lustre/mdt/mdt_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2010, 2013, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/mdt/mdt_handler.c
  37  *
  38  * Lustre Metadata Target (mdt) request handler
  39  *
  40  * Author: Peter Braam <braam@clusterfs.com>
  41  * Author: Andreas Dilger <adilger@clusterfs.com>
  42  * Author: Phil Schwan <phil@clusterfs.com>
  43  * Author: Mike Shaver <shaver@clusterfs.com>
  44  * Author: Nikita Danilov <nikita@clusterfs.com>
  45  * Author: Huang Hua <huanghua@clusterfs.com>
  46  * Author: Yury Umanets <umka@clusterfs.com>
  47  */
  48
  49 #define DEBUG_SUBSYSTEM S_MDS
  50
  51 #include <linux/module.h>
  52 /*
  53  * struct OBD_{ALLOC,FREE}*()
  54  */
  55 #include <obd_support.h>
  56 /* struct ptlrpc_request */
  57 #include <lustre_net.h>
  58 /* struct obd_export */
  59 #include <lustre_export.h>
  60 /* struct obd_device */
  61 #include <obd.h>
  62 /* lu2dt_dev() */
  63 #include <dt_object.h>
  64 #include <lustre_mds.h>
  65 #include <lustre_log.h>
  66 #include "mdt_internal.h"
  67 #include <lustre_acl.h>
  68 #include <lustre_param.h>
  69 #include <lustre_quota.h>
  70 #include <lustre_linkea.h>
  71 #include <lustre_lfsck.h>
  72
  73 mdl_mode_t mdt_mdl_lock_modes[] = {
  74         [LCK_MINMODE] = MDL_MINMODE,
  75         [LCK_EX]      = MDL_EX,
  76         [LCK_PW]      = MDL_PW,
  77         [LCK_PR]      = MDL_PR,
  78         [LCK_CW]      = MDL_CW,
  79         [LCK_CR]      = MDL_CR,
  80         [LCK_NL]      = MDL_NL,
  81         [LCK_GROUP]   = MDL_GROUP
  82 };
  83
  84 ldlm_mode_t mdt_dlm_lock_modes[] = {
  85         [MDL_MINMODE] = LCK_MINMODE,
  86         [MDL_EX]      = LCK_EX,
  87         [MDL_PW]      = LCK_PW,
  88         [MDL_PR]      = LCK_PR,
  89         [MDL_CW]      = LCK_CW,
  90         [MDL_CR]      = LCK_CR,
  91         [MDL_NL]      = LCK_NL,
  92         [MDL_GROUP]   = LCK_GROUP
  93 };
  94
  95 static struct mdt_device *mdt_dev(struct lu_device *d);
  96 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
  97
  98 static const struct lu_object_operations mdt_obj_ops;
  99
 100 /* Slab for MDT object allocation */
 101 static struct kmem_cache *mdt_object_kmem;
 102
 103 /* For HSM restore handles */
 104 struct kmem_cache *mdt_hsm_cdt_kmem;
 105
 106 /* For HSM request handles */
 107 struct kmem_cache *mdt_hsm_car_kmem;
 108
 109 static struct lu_kmem_descr mdt_caches[] = {
 110         {
 111                 .ckd_cache = &mdt_object_kmem,
 112                 .ckd_name  = "mdt_obj",
 113                 .ckd_size  = sizeof(struct mdt_object)
 114         },
 115         {
 116                 .ckd_cache      = &mdt_hsm_cdt_kmem,
 117                 .ckd_name       = "mdt_cdt_restore_handle",
 118                 .ckd_size       = sizeof(struct cdt_restore_handle)
 119         },
 120         {
 121                 .ckd_cache      = &mdt_hsm_car_kmem,
 122                 .ckd_name       = "mdt_cdt_agent_req",
 123                 .ckd_size       = sizeof(struct cdt_agent_req)
 124         },
 125         {
 126                 .ckd_cache = NULL
 127         }
 128 };
 129
 130 int mdt_get_disposition(struct ldlm_reply *rep, int flag)
 131 {
 132         if (!rep)
 133                 return 0;
 134         return (rep->lock_policy_res1 & flag);
 135 }
 136
 137 void mdt_clear_disposition(struct mdt_thread_info *info,
 138                            struct ldlm_reply *rep, int flag)
 139 {
 140         if (info) {
 141                 info->mti_opdata &= ~flag;
 142                 tgt_opdata_clear(info->mti_env, flag);
 143         }
 144         if (rep)
 145                 rep->lock_policy_res1 &= ~flag;
 146 }
 147
 148 void mdt_set_disposition(struct mdt_thread_info *info,
 149                          struct ldlm_reply *rep, int flag)
 150 {
 151         if (info) {
 152                 info->mti_opdata |= flag;
 153                 tgt_opdata_set(info->mti_env, flag);
 154         }
 155         if (rep)
 156                 rep->lock_policy_res1 |= flag;
 157 }
 158
 159 void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
 160 {
 161         lh->mlh_pdo_hash = 0;
 162         lh->mlh_reg_mode = lm;
 163         lh->mlh_rreg_mode = lm;
 164         lh->mlh_type = MDT_REG_LOCK;
 165 }
 166
 167 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lock_mode,
 168                        const struct lu_name *lname)
 169 {
 170         lh->mlh_reg_mode = lock_mode;
 171         lh->mlh_rreg_mode = lock_mode;
 172         lh->mlh_type = MDT_PDO_LOCK;
 173
 174         if (lu_name_is_valid(lname)) {
 175                 lh->mlh_pdo_hash = full_name_hash(lname->ln_name,
 176                                                   lname->ln_namelen);
 177                 /* XXX Workaround for LU-2856
 178                  *
 179                  * Zero is a valid return value of full_name_hash, but
 180                  * several users of mlh_pdo_hash assume a non-zero
 181                  * hash value. We therefore map zero onto an
 182                  * arbitrary, but consistent value (1) to avoid
 183                  * problems further down the road. */
 184                 if (unlikely(lh->mlh_pdo_hash == 0))
 185                         lh->mlh_pdo_hash = 1;
 186         } else {
 187                 lh->mlh_pdo_hash = 0;
 188         }
 189 }
 190
 191 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 192                               struct mdt_lock_handle *lh)
 193 {
 194         mdl_mode_t mode;
 195         ENTRY;
 196
 197         /*
 198          * Any dir access needs couple of locks:
 199          *
 200          * 1) on part of dir we gonna take lookup/modify;
 201          *
 202          * 2) on whole dir to protect it from concurrent splitting and/or to
 203          * flush client's cache for readdir().
 204          *
 205          * so, for a given mode and object this routine decides what lock mode
 206          * to use for lock #2:
 207          *
 208          * 1) if caller's gonna lookup in dir then we need to protect dir from
 209          * being splitted only - LCK_CR
 210          *
 211          * 2) if caller's gonna modify dir then we need to protect dir from
 212          * being splitted and to flush cache - LCK_CW
 213          *
 214          * 3) if caller's gonna modify dir and that dir seems ready for
 215          * splitting then we need to protect it from any type of access
 216          * (lookup/modify/split) - LCK_EX --bzzz
 217          */
 218
 219         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 220         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 221
 222         /*
 223          * Ask underlaying level its opinion about preferable PDO lock mode
 224          * having access type passed as regular lock mode:
 225          *
 226          * - MDL_MINMODE means that lower layer does not want to specify lock
 227          * mode;
 228          *
 229          * - MDL_NL means that no PDO lock should be taken. This is used in some
 230          * cases. Say, for non-splittable directories no need to use PDO locks
 231          * at all.
 232          */
 233         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 234                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 235
 236         if (mode != MDL_MINMODE) {
 237                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 238         } else {
 239                 /*
 240                  * Lower layer does not want to specify locking mode. We do it
 241                  * our selves. No special protection is needed, just flush
 242                  * client's cache on modification and allow concurrent
 243                  * mondification.
 244                  */
 245                 switch (lh->mlh_reg_mode) {
 246                 case LCK_EX:
 247                         lh->mlh_pdo_mode = LCK_EX;
 248                         break;
 249                 case LCK_PR:
 250                         lh->mlh_pdo_mode = LCK_CR;
 251                         break;
 252                 case LCK_PW:
 253                         lh->mlh_pdo_mode = LCK_CW;
 254                         break;
 255                 default:
 256                         CERROR("Not expected lock type (0x%x)\n",
 257                                (int)lh->mlh_reg_mode);
 258                         LBUG();
 259                 }
 260         }
 261
 262         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 263         EXIT;
 264 }
 265
 266 int mdt_getstatus(struct tgt_session_info *tsi)
 267 {
 268         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
 269         struct mdt_device       *mdt = info->mti_mdt;
 270         struct mdt_body         *repbody;
 271         int                      rc;
 272
 273         ENTRY;
 274
 275         rc = mdt_check_ucred(info);
 276         if (rc)
 277                 GOTO(out, rc = err_serious(rc));
 278
 279         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
 280                 GOTO(out, rc = err_serious(-ENOMEM));
 281
 282         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 283         repbody->fid1 = mdt->mdt_md_root_fid;
 284         repbody->valid |= OBD_MD_FLID;
 285
 286         if (tsi->tsi_tgt->lut_mds_capa &&
 287             exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) {
 288                 struct mdt_object       *root;
 289                 struct lustre_capa      *capa;
 290
 291                 root = mdt_object_find(info->mti_env, mdt, &repbody->fid1);
 292                 if (IS_ERR(root))
 293                         GOTO(out, rc = PTR_ERR(root));
 294
 295                 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
 296                 LASSERT(capa);
 297                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 298                 rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa,
 299                                  0);
 300                 mdt_object_put(info->mti_env, root);
 301                 if (rc == 0)
 302                         repbody->valid |= OBD_MD_FLMDSCAPA;
 303         }
 304         EXIT;
 305 out:
 306         mdt_thread_info_fini(info);
 307         return rc;
 308 }
 309
 310 int mdt_statfs(struct tgt_session_info *tsi)
 311 {
 312         struct ptlrpc_request           *req = tgt_ses_req(tsi);
 313         struct mdt_thread_info          *info = tsi2mdt_info(tsi);
 314         struct md_device                *next = info->mti_mdt->mdt_child;
 315         struct ptlrpc_service_part      *svcpt;
 316         struct obd_statfs               *osfs;
 317         int                             rc;
 318
 319         ENTRY;
 320
 321         svcpt = req->rq_rqbd->rqbd_svcpt;
 322
 323         /* This will trigger a watchdog timeout */
 324         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 325                          (MDT_SERVICE_WATCHDOG_FACTOR *
 326                           at_get(&svcpt->scp_at_estimate)) + 1);
 327
 328         rc = mdt_check_ucred(info);
 329         if (rc)
 330                 GOTO(out, rc = err_serious(rc));
 331
 332         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
 333                 GOTO(out, rc = err_serious(-ENOMEM));
 334
 335         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 336         if (!osfs)
 337                 GOTO(out, rc = -EPROTO);
 338
 339         /** statfs information are cached in the mdt_device */
 340         if (cfs_time_before_64(info->mti_mdt->mdt_osfs_age,
 341                                cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS))) {
 342                 /** statfs data is too old, get up-to-date one */
 343                 rc = next->md_ops->mdo_statfs(info->mti_env, next, osfs);
 344                 if (rc)
 345                         GOTO(out, rc);
 346                 spin_lock(&info->mti_mdt->mdt_osfs_lock);
 347                 info->mti_mdt->mdt_osfs = *osfs;
 348                 info->mti_mdt->mdt_osfs_age = cfs_time_current_64();
 349                 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
 350         } else {
 351                 /** use cached statfs data */
 352                 spin_lock(&info->mti_mdt->mdt_osfs_lock);
 353                 *osfs = info->mti_mdt->mdt_osfs;
 354                 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
 355         }
 356
 357         if (rc == 0)
 358                 mdt_counter_incr(req, LPROC_MDT_STATFS);
 359 out:
 360         mdt_thread_info_fini(info);
 361         RETURN(rc);
 362 }
 363
 364 /**
 365  * Pack SOM attributes into the reply.
 366  * Call under a DLM UPDATE lock.
 367  */
 368 static void mdt_pack_size2body(struct mdt_thread_info *info,
 369                                struct mdt_object *mo)
 370 {
 371         struct mdt_body *b;
 372         struct md_attr *ma = &info->mti_attr;
 373
 374         LASSERT(ma->ma_attr.la_valid & LA_MODE);
 375         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 376
 377         /* Check if Size-on-MDS is supported, if this is a regular file,
 378          * if SOM is enabled on the object and if SOM cache exists and valid.
 379          * Otherwise do not pack Size-on-MDS attributes to the reply. */
 380         if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) ||
 381             !S_ISREG(ma->ma_attr.la_mode) ||
 382             !mdt_object_is_som_enabled(mo) ||
 383             !(ma->ma_valid & MA_SOM))
 384                 return;
 385
 386         b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 387         b->size = ma->ma_som->msd_size;
 388         b->blocks = ma->ma_som->msd_blocks;
 389 }
 390
 391 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 392                         const struct lu_attr *attr, const struct lu_fid *fid)
 393 {
 394         struct md_attr *ma = &info->mti_attr;
 395
 396         LASSERT(ma->ma_valid & MA_INODE);
 397
 398         b->atime      = attr->la_atime;
 399         b->mtime      = attr->la_mtime;
 400         b->ctime      = attr->la_ctime;
 401         b->mode       = attr->la_mode;
 402         b->size       = attr->la_size;
 403         b->blocks     = attr->la_blocks;
 404         b->uid        = attr->la_uid;
 405         b->gid        = attr->la_gid;
 406         b->flags      = attr->la_flags;
 407         b->nlink      = attr->la_nlink;
 408         b->rdev       = attr->la_rdev;
 409
 410         /*XXX should pack the reply body according to lu_valid*/
 411         b->valid |= OBD_MD_FLCTIME | OBD_MD_FLUID   |
 412                     OBD_MD_FLGID   | OBD_MD_FLTYPE  |
 413                     OBD_MD_FLMODE  | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
 414                     OBD_MD_FLATIME | OBD_MD_FLMTIME ;
 415
 416         if (!S_ISREG(attr->la_mode)) {
 417                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 418         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
 419                 /* means no objects are allocated on osts. */
 420                 LASSERT(!(ma->ma_valid & MA_LOV));
 421                 /* just ignore blocks occupied by extend attributes on MDS */
 422                 b->blocks = 0;
 423                 /* if no object is allocated on osts, the size on mds is valid. b=22272 */
 424                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 425         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL &&
 426                    ma->ma_lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) {
 427                 /* A released file stores its size on MDS. */
 428                 /* But return 1 block for released file, unless tools like tar
 429                  * will consider it fully sparse. (LU-3864)
 430                  */
 431                 if (unlikely(b->size == 0))
 432                         b->blocks = 0;
 433                 else
 434                         b->blocks = 1;
 435                 b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 436         }
 437
 438         if (fid) {
 439                 b->fid1 = *fid;
 440                 b->valid |= OBD_MD_FLID;
 441                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n",
 442                                 PFID(fid), b->nlink, b->mode, b->size);
 443         }
 444
 445         if (info)
 446                 mdt_body_reverse_idmap(info, b);
 447
 448         if (fid != NULL && (b->valid & OBD_MD_FLSIZE))
 449                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 450                        PFID(fid), (unsigned long long)b->size);
 451 }
 452
 453 static inline int mdt_body_has_lov(const struct lu_attr *la,
 454                                    const struct mdt_body *body)
 455 {
 456         return ((S_ISREG(la->la_mode) && (body->valid & OBD_MD_FLEASIZE)) ||
 457                 (S_ISDIR(la->la_mode) && (body->valid & OBD_MD_FLDIREA )) );
 458 }
 459
 460 void mdt_client_compatibility(struct mdt_thread_info *info)
 461 {
 462         struct mdt_body       *body;
 463         struct ptlrpc_request *req = mdt_info_req(info);
 464         struct obd_export     *exp = req->rq_export;
 465         struct md_attr        *ma = &info->mti_attr;
 466         struct lu_attr        *la = &ma->ma_attr;
 467         ENTRY;
 468
 469         if (exp_connect_layout(exp))
 470                 /* the client can deal with 16-bit lmm_stripe_count */
 471                 RETURN_EXIT;
 472
 473         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 474
 475         if (!mdt_body_has_lov(la, body))
 476                 RETURN_EXIT;
 477
 478         /* now we have a reply with a lov for a client not compatible with the
 479          * layout lock so we have to clean the layout generation number */
 480         if (S_ISREG(la->la_mode))
 481                 ma->ma_lmm->lmm_layout_gen = 0;
 482         EXIT;
 483 }
 484
 485 static int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
 486                              char *name)
 487 {
 488         const struct lu_env *env = info->mti_env;
 489         int rc;
 490         ENTRY;
 491
 492         LASSERT(info->mti_big_lmm_used == 0);
 493         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
 494         if (rc < 0)
 495                 RETURN(rc);
 496
 497         /* big_lmm may need to be grown */
 498         if (info->mti_big_lmmsize < rc) {
 499                 int size = size_roundup_power2(rc);
 500
 501                 if (info->mti_big_lmmsize > 0) {
 502                         /* free old buffer */
 503                         LASSERT(info->mti_big_lmm);
 504                         OBD_FREE_LARGE(info->mti_big_lmm,
 505                                        info->mti_big_lmmsize);
 506                         info->mti_big_lmm = NULL;
 507                         info->mti_big_lmmsize = 0;
 508                 }
 509
 510                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
 511                 if (info->mti_big_lmm == NULL)
 512                         RETURN(-ENOMEM);
 513                 info->mti_big_lmmsize = size;
 514         }
 515         LASSERT(info->mti_big_lmmsize >= rc);
 516
 517         info->mti_buf.lb_buf = info->mti_big_lmm;
 518         info->mti_buf.lb_len = info->mti_big_lmmsize;
 519         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
 520
 521         RETURN(rc);
 522 }
 523
 524 int mdt_attr_get_lov(struct mdt_thread_info *info,
 525                      struct mdt_object *o, struct md_attr *ma)
 526 {
 527         struct md_object *next = mdt_object_child(o);
 528         struct lu_buf    *buf = &info->mti_buf;
 529         int rc;
 530
 531         buf->lb_buf = ma->ma_lmm;
 532         buf->lb_len = ma->ma_lmm_size;
 533         rc = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_LOV);
 534         if (rc > 0) {
 535                 ma->ma_lmm_size = rc;
 536                 ma->ma_valid |= MA_LOV;
 537                 rc = 0;
 538         } else if (rc == -ENODATA) {
 539                 /* no LOV EA */
 540                 rc = 0;
 541         } else if (rc == -ERANGE) {
 542                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LOV);
 543                 if (rc > 0) {
 544                         info->mti_big_lmm_used = 1;
 545                         ma->ma_valid |= MA_LOV;
 546                         ma->ma_lmm = info->mti_big_lmm;
 547                         ma->ma_lmm_size = rc;
 548                         /* update mdt_max_mdsize so all clients
 549                          * will be aware about that */
 550                         if (info->mti_mdt->mdt_max_mdsize < rc)
 551                                 info->mti_mdt->mdt_max_mdsize = rc;
 552                         rc = 0;
 553                 }
 554         }
 555
 556         return rc;
 557 }
 558
 559 int mdt_attr_get_pfid(struct mdt_thread_info *info,
 560                       struct mdt_object *o, struct lu_fid *pfid)
 561 {
 562         struct lu_buf           *buf = &info->mti_buf;
 563         struct link_ea_header   *leh;
 564         struct link_ea_entry    *lee;
 565         int                      rc;
 566         ENTRY;
 567
 568         buf->lb_buf = info->mti_big_lmm;
 569         buf->lb_len = info->mti_big_lmmsize;
 570         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
 571                           buf, XATTR_NAME_LINK);
 572         /* ignore errors, MA_PFID won't be set and it is
 573          * up to the caller to treat this as an error */
 574         if (rc == -ERANGE || buf->lb_len == 0) {
 575                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
 576                 buf->lb_buf = info->mti_big_lmm;
 577                 buf->lb_len = info->mti_big_lmmsize;
 578         }
 579
 580         if (rc < 0)
 581                 RETURN(rc);
 582         if (rc < sizeof(*leh)) {
 583                 CERROR("short LinkEA on "DFID": rc = %d\n",
 584                        PFID(mdt_object_fid(o)), rc);
 585                 RETURN(-ENODATA);
 586         }
 587
 588         leh = (struct link_ea_header *) buf->lb_buf;
 589         lee = (struct link_ea_entry *)(leh + 1);
 590         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
 591                 leh->leh_magic = LINK_EA_MAGIC;
 592                 leh->leh_reccount = __swab32(leh->leh_reccount);
 593                 leh->leh_len = __swab64(leh->leh_len);
 594         }
 595         if (leh->leh_magic != LINK_EA_MAGIC)
 596                 RETURN(-EINVAL);
 597         if (leh->leh_reccount == 0)
 598                 RETURN(-ENODATA);
 599
 600         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
 601         fid_be_to_cpu(pfid, pfid);
 602
 603         RETURN(0);
 604 }
 605
 606 int mdt_attr_get_complex(struct mdt_thread_info *info,
 607                          struct mdt_object *o, struct md_attr *ma)
 608 {
 609         const struct lu_env *env = info->mti_env;
 610         struct md_object    *next = mdt_object_child(o);
 611         struct lu_buf       *buf = &info->mti_buf;
 612         u32                  mode = lu_object_attr(&next->mo_lu);
 613         int                  need = ma->ma_need;
 614         int                  rc = 0, rc2;
 615         ENTRY;
 616
 617         ma->ma_valid = 0;
 618
 619         if (need & MA_INODE) {
 620                 ma->ma_need = MA_INODE;
 621                 rc = mo_attr_get(env, next, ma);
 622                 if (rc)
 623                         GOTO(out, rc);
 624                 ma->ma_valid |= MA_INODE;
 625         }
 626
 627         if (need & MA_PFID) {
 628                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
 629                 if (rc == 0)
 630                         ma->ma_valid |= MA_PFID;
 631                 /* ignore this error, parent fid is not mandatory */
 632                 rc = 0;
 633         }
 634
 635         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
 636                 rc = mdt_attr_get_lov(info, o, ma);
 637                 if (rc)
 638                         GOTO(out, rc);
 639         }
 640
 641         if (need & MA_LMV && S_ISDIR(mode)) {
 642                 buf->lb_buf = ma->ma_lmv;
 643                 buf->lb_len = ma->ma_lmv_size;
 644                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_LMV);
 645                 if (rc2 > 0) {
 646                         ma->ma_lmv_size = rc2;
 647                         ma->ma_valid |= MA_LMV;
 648                 } else if (rc2 == -ENODATA) {
 649                         /* no LMV EA */
 650                         ma->ma_lmv_size = 0;
 651                 } else
