356                 if (!mdt_object_exists(child)) {
1357                         LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1358                                         &child->mot_obj.mo_lu,
1359                                         "Object doesn't exist!\n");
1360                         GOTO(out_child, rc = -ENOENT);
1361                 }
1362
1363                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1364                       mdt_object_exists(child) && !mdt_object_remote(child)) {
1365                         struct md_attr *ma = &info->mti_attr;
1366
1367                         ma->ma_valid = 0;
1368                         ma->ma_need = MA_INODE;
1369                         rc = mdt_attr_get_complex(info, child, ma);
1370                         if (unlikely(rc != 0))
1371                                 GOTO(out_child, rc);
1372
1373                         /* If the file has not been changed for some time, we
1374                          * return not only a LOOKUP lock, but also an UPDATE
1375                          * lock and this might save us RPC on later STAT. For
1376                          * directories, it also let negative dentry starts
1377                          * working for this dir. */
1378                         if (ma->ma_valid & MA_INODE &&
1379                             ma->ma_attr.la_valid & LA_CTIME &&
1380                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1381                                 ma->ma_attr.la_ctime < cfs_time_current_sec())
1382                                 child_bits |= MDS_INODELOCK_UPDATE;
1383                 }
1384
1385                 /* layout lock must be granted in a best-effort way
1386                  * for IT operations */
1387                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1388                 if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1389                     exp_connect_layout(info->mti_exp) &&
1390                     S_ISREG(lu_object_attr(&child->mot_obj.mo_lu)) &&
1391                     ldlm_rep != NULL) {
1392                         /* try to grant layout lock for regular file. */
1393                         try_layout = true;
1394                 }
1395
1396                 rc = 0;
1397                 if (try_layout) {
1398                         child_bits |= MDS_INODELOCK_LAYOUT;
1399                         /* try layout lock, it may fail to be granted due to
1400                          * contention at LOOKUP or UPDATE */
1401                         if (!mdt_object_lock_try(info, child, lhc, child_bits,
1402                                                  MDT_CROSS_LOCK)) {
1403                                 child_bits &= ~MDS_INODELOCK_LAYOUT;
1404                                 LASSERT(child_bits != 0);
1405                                 rc = mdt_object_lock(info, child, lhc,
1406                                                 child_bits, MDT_CROSS_LOCK);
1407                         } else {
1408                                 ma_need |= MA_LOV;
1409                         }
1410                 } else {
1411                         rc = mdt_object_lock(info, child, lhc, child_bits,
1412                                                 MDT_CROSS_LOCK);
1413                 }
1414                 if (unlikely(rc != 0))
1415                         GOTO(out_child, rc);
1416         }
1417
1418         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1419         /* Get MA_SOM attributes if update lock is given. */
1420         if (lock &&
1421             lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_UPDATE &&
1422             S_ISREG(lu_object_attr(&mdt_object_child(child)->mo_lu)))
1423                 ma_need |= MA_SOM;
1424
1425         /* finally, we can get attr for child. */
1426         mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
1427         rc = mdt_getattr_internal(info, child, ma_need);
1428         if (unlikely(rc != 0)) {
1429                 mdt_object_unlock(info, child, lhc, 1);
1430         } else if (lock) {
1431                 /* Debugging code. */
1432                 res_id = &lock->l_resource->lr_name;
1433                 LDLM_DEBUG(lock, "Returning lock to client");
1434                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1435                                          &lock->l_resource->lr_name),
1436                          "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
1437                          (unsigned long)res_id->name[0],
1438                          (unsigned long)res_id->name[1],
1439                          (unsigned long)res_id->name[2],
1440                          PFID(mdt_object_fid(child)));
1441                 if (mdt_object_exists(child) && !mdt_object_remote(child))
1442                         mdt_pack_size2body(info, child);
1443         }
1444         if (lock)
1445                 LDLM_LOCK_PUT(lock);
1446
1447         EXIT;
1448 out_child:
1449         mdt_object_put(info->mti_env, child);
1450 out_parent:
1451         if (lhp)
1452                 mdt_object_unlock(info, parent, lhp, 1);
1453         return rc;
1454 }
1455
1456 /* normal handler: should release the child lock */
1457 int mdt_getattr_name(struct mdt_thread_info *info)
1458 {
1459         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1460         struct mdt_body        *reqbody;
1461         struct mdt_body        *repbody;
1462         int rc, rc2;
1463         ENTRY;
1464
1465         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1466         LASSERT(reqbody != NULL);
1467         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1468         LASSERT(repbody != NULL);
1469
1470         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1471         repbody->eadatasize = 0;
1472         repbody->aclsize = 0;
1473
1474         rc = mdt_init_ucred(info, reqbody);
1475         if (unlikely(rc))
1476                 GOTO(out_shrink, rc);
1477
1478         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1479         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1480                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1481                 lhc->mlh_reg_lh.cookie = 0;
1482         }
1483         mdt_exit_ucred(info);
1484         EXIT;
1485 out_shrink:
1486         mdt_client_compatibility(info);
1487         rc2 = mdt_fix_reply(info);
1488         if (rc == 0)
1489                 rc = rc2;
1490         return rc;
1491 }
1492
1493 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1494                          void *karg, void *uarg);
1495
1496 int mdt_set_info(struct mdt_thread_info *info)
1497 {
1498         struct ptlrpc_request *req = mdt_info_req(info);
1499         char *key;
1500         void *val;
1501         int keylen, vallen, rc = 0;
1502         ENTRY;
1503
1504         rc = req_capsule_server_pack(info->mti_pill);
1505         if (rc)
1506                 RETURN(rc);
1507
1508         key = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_KEY);
1509         if (key == NULL) {
1510                 DEBUG_REQ(D_HA, req, "no set_info key");
1511                 RETURN(-EFAULT);
1512         }
1513
1514         keylen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_KEY,
1515                                       RCL_CLIENT);
1516
1517         val = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_VAL);
1518         if (val == NULL) {
1519                 DEBUG_REQ(D_HA, req, "no set_info val");
1520                 RETURN(-EFAULT);
1521         }
1522
1523         vallen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_VAL,
1524                                       RCL_CLIENT);
1525
1526         /* Swab any part of val you need to here */
1527         if (KEY_IS(KEY_READ_ONLY)) {
1528                 req->rq_status = 0;
1529                 lustre_msg_set_status(req->rq_repmsg, 0);
1530
1531                 spin_lock(&req->rq_export->exp_lock);
1532                 if (*(__u32 *)val)
1533                         *exp_connect_flags_ptr(req->rq_export) |=
1534                                 OBD_CONNECT_RDONLY;
1535                 else
1536                         *exp_connect_flags_ptr(req->rq_export) &=
1537                                 ~OBD_CONNECT_RDONLY;
1538                 spin_unlock(&req->rq_export->exp_lock);
1539
1540         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1541                 struct changelog_setinfo *cs =
1542                         (struct changelog_setinfo *)val;
1543                 if (vallen != sizeof(*cs)) {
1544                         CERROR("Bad changelog_clear setinfo size %d\n", vallen);
1545                         RETURN(-EINVAL);
1546                 }
1547                 if (ptlrpc_req_need_swab(req)) {
1548                         __swab64s(&cs->cs_recno);
1549                         __swab32s(&cs->cs_id);
1550                 }
1551
1552                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, info->mti_exp,
1553                                    vallen, val, NULL);
1554                 lustre_msg_set_status(req->rq_repmsg, rc);
1555
1556         } else {
1557                 RETURN(-EINVAL);
1558         }
1559         RETURN(0);
1560 }
1561
1562 /**
1563  * Top-level handler for MDT connection requests.
1564  */
1565 int mdt_connect(struct mdt_thread_info *info)
1566 {
1567         int rc;
1568         struct obd_connect_data *reply;
1569         struct obd_export *exp;
1570         struct ptlrpc_request *req = mdt_info_req(info);
1571
1572         rc = target_handle_connect(req);
1573         if (rc != 0)
1574                 return err_serious(rc);
1575
1576         LASSERT(req->rq_export != NULL);
1577         info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
1578         rc = mdt_init_sec_level(info);
1579         if (rc != 0) {
1580                 obd_disconnect(class_export_get(req->rq_export));
1581                 return rc;
1582         }
1583
1584         /* To avoid exposing partially initialized connection flags, changes up
1585          * to this point have been staged in reply->ocd_connect_flags. Now that
1586          * connection handling has completed successfully, atomically update
1587          * the connect flags in the shared export data structure. LU-1623 */
1588         reply = req_capsule_server_get(info->mti_pill, &RMF_CONNECT_DATA);
1589         exp = req->rq_export;
1590         spin_lock(&exp->exp_lock);
1591         *exp_connect_flags_ptr(exp) = reply->ocd_connect_flags;
1592         spin_unlock(&exp->exp_lock);
1593
1594         rc = mdt_init_idmap(info);
1595         if (rc != 0)
1596                 obd_disconnect(class_export_get(req->rq_export));
1597
1598         return rc;
1599 }
1600
1601 int mdt_disconnect(struct mdt_thread_info *info)
1602 {
1603         int rc;
1604         ENTRY;
1605
1606         rc = target_handle_disconnect(mdt_info_req(info));
1607         if (rc)
1608                 rc = err_serious(rc);
1609         RETURN(rc);
1610 }
1611
1612 static int mdt_sendpage(struct mdt_thread_info *info,
1613                         struct lu_rdpg *rdpg, int nob)
1614 {
1615         struct ptlrpc_request   *req = mdt_info_req(info);
1616         struct obd_export       *exp = req->rq_export;
1617         struct ptlrpc_bulk_desc *desc;
1618         struct l_wait_info      *lwi = &info->mti_u.rdpg.mti_wait_info;
1619         int                      tmpcount;
1620         int                      tmpsize;
1621         int                      i;
1622         int                      rc;
1623         ENTRY;
1624
1625         desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, 1, BULK_PUT_SOURCE,
1626                                     MDS_BULK_PORTAL);
1627         if (desc == NULL)
1628                 RETURN(-ENOMEM);
1629
1630         if (!(exp_connect_flags(exp) & OBD_CONNECT_BRW_SIZE))
1631                 /* old client requires reply size in it's PAGE_SIZE,
1632                  * which is rdpg->rp_count */
1633                 nob = rdpg->rp_count;
1634
1635         for (i = 0, tmpcount = nob; i < rdpg->rp_npages && tmpcount > 0;
1636              i++, tmpcount -= tmpsize) {
1637                 tmpsize = min_t(int, tmpcount, PAGE_CACHE_SIZE);
1638                 ptlrpc_prep_bulk_page_pin(desc, rdpg->rp_pages[i], 0, tmpsize);
1639         }
1640
1641         LASSERT(desc->bd_nob == nob);
1642         rc = target_bulk_io(exp, desc, lwi);
1643         ptlrpc_free_bulk_pin(desc);
1644         RETURN(rc);
1645 }
1646
1647 int mdt_readpage(struct mdt_thread_info *info)
1648 {
1649         struct mdt_object *object = info->mti_object;
1650         struct lu_rdpg    *rdpg = &info->mti_u.rdpg.mti_rdpg;
1651         struct mdt_body   *reqbody;
1652         struct mdt_body   *repbody;
1653         int                rc;
1654         int                i;
1655         ENTRY;
1656
1657         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1658                 RETURN(err_serious(-ENOMEM));
1659
1660         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1661         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1662         if (reqbody == NULL || repbody == NULL)
1663                 RETURN(err_serious(-EFAULT));
1664
1665         /*
1666          * prepare @rdpg before calling lower layers and transfer itself. Here
1667          * reqbody->size contains offset of where to start to read and
1668          * reqbody->nlink contains number bytes to read.
1669          */
1670         rdpg->rp_hash = reqbody->size;
1671         if (rdpg->rp_hash != reqbody->size) {
1672                 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1673                        rdpg->rp_hash, reqbody->size);
1674                 RETURN(-EFAULT);
1675         }
1676
1677         rdpg->rp_attrs = reqbody->mode;
1678         if (exp_connect_flags(info->mti_exp) & OBD_CONNECT_64BITHASH)
1679                 rdpg->rp_attrs |= LUDA_64BITHASH;
1680         rdpg->rp_count  = min_t(unsigned int, reqbody->nlink,
1681                                 exp_max_brw_size(info->mti_exp));
1682         rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE - 1) >>
1683                           PAGE_CACHE_SHIFT;
1684         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1685         if (rdpg->rp_pages == NULL)
1686                 RETURN(-ENOMEM);
1687
1688         for (i = 0; i < rdpg->rp_npages; ++i) {
1689                 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1690                 if (rdpg->rp_pages[i] == NULL)
1691                         GOTO(free_rdpg, rc = -ENOMEM);
1692         }
1693
1694         /* call lower layers to fill allocated pages with directory data */
1695         rc = mo_readpage(info->mti_env, mdt_object_child(object), rdpg);
1696         if (rc < 0)
1697                 GOTO(free_rdpg, rc);
1698
1699         /* send pages to client */
1700         rc = mdt_sendpage(info, rdpg, rc);
1701
1702         EXIT;
1703 free_rdpg:
1704
1705         for (i = 0; i < rdpg->rp_npages; i++)
1706                 if (rdpg->rp_pages[i] != NULL)
1707                         __free_page(rdpg->rp_pages[i]);
1708         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1709
1710         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1711                 RETURN(0);
1712
1713         return rc;
1714 }
1715
1716 static int mdt_reint_internal(struct mdt_thread_info *info,
1717                               struct mdt_lock_handle *lhc,
1718                               __u32 op)
1719 {
1720         struct req_capsule      *pill = info->mti_pill;
1721         struct mdt_body         *repbody;
1722         int                      rc = 0, rc2;
1723         ENTRY;
1724
1725
1726         rc = mdt_reint_unpack(info, op);
1727         if (rc != 0) {
1728                 CERROR("Can't unpack reint, rc %d\n", rc);
1729                 RETURN(err_serious(rc));
1730         }
1731
1732         /* for replay (no_create) lmm is not needed, client has it already */
1733         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1734                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1735                                      info->mti_rr.rr_eadatalen);
1736
1737         /* llog cookies are always 0, the field is kept for compatibility */
1738         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1739                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
1740
1741         rc = req_capsule_server_pack(pill);
1742         if (rc != 0) {
1743                 CERROR("Can't pack response, rc %d\n", rc);
1744                 RETURN(err_serious(rc));
1745         }
1746
1747         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1748                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1749                 LASSERT(repbody);
1750                 repbody->eadatasize = 0;
1751                 repbody->aclsize = 0;
1752         }
1753
1754         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1755
1756         /* for replay no cookkie / lmm need, because client have this already */
1757         if (info->mti_spec.no_create)
1758                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1759                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1760
1761         rc = mdt_init_ucred_reint(info);
1762         if (rc)
1763                 GOTO(out_shrink, rc);
1764
1765         rc = mdt_fix_attr_ucred(info, op);
1766         if (rc != 0)
1767                 GOTO(out_ucred, rc = err_serious(rc));
1768
1769         if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1770                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1771                 GOTO(out_ucred, rc);
1772         }
1773         rc = mdt_reint_rec(info, lhc);
1774         EXIT;
1775 out_ucred:
1776         mdt_exit_ucred(info);
1777 out_shrink:
1778         mdt_client_compatibility(info);
1779         rc2 = mdt_fix_reply(info);
1780         if (rc == 0)
1781                 rc = rc2;
1782         return rc;
1783 }
1784
1785 static long mdt_reint_opcode(struct mdt_thread_info *info,
1786                              const struct req_format **fmt)
1787 {
1788         struct mdt_rec_reint *rec;
1789         long opc;
1790
1791         rec = req_capsule_client_get(info->mti_pill, &RMF_REC_REINT);
1792         if (rec != NULL) {
1793                 opc = rec->rr_opcode;
1794                 DEBUG_REQ(D_INODE, mdt_info_req(info), "reint opt = %ld", opc);
1795                 if (opc < REINT_MAX && fmt[opc] != NULL)
1796                         req_capsule_extend(info->mti_pill, fmt[opc]);
1797                 else {
1798                         CERROR("%s: Unsupported opcode '%ld' from client '%s': "
1799                                "rc = %d\n", mdt_obd_name(info->mti_mdt), opc,
1800                                info->mti_mdt->mdt_ldlm_client->cli_name,
1801                                -EFAULT);
1802                         opc = err_serious(-EFAULT);
1803                 }
1804         } else {
1805                 opc = err_serious(-EFAULT);
1806         }
1807         return opc;
1808 }
1809
1810 int mdt_reint(struct mdt_thread_info *info)
1811 {
1812         long opc;
1813         int  rc;
1814
1815         static const struct req_format *reint_fmts[REINT_MAX] = {
1816                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
1817                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
1818                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
1819                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
1820                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
1821                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
1822                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
1823                 [REINT_RMENTRY] = &RQF_MDS_REINT_UNLINK
1824         };
1825
1826         ENTRY;
1827
1828         opc = mdt_reint_opcode(info, reint_fmts);
1829         if (opc >= 0) {
1830                 /*
1831                  * No lock possible here from client to pass it to reint code
1832                  * path.
1833                  */
1834                 rc = mdt_reint_internal(info, NULL, opc);
1835         } else {
1836                 rc = opc;
1837         }
1838
1839         info->mti_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1840         RETURN(rc);
1841 }
1842
1843 /* this should sync the whole device */
1844 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1845 {
1846         struct dt_device *dt = mdt->mdt_bottom;
1847         int rc;
1848         ENTRY;
1849
1850         rc = dt->dd_ops->dt_sync(env, dt);
1851         RETURN(rc);
1852 }
1853
1854 /* this should sync this object */
1855 static int mdt_object_sync(struct mdt_thread_info *info)
1856 {
1857         struct md_object *next;
1858         int rc;
1859         ENTRY;
1860
1861         if (!mdt_object_exists(info->mti_object)) {
1862                 CWARN("Non existing object  "DFID"!\n",
1863                       PFID(mdt_object_fid(info->mti_object)));
1864                 RETURN(-ESTALE);
1865         }
1866         next = mdt_object_child(info->mti_object);
1867         rc = mo_object_sync(info->mti_env, next);
1868
1869         RETURN(rc);
1870 }
1871
1872 int mdt_sync(struct mdt_thread_info *info)
1873 {
1874         struct ptlrpc_request *req = mdt_info_req(info);
1875         struct req_capsule *pill = info->mti_pill;
1876         struct mdt_body *body;
1877         int rc;
1878         ENTRY;
1879
1880         /* The fid may be zero, so we req_capsule_set manually */
1881         req_capsule_set(pill, &RQF_MDS_SYNC);
1882
1883         body = req_capsule_client_get(pill, &RMF_MDT_BODY);
1884         if (body == NULL)
1885                 RETURN(err_serious(-EINVAL));
1886
1887         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1888                 RETURN(err_serious(-ENOMEM));
1889
1890         if (fid_seq(&body->fid1) == 0) {
1891                 /* sync the whole device */
1892                 rc = req_capsule_server_pack(pill);
1893                 if (rc == 0)
1894                         rc = mdt_device_sync(info->mti_env, info->mti_mdt);
1895                 else
1896                         rc = err_serious(rc);
1897         } else {
1898                 /* sync an object */
1899                 rc = mdt_unpack_req_pack_rep(info, HABEO_CORPUS|HABEO_REFERO);
1900                 if (rc == 0) {
1901                         rc = mdt_object_sync(info);
1902                         if (rc == 0) {
1903                                 const struct lu_fid *fid;
1904                                 struct lu_attr *la = &info->mti_attr.ma_attr;
1905
1906                                 info->mti_attr.ma_need = MA_INODE;
1907                                 info->mti_attr.ma_valid = 0;
1908                                 rc = mdt_attr_get_complex(info, info->mti_object,
1909                                                           &info->mti_attr);
1910                                 if (rc == 0) {
1911                                         body = req_capsule_server_get(pill,
1912                                                                 &RMF_MDT_BODY);
1913                                         fid = mdt_object_fid(info->mti_object);
1914                                         mdt_pack_attr2body(info, body, la, fid);
1915                                 }
1916                         }
1917                 } else
1918                         rc = err_serious(rc);
1919         }
1920         if (rc == 0)
1921                 mdt_counter_incr(req, LPROC_MDT_SYNC);
1922
1923         RETURN(rc);
1924 }
1925
1926 /*
1927  * Quotacheck handler.
1928  * in-kernel quotacheck isn't supported any more.