 652                         GOTO(out, rc = rc2);
 653         }
 654
 655         if (need & MA_SOM && S_ISREG(mode)) {
 656                 buf->lb_buf = info->mti_xattr_buf;
 657                 buf->lb_len = sizeof(info->mti_xattr_buf);
 658                 CLASSERT(sizeof(struct som_attrs) <=
 659                          sizeof(info->mti_xattr_buf));
 660                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_SOM);
 661                 rc2 = lustre_buf2som(info->mti_xattr_buf, rc2, ma->ma_som);
 662                 if (rc2 == 0)
 663                         ma->ma_valid |= MA_SOM;
 664                 else if (rc2 < 0 && rc2 != -ENODATA)
 665                         GOTO(out, rc = rc2);
 666         }
 667
 668         if (need & MA_HSM && S_ISREG(mode)) {
 669                 buf->lb_buf = info->mti_xattr_buf;
 670                 buf->lb_len = sizeof(info->mti_xattr_buf);
 671                 CLASSERT(sizeof(struct hsm_attrs) <=
 672                          sizeof(info->mti_xattr_buf));
 673                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
 674                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
 675                 if (rc2 == 0)
 676                         ma->ma_valid |= MA_HSM;
 677                 else if (rc2 < 0 && rc2 != -ENODATA)
 678                         GOTO(out, rc = rc2);
 679         }
 680
 681 #ifdef CONFIG_FS_POSIX_ACL
 682         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
 683                 buf->lb_buf = ma->ma_acl;
 684                 buf->lb_len = ma->ma_acl_size;
 685                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
 686                 if (rc2 > 0) {
 687                         ma->ma_acl_size = rc2;
 688                         ma->ma_valid |= MA_ACL_DEF;
 689                 } else if (rc2 == -ENODATA) {
 690                         /* no ACLs */
 691                         ma->ma_acl_size = 0;
 692                 } else
 693                         GOTO(out, rc = rc2);
 694         }
 695 #endif
 696 out:
 697         ma->ma_need = need;
 698         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
 699                rc, ma->ma_valid, ma->ma_lmm);
 700         RETURN(rc);
 701 }
 702
 703 static int mdt_getattr_internal(struct mdt_thread_info *info,
 704                                 struct mdt_object *o, int ma_need)
 705 {
 706         struct md_object        *next = mdt_object_child(o);
 707         const struct mdt_body   *reqbody = info->mti_body;
 708         struct ptlrpc_request   *req = mdt_info_req(info);
 709         struct md_attr          *ma = &info->mti_attr;
 710         struct lu_attr          *la = &ma->ma_attr;
 711         struct req_capsule      *pill = info->mti_pill;
 712         const struct lu_env     *env = info->mti_env;
 713         struct mdt_body         *repbody;
 714         struct lu_buf           *buffer = &info->mti_buf;
 715         int                     rc;
 716         int                     is_root;
 717         ENTRY;
 718
 719         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
 720                 RETURN(err_serious(-ENOMEM));
 721
 722         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 723
 724         ma->ma_valid = 0;
 725
 726         if (mdt_object_remote(o)) {
 727                 /* This object is located on remote node.*/
 728                 /* Return -EIO for old client */
 729                 if (!mdt_is_dne_client(req->rq_export))
 730                         GOTO(out, rc = -EIO);
 731
 732                 repbody->fid1 = *mdt_object_fid(o);
 733                 repbody->valid = OBD_MD_FLID | OBD_MD_MDS;
 734                 GOTO(out, rc = 0);
 735         }
 736
 737         buffer->lb_len = reqbody->eadatasize;
 738         if (buffer->lb_len > 0)
 739                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
 740         else
 741                 buffer->lb_buf = NULL;
 742
 743         /* If it is dir object and client require MEA, then we got MEA */
 744         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 745             reqbody->valid & OBD_MD_MEA) {
 746                 /* Assumption: MDT_MD size is enough for lmv size. */
 747                 ma->ma_lmv = buffer->lb_buf;
 748                 ma->ma_lmv_size = buffer->lb_len;
 749                 ma->ma_need = MA_LMV | MA_INODE;
 750         } else {
 751                 ma->ma_lmm = buffer->lb_buf;
 752                 ma->ma_lmm_size = buffer->lb_len;
 753                 ma->ma_need = MA_LOV | MA_INODE | MA_HSM;
 754         }
 755
 756         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 757             reqbody->valid & OBD_MD_FLDIREA  &&
 758             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
 759                 /* get default stripe info for this dir. */
 760                 ma->ma_need |= MA_LOV_DEF;
 761         }
 762         ma->ma_need |= ma_need;
 763         if (ma->ma_need & MA_SOM)
 764                 ma->ma_som = &info->mti_u.som.data;
 765
 766         rc = mdt_attr_get_complex(info, o, ma);
 767         if (unlikely(rc)) {
 768                 CERROR("%s: getattr error for "DFID": rc = %d\n",
 769                        mdt_obd_name(info->mti_mdt),
 770                        PFID(mdt_object_fid(o)), rc);
 771                 RETURN(rc);
 772         }
 773
 774         /* if file is released, check if a restore is running */
 775         if ((ma->ma_valid & MA_HSM) && (ma->ma_hsm.mh_flags & HS_RELEASED) &&
 776             mdt_hsm_restore_is_running(info, mdt_object_fid(o))) {
 777                 repbody->t_state = MS_RESTORE;
 778                 repbody->valid |= OBD_MD_TSTATE;
 779         }
 780
 781         is_root = lu_fid_eq(mdt_object_fid(o), &info->mti_mdt->mdt_md_root_fid);
 782
 783         /* the Lustre protocol supposes to return default striping
 784          * on the user-visible root if explicitly requested */
 785         if ((ma->ma_valid & MA_LOV) == 0 && S_ISDIR(la->la_mode) &&
 786             (ma->ma_need & MA_LOV_DEF && is_root) && ma->ma_need & MA_LOV) {
 787                 struct lu_fid      rootfid;
 788                 struct mdt_object *root;
 789                 struct mdt_device *mdt = info->mti_mdt;
 790
 791                 rc = dt_root_get(env, mdt->mdt_bottom, &rootfid);
 792                 if (rc)
 793                         RETURN(rc);
 794                 root = mdt_object_find(env, mdt, &rootfid);
 795                 if (IS_ERR(root))
 796                         RETURN(PTR_ERR(root));
 797                 rc = mdt_attr_get_lov(info, root, ma);
 798                 mdt_object_put(info->mti_env, root);
 799                 if (unlikely(rc)) {
 800                         CERROR("%s: getattr error for "DFID": rc = %d\n",
 801                                mdt_obd_name(info->mti_mdt),
 802                                PFID(mdt_object_fid(o)), rc);
 803                         RETURN(rc);
 804                 }
 805         }
 806
 807         if (likely(ma->ma_valid & MA_INODE))
 808                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
 809         else
 810                 RETURN(-EFAULT);
 811
 812         if (mdt_body_has_lov(la, reqbody)) {
 813                 if (ma->ma_valid & MA_LOV) {
 814                         LASSERT(ma->ma_lmm_size);
 815                         mdt_dump_lmm(D_INFO, ma->ma_lmm);
 816                         repbody->eadatasize = ma->ma_lmm_size;
 817                         if (S_ISDIR(la->la_mode))
 818                                 repbody->valid |= OBD_MD_FLDIREA;
 819                         else
 820                                 repbody->valid |= OBD_MD_FLEASIZE;
 821                 }
 822                 if (ma->ma_valid & MA_LMV) {
 823                         LASSERT(S_ISDIR(la->la_mode));
 824                         repbody->eadatasize = ma->ma_lmv_size;
 825                         repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
 826                 }
 827         } else if (S_ISLNK(la->la_mode) &&
 828                    reqbody->valid & OBD_MD_LINKNAME) {
 829                 buffer->lb_buf = ma->ma_lmm;
 830                 /* eadatasize from client includes NULL-terminator, so
 831                  * there is no need to read it */
 832                 buffer->lb_len = reqbody->eadatasize - 1;
 833                 rc = mo_readlink(env, next, buffer);
 834                 if (unlikely(rc <= 0)) {
 835                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
 836                                mdt_obd_name(info->mti_mdt),
 837                                PFID(mdt_object_fid(o)), rc);
 838                         rc = -EFAULT;
 839                 } else {
 840                         int print_limit = min_t(int, PAGE_CACHE_SIZE - 128, rc);
 841
 842                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
 843                                 rc -= 2;
 844                         repbody->valid |= OBD_MD_LINKNAME;
 845                         /* we need to report back size with NULL-terminator
 846                          * because client expects that */
 847                         repbody->eadatasize = rc + 1;
 848                         if (repbody->eadatasize != reqbody->eadatasize)
 849                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
 850                                        "on "DFID ", expected %d\n",
 851                                        mdt_obd_name(info->mti_mdt),
 852                                        rc, PFID(mdt_object_fid(o)),
 853                                        reqbody->eadatasize - 1);
 854                         /* NULL terminate */
 855                         ((char *)ma->ma_lmm)[rc] = 0;
 856
 857                         /* If the total CDEBUG() size is larger than a page, it
 858                          * will print a warning to the console, avoid this by
 859                          * printing just the last part of the symlink. */
 860                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
 861                                print_limit < rc ? "..." : "", print_limit,
 862                                (char *)ma->ma_lmm + rc - print_limit, rc);
 863                         rc = 0;
 864                 }
 865         }
 866
 867         if (reqbody->valid & OBD_MD_FLMODEASIZE) {
 868                 repbody->max_cookiesize = 0;
 869                 repbody->max_mdsize = info->mti_mdt->mdt_max_mdsize;
 870                 repbody->valid |= OBD_MD_FLMODEASIZE;
 871                 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
 872                        "MAX_COOKIE to : %d:%d\n", repbody->max_mdsize,
 873                        repbody->max_cookiesize);
 874         }
 875
 876         if (exp_connect_rmtclient(info->mti_exp) &&
 877             reqbody->valid & OBD_MD_FLRMTPERM) {
 878                 void *buf = req_capsule_server_get(pill, &RMF_ACL);
 879
 880                 /* mdt_getattr_lock only */
 881                 rc = mdt_pack_remote_perm(info, o, buf);
 882                 if (rc) {
 883                         repbody->valid &= ~OBD_MD_FLRMTPERM;
 884                         repbody->aclsize = 0;
 885                         RETURN(rc);
 886                 } else {
 887                         repbody->valid |= OBD_MD_FLRMTPERM;
 888                         repbody->aclsize = sizeof(struct mdt_remote_perm);
 889                 }
 890         }
 891 #ifdef CONFIG_FS_POSIX_ACL
 892         else if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
 893                  (reqbody->valid & OBD_MD_FLACL)) {
 894                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
 895                 buffer->lb_len = req_capsule_get_size(pill,
 896                                                       &RMF_ACL, RCL_SERVER);
 897                 if (buffer->lb_len > 0) {
 898                         rc = mo_xattr_get(env, next, buffer,
 899                                           XATTR_NAME_ACL_ACCESS);
 900                         if (rc < 0) {
 901                                 if (rc == -ENODATA) {
 902                                         repbody->aclsize = 0;
 903                                         repbody->valid |= OBD_MD_FLACL;
 904                                         rc = 0;
 905                                 } else if (rc == -EOPNOTSUPP) {
 906                                         rc = 0;
 907                                 } else {
 908                                         CERROR("%s: unable to read "DFID
 909                                                " ACL: rc = %d\n",
 910                                                mdt_obd_name(info->mti_mdt),
 911                                                PFID(mdt_object_fid(o)), rc);
 912                                 }
 913                         } else {
 914                                 repbody->aclsize = rc;
 915                                 repbody->valid |= OBD_MD_FLACL;
 916                                 rc = 0;
 917                         }
 918                 }
 919         }
 920 #endif
 921
 922         if (reqbody->valid & OBD_MD_FLMDSCAPA &&
 923             info->mti_mdt->mdt_lut.lut_mds_capa &&
 924             exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) {
 925                 struct lustre_capa *capa;
 926
 927                 capa = req_capsule_server_get(pill, &RMF_CAPA1);
 928                 LASSERT(capa);
 929                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 930                 rc = mo_capa_get(env, next, capa, 0);
 931                 if (rc)
 932                         RETURN(rc);
 933                 repbody->valid |= OBD_MD_FLMDSCAPA;
 934         }
 935
 936 out:
 937         if (rc == 0)
 938                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
 939
 940         RETURN(rc);
 941 }
 942
 943 static int mdt_renew_capa(struct mdt_thread_info *info)
 944 {
 945         struct mdt_object  *obj = info->mti_object;
 946         struct mdt_body    *body;
 947         struct lustre_capa *capa, *c;
 948         int rc;
 949         ENTRY;
 950
 951         /* if object doesn't exist, or server has disabled capability,
 952          * return directly, client will find body->valid OBD_MD_FLOSSCAPA
 953          * flag not set.
 954          */
 955         if (!obj || !info->mti_mdt->mdt_lut.lut_oss_capa ||
 956             !(exp_connect_flags(info->mti_exp) & OBD_CONNECT_OSS_CAPA))
 957                 RETURN(0);
 958
 959         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 960         LASSERT(body != NULL);
 961
 962         c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
 963         LASSERT(c);
 964
 965         capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
 966         LASSERT(capa);
 967
 968         *capa = *c;
 969         rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
 970         if (rc == 0)
 971                 body->valid |= OBD_MD_FLOSSCAPA;
 972         RETURN(rc);
 973 }
 974
 975 int mdt_getattr(struct tgt_session_info *tsi)
 976 {
 977         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
 978         struct mdt_object       *obj = info->mti_object;
 979         struct req_capsule      *pill = info->mti_pill;
 980         struct mdt_body         *reqbody;
 981         struct mdt_body         *repbody;
 982         mode_t                   mode;
 983         int rc, rc2;
 984         ENTRY;
 985
 986         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
 987         LASSERT(reqbody);
 988
 989         if (reqbody->valid & OBD_MD_FLOSSCAPA) {
 990                 rc = req_capsule_server_pack(pill);
 991                 if (unlikely(rc))
 992                         RETURN(err_serious(rc));
 993                 rc = mdt_renew_capa(info);
 994                 GOTO(out_shrink, rc);
 995         }
 996
 997         LASSERT(obj != NULL);
 998         LASSERT(lu_object_assert_exists(&obj->mot_obj));
 999
1000         mode = lu_object_attr(&obj->mot_obj);
1001
1002         /* old clients may not report needed easize, use max value then */
1003         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1004                              reqbody->eadatasize == 0 ?
1005                              info->mti_mdt->mdt_max_mdsize :
1006                              reqbody->eadatasize);
1007
1008         rc = req_capsule_server_pack(pill);
1009         if (unlikely(rc != 0))
1010                 GOTO(out, rc = err_serious(rc));
1011
1012         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1013         LASSERT(repbody != NULL);
1014         repbody->eadatasize = 0;
1015         repbody->aclsize = 0;
1016
1017         if (reqbody->valid & OBD_MD_FLRMTPERM)
1018                 rc = mdt_init_ucred(info, reqbody);
1019         else
1020                 rc = mdt_check_ucred(info);
1021         if (unlikely(rc))
1022                 GOTO(out_shrink, rc);
1023
1024         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1025
1026         /*
1027          * Don't check capability at all, because rename might getattr for
1028          * remote obj, and at that time no capability is available.
1029          */
1030         mdt_set_capainfo(info, 1, &reqbody->fid1, BYPASS_CAPA);
1031         rc = mdt_getattr_internal(info, obj, 0);
1032         if (reqbody->valid & OBD_MD_FLRMTPERM)
1033                 mdt_exit_ucred(info);
1034         EXIT;
1035 out_shrink:
1036         mdt_client_compatibility(info);
1037         rc2 = mdt_fix_reply(info);
1038         if (rc == 0)
1039                 rc = rc2;
1040 out:
1041         mdt_thread_info_fini(info);
1042         return rc;
1043 }
1044
1045 int mdt_is_subdir(struct tgt_session_info *tsi)
1046 {
1047         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1048         struct mdt_object     *o = info->mti_object;
1049         struct req_capsule    *pill = info->mti_pill;
1050         const struct mdt_body *body = info->mti_body;
1051         struct mdt_body       *repbody;
1052         int                    rc;
1053         ENTRY;
1054
1055         LASSERT(o != NULL);
1056
1057         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1058
1059         /*
1060          * We save last checked parent fid to @repbody->fid1 for remote
1061          * directory case.
1062          */
1063         LASSERT(fid_is_sane(&body->fid2));
1064         LASSERT(mdt_object_exists(o) && !mdt_object_remote(o));
1065         rc = mdo_is_subdir(info->mti_env, mdt_object_child(o),
1066                            &body->fid2, &repbody->fid1);
1067         if (rc == 0 || rc == -EREMOTE)
1068                 repbody->valid |= OBD_MD_FLID;
1069
1070         mdt_thread_info_fini(info);
1071         RETURN(rc);
1072 }
1073
1074 int mdt_swap_layouts(struct tgt_session_info *tsi)
1075 {
1076         struct mdt_thread_info  *info;
1077         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1078         struct obd_export       *exp = req->rq_export;
1079         struct mdt_object       *o1, *o2, *o;
1080         struct mdt_lock_handle  *lh1, *lh2;
1081         struct mdc_swap_layouts *msl;
1082         int                      rc;
1083         ENTRY;
1084
1085         /* client does not support layout lock, so layout swaping
1086          * is disabled.
1087          * FIXME: there is a problem for old clients which don't support
1088          * layout lock yet. If those clients have already opened the file
1089          * they won't be notified at all so that old layout may still be
1090          * used to do IO. This can be fixed after file release is landed by
1091          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1092         if (!exp_connect_layout(exp))
1093                 RETURN(-EOPNOTSUPP);
1094
1095         info = tsi2mdt_info(tsi);
1096         if (req_capsule_get_size(info->mti_pill, &RMF_CAPA1, RCL_CLIENT))
1097                 mdt_set_capainfo(info, 0, &info->mti_body->fid1,
1098                                  req_capsule_client_get(info->mti_pill,
1099                                                         &RMF_CAPA1));
1100
1101         if (req_capsule_get_size(info->mti_pill, &RMF_CAPA2, RCL_CLIENT))
1102                 mdt_set_capainfo(info, 1, &info->mti_body->fid2,
1103                                  req_capsule_client_get(info->mti_pill,
1104                                                         &RMF_CAPA2));
1105
1106         o1 = info->mti_object;
1107         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1108                                 &info->mti_body->fid2);
1109         if (IS_ERR(o))
1110                 GOTO(out, rc = PTR_ERR(o));
1111
1112         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1113                 GOTO(put, rc = -ENOENT);
1114
1115         rc = lu_fid_cmp(&info->mti_body->fid1, &info->mti_body->fid2);
1116         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1117                 GOTO(put, rc);
1118
1119         if (rc < 0)
1120                 swap(o1, o2);
1121
1122         /* permission check. Make sure the calling process having permission
1123          * to write both files. */
1124         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1125                                 MAY_WRITE);
1126         if (rc < 0)
1127                 GOTO(put, rc);
1128
1129         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1130                                 MAY_WRITE);
1131         if (rc < 0)
1132                 GOTO(put, rc);
1133
1134         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1135         if (msl == NULL)
1136                 GOTO(put, rc = -EPROTO);
1137
1138         lh1 = &info->mti_lh[MDT_LH_NEW];
1139         mdt_lock_reg_init(lh1, LCK_EX);
1140         lh2 = &info->mti_lh[MDT_LH_OLD];
1141         mdt_lock_reg_init(lh2, LCK_EX);
1142
1143         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1144                              MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK);
1145         if (rc < 0)
1146                 GOTO(put, rc);
1147
1148         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1149                              MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK);
1150         if (rc < 0)
1151                 GOTO(unlock1, rc);
1152
1153         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1154                              mdt_object_child(o2), msl->msl_flags);
1155         GOTO(unlock2, rc);
1156 unlock2:
1157         mdt_object_unlock(info, o2, lh2, rc);
1158 unlock1:
1159         mdt_object_unlock(info, o1, lh1, rc);
1160 put:
1161         mdt_object_put(info->mti_env, o);
1162 out:
1163         mdt_thread_info_fini(info);
1164         RETURN(rc);
1165 }
1166
1167 static int mdt_raw_lookup(struct mdt_thread_info *info,
1168                           struct mdt_object *parent,
1169                           const struct lu_name *lname,
1170                           struct ldlm_reply *ldlm_rep)
1171 {
1172         struct md_object *next = mdt_object_child(info->mti_object);
1173         const struct mdt_body *reqbody = info->mti_body;
1174         struct lu_fid *child_fid = &info->mti_tmp_fid1;
1175         struct mdt_body *repbody;
1176         int rc;
1177         ENTRY;
1178
1179         if (reqbody->valid != OBD_MD_FLID)
1180                 RETURN(0);
1181
1182         LASSERT(!info->mti_cross_ref);
1183
1184         /* Only got the fid of this obj by name */
1185         fid_zero(child_fid);
1186         rc = mdo_lookup(info->mti_env, next, lname, child_fid,
1187                         &info->mti_spec);
1188 #if 0
1189         /* XXX is raw_lookup possible as intent operation? */
1190         if (rc != 0) {
1191                 if (rc == -ENOENT)
1192                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1193                 RETURN(rc);
1194         } else
1195                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1196
1197         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1198 #endif
1199         if (rc == 0) {
1200                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1201                 repbody->fid1 = *child_fid;
1202                 repbody->valid = OBD_MD_FLID;
1203         }
1204         RETURN(1);
1205 }
1206
1207 /*
1208  * UPDATE lock should be taken against parent, and be release before exit;
1209  * child_bits lock should be taken against child, and be returned back:
1210  *            (1)normal request should release the child lock;
1211  *            (2)intent request will grant the lock to client.