1929  */
1930 int mdt_quotacheck(struct mdt_thread_info *info)
1931 {
1932         struct obd_quotactl     *oqctl;
1933         int                      rc;
1934         ENTRY;
1935
1936         oqctl = req_capsule_client_get(info->mti_pill, &RMF_OBD_QUOTACTL);
1937         if (oqctl == NULL)
1938                 RETURN(err_serious(-EPROTO));
1939
1940         rc = req_capsule_server_pack(info->mti_pill);
1941         if (rc)
1942                 RETURN(err_serious(rc));
1943
1944         /* deprecated, not used any more */
1945         RETURN(-EOPNOTSUPP);
1946 }
1947
1948 /*
1949  * Handle quota control requests to consult current usage/limit, but also
1950  * to configure quota enforcement
1951  */
1952 int mdt_quotactl(struct mdt_thread_info *info)
1953 {
1954         struct obd_export       *exp  = info->mti_exp;
1955         struct req_capsule      *pill = info->mti_pill;
1956         struct obd_quotactl     *oqctl, *repoqc;
1957         int                      id, rc;
1958         struct lu_device        *qmt = info->mti_mdt->mdt_qmt_dev;
1959         ENTRY;
1960
1961         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1962         if (oqctl == NULL)
1963                 RETURN(err_serious(-EPROTO));
1964
1965         rc = req_capsule_server_pack(pill);
1966         if (rc)
1967                 RETURN(err_serious(rc));
1968
1969         switch (oqctl->qc_cmd) {
1970         case Q_QUOTACHECK:
1971         case LUSTRE_Q_INVALIDATE:
1972         case LUSTRE_Q_FINVALIDATE:
1973         case Q_QUOTAON:
1974         case Q_QUOTAOFF:
1975         case Q_INITQUOTA:
1976                 /* deprecated, not used any more */
1977                 RETURN(-EOPNOTSUPP);
1978                 /* master quotactl */
1979         case Q_GETINFO:
1980         case Q_SETINFO:
1981         case Q_SETQUOTA:
1982         case Q_GETQUOTA:
1983                 if (qmt == NULL)
1984                         RETURN(-EOPNOTSUPP);
1985                 /* slave quotactl */
1986         case Q_GETOINFO:
1987         case Q_GETOQUOTA:
1988                 break;
1989         default:
1990                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
1991                 RETURN(-EFAULT);
1992         }
1993
1994         /* map uid/gid for remote client */
1995         id = oqctl->qc_id;
1996         if (exp_connect_rmtclient(exp)) {
1997                 struct lustre_idmap_table *idmap;
1998
1999                 idmap = mdt_req2med(mdt_info_req(info))->med_idmap;
2000
2001                 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
2002                              oqctl->qc_cmd != Q_GETINFO))
2003                         RETURN(-EPERM);
2004
2005                 if (oqctl->qc_type == USRQUOTA)
2006                         id = lustre_idmap_lookup_uid(NULL, idmap, 0,
2007                                                      oqctl->qc_id);
2008                 else if (oqctl->qc_type == GRPQUOTA)
2009                         id = lustre_idmap_lookup_gid(NULL, idmap, 0,
2010                                                      oqctl->qc_id);
2011                 else
2012                         RETURN(-EINVAL);
2013
2014                 if (id == CFS_IDMAP_NOTFOUND) {
2015                         CDEBUG(D_QUOTA, "no mapping for id %u\n", oqctl->qc_id);
2016                         RETURN(-EACCES);
2017                 }
2018         }
2019
2020         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2021         if (repoqc == NULL)
2022                 RETURN(err_serious(-EFAULT));
2023
2024         if (oqctl->qc_id != id)
2025                 swap(oqctl->qc_id, id);
2026
2027         switch (oqctl->qc_cmd) {
2028
2029         case Q_GETINFO:
2030         case Q_SETINFO:
2031         case Q_SETQUOTA:
2032         case Q_GETQUOTA:
2033                 /* forward quotactl request to QMT */
2034                 rc = qmt_hdls.qmth_quotactl(info->mti_env, qmt, oqctl);
2035                 break;
2036
2037         case Q_GETOINFO:
2038         case Q_GETOQUOTA:
2039                 /* slave quotactl */
2040                 rc = lquotactl_slv(info->mti_env, info->mti_mdt->mdt_bottom,
2041                                    oqctl);
2042                 break;
2043
2044         default:
2045                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2046                 RETURN(-EFAULT);
2047         }
2048
2049         if (oqctl->qc_id != id)
2050                 swap(oqctl->qc_id, id);
2051
2052         *repoqc = *oqctl;
2053         RETURN(rc);
2054 }
2055
2056 /*
2057  * OBD PING and other handlers.
2058  */
2059 int mdt_obd_ping(struct mdt_thread_info *info)
2060 {
2061         int rc;
2062         ENTRY;
2063
2064         req_capsule_set(info->mti_pill, &RQF_OBD_PING);
2065
2066         rc = target_handle_ping(mdt_info_req(info));
2067         if (rc < 0)
2068                 rc = err_serious(rc);
2069         RETURN(rc);
2070 }
2071
2072 /*
2073  * OBD_IDX_READ handler
2074  */
2075 int mdt_obd_idx_read(struct mdt_thread_info *info)
2076 {
2077         struct mdt_device       *mdt = info->mti_mdt;
2078         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
2079         struct idx_info         *req_ii, *rep_ii;
2080         int                      rc, i;
2081         ENTRY;
2082
2083         memset(rdpg, 0, sizeof(*rdpg));
2084         req_capsule_set(info->mti_pill, &RQF_OBD_IDX_READ);
2085
2086         /* extract idx_info buffer from request & reply */
2087         req_ii = req_capsule_client_get(info->mti_pill, &RMF_IDX_INFO);
2088         if (req_ii == NULL || req_ii->ii_magic != IDX_INFO_MAGIC)
2089                 RETURN(err_serious(-EPROTO));
2090
2091         rc = req_capsule_server_pack(info->mti_pill);
2092         if (rc)
2093                 RETURN(err_serious(rc));
2094
2095         rep_ii = req_capsule_server_get(info->mti_pill, &RMF_IDX_INFO);
2096         if (rep_ii == NULL)
2097                 RETURN(err_serious(-EFAULT));
2098         rep_ii->ii_magic = IDX_INFO_MAGIC;
2099
2100         /* extract hash to start with */
2101         rdpg->rp_hash = req_ii->ii_hash_start;
2102
2103         /* extract requested attributes */
2104         rdpg->rp_attrs = req_ii->ii_attrs;
2105
2106         /* check that fid packed in request is valid and supported */
2107         if (!fid_is_sane(&req_ii->ii_fid))
2108                 RETURN(-EINVAL);
2109         rep_ii->ii_fid = req_ii->ii_fid;
2110
2111         /* copy flags */
2112         rep_ii->ii_flags = req_ii->ii_flags;
2113
2114         /* compute number of pages to allocate, ii_count is the number of 4KB
2115          * containers */
2116         if (req_ii->ii_count <= 0)
2117                 GOTO(out, rc = -EFAULT);
2118         rdpg->rp_count = min_t(unsigned int, req_ii->ii_count << LU_PAGE_SHIFT,
2119                                exp_max_brw_size(info->mti_exp));
2120         rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE - 1) >>
2121                                 PAGE_CACHE_SHIFT;
2122
2123         /* allocate pages to store the containers */
2124         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
2125         if (rdpg->rp_pages == NULL)
2126                 GOTO(out, rc = -ENOMEM);
2127         for (i = 0; i < rdpg->rp_npages; i++) {
2128                 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
2129                 if (rdpg->rp_pages[i] == NULL)
2130                         GOTO(out, rc = -ENOMEM);
2131         }
2132
2133         /* populate pages with key/record pairs */
2134         rc = dt_index_read(info->mti_env, mdt->mdt_bottom, rep_ii, rdpg);
2135         if (rc < 0)
2136                 GOTO(out, rc);
2137
2138         LASSERTF(rc <= rdpg->rp_count, "dt_index_read() returned more than "
2139                  "asked %d > %d\n", rc, rdpg->rp_count);
2140
2141         /* send pages to client */
2142         rc = mdt_sendpage(info, rdpg, rc);
2143
2144         GOTO(out, rc);
2145 out:
2146         if (rdpg->rp_pages) {
2147                 for (i = 0; i < rdpg->rp_npages; i++)
2148                         if (rdpg->rp_pages[i])
2149                                 __free_page(rdpg->rp_pages[i]);
2150                 OBD_FREE(rdpg->rp_pages,
2151                          rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
2152         }
2153         return rc;
2154 }
2155
2156 int mdt_obd_log_cancel(struct mdt_thread_info *info)
2157 {
2158         return err_serious(-EOPNOTSUPP);
2159 }
2160
2161 int mdt_obd_qc_callback(struct mdt_thread_info *info)
2162 {
2163         return err_serious(-EOPNOTSUPP);
2164 }
2165
2166 /*
2167  * LLOG handlers.
2168  */
2169
2170 /** clone llog ctxt from child (mdd)
2171  * This allows remote llog (replicator) access.
2172  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2173  * context was originally set up, or we can handle them directly.
2174  * I choose the latter, but that means I need any llog
2175  * contexts set up by child to be accessable by the mdt.  So we clone the
2176  * context into our context list here.
2177  */
2178 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2179                                int idx)
2180 {
2181         struct md_device  *next = mdt->mdt_child;
2182         struct llog_ctxt *ctxt;
2183         int rc;
2184
2185         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2186                 return 0;
2187
2188         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2189         if (rc || ctxt == NULL) {
2190                 return 0;
2191         }
2192
2193         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2194         if (rc)
2195                 CERROR("Can't set mdt ctxt %d\n", rc);
2196
2197         return rc;
2198 }
2199
2200 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2201                                  struct mdt_device *mdt, int idx)
2202 {
2203         struct llog_ctxt *ctxt;
2204
2205         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2206         if (ctxt == NULL)
2207                 return 0;
2208         /* Put once for the get we just did, and once for the clone */
2209         llog_ctxt_put(ctxt);
2210         llog_ctxt_put(ctxt);
2211         return 0;
2212 }
2213
2214 int mdt_llog_create(struct mdt_thread_info *info)
2215 {
2216         int rc;
2217
2218         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_CREATE);
2219         rc = llog_origin_handle_open(mdt_info_req(info));
2220         return (rc < 0 ? err_serious(rc) : rc);
2221 }
2222
2223 int mdt_llog_destroy(struct mdt_thread_info *info)
2224 {
2225         int rc;
2226
2227         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_DESTROY);
2228         rc = llog_origin_handle_destroy(mdt_info_req(info));
2229         return (rc < 0 ? err_serious(rc) : rc);
2230 }
2231
2232 int mdt_llog_read_header(struct mdt_thread_info *info)
2233 {
2234         int rc;
2235
2236         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_READ_HEADER);
2237         rc = llog_origin_handle_read_header(mdt_info_req(info));
2238         return (rc < 0 ? err_serious(rc) : rc);
2239 }
2240
2241 int mdt_llog_next_block(struct mdt_thread_info *info)
2242 {
2243         int rc;
2244
2245         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_NEXT_BLOCK);
2246         rc = llog_origin_handle_next_block(mdt_info_req(info));
2247         return (rc < 0 ? err_serious(rc) : rc);
2248 }
2249
2250 int mdt_llog_prev_block(struct mdt_thread_info *info)
2251 {
2252         int rc;
2253
2254         req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_PREV_BLOCK);
2255         rc = llog_origin_handle_prev_block(mdt_info_req(info));
2256         return (rc < 0 ? err_serious(rc) : rc);
2257 }
2258
2259
2260 /*
2261  * DLM handlers.
2262  */
2263
2264 static struct ldlm_callback_suite cbs = {
2265         .lcs_completion = ldlm_server_completion_ast,
2266         .lcs_blocking   = ldlm_server_blocking_ast,
2267         .lcs_glimpse    = ldlm_server_glimpse_ast
2268 };
2269
2270 int mdt_enqueue(struct mdt_thread_info *info)
2271 {
2272         struct ptlrpc_request *req;
2273         int rc;
2274
2275         /*
2276          * info->mti_dlm_req already contains swapped and (if necessary)
2277          * converted dlm request.
2278          */
2279         LASSERT(info->mti_dlm_req != NULL);
2280
2281         req = mdt_info_req(info);
2282         rc = ldlm_handle_enqueue0(info->mti_mdt->mdt_namespace,
2283                                   req, info->mti_dlm_req, &cbs);
2284         info->mti_fail_id = OBD_FAIL_LDLM_REPLY;
2285         return rc ? err_serious(rc) : req->rq_status;
2286 }
2287
2288 int mdt_convert(struct mdt_thread_info *info)
2289 {
2290         int rc;
2291         struct ptlrpc_request *req;
2292
2293         LASSERT(info->mti_dlm_req);
2294         req = mdt_info_req(info);
2295         rc = ldlm_handle_convert0(req, info->mti_dlm_req);
2296         return rc ? err_serious(rc) : req->rq_status;
2297 }
2298
2299 int mdt_bl_callback(struct mdt_thread_info *info)
2300 {
2301         CERROR("bl callbacks should not happen on MDS\n");
2302         LBUG();
2303         return err_serious(-EOPNOTSUPP);
2304 }
2305
2306 int mdt_cp_callback(struct mdt_thread_info *info)
2307 {
2308         CERROR("cp callbacks should not happen on MDS\n");
2309         LBUG();
2310         return err_serious(-EOPNOTSUPP);
2311 }
2312
2313 /*
2314  * sec context handlers
2315  */
2316 int mdt_sec_ctx_handle(struct mdt_thread_info *info)
2317 {
2318         int rc;
2319
2320         rc = mdt_handle_idmap(info);
2321
2322         if (unlikely(rc)) {
2323                 struct ptlrpc_request *req = mdt_info_req(info);
2324                 __u32                  opc;
2325
2326                 opc = lustre_msg_get_opc(req->rq_reqmsg);
2327                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2328                         sptlrpc_svc_ctx_invalidate(req);
2329         }
2330
2331         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2332
2333         return rc;
2334 }
2335
2336 /*
2337  * quota request handlers
2338  */
2339 int mdt_quota_dqacq(struct mdt_thread_info *info)
2340 {
2341         struct lu_device        *qmt = info->mti_mdt->mdt_qmt_dev;
2342         int                      rc;
2343         ENTRY;
2344
2345         if (qmt == NULL)
2346                 RETURN(err_serious(-EOPNOTSUPP));
2347
2348         rc = qmt_hdls.qmth_dqacq(info->mti_env, qmt, mdt_info_req(info));
2349         RETURN(rc);
2350 }
2351
2352 static struct mdt_object *mdt_obj(struct lu_object *o)
2353 {
2354         LASSERT(lu_device_is_mdt(o->lo_dev));
2355         return container_of0(o, struct mdt_object, mot_obj.mo_lu);
2356 }
2357
2358 struct mdt_object *mdt_object_new(const struct lu_env *env,
2359                                   struct mdt_device *d,
2360                                   const struct lu_fid *f)
2361 {
2362         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2363         struct lu_object *o;
2364         struct mdt_object *m;
2365         ENTRY;
2366
2367         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2368         o = lu_object_find(env, &d->mdt_md_dev.md_lu_dev, f, &conf);
2369         if (unlikely(IS_ERR(o)))
2370                 m = (struct mdt_object *)o;
2371         else
2372                 m = mdt_obj(o);
2373         RETURN(m);
2374 }
2375
2376 struct mdt_object *mdt_object_find(const struct lu_env *env,
2377                                    struct mdt_device *d,
2378                                    const struct lu_fid *f)
2379 {
2380         struct lu_object *o;
2381         struct mdt_object *m;
2382         ENTRY;
2383
2384         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2385         o = lu_object_find(env, &d->mdt_md_dev.md_lu_dev, f, NULL);
2386         if (unlikely(IS_ERR(o)))
2387                 m = (struct mdt_object *)o;
2388         else
2389                 m = mdt_obj(o);
2390         RETURN(m);
2391 }
2392
2393 /**
2394  * Asyncronous commit for mdt device.
2395  *
2396  * Pass asynchonous commit call down the MDS stack.
2397  *
2398  * \param env environment
2399  * \param mdt the mdt device
2400  */
2401 static void mdt_device_commit_async(const struct lu_env *env,
2402                                     struct mdt_device *mdt)
2403 {
2404         struct dt_device *dt = mdt->mdt_bottom;
2405         int rc;
2406
2407         rc = dt->dd_ops->dt_commit_async(env, dt);
2408         if (unlikely(rc != 0))
2409                 CWARN("async commit start failed with rc = %d", rc);
2410 }
2411
2412 /**
2413  * Mark the lock as "synchonous".
2414  *
2415  * Mark the lock to deffer transaction commit to the unlock time.
2416  *
2417  * \param lock the lock to mark as "synchonous"
2418  *
2419  * \see mdt_is_lock_sync
2420  * \see mdt_save_lock
2421  */
2422 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2423 {
2424         lock->l_ast_data = (void*)1;
2425 }
2426
2427 /**
2428  * Check whehter the lock "synchonous" or not.
2429  *
2430  * \param lock the lock to check
2431  * \retval 1 the lock is "synchonous"
2432  * \retval 0 the lock isn't "synchronous"
2433  *
2434  * \see mdt_set_lock_sync
2435  * \see mdt_save_lock
2436  */
2437 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2438 {
2439         return lock->l_ast_data != NULL;
2440 }
2441
2442 /**
2443  * Blocking AST for mdt locks.
2444  *
2445  * Starts transaction commit if in case of COS lock conflict or
2446  * deffers such a commit to the mdt_save_lock.
2447  *
2448  * \param lock the lock which blocks a request or cancelling lock
2449  * \param desc unused
2450  * \param data unused
2451  * \param flag indicates whether this cancelling or blocking callback
2452  * \retval 0
2453  * \see ldlm_blocking_ast_nocheck
2454  */
2455 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2456                      void *data, int flag)
2457 {
2458         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2459         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2460         int rc;
2461         ENTRY;
2462
2463         if (flag == LDLM_CB_CANCELING)
2464                 RETURN(0);
2465         lock_res_and_lock(lock);
2466         if (lock->l_blocking_ast != mdt_blocking_ast) {
2467                 unlock_res_and_lock(lock);
2468                 RETURN(0);
2469         }
2470         if (mdt_cos_is_enabled(mdt) &&
2471             lock->l_req_mode & (LCK_PW | LCK_EX) &&
2472             lock->l_blocking_lock != NULL &&
2473             lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
2474                 mdt_set_lock_sync(lock);
2475         }
2476         rc = ldlm_blocking_ast_nocheck(lock);
2477
2478         /* There is no lock conflict if l_blocking_lock == NULL,
2479          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2480          * when the last reference to a local lock was released */
2481         if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2482                 struct lu_env env;
2483
2484                 rc = lu_env_init(&env, LCT_LOCAL);
2485                 if (unlikely(rc != 0))
2486                         CWARN("lu_env initialization failed with rc = %d,"
2487                               "cannot start asynchronous commit\n", rc);
2488                 else
2489                         mdt_device_commit_async(&env, mdt);
2490                 lu_env_fini(&env);
2491         }
2492         RETURN(rc);
2493 }
2494
2495 int mdt_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2496                         void *data, int flag)
2497 {
2498         struct lustre_handle lockh;
2499         int               rc;
2500
2501         switch (flag) {
2502         case LDLM_CB_BLOCKING:
2503                 ldlm_lock2handle(lock, &lockh);
2504                 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
2505                 if (rc < 0) {
2506                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2507                         RETURN(rc);
2508                 }
2509                 break;
2510         case LDLM_CB_CANCELING:
2511                 LDLM_DEBUG(lock, "Revoke remote lock\n");
2512                 break;
2513         default:
2514                 LBUG();
2515         }
2516         RETURN(0);
2517 }
2518
2519 int mdt_remote_object_lock(struct mdt_thread_info *mti,
2520                            struct mdt_object *o, struct lustre_handle *lh,
2521                            ldlm_mode_t mode, __u64 ibits)
2522 {
2523         struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
2524         ldlm_policy_data_t *policy = &mti->mti_policy;
2525         int rc = 0;
2526         ENTRY;
2527
2528         LASSERT(mdt_object_remote(o));
2529
2530         LASSERT((ibits & MDS_INODELOCK_UPDATE));
2531
2532         memset(einfo, 0, sizeof(*einfo));
2533         einfo->ei_type = LDLM_IBITS;
2534         einfo->ei_mode = mode;
2535         einfo->ei_cb_bl = mdt_md_blocking_ast;
2536         einfo->ei_cb_cp = ldlm_completion_ast;
2537
2538         memset(policy, 0, sizeof(*policy));
2539         policy->l_inodebits.bits = ibits;
2540
2541         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2542                             policy);
2543         RETURN(rc);
2544 }
2545
2546 static int mdt_object_lock0(struct mdt_thread_info *info, struct mdt_object *o,
2547                             struct mdt_lock_handle *lh, __u64 ibits,
2548                             bool nonblock, int locality)
2549 {
2550         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2551         ldlm_policy_data_t *policy = &info->mti_policy;
2552         struct ldlm_res_id *res_id = &info->mti_res_id;
2553         __u64 dlmflags;
2554         int rc;
2555         ENTRY;
2556
2557         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2558         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2559         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2560         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2561
2562         if (mdt_object_remote(o)) {
2563                 if (locality == MDT_CROSS_LOCK) {
2564                         ibits &= ~(MDS_INODELOCK_UPDATE | MDS_INODELOCK_PERM);
2565                         ibits |= MDS_INODELOCK_LOOKUP;
2566                 } else {
2567                         LASSERTF(!(ibits &
2568                                   (MDS_INODELOCK_UPDATE | MDS_INODELOCK_PERM)),
2569                                 "%s: wrong bit "LPX64" for remote obj "DFID"\n",
2570                                 mdt_obd_name(info->mti_mdt), ibits,
2571                                 PFID(mdt_object_fid(o)));
2572                         LASSERT(ibits & MDS_INODELOCK_LOOKUP);
2573                 }
2574                 /* No PDO lock on remote object */
2575                 LASSERT(lh->mlh_type != MDT_PDO_LOCK);
2576         }
2577
2578         if (lh->mlh_type == MDT_PDO_LOCK) {
2579                 /* check for exists after object is locked */
2580                 if (mdt_object_exists(o) == 0) {
2581                         /* Non-existent object shouldn't have PDO lock */
2582                         RETURN(-ESTALE);
2583                 } else {
2584                         /* Non-dir object shouldn't have PDO lock */
2585                         if (!S_ISDIR(lu_object_attr(&o->mot_obj.mo_lu)))
2586                                 RETURN(-ENOTDIR);
2587                 }
2588         }
2589
2590         memset(policy, 0, sizeof(*policy));
2591         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2592
2593         dlmflags = LDLM_FL_ATOMIC_CB;
2594         if (nonblock)
2595                 dlmflags |= LDLM_FL_BLOCK_NOWAIT;
2596
2597         /*
2598          * Take PDO lock on whole directory and build correct @res_id for lock
2599          * on part of directory.