1212  */
1213 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1214                                  struct mdt_lock_handle *lhc,
1215                                  __u64 child_bits,
1216                                  struct ldlm_reply *ldlm_rep)
1217 {
1218         struct ptlrpc_request  *req       = mdt_info_req(info);
1219         struct mdt_body        *reqbody   = NULL;
1220         struct mdt_object      *parent    = info->mti_object;
1221         struct mdt_object      *child;
1222         struct md_object       *next      = mdt_object_child(parent);
1223         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1224         struct lu_name         *lname     = NULL;
1225         struct mdt_lock_handle *lhp       = NULL;
1226         struct ldlm_lock       *lock;
1227         struct ldlm_res_id     *res_id;
1228         int                     is_resent;
1229         int                     ma_need = 0;
1230         int                     rc;
1231
1232         ENTRY;
1233
1234         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1235         LASSERT(ergo(is_resent,
1236                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1237
1238         LASSERT(parent != NULL);
1239
1240         lname = &info->mti_name;
1241         mdt_name_unpack(info->mti_pill, &RMF_NAME, lname, MNF_FIX_ANON);
1242
1243         if (!info->mti_cross_ref) {
1244                 /*
1245                  * XXX: Check for anonymous name is for getattr by fid
1246                  * (OBD_CONNECT_ATTRFID), otherwise do not allow empty name,
1247                  * that is the name must contain at least one character and
1248                  * the terminating '\0'.
1249                  */
1250                 if (!lu_name_is_valid(lname)) {
1251                         reqbody = req_capsule_client_get(info->mti_pill,
1252                                                          &RMF_MDT_BODY);
1253                         if (unlikely(reqbody == NULL))
1254                                 RETURN(err_serious(-EFAULT));
1255
1256                         if (unlikely(!fid_is_sane(&reqbody->fid2)))
1257                                 RETURN(err_serious(-EINVAL));
1258
1259                         CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1260                                "ldlm_rep = %p\n",
1261                                PFID(mdt_object_fid(parent)),
1262                                PFID(&reqbody->fid2), ldlm_rep);
1263                 } else {
1264                         CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1265                                "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1266                                PNAME(lname), ldlm_rep);
1267                 }
1268         }
1269
1270         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1271
1272         if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1273                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1274                                 &parent->mot_obj,
1275                                 "Parent doesn't exist!\n");
1276                 RETURN(-ESTALE);
1277         } else if (!info->mti_cross_ref) {
1278                 LASSERTF(!mdt_object_remote(parent),
1279                          "Parent "DFID" is on remote server\n",
1280                          PFID(mdt_object_fid(parent)));
1281         }
1282
1283         if (lu_name_is_valid(lname)) {
1284                 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1285                 if (rc != 0) {
1286                         if (rc > 0)
1287                                 rc = 0;
1288                         RETURN(rc);
1289                 }
1290         }
1291
1292         if (info->mti_cross_ref) {
1293                 /* Only getattr on the child. Parent is on another node. */
1294                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1295                 child = parent;
1296                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1297                        "ldlm_rep=%p\n", PFID(mdt_object_fid(child)), ldlm_rep);
1298
1299                 if (is_resent) {
1300                         /* Do not take lock for resent case. */
1301                         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1302                         LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
1303                                  lhc->mlh_reg_lh.cookie);
1304                         LASSERT(fid_res_name_eq(mdt_object_fid(child),
1305                                                 &lock->l_resource->lr_name));
1306                         LDLM_LOCK_PUT(lock);
1307                         rc = 0;
1308                 } else {
1309                         mdt_lock_handle_init(lhc);
1310                         mdt_lock_reg_init(lhc, LCK_PR);
1311
1312                         /*
1313                          * Object's name is on another MDS, no lookup or layout
1314                          * lock is needed here but update lock is.
1315                          */
1316                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
1317                                         MDS_INODELOCK_LAYOUT);
1318                         child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1319
1320                         rc = mdt_object_lock(info, child, lhc, child_bits,
1321                                              MDT_LOCAL_LOCK);
1322                 }
1323                 if (rc == 0) {
1324                         /* Finally, we can get attr for child. */
1325                         if (!mdt_object_exists(child)) {
1326                                 LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1327                                                 &child->mot_obj,
1328                                              "remote object doesn't exist.\n");
1329                                 mdt_object_unlock(info, child, lhc, 1);
1330                                 RETURN(-ENOENT);
1331                         }
1332
1333                         mdt_set_capainfo(info, 0, mdt_object_fid(child),
1334                                          BYPASS_CAPA);
1335                         rc = mdt_getattr_internal(info, child, 0);
1336                         if (unlikely(rc != 0))
1337                                 mdt_object_unlock(info, child, lhc, 1);
1338                 }
1339                 RETURN(rc);
1340         }
1341
1342         if (lu_name_is_valid(lname)) {
1343                 /* step 1: lock parent only if parent is a directory */
1344                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1345                         lhp = &info->mti_lh[MDT_LH_PARENT];
1346                         mdt_lock_pdo_init(lhp, LCK_PR, lname);
1347                         rc = mdt_object_lock(info, parent, lhp,
1348                                              MDS_INODELOCK_UPDATE,
1349                                              MDT_LOCAL_LOCK);
1350                         if (unlikely(rc != 0))
1351                                 RETURN(rc);
1352                 }
1353
1354                 /* step 2: lookup child's fid by name */
1355                 fid_zero(child_fid);
1356                 rc = mdo_lookup(info->mti_env, next, lname, child_fid,
1357                                 &info->mti_spec);
1358
1359                 if (rc != 0) {
1360                         if (rc == -ENOENT)
1361                                 mdt_set_disposition(info, ldlm_rep,
1362                                                     DISP_LOOKUP_NEG);
1363                         GOTO(out_parent, rc);
1364                 } else
1365                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1366         } else {
1367                 *child_fid = reqbody->fid2;
1368                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1369         }
1370
1371         /*
1372          *step 3: find the child object by fid & lock it.
1373          *        regardless if it is local or remote.
1374          *
1375          *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1376          *      set parent dir fid the same as child fid in getattr by fid case
1377          *      we should not lu_object_find() the object again, could lead
1378          *      to hung if there is a concurrent unlink destroyed the object.
1379          */
1380         if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1381                 mdt_object_get(info->mti_env, parent);
1382                 child = parent;
1383         } else {
1384                 child = mdt_object_find(info->mti_env, info->mti_mdt,
1385                                         child_fid);
1386         }
1387
1388         if (unlikely(IS_ERR(child)))
1389                 GOTO(out_parent, rc = PTR_ERR(child));
1390         if (is_resent) {
1391                 /* Do not take lock for resent case. */
1392                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1393                 LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
1394                          lhc->mlh_reg_lh.cookie);
1395
1396                 res_id = &lock->l_resource->lr_name;
1397                 if (!fid_res_name_eq(mdt_object_fid(child),
1398                                      &lock->l_resource->lr_name)) {
1399                         LASSERTF(fid_res_name_eq(mdt_object_fid(parent),
1400                                                  &lock->l_resource->lr_name),
1401                                  "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1402                                  PLDLMRES(lock->l_resource),
1403                                  PFID(mdt_object_fid(parent)));
1404                         CWARN("Although resent, but still not get child lock"
1405                               "parent:"DFID" child:"DFID"\n",
1406                               PFID(mdt_object_fid(parent)),
1407                               PFID(mdt_object_fid(child)));
1408                         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
1409                         LDLM_LOCK_PUT(lock);
1410                         GOTO(relock, 0);
1411                 }
1412                 LDLM_LOCK_PUT(lock);
1413                 rc = 0;
1414         } else {
1415                 bool try_layout = false;
1416
1417 relock:
1418                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout*2);
1419                 mdt_lock_handle_init(lhc);
1420                 mdt_lock_reg_init(lhc, LCK_PR);
1421
1422                 if (!mdt_object_exists(child)) {
1423                         LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1424                                         &child->mot_obj,
1425                                         "Object doesn't exist!\n");
1426                         GOTO(out_child, rc = -ENOENT);
1427                 }
1428
1429                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1430                       mdt_object_exists(child) && !mdt_object_remote(child)) {
1431                         struct md_attr *ma = &info->mti_attr;
1432
1433                         ma->ma_valid = 0;
1434                         ma->ma_need = MA_INODE;
1435                         rc = mdt_attr_get_complex(info, child, ma);
1436                         if (unlikely(rc != 0))
1437                                 GOTO(out_child, rc);
1438
1439                         /* If the file has not been changed for some time, we
1440                          * return not only a LOOKUP lock, but also an UPDATE
1441                          * lock and this might save us RPC on later STAT. For
1442                          * directories, it also let negative dentry cache start
1443                          * working for this dir. */
1444                         if (ma->ma_valid & MA_INODE &&
1445                             ma->ma_attr.la_valid & LA_CTIME &&
1446                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1447                                 ma->ma_attr.la_ctime < cfs_time_current_sec())
1448                                 child_bits |= MDS_INODELOCK_UPDATE;
1449                 }
1450
1451                 /* layout lock must be granted in a best-effort way
1452                  * for IT operations */
1453                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1454                 if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1455                     exp_connect_layout(info->mti_exp) &&
1456                     S_ISREG(lu_object_attr(&child->mot_obj)) &&
1457                     !mdt_object_remote(child) && ldlm_rep != NULL) {
1458                         /* try to grant layout lock for regular file. */
1459                         try_layout = true;
1460                 }
1461
1462                 rc = 0;
1463                 if (try_layout) {
1464                         child_bits |= MDS_INODELOCK_LAYOUT;
1465                         /* try layout lock, it may fail to be granted due to
1466                          * contention at LOOKUP or UPDATE */
1467                         if (!mdt_object_lock_try(info, child, lhc, child_bits,
1468                                                  MDT_CROSS_LOCK)) {
1469                                 child_bits &= ~MDS_INODELOCK_LAYOUT;
1470                                 LASSERT(child_bits != 0);
1471                                 rc = mdt_object_lock(info, child, lhc,
1472                                                 child_bits, MDT_CROSS_LOCK);
1473                         } else {
1474                                 ma_need |= MA_LOV;
1475                         }
1476                 } else {
1477                         rc = mdt_object_lock(info, child, lhc, child_bits,
1478                                                 MDT_CROSS_LOCK);
1479                 }
1480                 if (unlikely(rc != 0))
1481                         GOTO(out_child, rc);
1482         }
1483
1484         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1485         /* Get MA_SOM attributes if update lock is given. */
1486         if (lock &&
1487             lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_UPDATE &&
1488             S_ISREG(lu_object_attr(&mdt_object_child(child)->mo_lu)))
1489                 ma_need |= MA_SOM;
1490
1491         /* finally, we can get attr for child. */
1492         mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
1493         rc = mdt_getattr_internal(info, child, ma_need);
1494         if (unlikely(rc != 0)) {
1495                 mdt_object_unlock(info, child, lhc, 1);
1496         } else if (lock) {
1497                 /* Debugging code. */
1498                 res_id = &lock->l_resource->lr_name;
1499                 LDLM_DEBUG(lock, "Returning lock to client");
1500                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1501                                          &lock->l_resource->lr_name),
1502                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1503                          PLDLMRES(lock->l_resource),
1504                          PFID(mdt_object_fid(child)));
1505                 if (mdt_object_exists(child) && !mdt_object_remote(child))
1506                         mdt_pack_size2body(info, child);
1507         }
1508         if (lock)
1509                 LDLM_LOCK_PUT(lock);
1510
1511         EXIT;
1512 out_child:
1513         mdt_object_put(info->mti_env, child);
1514 out_parent:
1515         if (lhp)
1516                 mdt_object_unlock(info, parent, lhp, 1);
1517         return rc;
1518 }
1519
1520 /* normal handler: should release the child lock */
1521 int mdt_getattr_name(struct tgt_session_info *tsi)
1522 {
1523         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1524         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1525         struct mdt_body        *reqbody;
1526         struct mdt_body        *repbody;
1527         int rc, rc2;
1528         ENTRY;
1529
1530         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1531         LASSERT(reqbody != NULL);
1532         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1533         LASSERT(repbody != NULL);
1534
1535         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
1536         repbody->eadatasize = 0;
1537         repbody->aclsize = 0;
1538
1539         rc = mdt_init_ucred(info, reqbody);
1540         if (unlikely(rc))
1541                 GOTO(out_shrink, rc);
1542
1543         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1544         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1545                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1546                 lhc->mlh_reg_lh.cookie = 0;
1547         }
1548         mdt_exit_ucred(info);
1549         EXIT;
1550 out_shrink:
1551         mdt_client_compatibility(info);
1552         rc2 = mdt_fix_reply(info);
1553         if (rc == 0)
1554                 rc = rc2;
1555         mdt_thread_info_fini(info);
1556         return rc;
1557 }
1558
1559 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1560                          void *karg, void *uarg);
1561
1562 int mdt_set_info(struct tgt_session_info *tsi)
1563 {
1564         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1565         char                    *key;
1566         void                    *val;
1567         int                      keylen, vallen, rc = 0;
1568
1569         ENTRY;
1570
1571         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
1572         if (key == NULL) {
1573                 DEBUG_REQ(D_HA, req, "no set_info key");
1574                 RETURN(err_serious(-EFAULT));
1575         }
1576
1577         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
1578                                       RCL_CLIENT);
1579
1580         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
1581         if (val == NULL) {
1582                 DEBUG_REQ(D_HA, req, "no set_info val");
1583                 RETURN(err_serious(-EFAULT));
1584         }
1585
1586         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
1587                                       RCL_CLIENT);
1588
1589         /* Swab any part of val you need to here */
1590         if (KEY_IS(KEY_READ_ONLY)) {
1591                 spin_lock(&req->rq_export->exp_lock);
1592                 if (*(__u32 *)val)
1593                         *exp_connect_flags_ptr(req->rq_export) |=
1594                                 OBD_CONNECT_RDONLY;
1595                 else
1596                         *exp_connect_flags_ptr(req->rq_export) &=
1597                                 ~OBD_CONNECT_RDONLY;
1598                 spin_unlock(&req->rq_export->exp_lock);
1599         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1600                 struct changelog_setinfo *cs = val;
1601
1602                 if (vallen != sizeof(*cs)) {
1603                         CERROR("%s: bad changelog_clear setinfo size %d\n",
1604                                tgt_name(tsi->tsi_tgt), vallen);
1605                         RETURN(-EINVAL);
1606                 }
1607                 if (ptlrpc_req_need_swab(req)) {
1608                         __swab64s(&cs->cs_recno);
1609                         __swab32s(&cs->cs_id);
1610                 }
1611
1612                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
1613                                    vallen, val, NULL);
1614         } else {
1615                 RETURN(-EINVAL);
1616         }
1617         RETURN(rc);
1618 }
1619
1620 int mdt_readpage(struct tgt_session_info *tsi)
1621 {
1622         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
1623         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
1624         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
1625         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
1626         struct mdt_body         *repbody;
1627         int                      rc;
1628         int                      i;
1629
1630         ENTRY;
1631
1632         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1633                 RETURN(err_serious(-ENOMEM));
1634
1635         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
1636         if (repbody == NULL || reqbody == NULL)
1637                 RETURN(err_serious(-EFAULT));
1638
1639         /*
1640          * prepare @rdpg before calling lower layers and transfer itself. Here
1641          * reqbody->size contains offset of where to start to read and
1642          * reqbody->nlink contains number bytes to read.
1643          */
1644         rdpg->rp_hash = reqbody->size;
1645         if (rdpg->rp_hash != reqbody->size) {
1646                 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1647                        rdpg->rp_hash, reqbody->size);
1648                 RETURN(-EFAULT);
1649         }
1650
1651         rdpg->rp_attrs = reqbody->mode;
1652         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
1653                 rdpg->rp_attrs |= LUDA_64BITHASH;
1654         rdpg->rp_count  = min_t(unsigned int, reqbody->nlink,
1655                                 exp_max_brw_size(tsi->tsi_exp));
1656         rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE - 1) >>
1657                           PAGE_CACHE_SHIFT;
1658         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1659         if (rdpg->rp_pages == NULL)
1660                 RETURN(-ENOMEM);
1661
1662         for (i = 0; i < rdpg->rp_npages; ++i) {
1663                 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1664                 if (rdpg->rp_pages[i] == NULL)
1665                         GOTO(free_rdpg, rc = -ENOMEM);
1666         }
1667
1668         /* call lower layers to fill allocated pages with directory data */
1669         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
1670         if (rc < 0)
1671                 GOTO(free_rdpg, rc);
1672
1673         /* send pages to client */
1674         rc = tgt_sendpage(tsi, rdpg, rc);
1675
1676         EXIT;
1677 free_rdpg:
1678
1679         for (i = 0; i < rdpg->rp_npages; i++)
1680                 if (rdpg->rp_pages[i] != NULL)
1681                         __free_page(rdpg->rp_pages[i]);
1682         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1683
1684         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1685                 RETURN(0);
1686
1687         return rc;
1688 }
1689
1690 static int mdt_reint_internal(struct mdt_thread_info *info,
1691                               struct mdt_lock_handle *lhc,
1692                               __u32 op)
1693 {
1694         struct req_capsule      *pill = info->mti_pill;
1695         struct mdt_body         *repbody;
1696         int                      rc = 0, rc2;
1697
1698         ENTRY;
1699
1700         rc = mdt_reint_unpack(info, op);
1701         if (rc != 0) {
1702                 CERROR("Can't unpack reint, rc %d\n", rc);
1703                 RETURN(err_serious(rc));
1704         }
1705
1706         /* for replay (no_create) lmm is not needed, client has it already */
1707         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1708                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1709                                      info->mti_rr.rr_eadatalen);
1710
1711         /* llog cookies are always 0, the field is kept for compatibility */
1712         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1713                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
1714
1715         rc = req_capsule_server_pack(pill);
1716         if (rc != 0) {
1717                 CERROR("Can't pack response, rc %d\n", rc);
1718                 RETURN(err_serious(rc));
1719         }
1720
1721         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1722                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1723                 LASSERT(repbody);
1724                 repbody->eadatasize = 0;
1725                 repbody->aclsize = 0;
1726         }
1727
1728         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1729
1730         /* for replay no cookkie / lmm need, because client have this already */
1731         if (info->mti_spec.no_create)
1732                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1733                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1734
1735         rc = mdt_init_ucred_reint(info);
1736         if (rc)
1737                 GOTO(out_shrink, rc);
1738
1739         rc = mdt_fix_attr_ucred(info, op);
1740         if (rc != 0)
1741                 GOTO(out_ucred, rc = err_serious(rc));
1742
1743         if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1744                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1745                 GOTO(out_ucred, rc);
1746         }
1747         rc = mdt_reint_rec(info, lhc);
1748         EXIT;
1749 out_ucred:
1750         mdt_exit_ucred(info);
1751 out_shrink:
1752         mdt_client_compatibility(info);
1753         rc2 = mdt_fix_reply(info);
1754         if (rc == 0)
1755                 rc = rc2;
1756         return rc;
1757 }
1758
1759 static long mdt_reint_opcode(struct ptlrpc_request *req,
1760                              const struct req_format **fmt)
1761 {
1762         struct mdt_device       *mdt;
1763         struct mdt_rec_reint    *rec;
1764         long                     opc;
1765
1766         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
1767         if (rec != NULL) {
1768                 opc = rec->rr_opcode;
1769                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
1770                 if (opc < REINT_MAX && fmt[opc] != NULL)
1771                         req_capsule_extend(&req->rq_pill, fmt[opc]);
1772                 else {
1773                         mdt = mdt_exp2dev(req->rq_export);
1774                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
1775                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
1776                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
1777                         opc = err_serious(-EFAULT);
1778                 }
1779         } else {
1780                 opc = err_serious(-EFAULT);
1781         }
1782         return opc;
1783 }
1784
1785 int mdt_reint(struct tgt_session_info *tsi)
1786 {
1787         long opc;
1788         int  rc;
1789         static const struct req_format *reint_fmts[REINT_MAX] = {
1790                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
1791                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
1792                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
1793                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
1794                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
1795                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
1796                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
1797                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK
1798         };
1799
1800         ENTRY;
1801
1802         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
1803         if (opc >= 0) {
1804                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1805                 /*
1806                  * No lock possible here from client to pass it to reint code
1807                  * path.