2600          */
2601         if (lh->mlh_pdo_hash != 0) {
2602                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2603                 mdt_lock_pdo_mode(info, o, lh);
2604                 if (lh->mlh_pdo_mode != LCK_NL) {
2605                         /*
2606                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2607                          * is never going to be sent to client and we do not
2608                          * want it slowed down due to possible cancels.
2609                          */
2610                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2611                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2612                                           policy, res_id, dlmflags,
2613                                           &info->mti_exp->exp_handle.h_cookie);
2614                         if (unlikely(rc))
2615                                 RETURN(rc);
2616                 }
2617
2618                 /*
2619                  * Finish res_id initializing by name hash marking part of
2620                  * directory which is taking modification.
2621                  */
2622                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2623         }
2624
2625         policy->l_inodebits.bits = ibits;
2626
2627         /*
2628          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2629          * going to be sent to client. If it is - mdt_intent_policy() path will
2630          * fix it up and turn FL_LOCAL flag off.
2631          */
2632         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2633                           res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2634                           &info->mti_exp->exp_handle.h_cookie);
2635         if (rc)
2636                 mdt_object_unlock(info, o, lh, 1);
2637         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2638                  lh->mlh_pdo_hash != 0 &&
2639                  (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX)) {
2640                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2641         }
2642
2643         RETURN(rc);
2644 }
2645
2646 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2647                     struct mdt_lock_handle *lh, __u64 ibits, int locality)
2648 {
2649         return mdt_object_lock0(info, o, lh, ibits, false, locality);
2650 }
2651
2652 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
2653                         struct mdt_lock_handle *lh, __u64 ibits, int locality)
2654 {
2655         struct mdt_lock_handle tmp = *lh;
2656         int rc;
2657
2658         rc = mdt_object_lock0(info, o, &tmp, ibits, true, locality);
2659         if (rc == 0)
2660                 *lh = tmp;
2661
2662         return rc == 0;
2663 }
2664
2665 /**
2666  * Save a lock within request object.
2667  *
2668  * Keep the lock referenced until whether client ACK or transaction
2669  * commit happens or release the lock immediately depending on input
2670  * parameters. If COS is ON, a write lock is converted to COS lock
2671  * before saving.
2672  *
2673  * \param info thead info object
2674  * \param h lock handle
2675  * \param mode lock mode
2676  * \param decref force immediate lock releasing
2677  */
2678 static
2679 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2680                    ldlm_mode_t mode, int decref)
2681 {
2682         ENTRY;
2683
2684         if (lustre_handle_is_used(h)) {
2685                 if (decref || !info->mti_has_trans ||
2686                     !(mode & (LCK_PW | LCK_EX))){
2687                         mdt_fid_unlock(h, mode);
2688                 } else {
2689                         struct mdt_device *mdt = info->mti_mdt;
2690                         struct ldlm_lock *lock = ldlm_handle2lock(h);
2691                         struct ptlrpc_request *req = mdt_info_req(info);
2692                         int no_ack = 0;
2693
2694                         LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2695                                  h->cookie);
2696                         CDEBUG(D_HA, "request = %p reply state = %p"
2697                                " transno = "LPD64"\n",
2698                                req, req->rq_reply_state, req->rq_transno);
2699                         if (mdt_cos_is_enabled(mdt)) {
2700                                 no_ack = 1;
2701                                 ldlm_lock_downgrade(lock, LCK_COS);
2702                                 mode = LCK_COS;
2703                         }
2704                         ptlrpc_save_lock(req, h, mode, no_ack);
2705                         if (mdt_is_lock_sync(lock)) {
2706                                 CDEBUG(D_HA, "found sync-lock,"
2707                                        " async commit started\n");
2708                                 mdt_device_commit_async(info->mti_env,
2709                                                         mdt);
2710                         }
2711                         LDLM_LOCK_PUT(lock);
2712                 }
2713                 h->cookie = 0ull;
2714         }
2715
2716         EXIT;
2717 }
2718
2719 /**
2720  * Unlock mdt object.
2721  *
2722  * Immeditely release the regular lock and the PDO lock or save the
2723  * lock in reqeuest and keep them referenced until client ACK or
2724  * transaction commit.
2725  *
2726  * \param info thread info object
2727  * \param o mdt object
2728  * \param lh mdt lock handle referencing regular and PDO locks
2729  * \param decref force immediate lock releasing
2730  */
2731 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2732                        struct mdt_lock_handle *lh, int decref)
2733 {
2734         ENTRY;
2735
2736         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2737         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2738
2739         if (lustre_handle_is_used(&lh->mlh_rreg_lh))
2740                 ldlm_lock_decref(&lh->mlh_rreg_lh, lh->mlh_rreg_mode);
2741
2742         EXIT;
2743 }
2744
2745 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2746                                         const struct lu_fid *f,
2747                                         struct mdt_lock_handle *lh,
2748                                         __u64 ibits)
2749 {
2750         struct mdt_object *o;
2751
2752         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2753         if (!IS_ERR(o)) {
2754                 int rc;
2755
2756                 rc = mdt_object_lock(info, o, lh, ibits,
2757                                      MDT_LOCAL_LOCK);
2758                 if (rc != 0) {
2759                         mdt_object_put(info->mti_env, o);
2760                         o = ERR_PTR(rc);
2761                 }
2762         }
2763         return o;
2764 }
2765
2766 void mdt_object_unlock_put(struct mdt_thread_info * info,
2767                            struct mdt_object * o,
2768                            struct mdt_lock_handle *lh,
2769                            int decref)
2770 {
2771         mdt_object_unlock(info, o, lh, decref);
2772         mdt_object_put(info->mti_env, o);
2773 }
2774
2775 struct mdt_handler *mdt_handler_find(__u32 opc, struct mdt_opc_slice *supported)
2776 {
2777         struct mdt_opc_slice *s;
2778         struct mdt_handler   *h;
2779
2780         h = NULL;
2781         for (s = supported; s->mos_hs != NULL; s++) {
2782                 if (s->mos_opc_start <= opc && opc < s->mos_opc_end) {
2783                         h = s->mos_hs + (opc - s->mos_opc_start);
2784                         if (likely(h->mh_opc != 0))
2785                                 LASSERTF(h->mh_opc == opc,
2786                                          "opcode mismatch %d != %d\n",
2787                                          h->mh_opc, opc);
2788                         else
2789                                 h = NULL; /* unsupported opc */
2790                         break;
2791                 }
2792         }
2793         return h;
2794 }
2795
2796 static int mdt_lock_resname_compat(struct mdt_device *m,
2797                                    struct ldlm_request *req)
2798 {
2799         /* XXX something... later. */
2800         return 0;
2801 }
2802
2803 static int mdt_lock_reply_compat(struct mdt_device *m, struct ldlm_reply *rep)
2804 {
2805         /* XXX something... later. */
2806         return 0;
2807 }
2808
2809 /*
2810  * Generic code handling requests that have struct mdt_body passed in:
2811  *
2812  *  - extract mdt_body from request and save it in @info, if present;
2813  *
2814  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
2815  *  @info;
2816  *
2817  *  - if HABEO_CORPUS flag is set for this request type check whether object
2818  *  actually exists on storage (lu_object_exists()).
2819  *
2820  */
2821 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2822 {
2823         const struct mdt_body    *body;
2824         struct mdt_object        *obj;
2825         const struct lu_env      *env;
2826         struct req_capsule       *pill;
2827         int                       rc;
2828         ENTRY;
2829
2830         env = info->mti_env;
2831         pill = info->mti_pill;
2832
2833         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2834         if (body == NULL)
2835                 RETURN(-EFAULT);
2836
2837         if (!(body->valid & OBD_MD_FLID))
2838                 RETURN(0);
2839
2840         if (!fid_is_sane(&body->fid1)) {
2841                 CERROR("Invalid fid: "DFID"\n", PFID(&body->fid1));
2842                 RETURN(-EINVAL);
2843         }
2844
2845         /*
2846          * Do not get size or any capa fields before we check that request
2847          * contains capa actually. There are some requests which do not, for
2848          * instance MDS_IS_SUBDIR.
2849          */
2850         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2851             req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
2852                 mdt_set_capainfo(info, 0, &body->fid1,
2853                                  req_capsule_client_get(pill, &RMF_CAPA1));
2854
2855         obj = mdt_object_find(env, info->mti_mdt, &body->fid1);
2856         if (!IS_ERR(obj)) {
2857                 if ((flags & HABEO_CORPUS) &&
2858                     !mdt_object_exists(obj)) {
2859                         mdt_object_put(env, obj);
2860                         /* for capability renew ENOENT will be handled in
2861                          * mdt_renew_capa */
2862                         if (body->valid & OBD_MD_FLOSSCAPA)
2863                                 rc = 0;
2864                         else
2865                                 rc = -ENOENT;
2866                 } else {
2867                         info->mti_object = obj;
2868                         rc = 0;
2869                 }
2870         } else
2871                 rc = PTR_ERR(obj);
2872
2873         RETURN(rc);
2874 }
2875
2876 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2877 {
2878         struct req_capsule *pill = info->mti_pill;
2879         int rc;
2880         ENTRY;
2881
2882         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2883                 rc = mdt_body_unpack(info, flags);
2884         else
2885                 rc = 0;
2886
2887         if (rc == 0 && (flags & HABEO_REFERO)) {
2888                 /* Pack reply. */
2889                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2890                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2891                                              info->mti_body->eadatasize);
2892                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2893                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
2894                                              RCL_SERVER, 0);
2895
2896                 rc = req_capsule_server_pack(pill);
2897         }
2898         RETURN(rc);
2899 }
2900
2901 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2902 {
2903         struct md_device *next = m->mdt_child;
2904
2905         return next->md_ops->mdo_init_capa_ctxt(env, next,
2906                                                 m->mdt_opts.mo_mds_capa,
2907                                                 m->mdt_capa_timeout,
2908                                                 m->mdt_capa_alg,
2909                                                 m->mdt_capa_keys);
2910 }
2911
2912 /*
2913  * Invoke handler for this request opc. Also do necessary preprocessing
2914  * (according to handler ->mh_flags), and post-processing (setting of
2915  * ->last_{xid,committed}).
2916  */
2917 static int mdt_req_handle(struct mdt_thread_info *info,
2918                           struct mdt_handler *h, struct ptlrpc_request *req)
2919 {
2920         int   rc, serious = 0;
2921         __u32 flags;
2922
2923         ENTRY;
2924
2925         LASSERT(h->mh_act != NULL);
2926         LASSERT(h->mh_opc == lustre_msg_get_opc(req->rq_reqmsg));
2927         LASSERT(current->journal_info == NULL);
2928
2929         /*
2930          * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
2931          * to put same checks into handlers like mdt_close(), mdt_reint(),
2932          * etc., without talking to mdt authors first. Checking same thing
2933          * there again is useless and returning 0 error without packing reply
2934          * is buggy! Handlers either pack reply or return error.
2935          *
2936          * We return 0 here and do not send any reply in order to emulate
2937          * network failure. Do not send any reply in case any of NET related
2938          * fail_id has occured.
2939          */
2940         if (OBD_FAIL_CHECK_ORSET(h->mh_fail_id, OBD_FAIL_ONCE))
2941                 RETURN(0);
2942
2943         rc = 0;
2944         flags = h->mh_flags;
2945         LASSERT(ergo(flags & (HABEO_CORPUS|HABEO_REFERO), h->mh_fmt != NULL));
2946
2947         if (h->mh_fmt != NULL) {
2948                 req_capsule_set(info->mti_pill, h->mh_fmt);
2949                 rc = mdt_unpack_req_pack_rep(info, flags);
2950         }
2951
2952         if (rc == 0 && flags & MUTABOR &&
2953             exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
2954                 /* should it be rq_status? */
2955                 rc = -EROFS;
2956
2957         if (rc == 0 && flags & HABEO_CLAVIS) {
2958                 struct ldlm_request *dlm_req;
2959
2960                 LASSERT(h->mh_fmt != NULL);
2961
2962                 dlm_req = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
2963                 if (dlm_req != NULL) {
2964                         if (unlikely(dlm_req->lock_desc.l_resource.lr_type ==
2965                                         LDLM_IBITS &&
2966                                      dlm_req->lock_desc.l_policy_data.\
2967                                         l_inodebits.bits == 0)) {
2968                                 /*
2969                                  * Lock without inodebits makes no sense and
2970                                  * will oops later in ldlm. If client miss to
2971                                  * set such bits, do not trigger ASSERTION.
2972                                  *
2973                                  * For liblustre flock case, it maybe zero.
2974                                  */
2975                                 rc = -EPROTO;
2976                         } else {
2977                                 if (info->mti_mdt->mdt_opts.mo_compat_resname)
2978                                         rc = mdt_lock_resname_compat(
2979                                                                 info->mti_mdt,
2980                                                                 dlm_req);
2981                                 info->mti_dlm_req = dlm_req;
2982                         }
2983                 } else {
2984                         rc = -EFAULT;
2985                 }
2986         }
2987
2988         /* capability setting changed via /proc, needs reinitialize ctxt */
2989         if (info->mti_mdt && info->mti_mdt->mdt_capa_conf) {
2990                 mdt_init_capa_ctxt(info->mti_env, info->mti_mdt);
2991                 info->mti_mdt->mdt_capa_conf = 0;
2992         }
2993
2994         if (likely(rc == 0)) {
2995                 /*
2996                  * Process request, there can be two types of rc:
2997                  * 1) errors with msg unpack/pack, other failures outside the
2998                  * operation itself. This is counted as serious errors;
2999                  * 2) errors during fs operation, should be placed in rq_status
3000                  * only
3001                  */
3002                 rc = h->mh_act(info);
3003                 if (rc == 0 &&
3004                     !req->rq_no_reply && req->rq_reply_state == NULL) {
3005                         DEBUG_REQ(D_ERROR, req, "MDT \"handler\" %s did not "
3006                                   "pack reply and returned 0 error\n",
3007                                   h->mh_name);
3008                         LBUG();
3009                 }
3010                 serious = is_serious(rc);
3011                 rc = clear_serious(rc);
3012         } else
3013                 serious = 1;
3014
3015         req->rq_status = rc;
3016
3017         /*
3018          * ELDLM_* codes which > 0 should be in rq_status only as well as
3019          * all non-serious errors.
3020          */
3021         if (rc > 0 || !serious)
3022                 rc = 0;
3023
3024         LASSERT(current->journal_info == NULL);
3025
3026         if (rc == 0 && (flags & HABEO_CLAVIS) &&
3027             info->mti_mdt->mdt_opts.mo_compat_resname) {
3028                 struct ldlm_reply *dlmrep;
3029
3030                 dlmrep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3031                 if (dlmrep != NULL)
3032                         rc = mdt_lock_reply_compat(info->mti_mdt, dlmrep);
3033         }
3034
3035         /* If we're DISCONNECTing, the mdt_export_data is already freed */
3036         if (likely(rc == 0 && req->rq_export && h->mh_opc != MDS_DISCONNECT))
3037                 target_committed_to_req(req);
3038
3039         if (unlikely(req_is_replay(req) &&
3040                      lustre_msg_get_transno(req->rq_reqmsg) == 0)) {
3041                 DEBUG_REQ(D_ERROR, req, "transno is 0 during REPLAY");
3042                 LBUG();
3043         }
3044
3045         target_send_reply(req, rc, info->mti_fail_id);
3046         RETURN(0);
3047 }
3048
3049 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
3050 {
3051         lh->mlh_type = MDT_NUL_LOCK;
3052         lh->mlh_reg_lh.cookie = 0ull;
3053         lh->mlh_reg_mode = LCK_MINMODE;
3054         lh->mlh_pdo_lh.cookie = 0ull;
3055         lh->mlh_pdo_mode = LCK_MINMODE;
3056         lh->mlh_rreg_lh.cookie = 0ull;
3057         lh->mlh_rreg_mode = LCK_MINMODE;
3058 }
3059
3060 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
3061 {
3062         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
3063         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
3064 }
3065
3066 /*
3067  * Initialize fields of struct mdt_thread_info. Other fields are left in
3068  * uninitialized state, because it's too expensive to zero out whole
3069  * mdt_thread_info (> 1K) on each request arrival.
3070  */
3071 static void mdt_thread_info_init(struct ptlrpc_request *req,
3072                                  struct mdt_thread_info *info)
3073 {
3074         int i;
3075
3076         req_capsule_init(&req->rq_pill, req, RCL_SERVER);
3077         info->mti_pill = &req->rq_pill;
3078
3079         /* lock handle */
3080         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3081                 mdt_lock_handle_init(&info->mti_lh[i]);
3082
3083         /* mdt device: it can be NULL while CONNECT */
3084         if (req->rq_export) {
3085                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
3086                 info->mti_exp = req->rq_export;
3087         } else
3088                 info->mti_mdt = NULL;
3089         info->mti_env = req->rq_svc_thread->t_env;
3090         info->mti_fail_id = OBD_FAIL_MDS_ALL_REPLY_NET;
3091         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
3092         info->mti_mos = NULL;
3093
3094         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
3095         info->mti_big_buf = LU_BUF_NULL;
3096         info->mti_body = NULL;
3097         info->mti_object = NULL;
3098         info->mti_dlm_req = NULL;
3099         info->mti_has_trans = 0;
3100         info->mti_cross_ref = 0;
3101         info->mti_opdata = 0;
3102         info->mti_big_lmm_used = 0;
3103
3104         /* To not check for split by default. */
3105         info->mti_spec.no_create = 0;
3106         info->mti_spec.sp_rm_entry = 0;
3107 }
3108
3109 static void mdt_thread_info_fini(struct mdt_thread_info *info)
3110 {
3111         int i;
3112
3113         req_capsule_fini(info->mti_pill);
3114         if (info->mti_object != NULL) {
3115                 mdt_object_put(info->mti_env, info->mti_object);
3116                 info->mti_object = NULL;
3117         }
3118
3119         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3120                 mdt_lock_handle_fini(&info->mti_lh[i]);
3121         info->mti_env = NULL;
3122
3123         if (unlikely(info->mti_big_buf.lb_buf != NULL))
3124                 lu_buf_free(&info->mti_big_buf);
3125 }
3126
3127 static int mdt_filter_recovery_request(struct ptlrpc_request *req,
3128                                        struct obd_device *obd, int *process)
3129 {
3130         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
3131         case MDS_CONNECT: /* This will never get here, but for completeness. */
3132         case OST_CONNECT: /* This will never get here, but for completeness. */
3133         case MDS_DISCONNECT:
3134         case OST_DISCONNECT:
3135         case OBD_IDX_READ:
3136                *process = 1;
3137                RETURN(0);
3138
3139         case MDS_CLOSE:
3140         case MDS_DONE_WRITING:
3141         case MDS_SYNC: /* used in unmounting */
3142         case OBD_PING:
3143         case MDS_REINT:
3144         case UPDATE_OBJ:
3145         case SEQ_QUERY:
3146         case FLD_QUERY:
3147         case LDLM_ENQUEUE:
3148                 *process = target_queue_recovery_request(req, obd);
3149                 RETURN(0);
3150
3151         default:
3152                 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
3153                 *process = -EAGAIN;
3154                 RETURN(0);
3155         }
3156 }
3157
3158 /*
3159  * Handle recovery. Return:
3160  *        +1: continue request processing;
3161  *       -ve: abort immediately with the given error code;
3162  *         0: send reply with error code in req->rq_status;
3163  */
3164 static int mdt_recovery(struct mdt_thread_info *info)
3165 {
3166         struct ptlrpc_request *req = mdt_info_req(info);
3167         struct obd_device *obd;
3168
3169         ENTRY;
3170
3171         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
3172         case MDS_CONNECT:
3173         case SEC_CTX_INIT:
3174         case SEC_CTX_INIT_CONT:
3175         case SEC_CTX_FINI:
3176                 {
3177 #if 0
3178                         int rc;
3179
3180                         rc = mdt_handle_idmap(info);
3181                         if (rc)
3182                                 RETURN(rc);
3183                         else
3184 #endif
3185                                 RETURN(+1);
3186                 }
3187         }
3188
3189         if (unlikely(!class_connected_export(req->rq_export))) {
3190                 CDEBUG(D_HA, "operation %d on unconnected MDS from %s\n",
3191                        lustre_msg_get_opc(req->rq_reqmsg),
3192                        libcfs_id2str(req->rq_peer));
3193                 /* FIXME: For CMD cleanup, when mds_B stop, the req from
3194                  * mds_A will get -ENOTCONN(especially for ping req),
3195                  * which will cause that mds_A deactive timeout, then when
3196                  * mds_A cleanup, the cleanup process will be suspended since
3197                  * deactive timeout is not zero.