1808                  */
1809                 rc = mdt_reint_internal(info, NULL, opc);
1810                 mdt_thread_info_fini(info);
1811         } else {
1812                 rc = opc;
1813         }
1814
1815         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1816         RETURN(rc);
1817 }
1818
1819 /* this should sync the whole device */
1820 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1821 {
1822         struct dt_device *dt = mdt->mdt_bottom;
1823         int rc;
1824         ENTRY;
1825
1826         rc = dt->dd_ops->dt_sync(env, dt);
1827         RETURN(rc);
1828 }
1829
1830 /* this should sync this object */
1831 static int mdt_object_sync(struct mdt_thread_info *info)
1832 {
1833         struct md_object *next;
1834         int rc;
1835         ENTRY;
1836
1837         if (!mdt_object_exists(info->mti_object)) {
1838                 CWARN("Non existing object  "DFID"!\n",
1839                       PFID(mdt_object_fid(info->mti_object)));
1840                 RETURN(-ESTALE);
1841         }
1842         next = mdt_object_child(info->mti_object);
1843         rc = mo_object_sync(info->mti_env, next);
1844
1845         RETURN(rc);
1846 }
1847
1848 int mdt_sync(struct tgt_session_info *tsi)
1849 {
1850         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1851         struct req_capsule      *pill = tsi->tsi_pill;
1852         struct mdt_body         *body;
1853         int                      rc;
1854
1855         ENTRY;
1856
1857         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1858                 RETURN(err_serious(-ENOMEM));
1859
1860         if (fid_seq(&tsi->tsi_mdt_body->fid1) == 0) {
1861                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
1862         } else {
1863                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1864
1865                 /* sync an object */
1866                 rc = mdt_object_sync(info);
1867                 if (rc == 0) {
1868                         const struct lu_fid *fid;
1869                         struct lu_attr *la = &info->mti_attr.ma_attr;
1870
1871                         info->mti_attr.ma_need = MA_INODE;
1872                         info->mti_attr.ma_valid = 0;
1873                         rc = mdt_attr_get_complex(info, info->mti_object,
1874                                                   &info->mti_attr);
1875                         if (rc == 0) {
1876                                 body = req_capsule_server_get(pill,
1877                                                               &RMF_MDT_BODY);
1878                                 fid = mdt_object_fid(info->mti_object);
1879                                 mdt_pack_attr2body(info, body, la, fid);
1880                         }
1881                 }
1882                 mdt_thread_info_fini(info);
1883         }
1884         if (rc == 0)
1885                 mdt_counter_incr(req, LPROC_MDT_SYNC);
1886
1887         RETURN(rc);
1888 }
1889
1890 /*
1891  * Handle quota control requests to consult current usage/limit, but also
1892  * to configure quota enforcement
1893  */
1894 int mdt_quotactl(struct tgt_session_info *tsi)
1895 {
1896         struct obd_export       *exp  = tsi->tsi_exp;
1897         struct req_capsule      *pill = tsi->tsi_pill;
1898         struct obd_quotactl     *oqctl, *repoqc;
1899         int                      id, rc;
1900         struct mdt_device       *mdt = mdt_exp2dev(exp);
1901         struct lu_device        *qmt = mdt->mdt_qmt_dev;
1902         ENTRY;
1903
1904         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1905         if (oqctl == NULL)
1906                 RETURN(err_serious(-EPROTO));
1907
1908         rc = req_capsule_server_pack(pill);
1909         if (rc)
1910                 RETURN(err_serious(rc));
1911
1912         switch (oqctl->qc_cmd) {
1913         case Q_QUOTACHECK:
1914         case LUSTRE_Q_INVALIDATE:
1915         case LUSTRE_Q_FINVALIDATE:
1916         case Q_QUOTAON:
1917         case Q_QUOTAOFF:
1918         case Q_INITQUOTA:
1919                 /* deprecated, not used any more */
1920                 RETURN(-EOPNOTSUPP);
1921                 /* master quotactl */
1922         case Q_GETINFO:
1923         case Q_SETINFO:
1924         case Q_SETQUOTA:
1925         case Q_GETQUOTA:
1926                 if (qmt == NULL)
1927                         RETURN(-EOPNOTSUPP);
1928                 /* slave quotactl */
1929         case Q_GETOINFO:
1930         case Q_GETOQUOTA:
1931                 break;
1932         default:
1933                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
1934                 RETURN(-EFAULT);
1935         }
1936
1937         /* map uid/gid for remote client */
1938         id = oqctl->qc_id;
1939         if (exp_connect_rmtclient(exp)) {
1940                 struct lustre_idmap_table *idmap;
1941
1942                 idmap = exp->exp_mdt_data.med_idmap;
1943
1944                 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
1945                              oqctl->qc_cmd != Q_GETINFO))
1946                         RETURN(-EPERM);
1947
1948                 if (oqctl->qc_type == USRQUOTA)
1949                         id = lustre_idmap_lookup_uid(NULL, idmap, 0,
1950                                                      oqctl->qc_id);
1951                 else if (oqctl->qc_type == GRPQUOTA)
1952                         id = lustre_idmap_lookup_gid(NULL, idmap, 0,
1953                                                      oqctl->qc_id);
1954                 else
1955                         RETURN(-EINVAL);
1956
1957                 if (id == CFS_IDMAP_NOTFOUND) {
1958                         CDEBUG(D_QUOTA, "no mapping for id %u\n", oqctl->qc_id);
1959                         RETURN(-EACCES);
1960                 }
1961         }
1962
1963         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
1964         if (repoqc == NULL)
1965                 RETURN(err_serious(-EFAULT));
1966
1967         if (oqctl->qc_id != id)
1968                 swap(oqctl->qc_id, id);
1969
1970         switch (oqctl->qc_cmd) {
1971
1972         case Q_GETINFO:
1973         case Q_SETINFO:
1974         case Q_SETQUOTA:
1975         case Q_GETQUOTA:
1976                 /* forward quotactl request to QMT */
1977                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
1978                 break;
1979
1980         case Q_GETOINFO:
1981         case Q_GETOQUOTA:
1982                 /* slave quotactl */
1983                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
1984                                    oqctl);
1985                 break;
1986
1987         default:
1988                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
1989                 RETURN(-EFAULT);
1990         }
1991
1992         if (oqctl->qc_id != id)
1993                 swap(oqctl->qc_id, id);
1994
1995         *repoqc = *oqctl;
1996         RETURN(rc);
1997 }
1998
1999 /** clone llog ctxt from child (mdd)
2000  * This allows remote llog (replicator) access.
2001  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2002  * context was originally set up, or we can handle them directly.
2003  * I choose the latter, but that means I need any llog
2004  * contexts set up by child to be accessable by the mdt.  So we clone the
2005  * context into our context list here.
2006  */
2007 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2008                                int idx)
2009 {
2010         struct md_device  *next = mdt->mdt_child;
2011         struct llog_ctxt *ctxt;
2012         int rc;
2013
2014         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2015                 return 0;
2016
2017         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2018         if (rc || ctxt == NULL) {
2019                 return 0;
2020         }
2021
2022         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2023         if (rc)
2024                 CERROR("Can't set mdt ctxt %d\n", rc);
2025
2026         return rc;
2027 }
2028
2029 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2030                                  struct mdt_device *mdt, int idx)
2031 {
2032         struct llog_ctxt *ctxt;
2033
2034         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2035         if (ctxt == NULL)
2036                 return 0;
2037         /* Put once for the get we just did, and once for the clone */
2038         llog_ctxt_put(ctxt);
2039         llog_ctxt_put(ctxt);
2040         return 0;
2041 }
2042
2043 /*
2044  * sec context handlers
2045  */
2046 int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2047 {
2048         int rc;
2049
2050         rc = mdt_handle_idmap(tsi);
2051         if (unlikely(rc)) {
2052                 struct ptlrpc_request   *req = tgt_ses_req(tsi);
2053                 __u32                    opc;
2054
2055                 opc = lustre_msg_get_opc(req->rq_reqmsg);
2056                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2057                         sptlrpc_svc_ctx_invalidate(req);
2058         }
2059
2060         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2061
2062         return rc;
2063 }
2064
2065 /*
2066  * quota request handlers
2067  */
2068 int mdt_quota_dqacq(struct tgt_session_info *tsi)
2069 {
2070         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
2071         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2072         int                      rc;
2073         ENTRY;
2074
2075         if (qmt == NULL)
2076                 RETURN(err_serious(-EOPNOTSUPP));
2077
2078         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2079         RETURN(rc);
2080 }
2081
2082 struct mdt_object *mdt_object_new(const struct lu_env *env,
2083                                   struct mdt_device *d,
2084                                   const struct lu_fid *f)
2085 {
2086         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2087         struct lu_object *o;
2088         struct mdt_object *m;
2089         ENTRY;
2090
2091         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2092         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2093         if (unlikely(IS_ERR(o)))
2094                 m = (struct mdt_object *)o;
2095         else
2096                 m = mdt_obj(o);
2097         RETURN(m);
2098 }
2099
2100 struct mdt_object *mdt_object_find(const struct lu_env *env,
2101                                    struct mdt_device *d,
2102                                    const struct lu_fid *f)
2103 {
2104         struct lu_object *o;
2105         struct mdt_object *m;
2106         ENTRY;
2107
2108         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2109         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2110         if (unlikely(IS_ERR(o)))
2111                 m = (struct mdt_object *)o;
2112         else
2113                 m = mdt_obj(o);
2114
2115         RETURN(m);
2116 }
2117
2118 /**
2119  * Asyncronous commit for mdt device.
2120  *
2121  * Pass asynchonous commit call down the MDS stack.
2122  *
2123  * \param env environment
2124  * \param mdt the mdt device
2125  */
2126 static void mdt_device_commit_async(const struct lu_env *env,
2127                                     struct mdt_device *mdt)
2128 {
2129         struct dt_device *dt = mdt->mdt_bottom;
2130         int rc;
2131
2132         rc = dt->dd_ops->dt_commit_async(env, dt);
2133         if (unlikely(rc != 0))
2134                 CWARN("async commit start failed with rc = %d", rc);
2135 }
2136
2137 /**
2138  * Mark the lock as "synchonous".
2139  *
2140  * Mark the lock to deffer transaction commit to the unlock time.
2141  *
2142  * \param lock the lock to mark as "synchonous"
2143  *
2144  * \see mdt_is_lock_sync
2145  * \see mdt_save_lock
2146  */
2147 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2148 {
2149         lock->l_ast_data = (void*)1;
2150 }
2151
2152 /**
2153  * Check whehter the lock "synchonous" or not.
2154  *
2155  * \param lock the lock to check
2156  * \retval 1 the lock is "synchonous"
2157  * \retval 0 the lock isn't "synchronous"
2158  *
2159  * \see mdt_set_lock_sync
2160  * \see mdt_save_lock
2161  */
2162 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2163 {
2164         return lock->l_ast_data != NULL;
2165 }
2166
2167 /**
2168  * Blocking AST for mdt locks.
2169  *
2170  * Starts transaction commit if in case of COS lock conflict or
2171  * deffers such a commit to the mdt_save_lock.
2172  *
2173  * \param lock the lock which blocks a request or cancelling lock
2174  * \param desc unused
2175  * \param data unused
2176  * \param flag indicates whether this cancelling or blocking callback
2177  * \retval 0
2178  * \see ldlm_blocking_ast_nocheck
2179  */
2180 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2181                      void *data, int flag)
2182 {
2183         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2184         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2185         int rc;
2186         ENTRY;
2187
2188         if (flag == LDLM_CB_CANCELING)
2189                 RETURN(0);
2190         lock_res_and_lock(lock);
2191         if (lock->l_blocking_ast != mdt_blocking_ast) {
2192                 unlock_res_and_lock(lock);
2193                 RETURN(0);
2194         }
2195         if (mdt_cos_is_enabled(mdt) &&
2196             lock->l_req_mode & (LCK_PW | LCK_EX) &&
2197             lock->l_blocking_lock != NULL &&
2198             lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
2199                 mdt_set_lock_sync(lock);
2200         }
2201         rc = ldlm_blocking_ast_nocheck(lock);
2202
2203         /* There is no lock conflict if l_blocking_lock == NULL,
2204          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2205          * when the last reference to a local lock was released */
2206         if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2207                 struct lu_env env;
2208
2209                 rc = lu_env_init(&env, LCT_LOCAL);
2210                 if (unlikely(rc != 0))
2211                         CWARN("lu_env initialization failed with rc = %d,"
2212                               "cannot start asynchronous commit\n", rc);
2213                 else
2214                         mdt_device_commit_async(&env, mdt);
2215                 lu_env_fini(&env);
2216         }
2217         RETURN(rc);
2218 }
2219
2220 int mdt_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2221                         void *data, int flag)
2222 {
2223         struct lustre_handle lockh;
2224         int               rc;
2225
2226         switch (flag) {
2227         case LDLM_CB_BLOCKING:
2228                 ldlm_lock2handle(lock, &lockh);
2229                 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
2230                 if (rc < 0) {
2231                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2232                         RETURN(rc);
2233                 }
2234                 break;
2235         case LDLM_CB_CANCELING:
2236                 LDLM_DEBUG(lock, "Revoke remote lock\n");
2237                 break;
2238         default:
2239                 LBUG();
2240         }
2241         RETURN(0);
2242 }
2243
2244 int mdt_remote_object_lock(struct mdt_thread_info *mti,
2245                            struct mdt_object *o, struct lustre_handle *lh,
2246                            ldlm_mode_t mode, __u64 ibits)
2247 {
2248         struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
2249         ldlm_policy_data_t *policy = &mti->mti_policy;
2250         int rc = 0;
2251         ENTRY;
2252
2253         LASSERT(mdt_object_remote(o));
2254
2255         LASSERT(ibits & MDS_INODELOCK_UPDATE);
2256
2257         memset(einfo, 0, sizeof(*einfo));
2258         einfo->ei_type = LDLM_IBITS;
2259         einfo->ei_mode = mode;
2260         einfo->ei_cb_bl = mdt_md_blocking_ast;
2261         einfo->ei_cb_cp = ldlm_completion_ast;
2262
2263         memset(policy, 0, sizeof(*policy));
2264         policy->l_inodebits.bits = ibits;
2265
2266         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2267                             policy);
2268         RETURN(rc);
2269 }
2270
2271 static int mdt_object_lock0(struct mdt_thread_info *info, struct mdt_object *o,
2272                             struct mdt_lock_handle *lh, __u64 ibits,
2273                             bool nonblock, int locality)
2274 {
2275         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2276         ldlm_policy_data_t *policy = &info->mti_policy;
2277         struct ldlm_res_id *res_id = &info->mti_res_id;
2278         __u64 dlmflags;
2279         int rc;
2280         ENTRY;
2281
2282         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2283         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2284         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2285         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2286
2287         if (mdt_object_remote(o)) {
2288                 if (locality == MDT_CROSS_LOCK) {
2289                         ibits &= ~(MDS_INODELOCK_UPDATE | MDS_INODELOCK_PERM |
2290                                    MDS_INODELOCK_LAYOUT);
2291                         ibits |= MDS_INODELOCK_LOOKUP;
2292                 } else {
2293                         LASSERTF(!(ibits &
2294                                  (MDS_INODELOCK_UPDATE | MDS_INODELOCK_PERM |
2295                                   MDS_INODELOCK_LAYOUT)),
2296                                 "%s: wrong bit "LPX64" for remote obj "DFID"\n",
2297                                 mdt_obd_name(info->mti_mdt), ibits,
2298                                 PFID(mdt_object_fid(o)));
2299                         LASSERT(ibits & MDS_INODELOCK_LOOKUP);
2300                 }
2301                 /* No PDO lock on remote object */
2302                 LASSERT(lh->mlh_type != MDT_PDO_LOCK);
2303         }
2304
2305         if (lh->mlh_type == MDT_PDO_LOCK) {
2306                 /* check for exists after object is locked */
2307                 if (mdt_object_exists(o) == 0) {
2308                         /* Non-existent object shouldn't have PDO lock */
2309                         RETURN(-ESTALE);
2310                 } else {
2311                         /* Non-dir object shouldn't have PDO lock */
2312                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2313                                 RETURN(-ENOTDIR);
2314                 }
2315         }
2316
2317         memset(policy, 0, sizeof(*policy));
2318         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2319
2320         dlmflags = LDLM_FL_ATOMIC_CB;
2321         if (nonblock)
2322                 dlmflags |= LDLM_FL_BLOCK_NOWAIT;
2323
2324         /*
2325          * Take PDO lock on whole directory and build correct @res_id for lock
2326          * on part of directory.
2327          */
2328         if (lh->mlh_pdo_hash != 0) {
2329                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2330                 mdt_lock_pdo_mode(info, o, lh);
2331                 if (lh->mlh_pdo_mode != LCK_NL) {
2332                         /*
2333                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2334                          * is never going to be sent to client and we do not
2335                          * want it slowed down due to possible cancels.
2336                          */
2337                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2338                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2339                                           policy, res_id, dlmflags,
2340                                           &info->mti_exp->exp_handle.h_cookie);
2341                         if (unlikely(rc))
2342                                 RETURN(rc);
2343                 }
2344
2345                 /*
2346                  * Finish res_id initializing by name hash marking part of
2347                  * directory which is taking modification.
2348                  */
2349                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2350         }
2351
2352         policy->l_inodebits.bits = ibits;
2353
2354         /*
2355          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2356          * going to be sent to client. If it is - mdt_intent_policy() path will
2357          * fix it up and turn FL_LOCAL flag off.
2358          */
2359         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2360                           res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2361                           &info->mti_exp->exp_handle.h_cookie);
2362         if (rc)
2363                 mdt_object_unlock(info, o, lh, 1);
2364         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2365                  lh->mlh_pdo_hash != 0 &&
2366                  (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX)) {
2367                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2368         }
2369
2370         RETURN(rc);
2371 }
2372
2373 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2374                     struct mdt_lock_handle *lh, __u64 ibits, int locality)
2375 {
2376         return mdt_object_lock0(info, o, lh, ibits, false, locality);
2377 }
2378
2379 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
2380                         struct mdt_lock_handle *lh, __u64 ibits, int locality)
2381 {
2382         struct mdt_lock_handle tmp = *lh;
2383         int rc;
2384
2385         rc = mdt_object_lock0(info, o, &tmp, ibits, true, locality);
2386         if (rc == 0)
2387                 *lh = tmp;
2388
2389         return rc == 0;
2390 }
2391
2392 /**
2393  * Save a lock within request object.
2394  *
2395  * Keep the lock referenced until whether client ACK or transaction
2396  * commit happens or release the lock immediately depending on input
2397  * parameters. If COS is ON, a write lock is converted to COS lock
2398  * before saving.
2399  *
2400  * \param info thead info object
2401  * \param h lock handle
2402  * \param mode lock mode
2403  * \param decref force immediate lock releasing
2404  */
2405 static
2406 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2407                    ldlm_mode_t mode, int decref)
2408 {
2409         ENTRY;
2410
2411         if (lustre_handle_is_used(h)) {
2412                 if (decref || !info->mti_has_trans ||
2413                     !(mode & (LCK_PW | LCK_EX))){
2414                         mdt_fid_unlock(h, mode);
2415                 } else {
2416                         struct mdt_device *mdt = info->mti_mdt;
2417                         struct ldlm_lock *lock = ldlm_handle2lock(h);
2418                         struct ptlrpc_request *req = mdt_info_req(info);
2419                         int no_ack = 0;
2420
2421                         LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2422                                  h->cookie);
2423                         /* there is no request if mdt_object_unlock() is called
2424                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
2425                         if (likely(req != NULL)) {
2426                                 CDEBUG(D_HA, "request = %p reply state = %p"
2427                                        " transno = "LPD64"\n", req,
2428                                        req->rq_reply_state, req->rq_transno);
2429                                 if (mdt_cos_is_enabled(mdt)) {
2430                                         no_ack = 1;
2431                                         ldlm_lock_downgrade(lock, LCK_COS);
2432                                         mode = LCK_COS;
2433                                 }
2434                                 ptlrpc_save_lock(req, h, mode, no_ack);
2435                         } else {
2436                                 ldlm_lock_decref(h, mode);
2437                         }
2438                         if (mdt_is_lock_sync(lock)) {
2439                                 CDEBUG(D_HA, "found sync-lock,"
2440                                        " async commit started\n");
2441                                 mdt_device_commit_async(info->mti_env,
2442                                                         mdt);
2443                         }
2444                         LDLM_LOCK_PUT(lock);
2445                 }
2446                 h->cookie = 0ull;
2447         }
2448
2449         EXIT;
2450 }
2451
2452 /**
2453  * Unlock mdt object.
2454  *
2455  * Immeditely release the regular lock and the PDO lock or save the
2456  * lock in reqeuest and keep them referenced until client ACK or
2457  * transaction commit.
2458  *
2459  * \param info thread info object
2460  * \param o mdt object
2461  * \param lh mdt lock handle referencing regular and PDO locks
2462  * \param decref force immediate lock releasing
2463  */
2464 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2465                        struct mdt_lock_handle *lh, int decref)
2466 {
2467         ENTRY;
2468
2469         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2470         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2471
2472         if (lustre_handle_is_used(&lh->mlh_rreg_lh))
2473                 ldlm_lock_decref(&lh->mlh_rreg_lh, lh->mlh_rreg_mode);
2474
2475         EXIT;
2476 }
2477
2478 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2479                                         const struct lu_fid *f,
2480                                         struct mdt_lock_handle *lh,
2481                                         __u64 ibits)
2482 {
2483         struct mdt_object *o;
2484
2485         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2486         if (!IS_ERR(o)) {
2487                 int rc;
2488
2489                 rc = mdt_object_lock(info, o, lh, ibits,
2490                                      MDT_LOCAL_LOCK);
2491                 if (rc != 0) {
2492                         mdt_object_put(info->mti_env, o);
2493                         o = ERR_PTR(rc);
2494                 }
2495         }
2496         return o;
2497 }
2498
2499 void mdt_object_unlock_put(struct mdt_thread_info * info,
2500                            struct mdt_object * o,
2501                            struct mdt_lock_handle *lh,
2502                            int decref)
2503 {
2504         mdt_object_unlock(info, o, lh, decref);
2505         mdt_object_put(info->mti_env, o);
2506 }
2507
2508 /*
2509  * Generic code handling requests that have struct mdt_body passed in:
2510  *
2511  *  - extract mdt_body from request and save it in @info, if present;
2512  *
2513  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
2514  *  @info;
2515  *
2516  *  - if HABEO_CORPUS flag is set for this request type check whether object
2517  *  actually exists on storage (lu_object_exists()).
2518  *
2519  */
2520 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2521 {
2522         const struct mdt_body    *body;
2523         struct mdt_object        *obj;
2524         const struct lu_env      *env;
2525         struct req_capsule       *pill;
2526         int                       rc;
2527         ENTRY;
2528
2529         env = info->mti_env;
2530         pill = info->mti_pill;
2531
2532         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2533         if (body == NULL)
2534                 RETURN(-EFAULT);
2535
2536         if (!(body->valid & OBD_MD_FLID))
2537                 RETURN(0);
2538
2539         if (!fid_is_sane(&body->fid1)) {
2540                 CERROR("Invalid fid: "DFID"\n", PFID(&body->fid1));
2541                 RETURN(-EINVAL);
2542         }
2543
2544         /*
2545          * Do not get size or any capa fields before we check that request
2546          * contains capa actually. There are some requests which do not, for
2547          * instance MDS_IS_SUBDIR.
2548          */
2549         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2550             req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
2551                 mdt_set_capainfo(info, 0, &body->fid1,
2552                                  req_capsule_client_get(pill, &RMF_CAPA1));
2553
2554         obj = mdt_object_find(env, info->mti_mdt, &body->fid1);
2555         if (!IS_ERR(obj)) {
2556                 if ((flags & HABEO_CORPUS) &&
2557                     !mdt_object_exists(obj)) {
2558                         mdt_object_put(env, obj);
2559                         /* for capability renew ENOENT will be handled in
2560                          * mdt_renew_capa */
2561                         if (body->valid & OBD_MD_FLOSSCAPA)
2562                                 rc = 0;
2563                         else
2564                                 rc = -ENOENT;
2565                 } else {
2566                         info->mti_object = obj;
2567                         rc = 0;
2568                 }
2569         } else
2570                 rc = PTR_ERR(obj);
2571
2572         RETURN(rc);
2573 }
2574
2575 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2576 {
2577         struct req_capsule *pill = info->mti_pill;
2578         int rc;
2579         ENTRY;
2580
2581         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2582                 rc = mdt_body_unpack(info, flags);
2583         else
2584                 rc = 0;
2585
2586         if (rc == 0 && (flags & HABEO_REFERO)) {
2587                 /* Pack reply. */
2588                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2589                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2590                                              info->mti_body->eadatasize);
2591                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2592                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
2593                                              RCL_SERVER, 0);
2594
2595                 rc = req_capsule_server_pack(pill);
2596         }
2597         RETURN(rc);
2598 }
2599
2600 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2601 {
2602         struct md_device *next = m->mdt_child;
2603
2604         return next->md_ops->mdo_init_capa_ctxt(env, next,
2605                                                 m->mdt_lut.lut_mds_capa,
2606                                                 m->mdt_capa_timeout,
2607                                                 m->mdt_capa_alg,
2608                                                 m->mdt_capa_keys);
2609 }
2610
2611 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2612 {
2613         lh->mlh_type = MDT_NUL_LOCK;
2614         lh->mlh_reg_lh.cookie = 0ull;
2615         lh->mlh_reg_mode = LCK_MINMODE;
2616         lh->mlh_pdo_lh.cookie = 0ull;
2617         lh->mlh_pdo_mode = LCK_MINMODE;
2618         lh->mlh_rreg_lh.cookie = 0ull;
2619         lh->mlh_rreg_mode = LCK_MINMODE;
2620 }
2621
2622 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2623 {
2624         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2625         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2626 }
2627
2628 /*
2629  * Initialize fields of struct mdt_thread_info. Other fields are left in
2630  * uninitialized state, because it's too expensive to zero out whole
2631  * mdt_thread_info (> 1K) on each request arrival.
2632  */
2633 void mdt_thread_info_init(struct ptlrpc_request *req,
2634                           struct mdt_thread_info *info)
2635 {
2636         int i;
2637
2638         info->mti_pill = &req->rq_pill;
2639
2640         /* lock handle */
2641         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2642                 mdt_lock_handle_init(&info->mti_lh[i]);
2643
2644         /* mdt device: it can be NULL while CONNECT */
2645         if (req->rq_export) {
2646                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2647                 info->mti_exp = req->rq_export;
2648         } else
2649                 info->mti_mdt = NULL;
2650         info->mti_env = req->rq_svc_thread->t_env;
2651         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2652
2653         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2654         info->mti_big_buf = LU_BUF_NULL;
2655         info->mti_body = NULL;
2656         info->mti_object = NULL;
2657         info->mti_dlm_req = NULL;
2658         info->mti_has_trans = 0;
2659         info->mti_cross_ref = 0;
2660         info->mti_opdata = 0;
2661         info->mti_big_lmm_used = 0;
2662
2663         /* To not check for split by default. */
2664         info->mti_spec.no_create = 0;
2665         info->mti_spec.sp_rm_entry = 0;
2666 }
2667
2668 void mdt_thread_info_fini(struct mdt_thread_info *info)
2669 {
2670         int i;
2671
2672         if (info->mti_object != NULL) {
2673                 mdt_object_put(info->mti_env, info->mti_object);
2674                 info->mti_object = NULL;
2675         }
2676
2677         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2678                 mdt_lock_handle_fini(&info->mti_lh[i]);
2679         info->mti_env = NULL;
2680         info->mti_pill = NULL;
2681         info->mti_exp = NULL;
2682
2683         if (unlikely(info->mti_big_buf.lb_buf != NULL))
2684                 lu_buf_free(&info->mti_big_buf);
2685 }
2686
2687 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
2688 {
2689         struct mdt_thread_info  *mti;
2690         struct lustre_capa      *lc;
2691
2692         mti = mdt_th_info(tsi->tsi_env);
2693         LASSERT(mti != NULL);
2694
2695         mdt_thread_info_init(tgt_ses_req(tsi), mti);
2696         if (tsi->tsi_corpus != NULL) {
2697                 struct req_capsule *pill = tsi->tsi_pill;
2698
2699                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
2700                 lu_object_get(tsi->tsi_corpus);
2701
2702                 /*
2703                  * XXX: must be part of tgt_mdt_body_unpack but moved here
2704                  * due to mdt_set_capainfo().
2705                  */
2706                 if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2707                     req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT) > 0) {
2708                         lc = req_capsule_client_get(pill, &RMF_CAPA1);
2709                         mdt_set_capainfo(mti, 0, &tsi->tsi_mdt_body->fid1, lc);
2710                 }
2711         }
2712         mti->mti_body = tsi->tsi_mdt_body;
2713         mti->mti_dlm_req = tsi->tsi_dlm_req;
2714
2715         return mti;
2716 }
2717
2718 int mdt_tgt_connect(struct tgt_session_info *tsi)
2719 {
2720         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2721         int                      rc;
2722
2723         ENTRY;
2724
2725         rc = tgt_connect(tsi);
2726         if (rc != 0)
2727                 RETURN(rc);
2728
2729         rc = mdt_init_idmap(tsi);
2730         if (rc != 0)
2731                 GOTO(err, rc);
2732         RETURN(0);
2733 err:
2734         obd_disconnect(class_export_get(req->rq_export));
2735         return rc;
2736 }
2737
2738 enum mdt_it_code {
2739         MDT_IT_OPEN,
2740         MDT_IT_OCREAT,
2741         MDT_IT_CREATE,
2742         MDT_IT_GETATTR,
2743         MDT_IT_READDIR,
2744         MDT_IT_LOOKUP,
2745         MDT_IT_UNLINK,
2746         MDT_IT_TRUNC,
2747         MDT_IT_GETXATTR,
2748         MDT_IT_LAYOUT,
2749         MDT_IT_QUOTA,
2750         MDT_IT_NR
2751 };
2752
2753 static int mdt_intent_getattr(enum mdt_it_code opcode,
2754                               struct mdt_thread_info *info,
2755                               struct ldlm_lock **,
2756                               __u64);
2757
2758 static int mdt_intent_getxattr(enum mdt_it_code opcode,
2759                                 struct mdt_thread_info *info,
2760                                 struct ldlm_lock **lockp,
2761                                 __u64 flags);
2762
2763 static int mdt_intent_layout(enum mdt_it_code opcode,
2764                              struct mdt_thread_info *info,
2765                              struct ldlm_lock **,
2766                              __u64);
2767 static int mdt_intent_reint(enum mdt_it_code opcode,
2768                             struct mdt_thread_info *info,
2769                             struct ldlm_lock **,
2770                             __u64);
2771
2772 static struct mdt_it_flavor {
2773         const struct req_format *it_fmt;
2774         __u32                    it_flags;
2775         int                    (*it_act)(enum mdt_it_code ,
2776                                          struct mdt_thread_info *,
2777                                          struct ldlm_lock **,
2778                                          __u64);
2779         long                     it_reint;
2780 } mdt_it_flavor[] = {
2781         [MDT_IT_OPEN]     = {
2782                 .it_fmt   = &RQF_LDLM_INTENT,
2783                 /*.it_flags = HABEO_REFERO,*/
2784                 .it_flags = 0,
2785                 .it_act   = mdt_intent_reint,
2786                 .it_reint = REINT_OPEN
2787         },
2788         [MDT_IT_OCREAT]   = {
2789                 .it_fmt   = &RQF_LDLM_INTENT,
2790                 /*
2791                  * OCREAT is not a MUTABOR request as if the file
2792                  * already exists.
2793                  * We do the extra check of OBD_CONNECT_RDONLY in
2794                  * mdt_reint_open() when we really need to create
2795                  * the object.
2796                  */
2797                 .it_flags = 0,
2798                 .it_act   = mdt_intent_reint,
2799                 .it_reint = REINT_OPEN
2800         },
2801         [MDT_IT_CREATE]   = {
2802                 .it_fmt   = &RQF_LDLM_INTENT,
2803                 .it_flags = MUTABOR,
2804                 .it_act   = mdt_intent_reint,
2805                 .it_reint = REINT_CREATE
2806         },
2807         [MDT_IT_GETATTR]  = {
2808                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2809                 .it_flags = HABEO_REFERO,
2810                 .it_act   = mdt_intent_getattr
2811         },
2812         [MDT_IT_READDIR]  = {
2813                 .it_fmt   = NULL,
2814                 .it_flags = 0,
2815                 .it_act   = NULL
2816         },
2817         [MDT_IT_LOOKUP]   = {
2818                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2819                 .it_flags = HABEO_REFERO,
2820                 .it_act   = mdt_intent_getattr
2821         },
2822         [MDT_IT_UNLINK]   = {
2823                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
2824                 .it_flags = MUTABOR,
2825                 .it_act   = NULL,
2826                 .it_reint = REINT_UNLINK
2827         },
2828         [MDT_IT_TRUNC]    = {
2829                 .it_fmt   = NULL,
2830                 .it_flags = MUTABOR,
2831                 .it_act   = NULL
2832         },
2833         [MDT_IT_GETXATTR] = {
2834                 .it_fmt   = &RQF_LDLM_INTENT_GETXATTR,
2835                 .it_flags = HABEO_CORPUS,
2836                 .it_act   = mdt_intent_getxattr
2837         },
2838         [MDT_IT_LAYOUT] = {
2839                 .it_fmt   = &RQF_LDLM_INTENT_LAYOUT,
2840                 .it_flags = 0,
2841                 .it_act   = mdt_intent_layout
2842         }
2843 };
2844
2845 int mdt_intent_lock_replace(struct mdt_thread_info *info,
2846                             struct ldlm_lock **lockp,
2847                             struct ldlm_lock *new_lock,
2848                             struct mdt_lock_handle *lh,
2849                             __u64 flags)
2850 {
2851         struct ptlrpc_request  *req = mdt_info_req(info);
2852         struct ldlm_lock       *lock = *lockp;
2853
2854         /*
2855          * Get new lock only for cases when possible resent did not find any
2856          * lock.
2857          */
2858         if (new_lock == NULL)
2859                 new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
2860
2861         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
2862                 lh->mlh_reg_lh.cookie = 0;
2863                 RETURN(0);
2864         }
2865
2866         LASSERTF(new_lock != NULL,
2867                  "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
2868
2869         /*
2870          * If we've already given this lock to a client once, then we should
2871          * have no readers or writers.  Otherwise, we should have one reader
2872          * _or_ writer ref (which will be zeroed below) before returning the
2873          * lock to a client.
2874          */
2875         if (new_lock->l_export == req->rq_export) {
2876                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
2877         } else {
2878                 LASSERT(new_lock->l_export == NULL);
2879                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
2880         }
2881
2882         *lockp = new_lock;
2883
2884         if (new_lock->l_export == req->rq_export) {
2885                 /*
2886                  * Already gave this to the client, which means that we
2887                  * reconstructed a reply.
2888                  */
2889                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
2890                         MSG_RESENT);
2891                 lh->mlh_reg_lh.cookie = 0;
2892                 RETURN(ELDLM_LOCK_REPLACED);
2893         }
2894
2895         /*
2896          * Fixup the lock to be given to the client.
2897          */
2898         lock_res_and_lock(new_lock);
2899         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
2900          * possible blocking AST. */
2901         while (new_lock->l_readers > 0) {
2902                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
2903                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
2904                 new_lock->l_readers--;
2905         }
2906         while (new_lock->l_writers > 0) {
2907                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
2908                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
2909                 new_lock->l_writers--;
2910         }
2911
2912         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
2913         new_lock->l_blocking_ast = lock->l_blocking_ast;
2914         new_lock->l_completion_ast = lock->l_completion_ast;
2915         new_lock->l_remote_handle = lock->l_remote_handle;
2916         new_lock->l_flags &= ~LDLM_FL_LOCAL;
2917
2918         unlock_res_and_lock(new_lock);
2919
2920         cfs_hash_add(new_lock->l_export->exp_lock_hash,
2921                      &new_lock->l_remote_handle,
2922                      &new_lock->l_exp_hash);
2923
2924         LDLM_LOCK_RELEASE(new_lock);
2925         lh->mlh_reg_lh.cookie = 0;
2926
2927         RETURN(ELDLM_LOCK_REPLACED);
2928 }
2929
2930 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
2931                                     struct ldlm_lock *new_lock,
2932                                     struct ldlm_lock **old_lock,
2933                                     struct mdt_lock_handle *lh)
2934 {
2935         struct ptlrpc_request  *req = mdt_info_req(info);
2936         struct obd_export      *exp = req->rq_export;
2937         struct lustre_handle    remote_hdl;
2938         struct ldlm_request    *dlmreq;
2939         struct ldlm_lock       *lock;
2940
2941         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
2942                 return;
2943
2944         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
2945         remote_hdl = dlmreq->lock_handle[0];
2946
2947         /* In the function below, .hs_keycmp resolves to
2948          * ldlm_export_lock_keycmp() */
2949         /* coverity[overrun-buffer-val] */
2950         lock = cfs_hash_lookup(exp->exp_lock_hash, &remote_hdl);
2951         if (lock) {
2952                 if (lock != new_lock) {
2953                         lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie;
2954                         lh->mlh_reg_mode = lock->l_granted_mode;
2955
2956                         LDLM_DEBUG(lock, "Restoring lock cookie");
2957                         DEBUG_REQ(D_DLMTRACE, req,
2958                                   "restoring lock cookie "LPX64,
2959                                   lh->mlh_reg_lh.cookie);
2960                         if (old_lock)
2961                                 *old_lock = LDLM_LOCK_GET(lock);
2962                         cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
2963                         return;
2964                 }
2965
2966                 cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash);
2967         }
2968
2969         /*
2970          * If the xid matches, then we know this is a resent request, and allow
2971          * it. (It's probably an OPEN, for which we don't send a lock.
2972          */
2973         if (req_xid_is_last(req))
2974                 return;
2975
2976         /*
2977          * This remote handle isn't enqueued, so we never received or processed
2978          * this request.  Clear MSG_RESENT, because it can be handled like any
2979          * normal request now.
2980          */
2981         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
2982
2983         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
2984                   remote_hdl.cookie);
2985 }
2986
2987 static int mdt_intent_getxattr(enum mdt_it_code opcode,
2988                                 struct mdt_thread_info *info,
2989                                 struct ldlm_lock **lockp,
2990                                 __u64 flags)
2991 {
2992         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
2993         struct ldlm_reply      *ldlm_rep = NULL;
2994         int rc, grc;
2995
2996         /*
2997          * Initialize lhc->mlh_reg_lh either from a previously granted lock
2998          * (for the resend case) or a new lock. Below we will use it to
2999          * replace the original lock.
3000          */
3001         mdt_intent_fixup_resent(info, *lockp, NULL, lhc);
3002         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3003                 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3004                 rc = mdt_object_lock(info, info->mti_object, lhc,
3005                                         MDS_INODELOCK_XATTR,
3006                                         MDT_LOCAL_LOCK);
3007                 if (rc)
3008                         return rc;
3009         }
3010
3011         grc = mdt_getxattr(info);
3012
3013         rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3014
3015         if (mdt_info_req(info)->rq_repmsg != NULL)
3016                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3017         if (ldlm_rep == NULL)
3018                 RETURN(err_serious(-EFAULT));
3019
3020         ldlm_rep->lock_policy_res2 = grc;
3021
3022         return rc;
3023 }
3024
3025 static int mdt_intent_getattr(enum mdt_it_code opcode,
3026                               struct mdt_thread_info *info,
3027                               struct ldlm_lock **lockp,
3028                               __u64 flags)
3029 {
3030         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3031         struct ldlm_lock       *new_lock = NULL;
3032         __u64                   child_bits;
3033         struct ldlm_reply      *ldlm_rep;
3034         struct ptlrpc_request  *req;
3035         struct mdt_body        *reqbody;
3036         struct mdt_body        *repbody;
3037         int                     rc, rc2;
3038         ENTRY;
3039
3040         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3041         LASSERT(reqbody);
3042
3043         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3044         LASSERT(repbody);
3045
3046         info->mti_cross_ref = !!(reqbody->valid & OBD_MD_FLCROSSREF);
3047         repbody->eadatasize = 0;
3048         repbody->aclsize = 0;
3049
3050         switch (opcode) {
3051         case MDT_IT_LOOKUP:
3052                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3053                 break;
3054         case MDT_IT_GETATTR:
3055                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3056                              MDS_INODELOCK_PERM;
3057                 break;
3058         default:
3059                 CERROR("Unsupported intent (%d)\n", opcode);
3060                 GOTO(out_shrink, rc = -EINVAL);
3061         }
3062
3063         rc = mdt_init_ucred(info, reqbody);
3064         if (rc)
3065                 GOTO(out_shrink, rc);
3066
3067         req = info->mti_pill->rc_req;
3068         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3069         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3070
3071         /* Get lock from request for possible resent case. */
3072         mdt_intent_fixup_resent(info, *lockp, &new_lock, lhc);
3073
3074         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3075         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3076
3077         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3078                 ldlm_rep->lock_policy_res2 = 0;
3079         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3080             ldlm_rep->lock_policy_res2) {
3081                 lhc->mlh_reg_lh.cookie = 0ull;
3082                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3083         }
3084
3085         rc = mdt_intent_lock_replace(info, lockp, new_lock, lhc, flags);
3086         EXIT;
3087 out_ucred:
3088         mdt_exit_ucred(info);
3089 out_shrink:
3090         mdt_client_compatibility(info);
3091         rc2 = mdt_fix_reply(info);
3092         if (rc == 0)
3093                 rc = rc2;
3094         return rc;
3095 }
3096
3097 static int mdt_intent_layout(enum mdt_it_code opcode,
3098                              struct mdt_thread_info *info,
3099                              struct ldlm_lock **lockp,
3100                              __u64 flags)
3101 {
3102         struct layout_intent *layout;
3103         struct lu_fid *fid;
3104         struct mdt_object *obj = NULL;
3105         struct md_object *child = NULL;
3106         int rc;
3107         ENTRY;
3108
3109         if (opcode != MDT_IT_LAYOUT) {
3110                 CERROR("%s: Unknown intent (%d)\n", mdt_obd_name(info->mti_mdt),
3111                         opcode);
3112                 RETURN(-EINVAL);
3113         }
3114
3115         fid = &info->mti_tmp_fid2;
3116         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3117
3118         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3119         if (IS_ERR(obj))
3120                 RETURN(PTR_ERR(obj));
3121
3122         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3123                 child = mdt_object_child(obj);
3124
3125                 /* get the length of lsm */
3126                 rc = mo_xattr_get(info->mti_env, child, &LU_BUF_NULL,
3127                                   XATTR_NAME_LOV);
3128
3129                 if (rc > info->mti_mdt->mdt_max_mdsize)
3130                         info->mti_mdt->mdt_max_mdsize = rc;
3131         }
3132
3133         mdt_object_put(info->mti_env, obj);
3134
3135         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3136         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
3137                         ldlm_lvbo_size(*lockp));
3138         rc = req_capsule_server_pack(info->mti_pill);
3139         if (rc != 0)
3140                 RETURN(-EINVAL);
3141
3142         layout = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3143         LASSERT(layout != NULL);
3144         if (layout->li_opc == LAYOUT_INTENT_ACCESS)
3145                 /* return to normal ldlm handling */
3146                 RETURN(0);
3147
3148         CERROR("%s: Unsupported layout intent (%d)\n",
3149                 mdt_obd_name(info->mti_mdt), layout->li_opc);
3150         RETURN(-EINVAL);
3151 }
3152
3153 static int mdt_intent_reint(enum mdt_it_code opcode,
3154                             struct mdt_thread_info *info,
3155                             struct ldlm_lock **lockp,
3156                             __u64 flags)
3157 {
3158         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3159         struct ldlm_reply      *rep = NULL;
3160         long                    opc;
3161         int                     rc;
3162
3163         static const struct req_format *intent_fmts[REINT_MAX] = {
3164                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3165                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3166         };
3167
3168         ENTRY;
3169
3170         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
3171         if (opc < 0)
3172                 RETURN(opc);
3173
3174         if (mdt_it_flavor[opcode].it_reint != opc) {
3175                 CERROR("Reint code %ld doesn't match intent: %d\n",
3176                        opc, opcode);
3177                 RETURN(err_serious(-EPROTO));
3178         }
3179
3180         /* Get lock from request for possible resent case. */
3181         mdt_intent_fixup_resent(info, *lockp, NULL, lhc);
3182
3183         rc = mdt_reint_internal(info, lhc, opc);
3184
3185         /* Check whether the reply has been packed successfully. */
3186         if (mdt_info_req(info)->rq_repmsg != NULL)
3187                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3188         if (rep == NULL)
3189                 RETURN(err_serious(-EFAULT));
3190
3191         /* MDC expects this in any case */
3192         if (rc != 0)
3193                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3194
3195         /* the open lock or the lock for cross-ref object should be
3196          * returned to the client */
3197         if (rc == -EREMOTE || mdt_get_disposition(rep, DISP_OPEN_LOCK)) {
3198                 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
3199                 rep->lock_policy_res2 = 0;
3200                 rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3201                 RETURN(rc);
3202         }
3203
3204         rep->lock_policy_res2 = clear_serious(rc);
3205
3206         if (rep->lock_policy_res2 == -ENOENT &&
3207             mdt_get_disposition(rep, DISP_LOOKUP_NEG))
3208                 rep->lock_policy_res2 = 0;
3209
3210         if (rc == -ENOTCONN || rc == -ENODEV ||
3211             rc == -EOVERFLOW) { /**< if VBR failure then return error */
3212                 /*
3213                  * If it is the disconnect error (ENODEV & ENOCONN), the error
3214                  * will be returned by rq_status, and client at ptlrpc layer
3215                  * will detect this, then disconnect, reconnect the import
3216                  * immediately, instead of impacting the following the rpc.