3198                  */
3199                 req->rq_status = -ENOTCONN;
3200                 target_send_reply(req, -ENOTCONN, info->mti_fail_id);
3201                 RETURN(0);
3202         }
3203
3204         /* sanity check: if the xid matches, the request must be marked as a
3205          * resent or replayed */
3206         if (req_xid_is_last(req)) {
3207                 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
3208                       (MSG_RESENT | MSG_REPLAY))) {
3209                         DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches last_xid, "
3210                                   "expected REPLAY or RESENT flag (%x)", req->rq_xid,
3211                                   lustre_msg_get_flags(req->rq_reqmsg));
3212                         LBUG();
3213                         req->rq_status = -ENOTCONN;
3214                         RETURN(-ENOTCONN);
3215                 }
3216         }
3217
3218         /* else: note the opposite is not always true; a RESENT req after a
3219          * failover will usually not match the last_xid, since it was likely
3220          * never committed. A REPLAYed request will almost never match the
3221          * last xid, however it could for a committed, but still retained,
3222          * open. */
3223
3224         obd = req->rq_export->exp_obd;
3225
3226         /* Check for aborted recovery... */
3227         if (unlikely(obd->obd_recovering)) {
3228                 int rc;
3229                 int should_process;
3230                 DEBUG_REQ(D_INFO, req, "Got new replay");
3231                 rc = mdt_filter_recovery_request(req, obd, &should_process);
3232                 if (rc != 0 || !should_process)
3233                         RETURN(rc);
3234                 else if (should_process < 0) {
3235                         req->rq_status = should_process;
3236                         rc = ptlrpc_error(req);
3237                         RETURN(rc);
3238                 }
3239         }
3240         RETURN(+1);
3241 }
3242
3243 static int mdt_msg_check_version(struct lustre_msg *msg)
3244 {
3245         int rc;
3246
3247         switch (lustre_msg_get_opc(msg)) {
3248         case MDS_CONNECT:
3249         case MDS_DISCONNECT:
3250         case OBD_PING:
3251         case SEC_CTX_INIT:
3252         case SEC_CTX_INIT_CONT:
3253         case SEC_CTX_FINI:
3254         case OBD_IDX_READ:
3255                 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
3256                 if (rc)
3257                         CERROR("bad opc %u version %08x, expecting %08x\n",
3258                                lustre_msg_get_opc(msg),
3259                                lustre_msg_get_version(msg),
3260                                LUSTRE_OBD_VERSION);
3261                 break;
3262         case MDS_GETSTATUS:
3263         case MDS_GETATTR:
3264         case MDS_GETATTR_NAME:
3265         case MDS_STATFS:
3266         case MDS_READPAGE:
3267         case MDS_WRITEPAGE:
3268         case MDS_IS_SUBDIR:
3269         case MDS_REINT:
3270         case MDS_CLOSE:
3271         case MDS_DONE_WRITING:
3272         case MDS_PIN:
3273         case MDS_SYNC:
3274         case MDS_GETXATTR:
3275         case MDS_SETXATTR:
3276         case MDS_SET_INFO:
3277         case MDS_GET_INFO:
3278         case MDS_HSM_PROGRESS:
3279         case MDS_HSM_REQUEST:
3280         case MDS_HSM_CT_REGISTER:
3281         case MDS_HSM_CT_UNREGISTER:
3282         case MDS_HSM_STATE_GET:
3283         case MDS_HSM_STATE_SET:
3284         case MDS_HSM_ACTION:
3285         case MDS_QUOTACHECK:
3286         case MDS_QUOTACTL:
3287         case UPDATE_OBJ:
3288         case MDS_SWAP_LAYOUTS:
3289         case QUOTA_DQACQ:
3290         case QUOTA_DQREL:
3291         case SEQ_QUERY:
3292         case FLD_QUERY:
3293                 rc = lustre_msg_check_version(msg, LUSTRE_MDS_VERSION);
3294                 if (rc)
3295                         CERROR("bad opc %u version %08x, expecting %08x\n",
3296                                lustre_msg_get_opc(msg),
3297                                lustre_msg_get_version(msg),
3298                                LUSTRE_MDS_VERSION);
3299                 break;
3300         case LDLM_ENQUEUE:
3301         case LDLM_CONVERT:
3302         case LDLM_BL_CALLBACK:
3303         case LDLM_CP_CALLBACK:
3304                 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
3305                 if (rc)
3306                         CERROR("bad opc %u version %08x, expecting %08x\n",
3307                                lustre_msg_get_opc(msg),
3308                                lustre_msg_get_version(msg),
3309                                LUSTRE_DLM_VERSION);
3310                 break;
3311         case OBD_LOG_CANCEL:
3312         case LLOG_ORIGIN_HANDLE_CREATE:
3313         case LLOG_ORIGIN_HANDLE_NEXT_BLOCK:
3314         case LLOG_ORIGIN_HANDLE_READ_HEADER:
3315         case LLOG_ORIGIN_HANDLE_CLOSE:
3316         case LLOG_ORIGIN_HANDLE_DESTROY:
3317         case LLOG_ORIGIN_HANDLE_PREV_BLOCK:
3318         case LLOG_CATINFO:
3319                 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
3320                 if (rc)
3321                         CERROR("bad opc %u version %08x, expecting %08x\n",
3322                                lustre_msg_get_opc(msg),
3323                                lustre_msg_get_version(msg),
3324                                LUSTRE_LOG_VERSION);
3325                 break;
3326         default:
3327                 CERROR("MDS unknown opcode %d\n", lustre_msg_get_opc(msg));
3328                 rc = -ENOTSUPP;
3329         }
3330         return rc;
3331 }
3332
3333 static int mdt_handle0(struct ptlrpc_request *req,
3334                        struct mdt_thread_info *info,
3335                        struct mdt_opc_slice *supported)
3336 {
3337         struct mdt_handler *h;
3338         struct lustre_msg  *msg;
3339         int                 rc;
3340
3341         ENTRY;
3342
3343         if (OBD_FAIL_CHECK_ORSET(OBD_FAIL_MDS_ALL_REQUEST_NET, OBD_FAIL_ONCE))
3344                 RETURN(0);
3345
3346         LASSERT(current->journal_info == NULL);
3347
3348         msg = req->rq_reqmsg;
3349         rc = mdt_msg_check_version(msg);
3350         if (likely(rc == 0)) {
3351                 rc = mdt_recovery(info);
3352                 if (likely(rc == +1)) {
3353                         h = mdt_handler_find(lustre_msg_get_opc(msg),
3354                                              supported);
3355                         if (likely(h != NULL)) {
3356                                 rc = mdt_req_handle(info, h, req);
3357                         } else {
3358                                 CERROR("The unsupported opc: 0x%x\n",
3359                                        lustre_msg_get_opc(msg) );
3360                                 req->rq_status = -ENOTSUPP;
3361                                 rc = ptlrpc_error(req);
3362                                 RETURN(rc);
3363                         }
3364                 }
3365         } else
3366                 CERROR(LUSTRE_MDT_NAME" drops mal-formed request\n");
3367         RETURN(rc);
3368 }
3369
3370 /*
3371  * MDT handler function called by ptlrpc service thread when request comes.
3372  *
3373  * XXX common "target" functionality should be factored into separate module
3374  * shared by mdt, ost and stand-alone services like fld.
3375  */
3376 int mdt_handle_common(struct ptlrpc_request *req,
3377                       struct mdt_opc_slice *supported)
3378 {
3379         struct lu_env          *env;
3380         struct mdt_thread_info *info;
3381         int                     rc;
3382         ENTRY;
3383
3384         env = req->rq_svc_thread->t_env;
3385         /* Refill(initilize) the context(mdt_thread_info), in case it is
3386          * not initialized yet. Usually it happens during start up, after
3387          * MDS(ptlrpc threads) is start up, it gets the first CONNECT request,
3388          * before MDT_thread_info is initialized */
3389         lu_env_refill(env);
3390         LASSERT(env != NULL);
3391         LASSERT(env->le_ses != NULL);
3392         LASSERT(env->le_ctx.lc_thread == req->rq_svc_thread);
3393         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3394         LASSERT(info != NULL);
3395
3396         mdt_thread_info_init(req, info);
3397
3398         rc = mdt_handle0(req, info, supported);
3399
3400         mdt_thread_info_fini(info);
3401         RETURN(rc);
3402 }
3403
3404 /*
3405  * This is called from recovery code as handler of _all_ RPC types, FLD and SEQ
3406  * as well.
3407  */
3408 int mdt_recovery_handle(struct ptlrpc_request *req)
3409 {
3410         int rc;
3411         ENTRY;
3412
3413         switch (lustre_msg_get_opc(req->rq_reqmsg)) {
3414         case FLD_QUERY:
3415                 rc = mdt_handle_common(req, mdt_fld_handlers);
3416                 break;
3417         case SEQ_QUERY:
3418                 rc = mdt_handle_common(req, mdt_seq_handlers);
3419                 break;
3420         default:
3421                 rc = mdt_handle_common(req, mdt_regular_handlers);
3422                 break;
3423         }
3424
3425         RETURN(rc);
3426 }
3427
3428 enum mdt_it_code {
3429         MDT_IT_OPEN,
3430         MDT_IT_OCREAT,
3431         MDT_IT_CREATE,
3432         MDT_IT_GETATTR,
3433         MDT_IT_READDIR,
3434         MDT_IT_LOOKUP,
3435         MDT_IT_UNLINK,
3436         MDT_IT_TRUNC,
3437         MDT_IT_GETXATTR,
3438         MDT_IT_LAYOUT,
3439         MDT_IT_QUOTA,
3440         MDT_IT_NR
3441 };
3442
3443 static int mdt_intent_getattr(enum mdt_it_code opcode,
3444                               struct mdt_thread_info *info,
3445                               struct ldlm_lock **,
3446                               __u64);
3447 static int mdt_intent_layout(enum mdt_it_code opcode,
3448                              struct mdt_thread_info *info,
3449                              struct ldlm_lock **,
3450                              __u64);
3451 static int mdt_intent_reint(enum mdt_it_code opcode,
3452                             struct mdt_thread_info *info,
3453                             struct ldlm_lock **,
3454                             __u64);
3455
3456 static struct mdt_it_flavor {
3457         const struct req_format *it_fmt;
3458         __u32                    it_flags;
3459         int                    (*it_act)(enum mdt_it_code ,
3460                                          struct mdt_thread_info *,
3461                                          struct ldlm_lock **,
3462                                          __u64);
3463         long                     it_reint;
3464 } mdt_it_flavor[] = {
3465         [MDT_IT_OPEN]     = {
3466                 .it_fmt   = &RQF_LDLM_INTENT,
3467                 /*.it_flags = HABEO_REFERO,*/
3468                 .it_flags = 0,
3469                 .it_act   = mdt_intent_reint,
3470                 .it_reint = REINT_OPEN
3471         },
3472         [MDT_IT_OCREAT]   = {
3473                 .it_fmt   = &RQF_LDLM_INTENT,
3474                 .it_flags = MUTABOR,
3475                 .it_act   = mdt_intent_reint,
3476                 .it_reint = REINT_OPEN
3477         },
3478         [MDT_IT_CREATE]   = {
3479                 .it_fmt   = &RQF_LDLM_INTENT,
3480                 .it_flags = MUTABOR,
3481                 .it_act   = mdt_intent_reint,
3482                 .it_reint = REINT_CREATE
3483         },
3484         [MDT_IT_GETATTR]  = {
3485                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3486                 .it_flags = HABEO_REFERO,
3487                 .it_act   = mdt_intent_getattr
3488         },
3489         [MDT_IT_READDIR]  = {
3490                 .it_fmt   = NULL,
3491                 .it_flags = 0,
3492                 .it_act   = NULL
3493         },
3494         [MDT_IT_LOOKUP]   = {
3495                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3496                 .it_flags = HABEO_REFERO,
3497                 .it_act   = mdt_intent_getattr
3498         },
3499         [MDT_IT_UNLINK]   = {
3500                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
3501                 .it_flags = MUTABOR,
3502                 .it_act   = NULL,
3503                 .it_reint = REINT_UNLINK
3504         },
3505         [MDT_IT_TRUNC]    = {
3506                 .it_fmt   = NULL,
3507                 .it_flags = MUTABOR,
3508                 .it_act   = NULL
3509         },
3510         [MDT_IT_GETXATTR] = {
3511                 .it_fmt   = NULL,
3512                 .it_flags = 0,
3513                 .it_act   = NULL
3514         },
3515         [MDT_IT_LAYOUT] = {
3516                 .it_fmt   = &RQF_LDLM_INTENT_LAYOUT,
3517                 .it_flags = 0,
3518                 .it_act   = mdt_intent_layout
3519         }
3520 };
3521
3522 int mdt_intent_lock_replace(struct mdt_thread_info *info,
3523                             struct ldlm_lock **lockp,
3524                             struct ldlm_lock *new_lock,
3525                             struct mdt_lock_handle *lh,
3526                             __u64 flags)
3527 {
3528         struct ptlrpc_request  *req = mdt_info_req(info);
3529         struct ldlm_lock       *lock = *lockp;
3530
3531         /*
3532          * Get new lock only for cases when possible resent did not find any
3533          * lock.
3534          */
3535         if (new_lock == NULL)
3536                 new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3537
3538         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3539                 lh->mlh_reg_lh.cookie = 0;
3540                 RETURN(0);
3541         }
3542
3543         LASSERTF(new_lock != NULL,
3544                  "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
3545
3546         /*
3547          * If we've already given this lock to a client once, then we should
3548          * have no readers or writers.  Otherwise, we should have one reader
3549          * _or_ writer ref (which will be zeroed below) before returning the
3550          * lock to a client.
3551          */
3552         if (new_lock->l_export == req->rq_export) {
3553                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3554         } else {
3555                 LASSERT(new_lock->l_export == NULL);
3556                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3557         }
3558
3559         *lockp = new_lock;
3560
3561         if (new_lock->l_export == req->rq_export) {
3562                 /*
3563                  * Already gave this to the client, which means that we
3564                  * reconstructed a reply.
3565                  */
3566                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3567                         MSG_RESENT);
3568                 lh->mlh_reg_lh.cookie = 0;
3569                 RETURN(ELDLM_LOCK_REPLACED);
3570         }
3571
3572         /*
3573          * Fixup the lock to be given to the client.
3574          */
3575         lock_res_and_lock(new_lock);
3576         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3577          * possible blocking AST. */
3578         while (new_lock->l_readers > 0) {
3579                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3580                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3581                 new_lock->l_readers--;
3582         }
3583         while (new_lock->l_writers > 0) {
3584                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3585                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3586                 new_lock->l_writers--;
3587         }
3588
3589         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3590         new_lock->l_blocking_ast = lock->l_blocking_ast;
3591         new_lock->l_completion_ast = lock->l_completion_ast;
3592         new_lock->l_remote_handle = lock->l_remote_handle;
3593         new_lock->l_flags &= ~LDLM_FL_LOCAL;
3594
3595         unlock_res_and_lock(new_lock);
3596
3597         cfs_hash_add(new_lock->l_export->exp_lock_hash,
3598                      &new_lock->l_remote_handle,
3599                      &new_lock->l_exp_hash);
3600
3601         LDLM_LOCK_RELEASE(new_lock);
3602         lh->mlh_reg_lh.cookie = 0;
3603
3604         RETURN(ELDLM_LOCK_REPLACED);
3605 }
3606
3607 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3608                                     struct ldlm_lock *new_lock,
3609                                     struct ldlm_lock **old_lock,
3610                                     struct mdt_lock_handle *lh)
3611 {
3612         struct ptlrpc_request  *req = mdt_info_req(info);
3613         struct obd_export      *exp = req->rq_export;
3614         struct lustre_handle    remote_hdl;
3615         struct ldlm_request    *dlmreq;
3616         struct ldlm_lock       *lock;
3617
3618         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3619                 return;
3620
3621         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3622         remote_hdl = dlmreq->lock_handle[0];
3623
3624         /* In the function below, .hs_keycmp resolves to
3625          * ldlm_export_lock_keycmp() */
3626         /* coverity[overrun-buffer-val] */
3627         lock = cfs_hash_lookup(exp->exp_lock_hash, &remote_hdl);
3628         if (lock) {
3629                 if (lock != new_lock) {
3630                         lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie;
3631                         lh->mlh_reg_mode = lock->l_granted_mode;
3632
3633                         LDLM_DEBUG(lock, "Restoring lock cookie");
3634                         DEBUG_REQ(D_DLMTRACE, req,
3635                                   "restoring lock cookie "LPX64,
3636                                   lh->mlh_reg_lh.cookie);
3637                         if (old_lock)
3638                                 *old_lock = LDLM_LOCK_GET(lock);
3639                         cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
3640                         return;
3641                 }
3642
3643                 cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
3644         }
3645
3646         /*
3647          * If the xid matches, then we know this is a resent request, and allow
3648          * it. (It's probably an OPEN, for which we don't send a lock.
3649          */
3650         if (req_xid_is_last(req))
3651                 return;
3652
3653         /*
3654          * This remote handle isn't enqueued, so we never received or processed
3655          * this request.  Clear MSG_RESENT, because it can be handled like any
3656          * normal request now.
3657          */
3658         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3659
3660         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
3661                   remote_hdl.cookie);
3662 }
3663
3664 static int mdt_intent_getattr(enum mdt_it_code opcode,
3665                               struct mdt_thread_info *info,
3666                               struct ldlm_lock **lockp,
3667                               __u64 flags)
3668 {
3669         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3670         struct ldlm_lock       *new_lock = NULL;
3671         __u64                   child_bits;
3672         struct ldlm_reply      *ldlm_rep;
3673         struct ptlrpc_request  *req;
3674         struct mdt_body        *reqbody;
3675         struct mdt_body        *repbody;
3676         int                     rc, rc2;
3677         ENTRY;
3678
3679         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3680         LASSERT(reqbody);
3681
3682         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3683         LASSERT(repbody);
3684
3685         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
3686         repbody->eadatasize = 0;
3687         repbody->aclsize = 0;
3688
3689         switch (opcode) {
3690         case MDT_IT_LOOKUP:
3691                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3692                 break;
3693         case MDT_IT_GETATTR:
3694                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3695                              MDS_INODELOCK_PERM;
3696                 break;
3697         default:
3698                 CERROR("Unsupported intent (%d)\n", opcode);
3699                 GOTO(out_shrink, rc = -EINVAL);
3700         }
3701
3702         rc = mdt_init_ucred(info, reqbody);
3703         if (rc)
3704                 GOTO(out_shrink, rc);
3705
3706         req = info->mti_pill->rc_req;
3707         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3708         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3709
3710         /* Get lock from request for possible resent case. */
3711         mdt_intent_fixup_resent(info, *lockp, &new_lock, lhc);
3712
3713         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3714         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3715
3716         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3717                 ldlm_rep->lock_policy_res2 = 0;
3718         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3719             ldlm_rep->lock_policy_res2) {
3720                 lhc->mlh_reg_lh.cookie = 0ull;
3721                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3722         }
3723
3724         rc = mdt_intent_lock_replace(info, lockp, new_lock, lhc, flags);
3725         EXIT;
3726 out_ucred:
3727         mdt_exit_ucred(info);
3728 out_shrink:
3729         mdt_client_compatibility(info);
3730         rc2 = mdt_fix_reply(info);
3731         if (rc == 0)
3732                 rc = rc2;
3733         return rc;
3734 }
3735
3736 static int mdt_intent_layout(enum mdt_it_code opcode,
3737                              struct mdt_thread_info *info,
3738                              struct ldlm_lock **lockp,
3739                              __u64 flags)
3740 {
3741         struct layout_intent *layout;
3742         struct lu_fid *fid;
3743         struct mdt_object *obj = NULL;
3744         struct md_object *child = NULL;
3745         int rc;
3746         ENTRY;
3747
3748         if (opcode != MDT_IT_LAYOUT) {
3749                 CERROR("%s: Unknown intent (%d)\n", mdt_obd_name(info->mti_mdt),
3750                         opcode);
3751                 RETURN(-EINVAL);
3752         }
3753
3754         fid = &info->mti_tmp_fid2;
3755         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3756
3757         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3758         if (IS_ERR(obj))
3759                 RETURN(PTR_ERR(obj));
3760
3761         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3762                 child = mdt_object_child(obj);
3763
3764                 /* get the length of lsm */
3765                 rc = mo_xattr_get(info->mti_env, child, &LU_BUF_NULL,
3766                                   XATTR_NAME_LOV);
3767
3768                 if (rc > info->mti_mdt->mdt_max_mdsize)
3769                         info->mti_mdt->mdt_max_mdsize = rc;
3770         }
3771
3772         mdt_object_put(info->mti_env, obj);
3773
3774         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3775         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
3776                         ldlm_lvbo_size(*lockp));
3777         rc = req_capsule_server_pack(info->mti_pill);
3778         if (rc != 0)
3779                 RETURN(-EINVAL);
3780
3781         layout = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3782         LASSERT(layout != NULL);
3783         if (layout->li_opc == LAYOUT_INTENT_ACCESS)
3784                 /* return to normal ldlm handling */
3785                 RETURN(0);
3786
3787         CERROR("%s: Unsupported layout intent (%d)\n",
3788                 mdt_obd_name(info->mti_mdt), layout->li_opc);
3789         RETURN(-EINVAL);
3790 }
3791
3792 static int mdt_intent_reint(enum mdt_it_code opcode,
3793                             struct mdt_thread_info *info,
3794                             struct ldlm_lock **lockp,
3795                             __u64 flags)
3796 {
3797         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3798         struct ldlm_reply      *rep = NULL;
3799         long                    opc;
3800         int                     rc;
3801
3802         static const struct req_format *intent_fmts[REINT_MAX] = {
3803                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3804                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3805         };
3806
3807         ENTRY;
3808
3809         opc = mdt_reint_opcode(info, intent_fmts);
3810         if (opc < 0)
3811                 RETURN(opc);
3812
3813         if (mdt_it_flavor[opcode].it_reint != opc) {
3814                 CERROR("Reint code %ld doesn't match intent: %d\n",
3815                        opc, opcode);
3816                 RETURN(err_serious(-EPROTO));
3817         }
3818
3819         /* Get lock from request for possible resent case. */
3820         mdt_intent_fixup_resent(info, *lockp, NULL, lhc);
3821
3822         rc = mdt_reint_internal(info, lhc, opc);
3823
3824         /* Check whether the reply has been packed successfully. */
3825         if (mdt_info_req(info)->rq_repmsg != NULL)
3826                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3827         if (rep == NULL)
3828                 RETURN(err_serious(-EFAULT));
3829
3830         /* MDC expects this in any case */
3831         if (rc != 0)
3832                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3833
3834         /* the open lock or the lock for cross-ref object should be
3835          * returned to the client */
3836         if (rc == -EREMOTE || mdt_get_disposition(rep, DISP_OPEN_LOCK)) {
3837                 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
3838                 rep->lock_policy_res2 = 0;
3839                 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3840                 RETURN(rc);
3841         }
3842
3843         rep->lock_policy_res2 = clear_serious(rc);
3844
3845         if (rep->lock_policy_res2 == -ENOENT &&
3846             mdt_get_disposition(rep, DISP_LOOKUP_NEG))
3847                 rep->lock_policy_res2 = 0;
3848
3849         if (rc == -ENOTCONN || rc == -ENODEV ||
3850             rc == -EOVERFLOW) { /**< if VBR failure then return error */
3851                 /*
3852                  * If it is the disconnect error (ENODEV & ENOCONN), the error
3853                  * will be returned by rq_status, and client at ptlrpc layer
3854                  * will detect this, then disconnect, reconnect the import
3855                  * immediately, instead of impacting the following the rpc.