3217                  */
3218                 lhc->mlh_reg_lh.cookie = 0ull;
3219                 RETURN(rc);
3220         } else {
3221                 /*
3222                  * For other cases, the error will be returned by intent.
3223                  * and client will retrieve the result from intent.
3224                  */
3225                  /*
3226                   * FIXME: when open lock is finished, that should be
3227                   * checked here.
3228                   */
3229                 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3230                         LASSERTF(rc == 0, "Error occurred but lock handle "
3231                                  "is still in use, rc = %d\n", rc);
3232                         rep->lock_policy_res2 = 0;
3233                         rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
3234                         RETURN(rc);
3235                 } else {
3236                         lhc->mlh_reg_lh.cookie = 0ull;
3237                         RETURN(ELDLM_LOCK_ABORTED);
3238                 }
3239         }
3240 }
3241
3242 static int mdt_intent_code(long itcode)
3243 {
3244         int rc;
3245
3246         switch(itcode) {
3247         case IT_OPEN:
3248                 rc = MDT_IT_OPEN;
3249                 break;
3250         case IT_OPEN|IT_CREAT:
3251                 rc = MDT_IT_OCREAT;
3252                 break;
3253         case IT_CREAT:
3254                 rc = MDT_IT_CREATE;
3255                 break;
3256         case IT_READDIR:
3257                 rc = MDT_IT_READDIR;
3258                 break;
3259         case IT_GETATTR:
3260                 rc = MDT_IT_GETATTR;
3261                 break;
3262         case IT_LOOKUP:
3263                 rc = MDT_IT_LOOKUP;
3264                 break;
3265         case IT_UNLINK:
3266                 rc = MDT_IT_UNLINK;
3267                 break;
3268         case IT_TRUNC:
3269                 rc = MDT_IT_TRUNC;
3270                 break;
3271         case IT_GETXATTR:
3272                 rc = MDT_IT_GETXATTR;
3273                 break;
3274         case IT_LAYOUT:
3275                 rc = MDT_IT_LAYOUT;
3276                 break;
3277         case IT_QUOTA_DQACQ:
3278         case IT_QUOTA_CONN:
3279                 rc = MDT_IT_QUOTA;
3280                 break;
3281         default:
3282                 CERROR("Unknown intent opcode: %ld\n", itcode);
3283                 rc = -EINVAL;
3284                 break;
3285         }
3286         return rc;
3287 }
3288
3289 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
3290                           struct ldlm_lock **lockp, __u64 flags)
3291 {
3292         struct req_capsule   *pill;
3293         struct mdt_it_flavor *flv;
3294         int opc;
3295         int rc;
3296         ENTRY;
3297
3298         opc = mdt_intent_code(itopc);
3299         if (opc < 0)
3300                 RETURN(-EINVAL);
3301
3302         pill = info->mti_pill;
3303
3304         if (opc == MDT_IT_QUOTA) {
3305                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
3306
3307                 if (qmt == NULL)
3308                         RETURN(-EOPNOTSUPP);
3309
3310                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
3311                 /* pass the request to quota master */
3312                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
3313                                                  mdt_info_req(info), lockp,
3314                                                  flags);
3315                 RETURN(rc);
3316         }
3317
3318         flv  = &mdt_it_flavor[opc];
3319         if (flv->it_fmt != NULL)
3320                 req_capsule_extend(pill, flv->it_fmt);
3321
3322         rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3323         if (rc == 0) {
3324                 struct ptlrpc_request *req = mdt_info_req(info);
3325                 if (flv->it_flags & MUTABOR &&
3326                     exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
3327                         RETURN(-EROFS);
3328         }
3329         if (rc == 0 && flv->it_act != NULL) {
3330                 struct ldlm_reply *rep;
3331
3332                 /* execute policy */
3333                 rc = flv->it_act(opc, info, lockp, flags);
3334
3335                 /* Check whether the reply has been packed successfully. */
3336                 if (mdt_info_req(info)->rq_repmsg != NULL) {
3337                         rep = req_capsule_server_get(info->mti_pill,
3338                                                      &RMF_DLM_REP);
3339                         rep->lock_policy_res2 =
3340                                 ptlrpc_status_hton(rep->lock_policy_res2);
3341                 }
3342         } else {
3343                 rc = -EPROTO;
3344         }
3345         RETURN(rc);
3346 }
3347
3348 static int mdt_intent_policy(struct ldlm_namespace *ns,
3349                              struct ldlm_lock **lockp, void *req_cookie,
3350                              ldlm_mode_t mode, __u64 flags, void *data)
3351 {
3352         struct tgt_session_info *tsi;
3353         struct mdt_thread_info  *info;
3354         struct ptlrpc_request   *req  =  req_cookie;
3355         struct ldlm_intent      *it;
3356         struct req_capsule      *pill;
3357         int rc;
3358
3359         ENTRY;
3360
3361         LASSERT(req != NULL);
3362
3363         tsi = tgt_ses_info(req->rq_svc_thread->t_env);
3364
3365         info = tsi2mdt_info(tsi);
3366         LASSERT(info != NULL);
3367         pill = info->mti_pill;
3368         LASSERT(pill->rc_req == req);
3369
3370         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3371                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
3372                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3373                 if (it != NULL) {
3374                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
3375                         if (rc == 0)
3376                                 rc = ELDLM_OK;
3377
3378                         /* Lock without inodebits makes no sense and will oops
3379                          * later in ldlm. Let's check it now to see if we have
3380                          * ibits corrupted somewhere in mdt_intent_opc().
3381                          * The case for client miss to set ibits has been
3382                          * processed by others. */
3383                         LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
3384                                         lr_type == LDLM_IBITS,
3385                                      info->mti_dlm_req->lock_desc.\
3386                                         l_policy_data.l_inodebits.bits != 0));
3387                 } else
3388                         rc = err_serious(-EFAULT);
3389         } else {
3390                 /* No intent was provided */
3391                 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3392                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
3393                 rc = req_capsule_server_pack(pill);
3394                 if (rc)
3395                         rc = err_serious(rc);
3396         }
3397         mdt_thread_info_fini(info);
3398         RETURN(rc);
3399 }
3400
3401 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
3402 {
3403         struct seq_server_site  *ss = mdt_seq_site(mdt);
3404
3405         if (ss->ss_node_id == 0)
3406                 return;
3407
3408         if (ss->ss_client_seq != NULL) {
3409                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3410                 ss->ss_client_seq->lcs_exp = NULL;
3411         }
3412
3413         if (ss->ss_server_fld != NULL) {
3414                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
3415                 ss->ss_server_fld->lsf_control_exp = NULL;
3416         }
3417 }
3418
3419 static void mdt_seq_fini_cli(struct mdt_device *mdt)
3420 {
3421         struct seq_server_site *ss = mdt_seq_site(mdt);
3422
3423         if (ss == NULL)
3424                 return;
3425
3426         if (ss->ss_server_seq == NULL)
3427                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
3428
3429         return;
3430 }
3431
3432 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
3433 {
3434         mdt_seq_fini_cli(mdt);
3435         mdt_deregister_seq_exp(mdt);
3436
3437         return seq_site_fini(env, mdt_seq_site(mdt));
3438 }
3439
3440 /**
3441  * It will retrieve its FLDB entries from MDT0, and it only happens
3442  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
3443  * and it needs to refresh FLDB from the MDT0.
3444  **/
3445 static int mdt_register_lwp_callback(void *data)
3446 {
3447         struct lu_env           env;
3448         struct mdt_device       *mdt = data;
3449         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
3450         int                     rc;
3451         ENTRY;
3452
3453         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
3454
3455         if (!likely(fld->lsf_new))
3456                 RETURN(0);
3457
3458         rc = lu_env_init(&env, LCT_MD_THREAD);
3459         if (rc) {
3460                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
3461                 RETURN(rc);
3462         }
3463
3464         rc = fld_update_from_controller(&env, fld);
3465         if (rc != 0) {
3466                 CERROR("%s: cannot update controller: rc = %d\n",
3467                        mdt_obd_name(mdt), rc);
3468                 GOTO(out, rc);
3469         }
3470 out:
3471         lu_env_fini(&env);
3472         RETURN(rc);
3473 }
3474
3475 static int mdt_register_seq_exp(struct mdt_device *mdt)
3476 {
3477         struct seq_server_site  *ss = mdt_seq_site(mdt);
3478         char                    *lwp_name = NULL;
3479         int                     rc;
3480
3481         if (ss->ss_node_id == 0)
3482                 return 0;
3483
3484         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
3485         if (lwp_name == NULL)
3486                 GOTO(out_free, rc = -ENOMEM);
3487
3488         rc = tgt_name2lwpname(mdt_obd_name(mdt), lwp_name);
3489         if (rc != 0)
3490                 GOTO(out_free, rc);
3491
3492         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
3493                                       NULL, NULL);
3494         if (rc != 0)
3495                 GOTO(out_free, rc);
3496
3497         rc = lustre_register_lwp_item(lwp_name,
3498                                       &ss->ss_server_fld->lsf_control_exp,
3499                                       mdt_register_lwp_callback, mdt);
3500         if (rc != 0) {
3501                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3502                 ss->ss_client_seq->lcs_exp = NULL;
3503                 GOTO(out_free, rc);
3504         }
3505 out_free:
3506         if (lwp_name != NULL)
3507                 OBD_FREE(lwp_name, MAX_OBD_NAME);
3508
3509         return rc;
3510 }
3511
3512 /*
3513  * Init client sequence manager which is used by local MDS to talk to sequence
3514  * controller on remote node.
3515  */
3516 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
3517 {
3518         struct seq_server_site  *ss = mdt_seq_site(mdt);
3519         int                     rc;
3520         char                    *prefix;
3521         ENTRY;
3522
3523         /* check if this is adding the first MDC and controller is not yet
3524          * initialized. */
3525         OBD_ALLOC_PTR(ss->ss_client_seq);
3526         if (ss->ss_client_seq == NULL)
3527                 RETURN(-ENOMEM);
3528
3529         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3530         if (prefix == NULL) {
3531                 OBD_FREE_PTR(ss->ss_client_seq);
3532                 ss->ss_client_seq = NULL;
3533                 RETURN(-ENOMEM);
3534         }
3535
3536         /* Note: seq_client_fini will be called in seq_site_fini */
3537         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
3538         rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
3539                              prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
3540                                                             NULL);
3541         OBD_FREE(prefix, MAX_OBD_NAME + 5);
3542         if (rc != 0) {
3543                 OBD_FREE_PTR(ss->ss_client_seq);
3544                 ss->ss_client_seq = NULL;
3545                 RETURN(rc);
3546         }
3547
3548         rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
3549
3550         RETURN(rc);
3551 }
3552
3553 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
3554 {
3555         struct seq_server_site  *ss;
3556         int                     rc;
3557         ENTRY;
3558
3559         ss = mdt_seq_site(mdt);
3560         /* init sequence controller server(MDT0) */
3561         if (ss->ss_node_id == 0) {
3562                 OBD_ALLOC_PTR(ss->ss_control_seq);
3563                 if (ss->ss_control_seq == NULL)
3564                         RETURN(-ENOMEM);
3565
3566                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
3567                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
3568                                      ss);
3569                 if (rc)
3570                         GOTO(out_seq_fini, rc);
3571         }
3572
3573         /* Init normal sequence server */
3574         OBD_ALLOC_PTR(ss->ss_server_seq);
3575         if (ss->ss_server_seq == NULL)
3576                 GOTO(out_seq_fini, rc = -ENOMEM);
3577
3578         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
3579                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
3580         if (rc)
3581                 GOTO(out_seq_fini, rc);
3582
3583         /* init seq client for seq server to talk to seq controller(MDT0) */
3584         rc = mdt_seq_init_cli(env, mdt);
3585         if (rc != 0)
3586                 GOTO(out_seq_fini, rc);
3587
3588         if (ss->ss_node_id != 0)
3589                 /* register controler export through lwp */
3590                 rc = mdt_register_seq_exp(mdt);
3591
3592         EXIT;
3593 out_seq_fini:
3594         if (rc)
3595                 mdt_seq_fini(env, mdt);
3596
3597         return rc;
3598 }
3599
3600 /*
3601  * FLD wrappers
3602  */
3603 static int mdt_fld_fini(const struct lu_env *env,
3604                         struct mdt_device *m)
3605 {
3606         struct seq_server_site *ss = mdt_seq_site(m);
3607         ENTRY;
3608
3609         if (ss && ss->ss_server_fld) {
3610                 fld_server_fini(env, ss->ss_server_fld);
3611                 OBD_FREE_PTR(ss->ss_server_fld);
3612                 ss->ss_server_fld = NULL;
3613         }
3614
3615         RETURN(0);
3616 }
3617
3618 static int mdt_fld_init(const struct lu_env *env,
3619                         const char *uuid,
3620                         struct mdt_device *m)
3621 {
3622         struct seq_server_site *ss;
3623         int rc;
3624         ENTRY;
3625
3626         ss = mdt_seq_site(m);
3627
3628         OBD_ALLOC_PTR(ss->ss_server_fld);
3629         if (ss->ss_server_fld == NULL)
3630                 RETURN(rc = -ENOMEM);
3631
3632         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
3633                              LU_SEQ_RANGE_MDT);
3634         if (rc) {
3635                 OBD_FREE_PTR(ss->ss_server_fld);
3636                 ss->ss_server_fld = NULL;
3637                 RETURN(rc);
3638         }
3639
3640         RETURN(0);
3641 }
3642
3643 static void mdt_stack_pre_fini(const struct lu_env *env,
3644                            struct mdt_device *m, struct lu_device *top)
3645 {
3646         struct obd_device       *obd;
3647         struct lustre_cfg_bufs  *bufs;
3648         struct lustre_cfg       *lcfg;
3649         struct mdt_thread_info  *info;
3650         ENTRY;
3651
3652         LASSERT(top);
3653
3654         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3655         LASSERT(info != NULL);
3656
3657         bufs = &info->mti_u.bufs;
3658
3659         LASSERT(m->mdt_child_exp);
3660         LASSERT(m->mdt_child_exp->exp_obd);
3661         obd = m->mdt_child_exp->exp_obd;
3662
3663         /* process cleanup, pass mdt obd name to get obd umount flags */
3664         /* XXX: this is needed because all layers are referenced by
3665          * objects (some of them are pinned by osd, for example *
3666          * the proper solution should be a model where object used
3667          * by osd only doesn't have mdt/mdd slices -bzzz */
3668         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
3669         lustre_cfg_bufs_set_string(bufs, 1, NULL);
3670         lcfg = lustre_cfg_new(LCFG_PRE_CLEANUP, bufs);
3671         if (!lcfg) {
3672                 CERROR("%s:Cannot alloc lcfg!\n", mdt_obd_name(m));
3673                 return;
3674         }
3675         top->ld_ops->ldo_process_config(env, top, lcfg);
3676         lustre_cfg_free(lcfg);
3677         EXIT;
3678 }
3679
3680 static void mdt_stack_fini(const struct lu_env *env,
3681                            struct mdt_device *m, struct lu_device *top)
3682 {
3683         struct obd_device       *obd = mdt2obd_dev(m);
3684         struct lustre_cfg_bufs  *bufs;
3685         struct lustre_cfg       *lcfg;
3686         struct mdt_thread_info  *info;
3687         char                     flags[3] = "";
3688         ENTRY;
3689
3690         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3691         LASSERT(info != NULL);
3692
3693         lu_dev_del_linkage(top->ld_site, top);
3694
3695         lu_site_purge(env, top->ld_site, -1);
3696
3697         bufs = &info->mti_u.bufs;
3698         /* process cleanup, pass mdt obd name to get obd umount flags */
3699         /* another purpose is to let all layers to release their objects */
3700         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
3701         if (obd->obd_force)
3702                 strcat(flags, "F");
3703         if (obd->obd_fail)
3704                 strcat(flags, "A");
3705         lustre_cfg_bufs_set_string(bufs, 1, flags);
3706         lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
3707         if (!lcfg) {
3708                 CERROR("Cannot alloc lcfg!\n");
3709                 return;
3710         }
3711         LASSERT(top);
3712         top->ld_ops->ldo_process_config(env, top, lcfg);
3713         lustre_cfg_free(lcfg);
3714
3715         lu_site_purge(env, top->ld_site, -1);
3716
3717         m->mdt_child = NULL;
3718         m->mdt_bottom = NULL;
3719
3720         obd_disconnect(m->mdt_child_exp);
3721         m->mdt_child_exp = NULL;
3722
3723         obd_disconnect(m->mdt_bottom_exp);
3724         m->mdt_child_exp = NULL;
3725 }
3726
3727 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
3728                                const char *next, struct obd_export **exp)
3729 {
3730         struct obd_connect_data *data = NULL;
3731         struct obd_device       *obd;
3732         int                      rc;
3733         ENTRY;
3734
3735         OBD_ALLOC_PTR(data);
3736         if (data == NULL)
3737                 GOTO(out, rc = -ENOMEM);
3738
3739         obd = class_name2obd(next);
3740         if (obd == NULL) {
3741                 CERROR("%s: can't locate next device: %s\n",
3742                        mdt_obd_name(m), next);
3743                 GOTO(out, rc = -ENOTCONN);
3744         }
3745
3746         data->ocd_connect_flags = OBD_CONNECT_VERSION;
3747         data->ocd_version = LUSTRE_VERSION_CODE;
3748
3749         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
3750         if (rc) {
3751                 CERROR("%s: cannot connect to next dev %s (%d)\n",
3752                        mdt_obd_name(m), next, rc);
3753                 GOTO(out, rc);
3754         }
3755
3756 out:
3757         if (data)
3758                 OBD_FREE_PTR(data);
3759         RETURN(rc);
3760 }
3761
3762 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
3763                           struct lustre_cfg *cfg)
3764 {
3765         char                   *dev = lustre_cfg_string(cfg, 0);
3766         int                     rc, name_size, uuid_size;
3767         char                   *name, *uuid, *p;
3768         struct lustre_cfg_bufs *bufs;
3769         struct lustre_cfg      *lcfg;
3770         struct obd_device      *obd;
3771         struct lustre_profile  *lprof;
3772         struct lu_site         *site;
3773         ENTRY;
3774
3775         /* in 1.8 we had the only device in the stack - MDS.
3776          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
3777          * in 2.3 OSD is instantiated by obd_mount.c, so we need
3778          * to generate names and setup MDT, MDD. MDT will be using
3779          * generated name to connect to MDD. for MDD the next device
3780          * will be LOD with name taken from so called "profile" which
3781          * is generated by mount_option line
3782          *
3783          * 1.8 MGS generates config. commands like this:
3784          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
3785          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
3786          * 2.0 MGS generates config. commands like this:
3787          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
3788          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
3789          *                    3:lustre-MDT0000-mdtlov  4:f
3790          *
3791          * we generate MDD name from MDT one, just replacing T with D
3792          *
3793          * after all the preparations, the logical equivalent will be
3794          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
3795          *                    3:lustre-MDT0000-mdtlov  4:f
3796          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
3797          *                    3:lustre-MDD0000  4:f
3798          *
3799          *  notice we build the stack from down to top: MDD first, then MDT */
3800
3801         name_size = MAX_OBD_NAME;
3802         uuid_size = MAX_OBD_NAME;
3803
3804         OBD_ALLOC(name, name_size);
3805         OBD_ALLOC(uuid, uuid_size);
3806         if (name == NULL || uuid == NULL)
3807                 GOTO(cleanup_mem, rc = -ENOMEM);
3808
3809         OBD_ALLOC_PTR(bufs);
3810         if (!bufs)
3811                 GOTO(cleanup_mem, rc = -ENOMEM);
3812
3813         strcpy(name, dev);
3814         p = strstr(name, "-MDT");
3815         if (p == NULL)
3816                 GOTO(free_bufs, rc = -ENOMEM);
3817         p[3] = 'D';
3818
3819         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
3820
3821         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
3822         if (lprof == NULL || lprof->lp_dt == NULL) {
3823                 CERROR("can't find the profile: %s\n",
3824                        lustre_cfg_string(cfg, 0));
3825                 GOTO(free_bufs, rc = -EINVAL);
3826         }
3827
3828         lustre_cfg_bufs_reset(bufs, name);
3829         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
3830         lustre_cfg_bufs_set_string(bufs, 2, uuid);
3831         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
3832
3833         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
3834         if (!lcfg)
3835                 GOTO(free_bufs, rc = -ENOMEM);
3836
3837         rc = class_attach(lcfg);
3838         if (rc)
3839                 GOTO(lcfg_cleanup, rc);
3840
3841         obd = class_name2obd(name);
3842         if (!obd) {
3843                 CERROR("Can not find obd %s (%s in config)\n",
3844                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
3845                 GOTO(class_detach, rc = -EINVAL);
3846         }
3847
3848         lustre_cfg_free(lcfg);
3849
3850         lustre_cfg_bufs_reset(bufs, name);
3851         lustre_cfg_bufs_set_string(bufs, 1, uuid);
3852         lustre_cfg_bufs_set_string(bufs, 2, dev);
3853         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
3854
3855         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
3856
3857         rc = class_setup(obd, lcfg);
3858         if (rc)
3859                 GOTO(class_detach, rc);
3860
3861         /* connect to MDD we just setup */
3862         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
3863         if (rc)
3864                 GOTO(class_detach, rc);
3865
3866         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
3867         LASSERT(site);
3868         LASSERT(mdt_lu_site(mdt) == NULL);
3869         mdt->mdt_lu_dev.ld_site = site;
3870         site->ls_top_dev = &mdt->mdt_lu_dev;
3871         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
3872
3873
3874         /* now connect to bottom OSD */
3875         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
3876         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
3877         if (rc)
3878                 GOTO(class_detach, rc);
3879         mdt->mdt_bottom =
3880                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
3881
3882
3883         rc = lu_env_refill((struct lu_env *)env);
3884         if (rc != 0)
3885                 CERROR("Failure to refill session: '%d'\n", rc);
3886
3887         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
3888
3889         EXIT;
3890 class_detach:
3891         if (rc)
3892                 class_detach(obd, lcfg);
3893 lcfg_cleanup:
3894         lustre_cfg_free(lcfg);
3895 free_bufs:
3896         OBD_FREE_PTR(bufs);
3897 cleanup_mem:
3898         if (name)
3899                 OBD_FREE(name, name_size);
3900         if (uuid)
3901                 OBD_FREE(uuid, uuid_size);
3902         RETURN(rc);
3903 }
3904
3905 /* setup quota master target on MDT0 */
3906 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
3907                           struct lustre_cfg *cfg)
3908 {
3909         struct obd_device       *obd;
3910         char                    *dev = lustre_cfg_string(cfg, 0);
3911         char                    *qmtname, *uuid, *p;
3912         struct lustre_cfg_bufs  *bufs;
3913         struct lustre_cfg       *lcfg;
3914         struct lustre_profile   *lprof;
3915         struct obd_connect_data *data;
3916         int                      rc;
3917         ENTRY;
3918
3919         LASSERT(mdt->mdt_qmt_exp == NULL);
3920         LASSERT(mdt->mdt_qmt_dev == NULL);
3921
3922         /* quota master is on MDT0 only for now */
3923         if (mdt->mdt_seq_site.ss_node_id != 0)
3924                 RETURN(0);
3925
3926         /* MGS generates config commands which look as follows:
3927          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
3928          *                    3:lustre-MDT0000-mdtlov  4:f
3929          *
3930          * We generate the QMT name from the MDT one, just replacing MD with QM
3931          * after all the preparations, the logical equivalent will be:
3932          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
3933          *                    3:lustre-MDT0000-osd  4:f */
3934         OBD_ALLOC(qmtname, MAX_OBD_NAME);
3935         OBD_ALLOC(uuid, UUID_MAX);
3936         OBD_ALLOC_PTR(bufs);
3937         OBD_ALLOC_PTR(data);
3938         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
3939                 GOTO(cleanup_mem, rc = -ENOMEM);
3940
3941         strcpy(qmtname, dev);
3942         p = strstr(qmtname, "-MDT");
3943         if (p == NULL)
3944                 GOTO(cleanup_mem, rc = -ENOMEM);
3945         /* replace MD with QM */
3946         p[1] = 'Q';
3947         p[2] = 'M';
3948
3949         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
3950
3951         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
3952         if (lprof == NULL || lprof->lp_dt == NULL) {
3953                 CERROR("can't find profile for %s\n",
3954                        lustre_cfg_string(cfg, 0));
3955                 GOTO(cleanup_mem, rc = -EINVAL);
3956         }
3957
3958         lustre_cfg_bufs_reset(bufs, qmtname);
3959         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
3960         lustre_cfg_bufs_set_string(bufs, 2, uuid);
3961         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
3962
3963         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
3964         if (!lcfg)
3965                 GOTO(cleanup_mem, rc = -ENOMEM);
3966
3967         rc = class_attach(lcfg);
3968         if (rc)
3969                 GOTO(lcfg_cleanup, rc);
3970
3971         obd = class_name2obd(qmtname);
3972         if (!obd) {
3973                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
3974                        lustre_cfg_string(cfg, 0));
3975                 GOTO(class_detach, rc = -EINVAL);
3976         }
3977
3978         lustre_cfg_free(lcfg);
3979
3980         lustre_cfg_bufs_reset(bufs, qmtname);
3981         lustre_cfg_bufs_set_string(bufs, 1, uuid);
3982         lustre_cfg_bufs_set_string(bufs, 2, dev);
3983
3984         /* for quota, the next device should be the OSD device */
3985         lustre_cfg_bufs_set_string(bufs, 3,
3986                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
3987
3988         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
3989
3990         rc = class_setup(obd, lcfg);
3991         if (rc)
3992                 GOTO(class_detach, rc);
3993
3994         mdt->mdt_qmt_dev = obd->obd_lu_dev;
3995
3996         /* configure local quota objects */
3997         rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
3998                                                    &mdt->mdt_lu_dev,
3999                                                    mdt->mdt_qmt_dev);
4000         if (rc)
4001                 GOTO(class_cleanup, rc);
4002
4003         /* connect to quota master target */
4004         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4005         data->ocd_version = LUSTRE_VERSION_CODE;
4006         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
4007                          data, NULL);
4008         if (rc) {
4009                 CERROR("cannot connect to quota master device %s (%d)\n",
4010                        qmtname, rc);
4011                 GOTO(class_cleanup, rc);
4012         }
4013
4014         EXIT;
4015 class_cleanup:
4016         if (rc) {
4017                 class_manual_cleanup(obd);
4018                 mdt->mdt_qmt_dev = NULL;
4019         }
4020 class_detach:
4021         if (rc)
4022                 class_detach(obd, lcfg);
4023 lcfg_cleanup:
4024         lustre_cfg_free(lcfg);
4025 cleanup_mem:
4026         if (bufs)
4027                 OBD_FREE_PTR(bufs);
4028         if (qmtname)
4029                 OBD_FREE(qmtname, MAX_OBD_NAME);
4030         if (uuid)
4031                 OBD_FREE(uuid, UUID_MAX);
4032         if (data)
4033                 OBD_FREE_PTR(data);
4034         return rc;
4035 }
4036
4037 /* Shutdown quota master target associated with mdt */
4038 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
4039 {
4040         ENTRY;
4041
4042         if (mdt->mdt_qmt_exp == NULL)
4043                 RETURN_EXIT;
4044         LASSERT(mdt->mdt_qmt_dev != NULL);
4045
4046         /* the qmt automatically shuts down when the mdt disconnects */
4047         obd_disconnect(mdt->mdt_qmt_exp);
4048         mdt->mdt_qmt_exp = NULL;
4049         mdt->mdt_qmt_dev = NULL;
4050         EXIT;
4051 }
4052
4053 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
4054  * for now. This will be removed along with converting rest of MDT code
4055  * to use tgt_session_info */
4056 int mdt_tgt_getxattr(struct tgt_session_info *tsi)
4057 {
4058         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
4059         int                      rc;
4060
4061         rc = mdt_getxattr(info);
4062
4063         mdt_thread_info_fini(info);
4064         return rc;
4065 }
4066
4067 static struct tgt_handler mdt_tgt_handlers[] = {
4068 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4069                 0,                      MDS_CONNECT,    mdt_tgt_connect,
4070                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
4071 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4072                 0,                      MDS_DISCONNECT, tgt_disconnect,
4073                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
4074 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4075                 HABEO_REFERO,           MDS_SET_INFO,   mdt_set_info,
4076                 &RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION),
4077 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
4078 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_GETSTATUS,  mdt_getstatus),
4079 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETATTR,    mdt_getattr),
4080 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_GETATTR_NAME,
4081                                                         mdt_getattr_name),
4082 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETXATTR,   mdt_tgt_getxattr),
4083 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_STATFS,     mdt_statfs),
4084 TGT_MDT_HDL(0           | MUTABOR,      MDS_REINT,      mdt_reint),
4085 TGT_MDT_HDL(HABEO_CORPUS,               MDS_CLOSE,      mdt_close),
4086 TGT_MDT_HDL(HABEO_CORPUS,               MDS_DONE_WRITING,
4087                                                         mdt_done_writing),
4088 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_READPAGE,   mdt_readpage),
4089 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_SYNC,       mdt_sync),
4090 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_IS_SUBDIR,  mdt_is_subdir),
4091 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
4092 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_PROGRESS,
4093                                                         mdt_hsm_progress),
4094 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_REGISTER,
4095                                                         mdt_hsm_ct_register),
4096 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_UNREGISTER,
4097                                                         mdt_hsm_ct_unregister),
4098 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_STATE_GET,
4099                                                         mdt_hsm_state_get),
4100 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_STATE_SET,
4101                                                         mdt_hsm_state_set),
4102 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_ACTION, mdt_hsm_action),
4103 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_REQUEST,
4104                                                         mdt_hsm_request),
4105 TGT_MDT_HDL(HABEO_CORPUS|HABEO_REFERO | MUTABOR, MDS_SWAP_LAYOUTS,
4106                                                         mdt_swap_layouts)
4107 };
4108
4109 static struct tgt_handler mdt_sec_ctx_ops[] = {
4110 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
4111 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
4112 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
4113 };
4114
4115 static struct tgt_handler mdt_quota_ops[] = {
4116 TGT_QUOTA_HDL(HABEO_REFERO,             QUOTA_DQACQ,      mdt_quota_dqacq),
4117 };
4118
4119 static struct tgt_opc_slice mdt_common_slice[] = {
4120         {
4121                 .tos_opc_start  = MDS_FIRST_OPC,
4122                 .tos_opc_end    = MDS_LAST_OPC,
4123                 .tos_hs         = mdt_tgt_handlers
4124         },
4125         {
4126                 .tos_opc_start  = OBD_FIRST_OPC,
4127                 .tos_opc_end    = OBD_LAST_OPC,
4128                 .tos_hs         = tgt_obd_handlers
4129         },
4130         {
4131                 .tos_opc_start  = LDLM_FIRST_OPC,
4132                 .tos_opc_end    = LDLM_LAST_OPC,
4133                 .tos_hs         = tgt_dlm_handlers
4134         },
4135         {
4136                 .tos_opc_start  = SEC_FIRST_OPC,
4137                 .tos_opc_end    = SEC_LAST_OPC,
4138                 .tos_hs         = mdt_sec_ctx_ops
4139         },
4140         {
4141                 .tos_opc_start  = UPDATE_OBJ,
4142                 .tos_opc_end    = UPDATE_LAST_OPC,
4143                 .tos_hs         = tgt_out_handlers
4144         },
4145         {
4146                 .tos_opc_start  = FLD_FIRST_OPC,
4147                 .tos_opc_end    = FLD_LAST_OPC,
4148                 .tos_hs         = fld_handlers
4149         },
4150         {
4151                 .tos_opc_start  = SEQ_FIRST_OPC,
4152                 .tos_opc_end    = SEQ_LAST_OPC,
4153                 .tos_hs         = seq_handlers
4154         },
4155         {
4156                 .tos_opc_start  = QUOTA_DQACQ,
4157                 .tos_opc_end    = QUOTA_LAST_OPC,
4158                 .tos_hs         = mdt_quota_ops
4159         },
4160         {
4161                 .tos_opc_start  = LLOG_FIRST_OPC,
4162                 .tos_opc_end    = LLOG_LAST_OPC,
4163                 .tos_hs         = tgt_llog_handlers
4164         },
4165
4166         {
4167                 .tos_hs         = NULL
4168         }
4169 };
4170
4171 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4172 {
4173         struct md_device  *next = m->mdt_child;
4174         struct lu_device  *d    = &m->mdt_lu_dev;
4175         struct obd_device *obd = mdt2obd_dev(m);
4176         ENTRY;
4177
4178         target_recovery_fini(obd);
4179
4180         ping_evictor_stop();
4181
4182         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
4183
4184         if (m->mdt_opts.mo_coordinator)
4185                 mdt_hsm_cdt_stop(m);
4186
4187         mdt_hsm_cdt_fini(m);
4188
4189         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
4190         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4191         obd_exports_barrier(obd);
4192         obd_zombie_barrier();
4193
4194         mdt_procfs_fini(m);
4195
4196         tgt_fini(env, &m->mdt_lut);
4197         mdt_fs_cleanup(env, m);
4198         upcall_cache_cleanup(m->mdt_identity_cache);
4199         m->mdt_identity_cache = NULL;
4200
4201         if (m->mdt_namespace != NULL) {
4202                 ldlm_namespace_free(m->mdt_namespace, NULL,
4203                                     d->ld_obd->obd_force);
4204                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4205         }
4206
4207         mdt_quota_fini(env, m);
4208
4209         cfs_free_nidlist(&m->mdt_nosquash_nids);
4210         if (m->mdt_nosquash_str) {
4211                 OBD_FREE(m->mdt_nosquash_str, m->mdt_nosquash_strlen);
4212                 m->mdt_nosquash_str = NULL;
4213                 m->mdt_nosquash_strlen = 0;
4214         }
4215
4216         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_PAUSE_LFSCK, 0, NULL);
4217
4218         mdt_seq_fini(env, m);
4219         mdt_fld_fini(env, m);
4220
4221         next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL);
4222         cfs_timer_disarm(&m->mdt_ck_timer);
4223         mdt_ck_thread_stop(m);
4224
4225         /*
4226          * Finish the stack
4227          */
4228         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4229
4230         LASSERT(cfs_atomic_read(&d->ld_ref) == 0);
4231
4232         server_put_mount(mdt_obd_name(m));
4233
4234         EXIT;
4235 }
4236
4237 int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4238
4239 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4240                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
4241 {
4242         struct mdt_thread_info    *info;
4243         struct obd_device         *obd;
4244         const char                *dev = lustre_cfg_string(cfg, 0);
4245         const char                *num = lustre_cfg_string(cfg, 2);
4246         struct lustre_mount_info  *lmi = NULL;
4247         struct lustre_sb_info     *lsi;
4248         struct lu_site            *s;
4249         struct seq_server_site    *ss_site;
4250         const char                *identity_upcall = "NONE";
4251         struct md_device          *next;
4252         int                        rc;
4253         long                       node_id;
4254         mntopt_t                   mntopts;
4255         ENTRY;
4256
4257         lu_device_init(&m->mdt_lu_dev, ldt);
4258         /*
4259          * Environment (env) might be missing mdt_thread_key values at that
4260          * point, if device is allocated when mdt_thread_key is in QUIESCENT
4261          * mode.
4262          *
4263          * Usually device allocation path doesn't use module key values, but
4264          * mdt has to do a lot of work here, so allocate key value.
4265          */
4266         rc = lu_env_refill((struct lu_env *)env);
4267         if (rc != 0)
4268                 RETURN(rc);
4269
4270         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4271         LASSERT(info != NULL);
4272
4273         obd = class_name2obd(dev);
4274         LASSERT(obd != NULL);
4275
4276         m->mdt_max_mdsize = MAX_MD_SIZE; /* 4 stripes */
4277
4278         m->mdt_som_conf = 0;
4279
4280         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4281
4282         /* default is coordinator off, it is started through conf_param
4283          * or /proc */
4284         m->mdt_opts.mo_coordinator = 0;
4285
4286         lmi = server_get_mount(dev);
4287         if (lmi == NULL) {
4288                 CERROR("Cannot get mount info for %s!\n", dev);
4289                 RETURN(-EFAULT);
4290         } else {
4291                 lsi = s2lsi(lmi->lmi_sb);
4292                 /* CMD is supported only in IAM mode */
4293                 LASSERT(num);
4294                 node_id = simple_strtol(num, NULL, 10);
4295                 obd->u.obt.obt_magic = OBT_MAGIC;
4296         }
4297
4298         spin_lock_init(&m->mdt_ioepoch_lock);
4299         m->mdt_capa_timeout = CAPA_TIMEOUT;
4300         m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1;
4301         m->mdt_ck_timeout = CAPA_KEY_TIMEOUT;
4302         m->mdt_squash_uid = 0;
4303         m->mdt_squash_gid = 0;
4304         CFS_INIT_LIST_HEAD(&m->mdt_nosquash_nids);
4305         m->mdt_nosquash_str = NULL;
4306         m->mdt_nosquash_strlen = 0;
4307         init_rwsem(&m->mdt_squash_sem);
4308         spin_lock_init(&m->mdt_osfs_lock);
4309         m->mdt_osfs_age = cfs_time_shift_64(-1000);
4310         m->mdt_enable_remote_dir = 0;
4311         m->mdt_enable_remote_dir_gid = 0;
4312
4313         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
4314         m->mdt_lu_dev.ld_obd = obd;
4315         /* Set this lu_device to obd for error handling purposes. */
4316         obd->obd_lu_dev = &m->mdt_lu_dev;
4317
4318         /* init the stack */
4319         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
4320         if (rc) {
4321                 CERROR("%s: Can't init device stack, rc %d\n",
4322                        mdt_obd_name(m), rc);
4323                 GOTO(err_lmi, rc);
4324         }
4325
4326         s = mdt_lu_site(m);
4327         ss_site = mdt_seq_site(m);
4328         s->ld_seq_site = ss_site;
4329         ss_site->ss_lu = s;
4330
4331         /* set server index */
4332         ss_site->ss_node_id = node_id;
4333
4334         /* failover is the default
4335          * FIXME: we do not failout mds0/mgs, which may cause some problems.