3856                  */
3857                 lhc->mlh_reg_lh.cookie = 0ull;
3858                 RETURN(rc);
3859         } else {
3860                 /*
3861                  * For other cases, the error will be returned by intent.
3862                  * and client will retrieve the result from intent.
3863                  */
3864                  /*
3865                   * FIXME: when open lock is finished, that should be
3866                   * checked here.
3867                   */
3868                 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3869                         LASSERTF(rc == 0, "Error occurred but lock handle "
3870                                  "is still in use, rc = %d\n", rc);
3871                         rep->lock_policy_res2 = 0;
3872                         rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3873                         RETURN(rc);
3874                 } else {
3875                         lhc->mlh_reg_lh.cookie = 0ull;
3876                         RETURN(ELDLM_LOCK_ABORTED);
3877                 }
3878         }
3879 }
3880
3881 static int mdt_intent_code(long itcode)
3882 {
3883         int rc;
3884
3885         switch(itcode) {
3886         case IT_OPEN:
3887                 rc = MDT_IT_OPEN;
3888                 break;
3889         case IT_OPEN|IT_CREAT:
3890                 rc = MDT_IT_OCREAT;
3891                 break;
3892         case IT_CREAT:
3893                 rc = MDT_IT_CREATE;
3894                 break;
3895         case IT_READDIR:
3896                 rc = MDT_IT_READDIR;
3897                 break;
3898         case IT_GETATTR:
3899                 rc = MDT_IT_GETATTR;
3900                 break;
3901         case IT_LOOKUP:
3902                 rc = MDT_IT_LOOKUP;
3903                 break;
3904         case IT_UNLINK:
3905                 rc = MDT_IT_UNLINK;
3906                 break;
3907         case IT_TRUNC:
3908                 rc = MDT_IT_TRUNC;
3909                 break;
3910         case IT_GETXATTR:
3911                 rc = MDT_IT_GETXATTR;
3912                 break;
3913         case IT_LAYOUT:
3914                 rc = MDT_IT_LAYOUT;
3915                 break;
3916         case IT_QUOTA_DQACQ:
3917         case IT_QUOTA_CONN:
3918                 rc = MDT_IT_QUOTA;
3919                 break;
3920         default:
3921                 CERROR("Unknown intent opcode: %ld\n", itcode);
3922                 rc = -EINVAL;
3923                 break;
3924         }
3925         return rc;
3926 }
3927
3928 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
3929                           struct ldlm_lock **lockp, __u64 flags)
3930 {
3931         struct req_capsule   *pill;
3932         struct mdt_it_flavor *flv;
3933         int opc;
3934         int rc;
3935         ENTRY;
3936
3937         opc = mdt_intent_code(itopc);
3938         if (opc < 0)
3939                 RETURN(-EINVAL);
3940
3941         pill = info->mti_pill;
3942
3943         if (opc == MDT_IT_QUOTA) {
3944                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
3945
3946                 if (qmt == NULL)
3947                         RETURN(-EOPNOTSUPP);
3948
3949                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
3950                 /* pass the request to quota master */
3951                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
3952                                                  mdt_info_req(info), lockp,
3953                                                  flags);
3954                 RETURN(rc);
3955         }
3956
3957         flv  = &mdt_it_flavor[opc];
3958         if (flv->it_fmt != NULL)
3959                 req_capsule_extend(pill, flv->it_fmt);
3960
3961         rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3962         if (rc == 0) {
3963                 struct ptlrpc_request *req = mdt_info_req(info);
3964                 if (flv->it_flags & MUTABOR &&
3965                     exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
3966                         RETURN(-EROFS);
3967         }
3968         if (rc == 0 && flv->it_act != NULL) {
3969                 struct ldlm_reply *rep;
3970
3971                 /* execute policy */
3972                 rc = flv->it_act(opc, info, lockp, flags);
3973
3974                 /* Check whether the reply has been packed successfully. */
3975                 if (mdt_info_req(info)->rq_repmsg != NULL) {
3976                         rep = req_capsule_server_get(info->mti_pill,
3977                                                      &RMF_DLM_REP);
3978                         rep->lock_policy_res2 =
3979                                 ptlrpc_status_hton(rep->lock_policy_res2);
3980                 }
3981         } else {
3982                 rc = -EOPNOTSUPP;
3983         }
3984         RETURN(rc);
3985 }
3986
3987 static int mdt_intent_policy(struct ldlm_namespace *ns,
3988                              struct ldlm_lock **lockp, void *req_cookie,
3989                              ldlm_mode_t mode, __u64 flags, void *data)
3990 {
3991         struct mdt_thread_info *info;
3992         struct ptlrpc_request  *req  =  req_cookie;
3993         struct ldlm_intent     *it;
3994         struct req_capsule     *pill;
3995         int rc;
3996
3997         ENTRY;
3998
3999         LASSERT(req != NULL);
4000
4001         info = lu_context_key_get(&req->rq_svc_thread->t_env->le_ctx,
4002                                   &mdt_thread_key);
4003         LASSERT(info != NULL);
4004         pill = info->mti_pill;
4005         LASSERT(pill->rc_req == req);
4006
4007         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
4008                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
4009                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
4010                 if (it != NULL) {
4011                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
4012                         if (rc == 0)
4013                                 rc = ELDLM_OK;
4014
4015                         /* Lock without inodebits makes no sense and will oops
4016                          * later in ldlm. Let's check it now to see if we have
4017                          * ibits corrupted somewhere in mdt_intent_opc().
4018                          * The case for client miss to set ibits has been
4019                          * processed by others. */
4020                         LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
4021                                         lr_type == LDLM_IBITS,
4022                                      info->mti_dlm_req->lock_desc.\
4023                                         l_policy_data.l_inodebits.bits != 0));
4024                 } else
4025                         rc = err_serious(-EFAULT);
4026         } else {
4027                 /* No intent was provided */
4028                 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
4029                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
4030                 rc = req_capsule_server_pack(pill);
4031                 if (rc)
4032                         rc = err_serious(rc);
4033         }
4034         RETURN(rc);
4035 }
4036
4037 static int mdt_seq_fini(const struct lu_env *env,
4038                         struct mdt_device *m)
4039 {
4040         return seq_site_fini(env, mdt_seq_site(m));
4041 }
4042
4043 static int mdt_seq_init(const struct lu_env *env,
4044                         const char *uuid,
4045                         struct mdt_device *m)
4046 {
4047         struct seq_server_site *ss;
4048         char *prefix;
4049         int rc;
4050         ENTRY;
4051
4052         ss = mdt_seq_site(m);
4053
4054         /*
4055          * This is sequence-controller node. Init seq-controller server on local
4056          * MDT.
4057          */
4058         if (ss->ss_node_id == 0) {
4059                 LASSERT(ss->ss_control_seq == NULL);
4060
4061                 OBD_ALLOC_PTR(ss->ss_control_seq);
4062                 if (ss->ss_control_seq == NULL)
4063                         RETURN(-ENOMEM);
4064
4065                 rc = seq_server_init(ss->ss_control_seq,
4066                                      m->mdt_bottom, uuid,
4067                                      LUSTRE_SEQ_CONTROLLER,
4068                                      ss,
4069                                      env);
4070
4071                 if (rc)
4072                         GOTO(out_seq_fini, rc);
4073
4074                 OBD_ALLOC_PTR(ss->ss_client_seq);
4075                 if (ss->ss_client_seq == NULL)
4076                         GOTO(out_seq_fini, rc = -ENOMEM);
4077
4078                 OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
4079                 if (prefix == NULL) {
4080                         OBD_FREE_PTR(ss->ss_client_seq);
4081                         GOTO(out_seq_fini, rc = -ENOMEM);
4082                 }
4083
4084                 snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s",
4085                          uuid);
4086
4087                 /*
4088                  * Init seq-controller client after seq-controller server is
4089                  * ready. Pass ss->ss_control_seq to it for direct talking.
4090                  */
4091                 rc = seq_client_init(ss->ss_client_seq, NULL,
4092                                      LUSTRE_SEQ_METADATA, prefix,
4093                                      ss->ss_control_seq);
4094                 OBD_FREE(prefix, MAX_OBD_NAME + 5);
4095
4096                 if (rc)
4097                         GOTO(out_seq_fini, rc);
4098         }
4099
4100         /* Init seq-server on local MDT */
4101         LASSERT(ss->ss_server_seq == NULL);
4102
4103         OBD_ALLOC_PTR(ss->ss_server_seq);
4104         if (ss->ss_server_seq == NULL)
4105                 GOTO(out_seq_fini, rc = -ENOMEM);
4106
4107         rc = seq_server_init(ss->ss_server_seq,
4108                              m->mdt_bottom, uuid,
4109                              LUSTRE_SEQ_SERVER,
4110                              ss,
4111                              env);
4112         if (rc)
4113                 GOTO(out_seq_fini, rc = -ENOMEM);
4114
4115         /* Assign seq-controller client to local seq-server. */
4116         if (ss->ss_node_id == 0) {
4117                 LASSERT(ss->ss_client_seq != NULL);
4118
4119                 rc = seq_server_set_cli(ss->ss_server_seq,
4120                                         ss->ss_client_seq,
4121                                         env);
4122         }
4123
4124         EXIT;
4125 out_seq_fini:
4126         if (rc)
4127                 mdt_seq_fini(env, m);
4128
4129         return rc;
4130 }
4131
4132 /*
4133  * FLD wrappers
4134  */
4135 static int mdt_fld_fini(const struct lu_env *env,
4136                         struct mdt_device *m)
4137 {
4138         struct seq_server_site *ss = mdt_seq_site(m);
4139         ENTRY;
4140
4141         if (ss && ss->ss_server_fld) {
4142                 fld_server_fini(env, ss->ss_server_fld);
4143                 OBD_FREE_PTR(ss->ss_server_fld);
4144                 ss->ss_server_fld = NULL;
4145         }
4146
4147         RETURN(0);
4148 }
4149
4150 static int mdt_fld_init(const struct lu_env *env,
4151                         const char *uuid,
4152                         struct mdt_device *m)
4153 {
4154         struct seq_server_site *ss;
4155         int rc;
4156         ENTRY;
4157
4158         ss = mdt_seq_site(m);
4159
4160         OBD_ALLOC_PTR(ss->ss_server_fld);
4161         if (ss->ss_server_fld == NULL)
4162                 RETURN(rc = -ENOMEM);
4163
4164         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
4165                              ss->ss_node_id, LU_SEQ_RANGE_MDT);
4166         if (rc) {
4167                 OBD_FREE_PTR(ss->ss_server_fld);
4168                 ss->ss_server_fld = NULL;
4169                 RETURN(rc);
4170         }
4171
4172         RETURN(0);
4173 }
4174
4175 static void mdt_stack_pre_fini(const struct lu_env *env,
4176                            struct mdt_device *m, struct lu_device *top)
4177 {
4178         struct obd_device       *obd;
4179         struct lustre_cfg_bufs  *bufs;
4180         struct lustre_cfg       *lcfg;
4181         struct mdt_thread_info  *info;
4182         ENTRY;
4183
4184         LASSERT(top);
4185
4186         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4187         LASSERT(info != NULL);
4188
4189         bufs = &info->mti_u.bufs;
4190
4191         LASSERT(m->mdt_child_exp);
4192         LASSERT(m->mdt_child_exp->exp_obd);
4193         obd = m->mdt_child_exp->exp_obd;
4194
4195         /* process cleanup, pass mdt obd name to get obd umount flags */
4196         /* XXX: this is needed because all layers are referenced by
4197          * objects (some of them are pinned by osd, for example *
4198          * the proper solution should be a model where object used
4199          * by osd only doesn't have mdt/mdd slices -bzzz */
4200         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4201         lustre_cfg_bufs_set_string(bufs, 1, NULL);
4202         lcfg = lustre_cfg_new(LCFG_PRE_CLEANUP, bufs);
4203         if (!lcfg) {
4204                 CERROR("%s:Cannot alloc lcfg!\n", mdt_obd_name(m));
4205                 return;
4206         }
4207         top->ld_ops->ldo_process_config(env, top, lcfg);
4208         lustre_cfg_free(lcfg);
4209         EXIT;
4210 }
4211
4212 static void mdt_stack_fini(const struct lu_env *env,
4213                            struct mdt_device *m, struct lu_device *top)
4214 {
4215         struct obd_device       *obd = mdt2obd_dev(m);
4216         struct lustre_cfg_bufs  *bufs;
4217         struct lustre_cfg       *lcfg;
4218         struct mdt_thread_info  *info;
4219         char                     flags[3] = "";
4220         ENTRY;
4221
4222         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4223         LASSERT(info != NULL);
4224
4225         lu_dev_del_linkage(top->ld_site, top);
4226
4227         lu_site_purge(env, top->ld_site, -1);
4228
4229         bufs = &info->mti_u.bufs;
4230         /* process cleanup, pass mdt obd name to get obd umount flags */
4231         /* another purpose is to let all layers to release their objects */
4232         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4233         if (obd->obd_force)
4234                 strcat(flags, "F");
4235         if (obd->obd_fail)
4236                 strcat(flags, "A");
4237         lustre_cfg_bufs_set_string(bufs, 1, flags);
4238         lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
4239         if (!lcfg) {
4240                 CERROR("Cannot alloc lcfg!\n");
4241                 return;
4242         }
4243         LASSERT(top);
4244         top->ld_ops->ldo_process_config(env, top, lcfg);
4245         lustre_cfg_free(lcfg);
4246
4247         lu_site_purge(env, top->ld_site, -1);
4248
4249         m->mdt_child = NULL;
4250         m->mdt_bottom = NULL;
4251
4252         obd_disconnect(m->mdt_child_exp);
4253         m->mdt_child_exp = NULL;
4254
4255         obd_disconnect(m->mdt_bottom_exp);
4256         m->mdt_child_exp = NULL;
4257 }
4258
4259 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
4260                                const char *next, struct obd_export **exp)
4261 {
4262         struct obd_connect_data *data = NULL;
4263         struct obd_device       *obd;
4264         int                      rc;
4265         ENTRY;
4266
4267         OBD_ALLOC_PTR(data);
4268         if (data == NULL)
4269                 GOTO(out, rc = -ENOMEM);
4270
4271         obd = class_name2obd(next);
4272         if (obd == NULL) {
4273                 CERROR("%s: can't locate next device: %s\n",
4274                        mdt_obd_name(m), next);
4275                 GOTO(out, rc = -ENOTCONN);
4276         }
4277
4278         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4279         data->ocd_version = LUSTRE_VERSION_CODE;
4280
4281         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
4282         if (rc) {
4283                 CERROR("%s: cannot connect to next dev %s (%d)\n",
4284                        mdt_obd_name(m), next, rc);
4285                 GOTO(out, rc);
4286         }
4287
4288 out:
4289         if (data)
4290                 OBD_FREE_PTR(data);
4291         RETURN(rc);
4292 }
4293
4294 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
4295                           struct lustre_cfg *cfg)
4296 {
4297         char                   *dev = lustre_cfg_string(cfg, 0);
4298         int                     rc, name_size, uuid_size;
4299         char                   *name, *uuid, *p;
4300         struct lustre_cfg_bufs *bufs;
4301         struct lustre_cfg      *lcfg;
4302         struct obd_device      *obd;
4303         struct lustre_profile  *lprof;
4304         struct lu_site         *site;
4305         ENTRY;
4306
4307         /* in 1.8 we had the only device in the stack - MDS.
4308          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
4309          * in 2.3 OSD is instantiated by obd_mount.c, so we need
4310          * to generate names and setup MDT, MDD. MDT will be using
4311          * generated name to connect to MDD. for MDD the next device
4312          * will be LOD with name taken from so called "profile" which
4313          * is generated by mount_option line
4314          *
4315          * 1.8 MGS generates config. commands like this:
4316          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
4317          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
4318          * 2.0 MGS generates config. commands like this:
4319          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
4320          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4321          *                    3:lustre-MDT0000-mdtlov  4:f
4322          *
4323          * we generate MDD name from MDT one, just replacing T with D
4324          *
4325          * after all the preparations, the logical equivalent will be
4326          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
4327          *                    3:lustre-MDT0000-mdtlov  4:f
4328          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4329          *                    3:lustre-MDD0000  4:f
4330          *
4331          *  notice we build the stack from down to top: MDD first, then MDT */
4332
4333         name_size = MAX_OBD_NAME;
4334         uuid_size = MAX_OBD_NAME;
4335
4336         OBD_ALLOC(name, name_size);
4337         OBD_ALLOC(uuid, uuid_size);
4338         if (name == NULL || uuid == NULL)
4339                 GOTO(cleanup_mem, rc = -ENOMEM);
4340
4341         OBD_ALLOC_PTR(bufs);
4342         if (!bufs)
4343                 GOTO(cleanup_mem, rc = -ENOMEM);
4344
4345         strcpy(name, dev);
4346         p = strstr(name, "-MDT");
4347         if (p == NULL)
4348                 GOTO(cleanup_mem, rc = -ENOMEM);
4349         p[3] = 'D';
4350
4351         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
4352
4353         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4354         if (lprof == NULL || lprof->lp_dt == NULL) {
4355                 CERROR("can't find the profile: %s\n",
4356                        lustre_cfg_string(cfg, 0));
4357                 GOTO(cleanup_mem, rc = -EINVAL);
4358         }
4359
4360         lustre_cfg_bufs_reset(bufs, name);
4361         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
4362         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4363         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4364
4365         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4366         if (!lcfg)
4367                 GOTO(free_bufs, rc = -ENOMEM);
4368
4369         rc = class_attach(lcfg);
4370         if (rc)
4371                 GOTO(lcfg_cleanup, rc);
4372
4373         obd = class_name2obd(name);
4374         if (!obd) {
4375                 CERROR("Can not find obd %s (%s in config)\n",
4376                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
4377                 GOTO(class_detach, rc = -EINVAL);
4378         }
4379
4380         lustre_cfg_free(lcfg);
4381
4382         lustre_cfg_bufs_reset(bufs, name);
4383         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4384         lustre_cfg_bufs_set_string(bufs, 2, dev);
4385         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4386
4387         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4388
4389         rc = class_setup(obd, lcfg);
4390         if (rc)
4391                 GOTO(class_detach, rc);
4392
4393         /* connect to MDD we just setup */
4394         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
4395         if (rc)
4396                 RETURN(rc);
4397
4398         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
4399         LASSERT(site);
4400         LASSERT(mdt_lu_site(mdt) == NULL);
4401         mdt->mdt_md_dev.md_lu_dev.ld_site = site;
4402         site->ls_top_dev = &mdt->mdt_md_dev.md_lu_dev;
4403         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
4404
4405
4406         /* now connect to bottom OSD */
4407         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
4408         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
4409         if (rc)
4410                 RETURN(rc);
4411         mdt->mdt_bottom =
4412                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
4413
4414
4415         rc = lu_env_refill((struct lu_env *)env);
4416         if (rc != 0)
4417                 CERROR("Failure to refill session: '%d'\n", rc);
4418
4419         lu_dev_add_linkage(site, &mdt->mdt_md_dev.md_lu_dev);
4420
4421         EXIT;
4422 class_detach:
4423         if (rc)
4424                 class_detach(obd, lcfg);
4425 lcfg_cleanup:
4426         lustre_cfg_free(lcfg);
4427 free_bufs:
4428         OBD_FREE_PTR(bufs);
4429 cleanup_mem:
4430         if (name)
4431                 OBD_FREE(name, name_size);
4432         if (uuid)
4433                 OBD_FREE(uuid, uuid_size);
4434         RETURN(rc);
4435 }
4436
4437 /* setup quota master target on MDT0 */
4438 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
4439                           struct lustre_cfg *cfg)
4440 {
4441         struct obd_device       *obd;
4442         char                    *dev = lustre_cfg_string(cfg, 0);
4443         char                    *qmtname, *uuid, *p;
4444         struct lustre_cfg_bufs  *bufs;
4445         struct lustre_cfg       *lcfg;
4446         struct lustre_profile   *lprof;
4447         struct obd_connect_data *data;
4448         int                      rc;
4449         ENTRY;
4450
4451         LASSERT(mdt->mdt_qmt_exp == NULL);
4452         LASSERT(mdt->mdt_qmt_dev == NULL);
4453
4454         /* quota master is on MDT0 only for now */
4455         if (mdt->mdt_seq_site.ss_node_id != 0)
4456                 RETURN(0);
4457
4458         /* MGS generates config commands which look as follows:
4459          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4460          *                    3:lustre-MDT0000-mdtlov  4:f
4461          *
4462          * We generate the QMT name from the MDT one, just replacing MD with QM
4463          * after all the preparations, the logical equivalent will be:
4464          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
4465          *                    3:lustre-MDT0000-osd  4:f */
4466         OBD_ALLOC(qmtname, MAX_OBD_NAME);
4467         OBD_ALLOC(uuid, UUID_MAX);
4468         OBD_ALLOC_PTR(bufs);
4469         OBD_ALLOC_PTR(data);
4470         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
4471                 GOTO(cleanup_mem, rc = -ENOMEM);
4472
4473         strcpy(qmtname, dev);
4474         p = strstr(qmtname, "-MDT");
4475         if (p == NULL)
4476                 GOTO(cleanup_mem, rc = -ENOMEM);
4477         /* replace MD with QM */
4478         p[1] = 'Q';
4479         p[2] = 'M';
4480
4481         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
4482
4483         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4484         if (lprof == NULL || lprof->lp_dt == NULL) {
4485                 CERROR("can't find profile for %s\n",
4486                        lustre_cfg_string(cfg, 0));
4487                 GOTO(cleanup_mem, rc = -EINVAL);
4488         }
4489
4490         lustre_cfg_bufs_reset(bufs, qmtname);
4491         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
4492         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4493         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4494
4495         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4496         if (!lcfg)
4497                 GOTO(cleanup_mem, rc = -ENOMEM);
4498
4499         rc = class_attach(lcfg);
4500         if (rc)
4501                 GOTO(lcfg_cleanup, rc);
4502
4503         obd = class_name2obd(qmtname);
4504         if (!obd) {
4505                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
4506                        lustre_cfg_string(cfg, 0));
4507                 GOTO(class_detach, rc = -EINVAL);
4508         }
4509
4510         lustre_cfg_free(lcfg);
4511
4512         lustre_cfg_bufs_reset(bufs, qmtname);
4513         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4514         lustre_cfg_bufs_set_string(bufs, 2, dev);
4515
4516         /* for quota, the next device should be the OSD device */
4517         lustre_cfg_bufs_set_string(bufs, 3,
4518                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
4519
4520         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4521
4522         rc = class_setup(obd, lcfg);
4523         if (rc)
4524                 GOTO(class_detach, rc);
4525
4526         mdt->mdt_qmt_dev = obd->obd_lu_dev;
4527
4528         /* configure local quota objects */
4529         rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
4530                                                    &mdt->mdt_md_dev.md_lu_dev,
4531                                                    mdt->mdt_qmt_dev);
4532         if (rc)
4533                 GOTO(class_cleanup, rc);
4534
4535         /* connect to quota master target */
4536         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4537         data->ocd_version = LUSTRE_VERSION_CODE;
4538         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
4539                          data, NULL);
4540         if (rc) {
4541                 CERROR("cannot connect to quota master device %s (%d)\n",
4542                        qmtname, rc);
4543                 GOTO(class_cleanup, rc);
4544         }
4545
4546         EXIT;
4547 class_cleanup:
4548         if (rc) {
4549                 class_manual_cleanup(obd);
4550                 mdt->mdt_qmt_dev = NULL;
4551         }
4552 class_detach:
4553         if (rc)
4554                 class_detach(obd, lcfg);
4555 lcfg_cleanup:
4556         lustre_cfg_free(lcfg);
4557 cleanup_mem:
4558         if (bufs)
4559                 OBD_FREE_PTR(bufs);
4560         if (qmtname)
4561                 OBD_FREE(qmtname, MAX_OBD_NAME);
4562         if (uuid)
4563                 OBD_FREE(uuid, UUID_MAX);
4564         if (data)
4565                 OBD_FREE_PTR(data);
4566         return rc;
4567 }
4568
4569 /* Shutdown quota master target associated with mdt */
4570 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
4571 {
4572         ENTRY;
4573
4574         if (mdt->mdt_qmt_exp == NULL)
4575                 RETURN_EXIT;
4576         LASSERT(mdt->mdt_qmt_dev != NULL);
4577
4578         /* the qmt automatically shuts down when the mdt disconnects */
4579         obd_disconnect(mdt->mdt_qmt_exp);
4580         mdt->mdt_qmt_exp = NULL;
4581         mdt->mdt_qmt_dev = NULL;
4582         EXIT;
4583 }
4584
4585 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4586 {
4587         struct md_device  *next = m->mdt_child;
4588         struct lu_device  *d    = &m->mdt_md_dev.md_lu_dev;
4589         struct obd_device *obd = mdt2obd_dev(m);
4590         ENTRY;
4591
4592         target_recovery_fini(obd);
4593
4594         ping_evictor_stop();
4595
4596
4597         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
4598         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4599         obd_exports_barrier(obd);
4600         obd_zombie_barrier();
4601
4602         mdt_procfs_fini(m);
4603
4604         tgt_fini(env, &m->mdt_lut);
4605         mdt_fs_cleanup(env, m);
4606         upcall_cache_cleanup(m->mdt_identity_cache);
4607         m->mdt_identity_cache = NULL;
4608
4609         if (m->mdt_namespace != NULL) {
4610                 ldlm_namespace_free(m->mdt_namespace, NULL,
4611                                     d->ld_obd->obd_force);
4612                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4613         }
4614
4615         mdt_quota_fini(env, m);
4616
4617         cfs_free_nidlist(&m->mdt_nosquash_nids);
4618         if (m->mdt_nosquash_str) {
4619                 OBD_FREE(m->mdt_nosquash_str, m->mdt_nosquash_strlen);
4620                 m->mdt_nosquash_str = NULL;
4621                 m->mdt_nosquash_strlen = 0;
4622         }
4623
4624         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_PAUSE_LFSCK, 0, NULL);
4625
4626         mdt_seq_fini(env, m);
4627         mdt_fld_fini(env, m);
4628         sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
4629
4630         next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL);
4631         cfs_timer_disarm(&m->mdt_ck_timer);
4632         mdt_ck_thread_stop(m);
4633
4634         /*
4635          * Finish the stack
4636          */
4637         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4638
4639         LASSERT(cfs_atomic_read(&d->ld_ref) == 0);
4640
4641         server_put_mount(mdt_obd_name(m), NULL);
4642
4643         EXIT;
4644 }
4645
4646 static int mdt_adapt_sptlrpc_conf(struct obd_device *obd, int initial)
4647 {
4648         struct mdt_device       *m = mdt_dev(obd->obd_lu_dev);
4649         struct sptlrpc_rule_set  tmp_rset;
4650         int                      rc;
4651
4652         sptlrpc_rule_set_init(&tmp_rset);
4653         rc = sptlrpc_conf_target_get_rules(obd, &tmp_rset, initial);
4654         if (rc) {
4655                 CERROR("mdt %s: failed get sptlrpc rules: %d\n",
4656                        mdt_obd_name(m), rc);
4657                 return rc;
4658         }
4659
4660         sptlrpc_target_update_exp_flavor(obd, &tmp_rset);
4661
4662         write_lock(&m->mdt_sptlrpc_lock);
4663         sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
4664         m->mdt_sptlrpc_rset = tmp_rset;
4665         write_unlock(&m->mdt_sptlrpc_lock);
4666
4667         return 0;
4668 }
4669
4670 int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4671
4672 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4673                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
4674 {
4675         struct mdt_thread_info    *info;
4676         struct obd_device         *obd;
4677         const char                *dev = lustre_cfg_string(cfg, 0);
4678         const char                *num = lustre_cfg_string(cfg, 2);
4679         struct lustre_mount_info  *lmi = NULL;
4680         struct lustre_sb_info     *lsi;
4681         struct lu_site            *s;
4682         struct seq_server_site    *ss_site;
4683         const char                *identity_upcall = "NONE";
4684         struct md_device          *next;
4685         int                        rc;
4686         int                        node_id;
4687         mntopt_t                   mntopts;
4688         ENTRY;
4689
4690         md_device_init(&m->mdt_md_dev, ldt);
4691         /*
4692          * Environment (env) might be missing mdt_thread_key values at that
4693          * point, if device is allocated when mdt_thread_key is in QUIESCENT
4694          * mode.
4695          *
4696          * Usually device allocation path doesn't use module key values, but
4697          * mdt has to do a lot of work here, so allocate key value.
4698          */
4699         rc = lu_env_refill((struct lu_env *)env);
4700         if (rc != 0)
4701                 RETURN(rc);
4702
4703         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4704         LASSERT(info != NULL);
4705
4706         obd = class_name2obd(dev);
4707         LASSERT(obd != NULL);
4708
4709         m->mdt_max_mdsize = MAX_MD_SIZE; /* 4 stripes */
4710
4711         m->mdt_som_conf = 0;
4712
4713         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4714         lmi = server_get_mount(dev);
4715         if (lmi == NULL) {
4716                 CERROR("Cannot get mount info for %s!\n", dev);
4717                 RETURN(-EFAULT);
4718         } else {
4719                 lsi = s2lsi(lmi->lmi_sb);
4720                 /* CMD is supported only in IAM mode */
4721                 LASSERT(num);
4722                 node_id = simple_strtol(num, NULL, 10);
4723                 obd->u.obt.obt_magic = OBT_MAGIC;
4724         }
4725
4726         rwlock_init(&m->mdt_sptlrpc_lock);
4727         sptlrpc_rule_set_init(&m->mdt_sptlrpc_rset);
4728
4729         spin_lock_init(&m->mdt_ioepoch_lock);
4730         m->mdt_opts.mo_compat_resname = 0;
4731         m->mdt_opts.mo_mds_capa = 1;
4732         m->mdt_opts.mo_oss_capa = 1;
4733         m->mdt_capa_timeout = CAPA_TIMEOUT;
4734         m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1;
4735         m->mdt_ck_timeout = CAPA_KEY_TIMEOUT;
4736         m->mdt_squash_uid = 0;
4737         m->mdt_squash_gid = 0;
4738         CFS_INIT_LIST_HEAD(&m->mdt_nosquash_nids);
4739         m->mdt_nosquash_str = NULL;
4740         m->mdt_nosquash_strlen = 0;
4741         init_rwsem(&m->mdt_squash_sem);
4742         spin_lock_init(&m->mdt_osfs_lock);
4743         m->mdt_osfs_age = cfs_time_shift_64(-1000);
4744         m->mdt_enable_remote_dir = 0;
4745         m->mdt_enable_remote_dir_gid = 0;
4746
4747         m->mdt_md_dev.md_lu_dev.ld_ops = &mdt_lu_ops;
4748         m->mdt_md_dev.md_lu_dev.ld_obd = obd;
4749         /* set this lu_device to obd, because error handling need it */
4750         obd->obd_lu_dev = &m->mdt_md_dev.md_lu_dev;
4751
4752         /* init the stack */
4753         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
4754         if (rc) {
4755                 CERROR("%s: Can't init device stack, rc %d\n",
4756                        mdt_obd_name(m), rc);
4757                 GOTO(err_lmi, rc);
4758         }
4759
4760         s = mdt_lu_site(m);
4761         ss_site = mdt_seq_site(m);
4762         s->ld_seq_site = ss_site;
4763         ss_site->ss_lu = s;
4764
4765         /* set server index */
4766         ss_site->ss_node_id = node_id;
4767
4768         /* failover is the default
4769          * FIXME: we do not failout mds0/mgs, which may cause some problems.
4770          * assumed whose ss_node_id == 0 XXX
4771          * */
4772         obd->obd_replayable = 1;
4773         /* No connection accepted until configurations will finish */
4774         obd->obd_no_conn = 1;
4775
4776         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4777                 char *str = lustre_cfg_string(cfg, 4);
4778                 if (strchr(str, 'n')) {
4779                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
4780                         obd->obd_replayable = 0;
4781                 }
4782         }
4783
4784         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom);
4785         if (rc)
4786                 GOTO(err_fini_stack, rc);
4787
4788         rc = mdt_fld_init(env, mdt_obd_name(m), m);
4789         if (rc)
4790                 GOTO(err_lut, rc);
4791
4792         rc = mdt_seq_init(env, mdt_obd_name(m), m);
4793         if (rc)
4794                 GOTO(err_fini_fld, rc);
4795
4796         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
4797                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
4798         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4799                                               LDLM_NAMESPACE_SERVER,
4800                                               LDLM_NAMESPACE_GREEDY,
4801                                               LDLM_NS_TYPE_MDT);
4802         if (m->mdt_namespace == NULL)
4803                 GOTO(err_fini_seq, rc = -ENOMEM);
4804
4805         m->mdt_namespace->ns_lvbp = m;
4806         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
4807
4808         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4809         /* set obd_namespace for compatibility with old code */
4810         obd->obd_namespace = m->mdt_namespace;
4811
4812         cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m);
4813
4814         rc = mdt_ck_thread_start(m);
4815         if (rc)
4816                 GOTO(err_free_ns, rc);
4817
4818         rc = mdt_fs_setup(env, m, obd, lsi);
4819         if (rc)
4820                 GOTO(err_capa, rc);
4821
4822         mdt_adapt_sptlrpc_conf(obd, 1);
4823
4824         next = m->mdt_child;
4825         rc = next->md_ops->mdo_iocontrol(env, next, OBD_IOC_GET_MNTOPT, 0,
4826                                          &mntopts);
4827         if (rc)
4828                 GOTO(err_llog_cleanup, rc);
4829
4830         if (mntopts & MNTOPT_USERXATTR)
4831                 m->mdt_opts.mo_user_xattr = 1;
4832         else
4833                 m->mdt_opts.mo_user_xattr = 0;
4834
4835         if (mntopts & MNTOPT_ACL)
4836                 m->mdt_opts.mo_acl = 1;
4837         else
4838                 m->mdt_opts.mo_acl = 0;
4839
4840         /* XXX: to support suppgid for ACL, we enable identity_upcall
4841          * by default, otherwise, maybe got unexpected -EACCESS. */
4842         if (m->mdt_opts.mo_acl)
4843                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4844
4845         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
4846                                                 identity_upcall,
4847                                                 &mdt_identity_upcall_cache_ops);
4848         if (IS_ERR(m->mdt_identity_cache)) {
4849                 rc = PTR_ERR(m->mdt_identity_cache);
4850                 m->mdt_identity_cache = NULL;
4851                 GOTO(err_llog_cleanup, rc);
4852         }
4853
4854         rc = mdt_procfs_init(m, dev);
4855         if (rc) {
4856                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4857                 GOTO(err_recovery, rc);
4858         }
4859
4860         rc = mdt_quota_init(env, m, cfg);
4861         if (rc)
4862                 GOTO(err_procfs, rc);
4863
4864         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
4865         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
4866                            "mdt_ldlm_client", m->mdt_ldlm_client);
4867
4868         ping_evictor_start();
4869
4870         /* recovery will be started upon mdt_prepare()
4871          * when the whole stack is complete and ready
4872          * to serve the requests */
4873
4874         mdt_init_capa_ctxt(env, m);
4875
4876         /* Reduce the initial timeout on an MDS because it doesn't need such
4877          * a long timeout as an OST does. Adaptive timeouts will adjust this
4878          * value appropriately. */
4879         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4880                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4881
4882         RETURN(0);
4883
4884 err_procfs:
4885         mdt_procfs_fini(m);
4886 err_recovery:
4887         target_recovery_fini(obd);
4888         upcall_cache_cleanup(m->mdt_identity_cache);
4889         m->mdt_identity_cache = NULL;
4890 err_llog_cleanup:
4891         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4892         mdt_fs_cleanup(env, m);
4893 err_capa:
4894         cfs_timer_disarm(&m->mdt_ck_timer);
4895         mdt_ck_thread_stop(m);
4896 err_free_ns:
4897         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4898         obd->obd_namespace = m->mdt_namespace = NULL;
4899 err_fini_seq:
4900         mdt_seq_fini(env, m);
4901 err_fini_fld:
4902         mdt_fld_fini(env, m);
4903 err_lut:
4904         tgt_fini(env, &m->mdt_lut);
4905 err_fini_stack:
4906         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4907 err_lmi:
4908         if (lmi)
4909                 server_put_mount(dev, lmi->lmi_mnt);
4910         return (rc);
4911 }
4912
4913 /* For interoperability, the left element is old parameter, the right one
4914  * is the new version of the parameter, if some parameter is deprecated,
4915  * the new version should be set as NULL. */
4916 static struct cfg_interop_param mdt_interop_param[] = {
4917         { "mdt.group_upcall",   NULL },
4918         { "mdt.quota_type",     NULL },
4919         { "mdd.quota_type",     NULL },
4920         { "mdt.rootsquash",     "mdt.root_squash" },
4921         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
4922         { NULL }
4923 };
4924
4925 /* used by MGS to process specific configurations */
4926 static int mdt_process_config(const struct lu_env *env,
4927                               struct lu_device *d, struct lustre_cfg *cfg)
4928 {
4929         struct mdt_device *m = mdt_dev(d);
4930         struct md_device *md_next = m->mdt_child;
4931         struct lu_device *next = md2lu_dev(md_next);
4932         int rc;
4933         ENTRY;
4934
4935         switch (cfg->lcfg_command) {
4936         case LCFG_PARAM: {
4937                 struct lprocfs_static_vars  lvars;
4938                 struct obd_device          *obd = d->ld_obd;
4939
4940                 /* For interoperability */
4941                 struct cfg_interop_param   *ptr = NULL;
4942                 struct lustre_cfg          *old_cfg = NULL;
4943                 char                       *param = NULL;
4944
4945                 param = lustre_cfg_string(cfg, 1);
4946                 if (param == NULL) {
4947                         CERROR("param is empty\n");
4948                         rc = -EINVAL;
4949                         break;
4950                 }
4951
4952                 ptr = class_find_old_param(param, mdt_interop_param);
4953                 if (ptr != NULL) {
4954                         if (ptr->new_param == NULL) {
4955                                 rc = 0;
4956                                 CWARN("For interoperability, skip this %s."
4957                                       " It is obsolete.\n", ptr->old_param);
4958                                 break;
4959                         }
4960
4961                         CWARN("Found old param %s, changed it to %s.\n",
4962                               ptr->old_param, ptr->new_param);
4963
4964                         old_cfg = cfg;
4965                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
4966                         if (IS_ERR(cfg)) {
4967                                 rc = PTR_ERR(cfg);
4968                                 break;
4969                         }
4970                 }
4971
4972                 lprocfs_mdt_init_vars(&lvars);
4973                 rc = class_process_proc_param(PARAM_MDT, lvars.obd_vars,
4974                                               cfg, obd);
4975                 if (rc > 0 || rc == -ENOSYS)
4976                         /* we don't understand; pass it on */
4977                         rc = next->ld_ops->ldo_process_config(env, next, cfg);
4978
4979                 if (old_cfg != NULL)
4980                         lustre_cfg_free(cfg);
4981
4982                 break;
4983         }
4984         default:
4985                 /* others are passed further */
4986                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4987                 break;
4988         }
4989         RETURN(rc);
4990 }
4991
4992 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4993                                           const struct lu_object_header *hdr,
4994                                           struct lu_device *d)
4995 {
4996         struct mdt_object *mo;
4997
4998         ENTRY;
4999
5000         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, __GFP_IO);
5001         if (mo != NULL) {
5002                 struct lu_object *o;
5003                 struct lu_object_header *h;
5004
5005                 o = &mo->mot_obj.mo_lu;
5006                 h = &mo->mot_header;
5007                 lu_object_header_init(h);
5008                 lu_object_init(o, h, d);
5009                 lu_object_add_top(h, o);
5010                 o->lo_ops = &mdt_obj_ops;
5011                 mutex_init(&mo->mot_ioepoch_mutex);
5012                 mutex_init(&mo->mot_lov_mutex);
5013                 RETURN(o);
5014         } else
5015                 RETURN(NULL);
5016 }
5017
5018 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
5019                            const struct lu_object_conf *unused)
5020 {
5021         struct mdt_device *d = mdt_dev(o->lo_dev);
5022         struct lu_device  *under;
5023         struct lu_object  *below;
5024         int                rc = 0;
5025         ENTRY;
5026
5027         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
5028                PFID(lu_object_fid(o)));
5029
5030         under = &d->mdt_child->md_lu_dev;
5031         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
5032         if (below != NULL) {
5033                 lu_object_add(o, below);
5034         } else
5035                 rc = -ENOMEM;
5036
5037         RETURN(rc);
5038 }
5039
5040 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
5041 {
5042         struct mdt_object *mo = mdt_obj(o);
5043         struct lu_object_header *h;
5044         ENTRY;
5045
5046         h = o->lo_header;
5047         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
5048                PFID(lu_object_fid(o)));
5049
5050         lu_object_fini(o);
5051         lu_object_header_fini(h);
5052         OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
5053
5054         EXIT;
5055 }
5056
5057 static int mdt_object_print(const struct lu_env *env, void *cookie,
5058                             lu_printer_t p, const struct lu_object *o)
5059 {
5060         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
5061         return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p(ioepoch="LPU64" "
5062                     "flags="LPX64", epochcount=%d, writecount=%d)",
5063                     mdto, mdto->mot_ioepoch, mdto->mot_flags,
5064                     mdto->mot_ioepoch_count, mdto->mot_writecount);
5065 }
5066
5067 static int mdt_prepare(const struct lu_env *env,
5068                 struct lu_device *pdev,
5069                 struct lu_device *cdev)
5070 {
5071         struct mdt_device *mdt = mdt_dev(cdev);
5072         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
5073         struct obd_device *obd = cdev->ld_obd;
5074         struct lfsck_start_param lsp;
5075         int rc;
5076
5077         ENTRY;
5078
5079         LASSERT(obd);
5080
5081         rc = next->ld_ops->ldo_prepare(env, cdev, next);
5082         if (rc)
5083                 RETURN(rc);
5084
5085         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
5086         if (rc)
5087                 RETURN(rc);
5088
5089         lsp.lsp_start = NULL;
5090         lsp.lsp_namespace = mdt->mdt_namespace;
5091         rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
5092                                                    OBD_IOC_START_LFSCK,
5093                                                    0, &lsp);
5094         if (rc != 0) {
5095                 CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
5096                       mdt_obd_name(mdt), rc);
5097                 rc = 0;
5098         }
5099
5100         if (mdt->mdt_seq_site.ss_node_id == 0) {
5101                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
5102                                                          &mdt->mdt_md_root_fid);
5103                 if (rc)
5104                         RETURN(rc);
5105         }
5106         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
5107         target_recovery_init(&mdt->mdt_lut, mdt_recovery_handle);
5108         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
5109         LASSERT(obd->obd_no_conn);
5110         spin_lock(&obd->obd_dev_lock);
5111         obd->obd_no_conn = 0;
5112         spin_unlock(&obd->obd_dev_lock);
5113
5114         if (obd->obd_recovering == 0)
5115                 mdt_postrecov(env, mdt);
5116
5117         RETURN(rc);
5118 }
5119
5120 const struct lu_device_operations mdt_lu_ops = {
5121         .ldo_object_alloc   = mdt_object_alloc,
5122         .ldo_process_config = mdt_process_config,
5123         .ldo_prepare        = mdt_prepare,
5124 };
5125
5126 static const struct lu_object_operations mdt_obj_ops = {
5127         .loo_object_init    = mdt_object_init,
5128         .loo_object_free    = mdt_object_free,
5129         .loo_object_print   = mdt_object_print
5130 };
5131
5132 static int mdt_obd_set_info_async(const struct lu_env *env,
5133                                   struct obd_export *exp,
5134                                   __u32 keylen, void *key,
5135                                   __u32 vallen, void *val,
5136                                   struct ptlrpc_request_set *set)
5137 {
5138         struct obd_device     *obd = exp->exp_obd;
5139         int                    rc;
5140         ENTRY;
5141
5142         LASSERT(obd);
5143
5144         if (KEY_IS(KEY_SPTLRPC_CONF)) {
5145                 rc = mdt_adapt_sptlrpc_conf(obd, 0);
5146                 RETURN(rc);
5147         }
5148
5149         RETURN(0);
5150 }
5151
5152 /**
5153  * Match client and server connection feature flags.
5154  *
5155  * Compute the compatibility flags for a connection request based on
5156  * features mutually supported by client and server.
5157  *
5158  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
5159  * must not be updated here, otherwise a partially initialized value may
5160  * be exposed. After the connection request is successfully processed,
5161  * the top-level MDT connect request handler atomically updates the export
5162  * connect flags from the obd_connect_data::ocd_connect_flags field of the
5163  * reply. \see mdt_connect().
5164  *
5165  * \param exp   the obd_export associated with this client/target pair
5166  * \param mdt   the target device for the connection
5167  * \param data  stores data for this connect request
5168  *
5169  * \retval 0       success
5170  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
5171  * \retval -EBADE  client and server feature requirements are incompatible
5172  */
5173 static int mdt_connect_internal(struct obd_export *exp,
5174                                 struct mdt_device *mdt,
5175                                 struct obd_connect_data *data)
5176 {
5177         LASSERT(data != NULL);
5178
5179         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
5180         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
5181
5182         /* If no known bits (which should not happen, probably,
5183            as everybody should support LOOKUP and UPDATE bits at least)
5184            revert to compat mode with plain locks. */
5185         if (!data->ocd_ibits_known &&
5186             data->ocd_connect_flags & OBD_CONNECT_IBITS)
5187                 data->ocd_connect_flags &= ~OBD_CONNECT_IBITS;
5188
5189         if (!mdt->mdt_opts.mo_acl)
5190                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
5191
5192         if (!mdt->mdt_opts.mo_user_xattr)
5193                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
5194
5195         if (!mdt->mdt_som_conf)
5196                 data->ocd_connect_flags &= ~OBD_CONNECT_SOM;
5197
5198         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
5199                 data->ocd_brw_size = min(data->ocd_brw_size,
5200                                          (__u32)MD_MAX_BRW_SIZE);
5201                 if (data->ocd_brw_size == 0) {
5202                         CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
5203                                " ocd_version: %x ocd_grant: %d "
5204                                "ocd_index: %u ocd_brw_size is "
5205                                "unexpectedly zero, network data "
5206                                "corruption? Refusing connection of this"
5207                                " client\n",
5208                                mdt_obd_name(mdt),
5209                                exp->exp_client_uuid.uuid,
5210                                exp, data->ocd_connect_flags, data->ocd_version,
5211                                data->ocd_grant, data->ocd_index);
5212                         return -EPROTO;
5213                 }
5214         }
5215
5216         /* NB: Disregard the rule against updating
5217          * exp_connect_data.ocd_connect_flags in this case, since
5218          * tgt_client_new() needs to know if this is a lightweight
5219          * connection, and it is safe to expose this flag before
5220          * connection processing completes. */
5221         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
5222                 spin_lock(&exp->exp_lock);
5223                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
5224                 spin_unlock(&exp->exp_lock);
5225         }
5226
5227         data->ocd_version = LUSTRE_VERSION_CODE;
5228         exp->exp_connect_data = *data;
5229         exp->exp_mdt_data.med_ibits_known = data->ocd_ibits_known;
5230
5231         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
5232                 CWARN("%s: MDS requires FID support, but client not\n",
5233                       mdt_obd_name(mdt));
5234                 return -EBADE;
5235         }
5236
5237         if (mdt->mdt_som_conf &&
5238             !(data->ocd_connect_flags & (OBD_CONNECT_LIGHTWEIGHT |
5239                                          OBD_CONNECT_MDS_MDS |
5240                                          OBD_CONNECT_SOM))) {
5241                 CWARN("%s: MDS has SOM enabled, but client does not support "
5242                       "it\n", mdt_obd_name(mdt));
5243                 return -EBADE;
5244         }
5245
5246         if (OCD_HAS_FLAG(data, PINGLESS)) {
5247                 if (ptlrpc_pinger_suppress_pings()) {
5248                         spin_lock(&exp->exp_obd->obd_dev_lock);
5249                         list_del_init(&exp->exp_obd_chain_timed);
5250                         spin_unlock(&exp->exp_obd->obd_dev_lock);
5251                 } else {
5252                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
5253                 }
5254         }
5255
5256         return 0;
5257 }
5258
5259 static int mdt_connect_check_sptlrpc(struct mdt_device *mdt,
5260                                      struct obd_export *exp,
5261                                      struct ptlrpc_request *req)
5262 {
5263         struct sptlrpc_flavor   flvr;
5264         int                     rc = 0;
5265
5266         if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
5267                 read_lock(&mdt->mdt_sptlrpc_lock);
5268                 sptlrpc_target_choose_flavor(&mdt->mdt_sptlrpc_rset,
5269                                              req->rq_sp_from,
5270                                              req->rq_peer.nid,
5271                                              &flvr);
5272                 read_unlock(&mdt->mdt_sptlrpc_lock);
5273
5274                 spin_lock(&exp->exp_lock);
5275
5276                 exp->exp_sp_peer = req->rq_sp_from;
5277                 exp->exp_flvr = flvr;
5278
5279                 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
5280                     exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
5281                         CERROR("unauthorized rpc flavor %x from %s, "
5282                                "expect %x\n", req->rq_flvr.sf_rpc,
5283                                libcfs_nid2str(req->rq_peer.nid),
5284                                exp->exp_flvr.sf_rpc);
5285                         rc = -EACCES;
5286                 }
5287
5288                 spin_unlock(&exp->exp_lock);
5289         } else {
5290                 if (exp->exp_sp_peer != req->rq_sp_from) {
5291                         CERROR("RPC source %s doesn't match %s\n",
5292                                sptlrpc_part2name(req->rq_sp_from),
5293                                sptlrpc_part2name(exp->exp_sp_peer));
5294                         rc = -EACCES;
5295                 } else {
5296                         rc = sptlrpc_target_export_check(exp, req);
5297                 }
5298         }
5299
5300         return rc;
5301 }
5302
5303 /* mds_connect copy */
5304 static int mdt_obd_connect(const struct lu_env *env,
5305                            struct obd_export **exp, struct obd_device *obd,
5306                            struct obd_uuid *cluuid,
5307                            struct obd_connect_data *data,
5308                            void *localdata)
5309 {
5310         struct mdt_thread_info *info;
5311         struct obd_export      *lexp;
5312         struct lustre_handle    conn = { 0 };
5313         struct mdt_device      *mdt;
5314         struct ptlrpc_request  *req;
5315         int                     rc;
5316         ENTRY;
5317
5318         LASSERT(env != NULL);
5319         if (!exp || !obd || !cluuid)
5320                 RETURN(-EINVAL);
5321
5322         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5323         req = info->mti_pill->rc_req;
5324         mdt = mdt_dev(obd->obd_lu_dev);
5325
5326         /*
5327          * first, check whether the stack is ready to handle requests
5328          * XXX: probably not very appropriate method is used now
5329          *      at some point we should find a better one
5330          */
5331         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) &&
5332             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
5333                 rc = obd_health_check(env, mdt->mdt_child_exp->exp_obd);
5334                 if (rc)
5335                         RETURN(-EAGAIN);
5336                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
5337         }
5338
5339         rc = class_connect(&conn, obd, cluuid);
5340         if (rc)
5341                 RETURN(rc);
5342
5343         lexp = class_conn2export(&conn);
5344         LASSERT(lexp != NULL);
5345
5346         rc = mdt_connect_check_sptlrpc(mdt, lexp, req);
5347         if (rc)
5348                 GOTO(out, rc);
5349
5350         if (OBD_FAIL_CHECK(OBD_FAIL_TGT_RCVG_FLAG))
5351                 lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_RECOVERING);
5352
5353         rc = mdt_connect_internal(lexp, mdt, data);
5354         if (rc == 0) {
5355                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5356
5357                 LASSERT(lcd);
5358                 info->mti_exp = lexp;
5359                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5360                 rc = tgt_client_new(env, lexp);
5361                 if (rc == 0)
5362                         mdt_export_stats_init(obd, lexp, localdata);
5363         }
5364
5365 out:
5366         if (rc != 0) {
5367                 class_disconnect(lexp);
5368                 *exp = NULL;
5369         } else {
5370                 *exp = lexp;
5371         }
5372
5373         RETURN(rc);
5374 }
5375
5376 static int mdt_obd_reconnect(const struct lu_env *env,
5377                              struct obd_export *exp, struct obd_device *obd,
5378                              struct obd_uuid *cluuid,
5379                              struct obd_connect_data *data,
5380                              void *localdata)
5381 {
5382         struct mdt_thread_info *info;
5383         struct mdt_device      *mdt;
5384         struct ptlrpc_request  *req;
5385         int                     rc;
5386         ENTRY;
5387
5388         if (exp == NULL || obd == NULL || cluuid == NULL)
5389                 RETURN(-EINVAL);
5390
5391         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5392         req = info->mti_pill->rc_req;
5393         mdt = mdt_dev(obd->obd_lu_dev);
5394
5395         rc = mdt_connect_check_sptlrpc(mdt, exp, req);
5396         if (rc)
5397                 RETURN(rc);
5398
5399         rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
5400         if (rc == 0)
5401                 mdt_export_stats_init(obd, exp, localdata);
5402
5403         RETURN(rc);
5404 }
5405
5406 static int mdt_export_cleanup(struct obd_export *exp)
5407 {
5408         struct mdt_export_data *med = &exp->exp_mdt_data;
5409         struct obd_device      *obd = exp->exp_obd;
5410         struct mdt_device      *mdt;
5411         struct mdt_thread_info *info;
5412         struct lu_env           env;
5413         CFS_LIST_HEAD(closing_list);
5414         struct mdt_file_data *mfd, *n;
5415         int rc = 0;
5416         ENTRY;
5417
5418         spin_lock(&med->med_open_lock);
5419         while (!cfs_list_empty(&med->med_open_head)) {
5420                 cfs_list_t *tmp = med->med_open_head.next;
5421                 mfd = cfs_list_entry(tmp, struct mdt_file_data, mfd_list);
5422
5423                 /* Remove mfd handle so it can't be found again.
5424                  * We are consuming the mfd_list reference here. */
5425                 class_handle_unhash(&mfd->mfd_handle);
5426                 cfs_list_move_tail(&mfd->mfd_list, &closing_list);
5427         }
5428         spin_unlock(&med->med_open_lock);
5429         mdt = mdt_dev(obd->obd_lu_dev);
5430         LASSERT(mdt != NULL);
5431
5432         rc = lu_env_init(&env, LCT_MD_THREAD);
5433         if (rc)
5434                 RETURN(rc);
5435
5436         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5437         LASSERT(info != NULL);
5438         memset(info, 0, sizeof *info);
5439         info->mti_env = &env;
5440         info->mti_mdt = mdt;
5441         info->mti_exp = exp;
5442
5443         if (!cfs_list_empty(&closing_list)) {
5444                 struct md_attr *ma = &info->mti_attr;
5445
5446                 /* Close any open files (which may also cause orphan unlinking). */
5447                 cfs_list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5448                         cfs_list_del_init(&mfd->mfd_list);
5449                         ma->ma_need = ma->ma_valid = 0;
5450                         /* Don't unlink orphan on failover umount, LU-184 */
5451                         if (exp->exp_flags & OBD_OPT_FAILOVER) {
5452                                 ma->ma_valid = MA_FLAGS;
5453                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5454                         }
5455                         mdt_mfd_close(info, mfd);
5456                 }
5457         }
5458         info->mti_mdt = NULL;
5459         /* cleanup client slot early */
5460         /* Do not erase record for recoverable client. */
5461         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5462                 tgt_client_del(&env, exp);
5463         lu_env_fini(&env);
5464
5465         RETURN(rc);
5466 }
5467
5468 static int mdt_obd_disconnect(struct obd_export *exp)
5469 {
5470         int rc;
5471         ENTRY;
5472
5473         LASSERT(exp);
5474         class_export_get(exp);
5475
5476         rc = server_disconnect_export(exp);
5477         if (rc != 0)
5478                 CDEBUG(D_IOCTL, "server disconnect error: %d\n", rc);
5479
5480         rc = mdt_export_cleanup(exp);
5481         class_export_put(exp);
5482         RETURN(rc);
5483 }
5484
5485 /* FIXME: Can we avoid using these two interfaces? */
5486 static int mdt_init_export(struct obd_export *exp)
5487 {
5488         struct mdt_export_data *med = &exp->exp_mdt_data;
5489         int                     rc;
5490         ENTRY;
5491
5492         CFS_INIT_LIST_HEAD(&med->med_open_head);
5493         spin_lock_init(&med->med_open_lock);
5494         mutex_init(&med->med_idmap_mutex);
5495         med->med_idmap = NULL;
5496         spin_lock(&exp->exp_lock);
5497         exp->exp_connecting = 1;
5498         spin_unlock(&exp->exp_lock);
5499
5500         /* self-export doesn't need client data and ldlm initialization */
5501         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5502                                      &exp->exp_client_uuid)))
5503                 RETURN(0);
5504
5505         rc = tgt_client_alloc(exp);
5506         if (rc)
5507                 GOTO(err, rc);
5508
5509         rc = ldlm_init_export(exp);
5510         if (rc)
5511                 GOTO(err_free, rc);
5512
5513         RETURN(rc);
5514
5515 err_free:
5516         tgt_client_free(exp);
5517 err:
5518         CERROR("%s: Failed to initialize export: rc = %d\n",
5519                exp->exp_obd->obd_name, rc);
5520         return rc;
5521 }
5522
5523 static int mdt_destroy_export(struct obd_export *exp)
5524 {
5525         ENTRY;
5526
5527         if (exp_connect_rmtclient(exp))
5528                 mdt_cleanup_idmap(&exp->exp_mdt_data);
5529
5530         target_destroy_export(exp);
5531         /* destroy can be called from failed obd_setup, so
5532          * checking uuid is safer than obd_self_export */
5533         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5534                                      &exp->exp_client_uuid)))
5535                 RETURN(0);
5536
5537         ldlm_destroy_export(exp);
5538         tgt_client_free(exp);
5539
5540         LASSERT(cfs_list_empty(&exp->exp_outstanding_replies));
5541         LASSERT(cfs_list_empty(&exp->exp_mdt_data.med_open_head));
5542
5543         RETURN(0);
5544 }
5545
5546 /** The maximum depth that fid2path() will search.
5547  * This is limited only because we want to store the fids for
5548  * historical path lookup purposes.
5549  */
5550 #define MAX_PATH_DEPTH 100
5551
5552 /** mdt_path() lookup structure. */
5553 struct path_lookup_info {
5554         __u64                   pli_recno;      /**< history point */
5555         __u64                   pli_currec;     /**< current record */
5556         struct lu_fid           pli_fid;
5557         struct lu_fid           pli_fids[MAX_PATH_DEPTH]; /**< path, in fids */
5558         struct mdt_object       *pli_mdt_obj;
5559         char                    *pli_path;      /**< full path */
5560         int                     pli_pathlen;
5561         int                     pli_linkno;     /**< which hardlink to follow */
5562         int                     pli_fidcount;   /**< number of \a pli_fids */
5563 };
5564
5565 static int mdt_links_read(struct mdt_thread_info *info,
5566                           struct mdt_object *mdt_obj, struct linkea_data *ldata)
5567 {
5568         int rc;
5569
5570         LASSERT(ldata->ld_buf->lb_buf != NULL);
5571
5572         if (!mdt_object_exists(mdt_obj))
5573                 return -ENODATA;
5574
5575         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5576                           ldata->ld_buf, XATTR_NAME_LINK);
5577         if (rc == -ERANGE) {
5578                 /* Buf was too small, figure out what we need. */
5579                 lu_buf_free(ldata->ld_buf);
5580                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5581                                   ldata->ld_buf, XATTR_NAME_LINK);
5582                 if (rc < 0)
5583                         return rc;
5584                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
5585                 if (ldata->ld_buf->lb_buf == NULL)
5586                         return -ENOMEM;
5587                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5588                                   ldata->ld_buf, XATTR_NAME_LINK);
5589         }
5590         if (rc < 0)
5591                 return rc;
5592
5593         return linkea_init(ldata);
5594 }
5595
5596 static int mdt_path_current(struct mdt_thread_info *info,
5597                             struct path_lookup_info *pli)
5598 {
5599         struct mdt_device       *mdt = info->mti_mdt;
5600         struct mdt_object       *mdt_obj;
5601         struct link_ea_header   *leh;
5602         struct link_ea_entry    *lee;
5603         struct lu_name          *tmpname = &info->mti_name;
5604         struct lu_fid           *tmpfid = &info->mti_tmp_fid1;
5605         struct lu_buf           *buf = &info->mti_big_buf;
5606         char                    *ptr;
5607         int                     reclen;
5608         struct linkea_data      ldata = { 0 };
5609         int                     rc = 0;
5610         ENTRY;
5611
5612         /* temp buffer for path element, the buffer will be finally freed
5613          * in mdt_thread_info_fini */
5614         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
5615         if (buf->lb_buf == NULL)
5616                 RETURN(-ENOMEM);
5617
5618         ldata.ld_buf = buf;
5619         ptr = pli->pli_path + pli->pli_pathlen - 1;
5620         *ptr = 0;
5621         --ptr;
5622         pli->pli_fidcount = 0;
5623         pli->pli_fids[0] = *(struct lu_fid *)mdt_object_fid(pli->pli_mdt_obj);
5624
5625         /* root FID only exists on MDT0, and fid2path should also ends at MDT0,
5626          * so checking root_fid can only happen on MDT0. */
5627         while (!lu_fid_eq(&mdt->mdt_md_root_fid,
5628                           &pli->pli_fids[pli->pli_fidcount])) {
5629                 mdt_obj = mdt_object_find(info->mti_env, mdt,
5630                                           &pli->pli_fids[pli->pli_fidcount]);
5631                 if (IS_ERR(mdt_obj))
5632                         GOTO(out, rc = PTR_ERR(mdt_obj));
5633                 if (mdt_object_remote(mdt_obj)) {
5634                         mdt_object_put(info->mti_env, mdt_obj);
5635                         GOTO(remote_out, rc = -EREMOTE);
5636                 }
5637                 if (!mdt_object_exists(mdt_obj)) {
5638                         mdt_object_put(info->mti_env, mdt_obj);
5639                         GOTO(out, rc = -ENOENT);
5640                 }
5641
5642                 rc = mdt_links_read(info, mdt_obj, &ldata);
5643                 mdt_object_put(info->mti_env, mdt_obj);
5644                 if (rc != 0)
5645                         GOTO(out, rc);
5646
5647                 leh = buf->lb_buf;
5648                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
5649                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
5650                 /* If set, use link #linkno for path lookup, otherwise use
5651                    link #0.  Only do this for the final path element. */
5652                 if ((pli->pli_fidcount == 0) &&
5653                     (pli->pli_linkno < leh->leh_reccount)) {
5654                         int count;
5655                         for (count = 0; count < pli->pli_linkno; count++) {
5656                                 lee = (struct link_ea_entry *)
5657                                      ((char *)lee + reclen);
5658                                 linkea_entry_unpack(lee, &reclen, tmpname,
5659                                                     tmpfid);
5660                         }
5661                         if (pli->pli_linkno < leh->leh_reccount - 1)
5662                                 /* indicate to user there are more links */
5663                                 pli->pli_linkno++;
5664                 }
5665
5666                 /* Pack the name in the end of the buffer */
5667                 ptr -= tmpname->ln_namelen;
5668                 if (ptr - 1 <= pli->pli_path)
5669                         GOTO(out, rc = -EOVERFLOW);
5670                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
5671                 *(--ptr) = '/';
5672
5673                 /* Store the parent fid for historic lookup */
5674                 if (++pli->pli_fidcount >= MAX_PATH_DEPTH)
5675                         GOTO(out, rc = -EOVERFLOW);
5676                 pli->pli_fids[pli->pli_fidcount] = *tmpfid;
5677         }
5678
5679 remote_out:
5680         ptr++; /* skip leading / */
5681         memmove(pli->pli_path, ptr, pli->pli_path + pli->pli_pathlen - ptr);
5682
5683         EXIT;
5684 out:
5685         return rc;
5686 }
5687
5688 /* Returns the full path to this fid, as of changelog record recno. */
5689 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
5690                     char *path, int pathlen, __u64 *recno, int *linkno,
5691                     struct lu_fid *fid)
5692 {
5693         struct mdt_device       *mdt = info->mti_mdt;
5694         struct path_lookup_info *pli;
5695         int                     tries = 3;
5696         int                     rc = -EAGAIN;
5697         ENTRY;
5698
5699         if (pathlen < 3)
5700                 RETURN(-EOVERFLOW);
5701
5702         if (lu_fid_eq(&mdt->mdt_md_root_fid, mdt_object_fid(obj))) {
5703                 path[0] = '\0';
5704                 RETURN(0);
5705         }
5706
5707         OBD_ALLOC_PTR(pli);
5708         if (pli == NULL)
5709                 RETURN(-ENOMEM);
5710
5711         pli->pli_mdt_obj = obj;
5712         pli->pli_recno = *recno;
5713         pli->pli_path = path;
5714         pli->pli_pathlen = pathlen;
5715         pli->pli_linkno = *linkno;
5716
5717         /* Retry multiple times in case file is being moved */
5718         while (tries-- && rc == -EAGAIN)
5719                 rc = mdt_path_current(info, pli);
5720
5721         /* return the last resolved fids to the client, so the client will
5722          * build the left path on another MDT for remote object */
5723         *fid = pli->pli_fids[pli->pli_fidcount];
5724
5725         *recno = pli->pli_currec;
5726         /* Return next link index to caller */
5727         *linkno = pli->pli_linkno;
5728
5729         OBD_FREE_PTR(pli);
5730
5731         RETURN(rc);
5732 }
5733
5734 static int mdt_fid2path(struct mdt_thread_info *info,
5735                         struct getinfo_fid2path *fp)
5736 {
5737         struct mdt_device *mdt = info->mti_mdt;
5738         struct mdt_object *obj;
5739         int    rc;
5740         ENTRY;
5741
5742         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
5743                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
5744
5745         if (!fid_is_sane(&fp->gf_fid))
5746                 RETURN(-EINVAL);
5747
5748         if (!fid_is_namespace_visible(&fp->gf_fid)) {
5749                 CWARN("%s: "DFID" is invalid, sequence should be "
5750                       ">= "LPX64"\n", mdt_obd_name(mdt),
5751                       PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
5752                 RETURN(-EINVAL);
5753         }
5754
5755         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
5756         if (obj == NULL || IS_ERR(obj)) {
5757                 CDEBUG(D_IOCTL, "no object "DFID": %ld\n", PFID(&fp->gf_fid),
5758                        PTR_ERR(obj));
5759                 RETURN(-EINVAL);
5760         }
5761
5762         if (mdt_object_remote(obj))
5763                 rc = -EREMOTE;
5764         else if (!mdt_object_exists(obj))
5765                 rc = -ENOENT;
5766         else
5767                 rc = 0;
5768
5769         if (rc < 0) {
5770                 mdt_object_put(info->mti_env, obj);
5771                 CDEBUG(D_IOCTL, "nonlocal object "DFID": %d\n",
5772                        PFID(&fp->gf_fid), rc);
5773                 RETURN(rc);
5774         }
5775
5776         rc = mdt_path(info, obj, fp->gf_path, fp->gf_pathlen, &fp->gf_recno,
5777                       &fp->gf_linkno, &fp->gf_fid);
5778
5779         CDEBUG(D_INFO, "fid "DFID", path %s recno "LPX64" linkno %u\n",
5780                PFID(&fp->gf_fid), fp->gf_path, fp->gf_recno, fp->gf_linkno);
5781
5782         mdt_object_put(info->mti_env, obj);
5783
5784         RETURN(rc);
5785 }
5786
5787 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key,
5788                             void *val, int vallen)
5789 {
5790         struct getinfo_fid2path *fpout, *fpin;
5791         int rc = 0;
5792
5793         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
5794         fpout = val;
5795
5796         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5797                 lustre_swab_fid2path(fpin);
5798
5799         memcpy(fpout, fpin, sizeof(*fpin));
5800         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
5801                 RETURN(-EINVAL);
5802
5803         rc = mdt_fid2path(info, fpout);
5804         RETURN(rc);
5805 }
5806
5807 int mdt_get_info(struct mdt_thread_info *info)
5808 {
5809         struct ptlrpc_request *req = mdt_info_req(info);
5810         char *key;
5811         int keylen;
5812         __u32 *vallen;
5813         void *valout;
5814         int rc;
5815         ENTRY;
5816
5817         key = req_capsule_client_get(info->mti_pill, &RMF_GETINFO_KEY);
5818         if (key == NULL) {
5819                 CDEBUG(D_IOCTL, "No GETINFO key");
5820                 RETURN(-EFAULT);
5821         }
5822         keylen = req_capsule_get_size(info->mti_pill, &RMF_GETINFO_KEY,
5823                                       RCL_CLIENT);
5824
5825         vallen = req_capsule_client_get(info->mti_pill, &RMF_GETINFO_VALLEN);
5826         if (vallen == NULL) {
5827                 CDEBUG(D_IOCTL, "Unable to get RMF_GETINFO_VALLEN buffer");
5828                 RETURN(-EFAULT);
5829         }
5830
5831         req_capsule_set_size(info->mti_pill, &RMF_GETINFO_VAL, RCL_SERVER,
5832                              *vallen);
5833         rc = req_capsule_server_pack(info->mti_pill);
5834         valout = req_capsule_server_get(info->mti_pill, &RMF_GETINFO_VAL);
5835         if (valout == NULL) {
5836                 CDEBUG(D_IOCTL, "Unable to get get-info RPC out buffer");
5837                 RETURN(-EFAULT);
5838         }
5839
5840         if (KEY_IS(KEY_FID2PATH))
5841                 rc = mdt_rpc_fid2path(info, key, valout, *vallen);
5842         else
5843                 rc = -EINVAL;
5844
5845         lustre_msg_set_status(req->rq_repmsg, rc);
5846
5847         RETURN(rc);
5848 }
5849
5850 /* Pass the ioc down */
5851 static int mdt_ioc_child(struct lu_env *env, struct mdt_device *mdt,
5852                          unsigned int cmd, int len, void *data)
5853 {
5854         struct lu_context ioctl_session;
5855         struct md_device *next = mdt->mdt_child;
5856         int rc;
5857         ENTRY;
5858
5859         rc = lu_context_init(&ioctl_session, LCT_SESSION);
5860         if (rc)
5861                 RETURN(rc);
5862         ioctl_session.lc_thread = (struct ptlrpc_thread *)cfs_current();
5863         lu_context_enter(&ioctl_session);
5864         env->le_ses = &ioctl_session;
5865
5866         LASSERT(next->md_ops->mdo_iocontrol);
5867         rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data);
5868
5869         lu_context_exit(&ioctl_session);
5870         lu_context_fini(&ioctl_session);
5871         RETURN(rc);
5872 }
5873
5874 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
5875 {
5876         struct obd_ioctl_data *data = karg;
5877         struct lu_fid *fid;
5878         __u64 version;
5879         struct mdt_object *obj;
5880         struct mdt_lock_handle  *lh;
5881         int rc;
5882         ENTRY;
5883
5884         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
5885             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
5886                 RETURN(-EINVAL);
5887
5888         fid = (struct lu_fid *)data->ioc_inlbuf1;
5889
5890         if (!fid_is_sane(fid))
5891                 RETURN(-EINVAL);
5892
5893         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
5894
5895         lh = &mti->mti_lh[MDT_LH_PARENT];
5896         mdt_lock_reg_init(lh, LCK_CR);
5897
5898         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
5899         if (IS_ERR(obj))
5900                 RETURN(PTR_ERR(obj));
5901
5902         if (mdt_object_remote(obj)) {
5903                 rc = -EREMOTE;
5904                 /**
5905                  * before calling version get the correct MDS should be
5906                  * fid, this is error to find remote object here
5907                  */
5908                 CERROR("nonlocal object "DFID"\n", PFID(fid));
5909         } else if (!mdt_object_exists(obj)) {
5910                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
5911                 rc = -ENOENT;
5912         } else {
5913                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
5914                *(__u64 *)data->ioc_inlbuf2 = version;
5915                 rc = 0;
5916         }
5917         mdt_object_unlock_put(mti, obj, lh, 1);
5918         RETURN(rc);
5919 }
5920
5921 /* ioctls on obd dev */
5922 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
5923                          void *karg, void *uarg)
5924 {
5925         struct lu_env      env;
5926         struct obd_device *obd = exp->exp_obd;
5927         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
5928         struct dt_device  *dt = mdt->mdt_bottom;
5929         int rc;
5930
5931         ENTRY;
5932         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
5933         rc = lu_env_init(&env, LCT_MD_THREAD);
5934         if (rc)
5935                 RETURN(rc);
5936
5937         switch (cmd) {
5938         case OBD_IOC_SYNC:
5939                 rc = mdt_device_sync(&env, mdt);
5940                 break;
5941         case OBD_IOC_SET_READONLY:
5942                 rc = dt->dd_ops->dt_ro(&env, dt);
5943                 break;
5944         case OBD_IOC_ABORT_RECOVERY:
5945                 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
5946                 target_stop_recovery_thread(obd);
5947                 rc = 0;
5948                 break;
5949         case OBD_IOC_CHANGELOG_REG:
5950         case OBD_IOC_CHANGELOG_DEREG:
5951         case OBD_IOC_CHANGELOG_CLEAR:
5952                 rc = mdt_ioc_child(&env, mdt, cmd, len, karg);
5953                 break;
5954         case OBD_IOC_START_LFSCK: {
5955                 struct md_device *next = mdt->mdt_child;
5956                 struct obd_ioctl_data *data = karg;
5957                 struct lfsck_start_param lsp;
5958
5959                 if (unlikely(data == NULL)) {
5960                         rc = -EINVAL;
5961                         break;
5962                 }
5963
5964                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
5965                 lsp.lsp_namespace = mdt->mdt_namespace;
5966                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
5967                 break;
5968         }
5969         case OBD_IOC_STOP_LFSCK: {
5970                 struct md_device *next = mdt->mdt_child;
5971
5972                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, NULL);
5973                 break;
5974         }
5975         case OBD_IOC_GET_OBJ_VERSION: {
5976                 struct mdt_thread_info *mti;
5977                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5978                 memset(mti, 0, sizeof *mti);
5979                 mti->mti_env = &env;
5980                 mti->mti_mdt = mdt;
5981                 mti->mti_exp = exp;
5982
5983                 rc = mdt_ioc_version_get(mti, karg);
5984                 break;
5985         }
5986         default:
5987                 rc = -EOPNOTSUPP;
5988                 CERROR("%s: Not supported cmd = %d, rc = %d\n",
5989                         mdt_obd_name(mdt), cmd, rc);
5990         }
5991
5992         lu_env_fini(&env);
5993         RETURN(rc);
5994 }
5995
5996 int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
5997 {
5998         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
5999         int rc;
6000         ENTRY;
6001
6002         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
6003         RETURN(rc);
6004 }
6005
6006 int mdt_obd_postrecov(struct obd_device *obd)
6007 {
6008         struct lu_env env;
6009         int rc;
6010
6011         rc = lu_env_init(&env, LCT_MD_THREAD);
6012         if (rc)
6013                 RETURN(rc);
6014         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
6015         lu_env_fini(&env);
6016         return rc;
6017 }
6018
6019 static struct obd_ops mdt_obd_device_ops = {
6020         .o_owner          = THIS_MODULE,
6021         .o_set_info_async = mdt_obd_set_info_async,
6022         .o_connect        = mdt_obd_connect,
6023         .o_reconnect      = mdt_obd_reconnect,
6024         .o_disconnect     = mdt_obd_disconnect,
6025         .o_init_export    = mdt_init_export,
6026         .o_destroy_export = mdt_destroy_export,
6027         .o_iocontrol      = mdt_iocontrol,
6028         .o_postrecov      = mdt_obd_postrecov,
6029 };
6030
6031 static struct lu_device* mdt_device_fini(const struct lu_env *env,
6032                                          struct lu_device *d)
6033 {
6034         struct mdt_device *m = mdt_dev(d);
6035         ENTRY;
6036
6037         mdt_fini(env, m);
6038         RETURN(NULL);
6039 }
6040
6041 static struct lu_device *mdt_device_free(const struct lu_env *env,
6042                                          struct lu_device *d)
6043 {
6044         struct mdt_device *m = mdt_dev(d);
6045         ENTRY;
6046
6047         md_device_fini(&m->mdt_md_dev);
6048         OBD_FREE_PTR(m);
6049         RETURN(NULL);
6050 }
6051
6052 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
6053                                           struct lu_device_type *t,
6054                                           struct lustre_cfg *cfg)
6055 {
6056         struct lu_device  *l;
6057         struct mdt_device *m;
6058
6059         OBD_ALLOC_PTR(m);
6060         if (m != NULL) {
6061                 int rc;
6062
6063                 l = &m->mdt_md_dev.md_lu_dev;
6064                 rc = mdt_init0(env, m, t, cfg);
6065                 if (rc != 0) {
6066                         mdt_device_free(env, l);
6067                         l = ERR_PTR(rc);
6068                         return l;
6069                 }
6070         } else
6071                 l = ERR_PTR(-ENOMEM);
6072         return l;
6073 }
6074
6075 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
6076 LU_KEY_INIT(mdt, struct mdt_thread_info);
6077
6078 static void mdt_key_fini(const struct lu_context *ctx,
6079                          struct lu_context_key *key, void* data)
6080 {
6081         struct mdt_thread_info *info = data;
6082
6083         if (info->mti_big_lmm) {
6084                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
6085                 info->mti_big_lmm = NULL;
6086                 info->mti_big_lmmsize = 0;
6087         }
6088         OBD_FREE_PTR(info);
6089 }
6090
6091 /* context key: mdt_thread_key */
6092 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
6093
6094 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
6095 {
6096         return lu_ucred(info->mti_env);
6097 }
6098
6099 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
6100 {
6101         return lu_ucred_check(info->mti_env);
6102 }
6103
6104 /**
6105  * Enable/disable COS (Commit On Sharing).
6106  *
6107  * Set/Clear the COS flag in mdt options.
6108  *
6109  * \param mdt mdt device
6110  * \param val 0 disables COS, other values enable COS
6111  */
6112 void mdt_enable_cos(struct mdt_device *mdt, int val)
6113 {
6114         struct lu_env env;
6115         int rc;
6116
6117         mdt->mdt_opts.mo_cos = !!val;
6118         rc = lu_env_init(&env, LCT_LOCAL);
6119         if (unlikely(rc != 0)) {
6120                 CWARN("lu_env initialization failed with rc = %d,"
6121                       "cannot sync\n", rc);
6122                 return;
6123         }
6124         mdt_device_sync(&env, mdt);
6125         lu_env_fini(&env);
6126 }
6127
6128 /**
6129  * Check COS (Commit On Sharing) status.
6130  *
6131  * Return COS flag status.
6132  *
6133  * \param mdt mdt device
6134  */
6135 int mdt_cos_is_enabled(struct mdt_device *mdt)
6136 {
6137         return mdt->mdt_opts.mo_cos != 0;
6138 }
6139
6140 static struct lu_device_type_operations mdt_device_type_ops = {
6141         .ldto_device_alloc = mdt_device_alloc,
6142         .ldto_device_free  = mdt_device_free,
6143         .ldto_device_fini  = mdt_device_fini
6144 };
6145
6146 static struct lu_device_type mdt_device_type = {
6147         .ldt_tags     = LU_DEVICE_MD,
6148         .ldt_name     = LUSTRE_MDT_NAME,
6149         .ldt_ops      = &mdt_device_type_ops,
6150         .ldt_ctx_tags = LCT_MD_THREAD
6151 };
6152
6153 static int __init mdt_mod_init(void)
6154 {
6155         struct lprocfs_static_vars lvars;
6156         int rc;
6157
6158         CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") ==
6159                  FID_NOBRACE_LEN + 1);
6160         CLASSERT(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") ==
6161                  FID_LEN + 1);
6162         rc = lu_kmem_init(mdt_caches);
6163         if (rc)
6164                 return rc;
6165
6166         rc = mds_mod_init();
6167         if (rc)
6168                 GOTO(lu_fini, rc);
6169
6170         lprocfs_mdt_init_vars(&lvars);
6171         rc = class_register_type(&mdt_obd_device_ops, NULL,
6172                                  lvars.module_vars, LUSTRE_MDT_NAME,
6173                                  &mdt_device_type);
6174         if (rc)
6175                 GOTO(mds_fini, rc);
6176 lu_fini:
6177         if (rc)
6178                 lu_kmem_fini(mdt_caches);
6179 mds_fini:
6180         if (rc)
6181                 mds_mod_exit();
6182         return rc;
6183 }
6184
6185 static void __exit mdt_mod_exit(void)
6186 {
6187         class_unregister_type(LUSTRE_MDT_NAME);
6188         mds_mod_exit();
6189         lu_kmem_fini(mdt_caches);
6190 }
6191
6192 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
6193 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
6194 MODULE_LICENSE("GPL");
6195
6196 cfs_module(mdt, LUSTRE_VERSION_STRING, mdt_mod_init, mdt_mod_exit);