4336          * assumed whose ss_node_id == 0 XXX
4337          * */
4338         obd->obd_replayable = 1;
4339         /* No connection accepted until configurations will finish */
4340         obd->obd_no_conn = 1;
4341
4342         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4343                 char *str = lustre_cfg_string(cfg, 4);
4344                 if (strchr(str, 'n')) {
4345                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
4346                         obd->obd_replayable = 0;
4347                 }
4348         }
4349
4350         rc = mdt_fld_init(env, mdt_obd_name(m), m);
4351         if (rc)
4352                 GOTO(err_fini_stack, rc);
4353
4354         rc = mdt_seq_init(env, m);
4355         if (rc)
4356                 GOTO(err_fini_fld, rc);
4357
4358         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
4359                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
4360         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4361                                               LDLM_NAMESPACE_SERVER,
4362                                               LDLM_NAMESPACE_GREEDY,
4363                                               LDLM_NS_TYPE_MDT);
4364         if (m->mdt_namespace == NULL)
4365                 GOTO(err_fini_seq, rc = -ENOMEM);
4366
4367         m->mdt_namespace->ns_lvbp = m;
4368         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
4369
4370         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4371         /* set obd_namespace for compatibility with old code */
4372         obd->obd_namespace = m->mdt_namespace;
4373
4374         cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m);
4375
4376         rc = mdt_hsm_cdt_init(m);
4377         if (rc != 0) {
4378                 CERROR("%s: error initializing coordinator, rc %d\n",
4379                        mdt_obd_name(m), rc);
4380                 GOTO(err_free_ns, rc);
4381         }
4382
4383         rc = mdt_ck_thread_start(m);
4384         if (rc)
4385                 GOTO(err_free_hsm, rc);
4386
4387         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
4388                       OBD_FAIL_MDS_ALL_REQUEST_NET,
4389                       OBD_FAIL_MDS_ALL_REPLY_NET);
4390         if (rc)
4391                 GOTO(err_capa, rc);
4392
4393         rc = mdt_fs_setup(env, m, obd, lsi);
4394         if (rc)
4395                 GOTO(err_tgt, rc);
4396
4397         tgt_adapt_sptlrpc_conf(&m->mdt_lut, 1);
4398
4399         next = m->mdt_child;
4400         rc = next->md_ops->mdo_iocontrol(env, next, OBD_IOC_GET_MNTOPT, 0,
4401                                          &mntopts);
4402         if (rc)
4403                 GOTO(err_fs_cleanup, rc);
4404
4405         if (mntopts & MNTOPT_USERXATTR)
4406                 m->mdt_opts.mo_user_xattr = 1;
4407         else
4408                 m->mdt_opts.mo_user_xattr = 0;
4409
4410         rc = next->md_ops->mdo_maxeasize_get(env, next, &m->mdt_max_ea_size);
4411         if (rc)
4412                 GOTO(err_fs_cleanup, rc);
4413
4414         if (mntopts & MNTOPT_ACL)
4415                 m->mdt_opts.mo_acl = 1;
4416         else
4417                 m->mdt_opts.mo_acl = 0;
4418
4419         /* XXX: to support suppgid for ACL, we enable identity_upcall
4420          * by default, otherwise, maybe got unexpected -EACCESS. */
4421         if (m->mdt_opts.mo_acl)
4422                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4423
4424         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
4425                                                 identity_upcall,
4426                                                 &mdt_identity_upcall_cache_ops);
4427         if (IS_ERR(m->mdt_identity_cache)) {
4428                 rc = PTR_ERR(m->mdt_identity_cache);
4429                 m->mdt_identity_cache = NULL;
4430                 GOTO(err_fs_cleanup, rc);
4431         }
4432
4433         rc = mdt_procfs_init(m, dev);
4434         if (rc) {
4435                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4436                 GOTO(err_recovery, rc);
4437         }
4438
4439         rc = mdt_quota_init(env, m, cfg);
4440         if (rc)
4441                 GOTO(err_procfs, rc);
4442
4443         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
4444         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
4445                            "mdt_ldlm_client", m->mdt_ldlm_client);
4446
4447         ping_evictor_start();
4448
4449         /* recovery will be started upon mdt_prepare()
4450          * when the whole stack is complete and ready
4451          * to serve the requests */
4452
4453         mdt_init_capa_ctxt(env, m);
4454
4455         /* Reduce the initial timeout on an MDS because it doesn't need such
4456          * a long timeout as an OST does. Adaptive timeouts will adjust this
4457          * value appropriately. */
4458         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4459                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4460
4461         RETURN(0);
4462 err_procfs:
4463         mdt_procfs_fini(m);
4464 err_recovery:
4465         target_recovery_fini(obd);
4466         upcall_cache_cleanup(m->mdt_identity_cache);
4467         m->mdt_identity_cache = NULL;
4468 err_fs_cleanup:
4469         mdt_fs_cleanup(env, m);
4470 err_tgt:
4471         tgt_fini(env, &m->mdt_lut);
4472 err_capa:
4473         cfs_timer_disarm(&m->mdt_ck_timer);
4474         mdt_ck_thread_stop(m);
4475 err_free_hsm:
4476         mdt_hsm_cdt_fini(m);
4477 err_free_ns:
4478         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4479         obd->obd_namespace = m->mdt_namespace = NULL;
4480 err_fini_seq:
4481         mdt_seq_fini(env, m);
4482 err_fini_fld:
4483         mdt_fld_fini(env, m);
4484 err_fini_stack:
4485         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4486 err_lmi:
4487         if (lmi)
4488                 server_put_mount(dev);
4489         return(rc);
4490 }
4491
4492 /* For interoperability, the left element is old parameter, the right one
4493  * is the new version of the parameter, if some parameter is deprecated,
4494  * the new version should be set as NULL. */
4495 static struct cfg_interop_param mdt_interop_param[] = {
4496         { "mdt.group_upcall",   NULL },
4497         { "mdt.quota_type",     NULL },
4498         { "mdd.quota_type",     NULL },
4499         { "mdt.rootsquash",     "mdt.root_squash" },
4500         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
4501         { NULL }
4502 };
4503
4504 /* used by MGS to process specific configurations */
4505 static int mdt_process_config(const struct lu_env *env,
4506                               struct lu_device *d, struct lustre_cfg *cfg)
4507 {
4508         struct mdt_device *m = mdt_dev(d);
4509         struct md_device *md_next = m->mdt_child;
4510         struct lu_device *next = md2lu_dev(md_next);
4511         int rc;
4512         ENTRY;
4513
4514         switch (cfg->lcfg_command) {
4515         case LCFG_PARAM: {
4516                 struct lprocfs_static_vars  lvars;
4517                 struct obd_device          *obd = d->ld_obd;
4518
4519                 /* For interoperability */
4520                 struct cfg_interop_param   *ptr = NULL;
4521                 struct lustre_cfg          *old_cfg = NULL;
4522                 char                       *param = NULL;
4523
4524                 param = lustre_cfg_string(cfg, 1);
4525                 if (param == NULL) {
4526                         CERROR("param is empty\n");
4527                         rc = -EINVAL;
4528                         break;
4529                 }
4530
4531                 ptr = class_find_old_param(param, mdt_interop_param);
4532                 if (ptr != NULL) {
4533                         if (ptr->new_param == NULL) {
4534                                 rc = 0;
4535                                 CWARN("For interoperability, skip this %s."
4536                                       " It is obsolete.\n", ptr->old_param);
4537                                 break;
4538                         }
4539
4540                         CWARN("Found old param %s, changed it to %s.\n",
4541                               ptr->old_param, ptr->new_param);
4542
4543                         old_cfg = cfg;
4544                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
4545                         if (IS_ERR(cfg)) {
4546                                 rc = PTR_ERR(cfg);
4547                                 break;
4548                         }
4549                 }
4550
4551                 lprocfs_mdt_init_vars(&lvars);
4552                 rc = class_process_proc_param(PARAM_MDT, lvars.obd_vars,
4553                                               cfg, obd);
4554                 if (rc > 0 || rc == -ENOSYS) {
4555                         /* is it an HSM var ? */
4556                         rc = class_process_proc_param(PARAM_HSM,
4557                                                       hsm_cdt_get_proc_vars(),
4558                                                       cfg, obd);
4559                         if (rc > 0 || rc == -ENOSYS)
4560                                 /* we don't understand; pass it on */
4561                                 rc = next->ld_ops->ldo_process_config(env, next,
4562                                                                       cfg);
4563                 }
4564
4565                 if (old_cfg != NULL)
4566                         lustre_cfg_free(cfg);
4567
4568                 break;
4569         }
4570         default:
4571                 /* others are passed further */
4572                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4573                 break;
4574         }
4575         RETURN(rc);
4576 }
4577
4578 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4579                                           const struct lu_object_header *hdr,
4580                                           struct lu_device *d)
4581 {
4582         struct mdt_object *mo;
4583
4584         ENTRY;
4585
4586         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, __GFP_IO);
4587         if (mo != NULL) {
4588                 struct lu_object *o;
4589                 struct lu_object_header *h;
4590
4591                 o = &mo->mot_obj;
4592                 h = &mo->mot_header;
4593                 lu_object_header_init(h);
4594                 lu_object_init(o, h, d);
4595                 lu_object_add_top(h, o);
4596                 o->lo_ops = &mdt_obj_ops;
4597                 mutex_init(&mo->mot_ioepoch_mutex);
4598                 mutex_init(&mo->mot_lov_mutex);
4599                 init_rwsem(&mo->mot_open_sem);
4600                 RETURN(o);
4601         }
4602         RETURN(NULL);
4603 }
4604
4605 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
4606                            const struct lu_object_conf *unused)
4607 {
4608         struct mdt_device *d = mdt_dev(o->lo_dev);
4609         struct lu_device  *under;
4610         struct lu_object  *below;
4611         int                rc = 0;
4612         ENTRY;
4613
4614         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4615                PFID(lu_object_fid(o)));
4616
4617         under = &d->mdt_child->md_lu_dev;
4618         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4619         if (below != NULL) {
4620                 lu_object_add(o, below);
4621         } else
4622                 rc = -ENOMEM;
4623
4624         RETURN(rc);
4625 }
4626
4627 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4628 {
4629         struct mdt_object *mo = mdt_obj(o);
4630         struct lu_object_header *h;
4631         ENTRY;
4632
4633         h = o->lo_header;
4634         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4635                PFID(lu_object_fid(o)));
4636
4637         LASSERT(atomic_read(&mo->mot_open_count) == 0);
4638         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
4639
4640         lu_object_fini(o);
4641         lu_object_header_fini(h);
4642         OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
4643
4644         EXIT;
4645 }
4646
4647 static int mdt_object_print(const struct lu_env *env, void *cookie,
4648                             lu_printer_t p, const struct lu_object *o)
4649 {
4650         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
4651         return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p(ioepoch="LPU64" "
4652                     "flags="LPX64", epochcount=%d, writecount=%d)",
4653                     mdto, mdto->mot_ioepoch, mdto->mot_flags,
4654                     mdto->mot_ioepoch_count, mdto->mot_writecount);
4655 }
4656
4657 static int mdt_prepare(const struct lu_env *env,
4658                 struct lu_device *pdev,
4659                 struct lu_device *cdev)
4660 {
4661         struct mdt_device *mdt = mdt_dev(cdev);
4662         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
4663         struct obd_device *obd = cdev->ld_obd;
4664         struct lfsck_start_param lsp;
4665         int rc;
4666
4667         ENTRY;
4668
4669         LASSERT(obd);
4670
4671         rc = next->ld_ops->ldo_prepare(env, cdev, next);
4672         if (rc)
4673                 RETURN(rc);
4674
4675         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
4676         if (rc)
4677                 RETURN(rc);
4678
4679         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
4680         if (rc)
4681                 RETURN(rc);
4682
4683         lsp.lsp_start = NULL;
4684         lsp.lsp_namespace = mdt->mdt_namespace;
4685         rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
4686                                                    OBD_IOC_START_LFSCK,
4687                                                    0, &lsp);
4688         if (rc != 0) {
4689                 CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
4690                       mdt_obd_name(mdt), rc);
4691                 rc = 0;
4692         }
4693
4694         if (mdt->mdt_seq_site.ss_node_id == 0) {
4695                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
4696                                                          &mdt->mdt_md_root_fid);
4697                 if (rc)
4698                         RETURN(rc);
4699         }
4700
4701         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
4702         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
4703         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
4704         LASSERT(obd->obd_no_conn);
4705         spin_lock(&obd->obd_dev_lock);
4706         obd->obd_no_conn = 0;
4707         spin_unlock(&obd->obd_dev_lock);
4708
4709         if (obd->obd_recovering == 0)
4710                 mdt_postrecov(env, mdt);
4711
4712         RETURN(rc);
4713 }
4714
4715 const struct lu_device_operations mdt_lu_ops = {
4716         .ldo_object_alloc   = mdt_object_alloc,
4717         .ldo_process_config = mdt_process_config,
4718         .ldo_prepare        = mdt_prepare,
4719 };
4720
4721 static const struct lu_object_operations mdt_obj_ops = {
4722         .loo_object_init    = mdt_object_init,
4723         .loo_object_free    = mdt_object_free,
4724         .loo_object_print   = mdt_object_print
4725 };
4726
4727 static int mdt_obd_set_info_async(const struct lu_env *env,
4728                                   struct obd_export *exp,
4729                                   __u32 keylen, void *key,
4730                                   __u32 vallen, void *val,
4731                                   struct ptlrpc_request_set *set)
4732 {
4733         int rc;
4734
4735         ENTRY;
4736
4737         if (KEY_IS(KEY_SPTLRPC_CONF)) {
4738                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp), 0);
4739                 RETURN(rc);
4740         }
4741
4742         RETURN(0);
4743 }
4744
4745 /**
4746  * Match client and server connection feature flags.
4747  *
4748  * Compute the compatibility flags for a connection request based on
4749  * features mutually supported by client and server.
4750  *
4751  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
4752  * must not be updated here, otherwise a partially initialized value may
4753  * be exposed. After the connection request is successfully processed,
4754  * the top-level MDT connect request handler atomically updates the export
4755  * connect flags from the obd_connect_data::ocd_connect_flags field of the
4756  * reply. \see mdt_connect().
4757  *
4758  * \param exp   the obd_export associated with this client/target pair
4759  * \param mdt   the target device for the connection
4760  * \param data  stores data for this connect request
4761  *
4762  * \retval 0       success
4763  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
4764  * \retval -EBADE  client and server feature requirements are incompatible
4765  */
4766 static int mdt_connect_internal(struct obd_export *exp,
4767                                 struct mdt_device *mdt,
4768                                 struct obd_connect_data *data)
4769 {
4770         LASSERT(data != NULL);
4771
4772         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
4773         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
4774
4775         /* If no known bits (which should not happen, probably,
4776            as everybody should support LOOKUP and UPDATE bits at least)
4777            revert to compat mode with plain locks. */
4778         if (!data->ocd_ibits_known &&
4779             data->ocd_connect_flags & OBD_CONNECT_IBITS)
4780                 data->ocd_connect_flags &= ~OBD_CONNECT_IBITS;
4781
4782         if (!mdt->mdt_opts.mo_acl)
4783                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
4784
4785         if (!mdt->mdt_opts.mo_user_xattr)
4786                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
4787
4788         if (!mdt->mdt_som_conf)
4789                 data->ocd_connect_flags &= ~OBD_CONNECT_SOM;
4790
4791         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
4792                 data->ocd_brw_size = min(data->ocd_brw_size,
4793                                          (__u32)MD_MAX_BRW_SIZE);
4794                 if (data->ocd_brw_size == 0) {
4795                         CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
4796                                " ocd_version: %x ocd_grant: %d "
4797                                "ocd_index: %u ocd_brw_size is "
4798                                "unexpectedly zero, network data "
4799                                "corruption? Refusing connection of this"
4800                                " client\n",
4801                                mdt_obd_name(mdt),
4802                                exp->exp_client_uuid.uuid,
4803                                exp, data->ocd_connect_flags, data->ocd_version,
4804                                data->ocd_grant, data->ocd_index);
4805                         return -EPROTO;
4806                 }
4807         }
4808
4809         /* NB: Disregard the rule against updating
4810          * exp_connect_data.ocd_connect_flags in this case, since
4811          * tgt_client_new() needs to know if this is a lightweight
4812          * connection, and it is safe to expose this flag before
4813          * connection processing completes. */
4814         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
4815                 spin_lock(&exp->exp_lock);
4816                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
4817                 spin_unlock(&exp->exp_lock);
4818         }
4819
4820         data->ocd_version = LUSTRE_VERSION_CODE;
4821
4822         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
4823                 CWARN("%s: MDS requires FID support, but client not\n",
4824                       mdt_obd_name(mdt));
4825                 return -EBADE;
4826         }
4827
4828         if (mdt->mdt_som_conf &&
4829             !(data->ocd_connect_flags & (OBD_CONNECT_LIGHTWEIGHT |
4830                                          OBD_CONNECT_MDS_MDS |
4831                                          OBD_CONNECT_SOM))) {
4832                 CWARN("%s: MDS has SOM enabled, but client does not support "
4833                       "it\n", mdt_obd_name(mdt));
4834                 return -EBADE;
4835         }
4836
4837         if (OCD_HAS_FLAG(data, PINGLESS)) {
4838                 if (ptlrpc_pinger_suppress_pings()) {
4839                         spin_lock(&exp->exp_obd->obd_dev_lock);
4840                         list_del_init(&exp->exp_obd_chain_timed);
4841                         spin_unlock(&exp->exp_obd->obd_dev_lock);
4842                 } else {
4843                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
4844                 }
4845         }
4846
4847         data->ocd_max_easize = mdt->mdt_max_ea_size;
4848
4849         return 0;
4850 }
4851
4852 /* mds_connect copy */
4853 static int mdt_obd_connect(const struct lu_env *env,
4854                            struct obd_export **exp, struct obd_device *obd,
4855                            struct obd_uuid *cluuid,
4856                            struct obd_connect_data *data,
4857                            void *localdata)
4858 {
4859         struct obd_export      *lexp;
4860         struct lustre_handle    conn = { 0 };
4861         struct mdt_device      *mdt;
4862         int                     rc;
4863         ENTRY;
4864
4865         LASSERT(env != NULL);
4866         if (!exp || !obd || !cluuid)
4867                 RETURN(-EINVAL);
4868
4869         mdt = mdt_dev(obd->obd_lu_dev);
4870
4871         /*
4872          * first, check whether the stack is ready to handle requests
4873          * XXX: probably not very appropriate method is used now
4874          *      at some point we should find a better one
4875          */
4876         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) && data != NULL &&
4877             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
4878                 rc = obd_get_info(env, mdt->mdt_child_exp,
4879                                   sizeof(KEY_OSP_CONNECTED),
4880                                   KEY_OSP_CONNECTED, NULL, NULL, NULL);
4881                 if (rc)
4882                         RETURN(-EAGAIN);
4883                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
4884         }
4885
4886         rc = class_connect(&conn, obd, cluuid);
4887         if (rc)
4888                 RETURN(rc);
4889
4890         lexp = class_conn2export(&conn);
4891         LASSERT(lexp != NULL);
4892
4893         rc = mdt_connect_internal(lexp, mdt, data);
4894         if (rc == 0) {
4895                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
4896
4897                 LASSERT(lcd);
4898                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
4899                 rc = tgt_client_new(env, lexp);
4900                 if (rc == 0)
4901                         mdt_export_stats_init(obd, lexp, localdata);
4902         }
4903
4904         if (rc != 0) {
4905                 class_disconnect(lexp);
4906                 *exp = NULL;
4907         } else {
4908                 *exp = lexp;
4909         }
4910
4911         RETURN(rc);
4912 }
4913
4914 static int mdt_obd_reconnect(const struct lu_env *env,
4915                              struct obd_export *exp, struct obd_device *obd,
4916                              struct obd_uuid *cluuid,
4917                              struct obd_connect_data *data,
4918                              void *localdata)
4919 {
4920         int                     rc;
4921         ENTRY;
4922
4923         if (exp == NULL || obd == NULL || cluuid == NULL)
4924                 RETURN(-EINVAL);
4925
4926         rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
4927         if (rc == 0)
4928                 mdt_export_stats_init(obd, exp, localdata);
4929
4930         RETURN(rc);
4931 }
4932
4933 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
4934                                    struct mdt_thread_info *info,
4935                                    struct mdt_file_data *mfd)
4936 {
4937         struct lu_context ses;
4938         int rc;
4939         ENTRY;
4940
4941         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
4942         if (rc)
4943                 RETURN(rc);
4944
4945         env->le_ses = &ses;
4946         lu_context_enter(&ses);
4947
4948         mdt_ucred(info)->uc_valid = UCRED_OLD;
4949         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
4950
4951         lu_context_exit(&ses);
4952         lu_context_fini(&ses);
4953         env->le_ses = NULL;
4954
4955         RETURN(rc);
4956 }
4957
4958 static int mdt_export_cleanup(struct obd_export *exp)
4959 {
4960         struct mdt_export_data *med = &exp->exp_mdt_data;
4961         struct obd_device      *obd = exp->exp_obd;
4962         struct mdt_device      *mdt;
4963         struct mdt_thread_info *info;
4964         struct lu_env           env;
4965         CFS_LIST_HEAD(closing_list);
4966         struct mdt_file_data *mfd, *n;
4967         int rc = 0;
4968         ENTRY;
4969
4970         spin_lock(&med->med_open_lock);
4971         while (!cfs_list_empty(&med->med_open_head)) {
4972                 cfs_list_t *tmp = med->med_open_head.next;
4973                 mfd = cfs_list_entry(tmp, struct mdt_file_data, mfd_list);
4974
4975                 /* Remove mfd handle so it can't be found again.
4976                  * We are consuming the mfd_list reference here. */
4977                 class_handle_unhash(&mfd->mfd_handle);
4978                 cfs_list_move_tail(&mfd->mfd_list, &closing_list);
4979         }
4980         spin_unlock(&med->med_open_lock);
4981         mdt = mdt_dev(obd->obd_lu_dev);
4982         LASSERT(mdt != NULL);
4983
4984         rc = lu_env_init(&env, LCT_MD_THREAD);
4985         if (rc)
4986                 RETURN(rc);
4987
4988         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
4989         LASSERT(info != NULL);
4990         memset(info, 0, sizeof *info);
4991         info->mti_env = &env;
4992         info->mti_mdt = mdt;
4993         info->mti_exp = exp;
4994
4995         if (!cfs_list_empty(&closing_list)) {
4996                 struct md_attr *ma = &info->mti_attr;
4997
4998                 /* Close any open files (which may also cause orphan unlinking). */
4999                 cfs_list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5000                         cfs_list_del_init(&mfd->mfd_list);
5001                         ma->ma_need = ma->ma_valid = 0;
5002
5003                         /* This file is being closed due to an eviction, it
5004                          * could have been modified and now dirty regarding to
5005                          * HSM archive, check this!
5006                          * The logic here is to mark a file dirty if there's a
5007                          * chance it was dirtied before the client was evicted,
5008                          * so that we don't have to wait for a release attempt
5009                          * before finding out the file was actually dirty and
5010                          * fail the release. Aggressively marking it dirty here
5011                          * will cause the policy engine to attempt to
5012                          * re-archive it; when rearchiving, we can compare the
5013                          * current version to the HSM data_version and make the
5014                          * archive request into a noop if it's not actually
5015                          * dirty.
5016                          */
5017                         if (mfd->mfd_mode & (FMODE_WRITE|MDS_FMODE_TRUNC))
5018                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
5019
5020                         /* Don't unlink orphan on failover umount, LU-184 */
5021                         if (exp->exp_flags & OBD_OPT_FAILOVER) {
5022                                 ma->ma_valid = MA_FLAGS;
5023                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5024                         }
5025                         mdt_mfd_close(info, mfd);
5026                 }
5027         }
5028         info->mti_mdt = NULL;
5029         /* cleanup client slot early */
5030         /* Do not erase record for recoverable client. */
5031         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5032                 tgt_client_del(&env, exp);
5033         lu_env_fini(&env);
5034
5035         RETURN(rc);
5036 }
5037
5038 static int mdt_obd_disconnect(struct obd_export *exp)
5039 {
5040         int rc;
5041         ENTRY;
5042
5043         LASSERT(exp);
5044         class_export_get(exp);
5045
5046         rc = server_disconnect_export(exp);
5047         if (rc != 0)
5048                 CDEBUG(D_IOCTL, "server disconnect error: %d\n", rc);
5049
5050         rc = mdt_export_cleanup(exp);
5051         class_export_put(exp);
5052         RETURN(rc);
5053 }
5054
5055 /* FIXME: Can we avoid using these two interfaces? */
5056 static int mdt_init_export(struct obd_export *exp)
5057 {
5058         struct mdt_export_data *med = &exp->exp_mdt_data;
5059         int                     rc;
5060         ENTRY;
5061
5062         CFS_INIT_LIST_HEAD(&med->med_open_head);
5063         spin_lock_init(&med->med_open_lock);
5064         mutex_init(&med->med_idmap_mutex);
5065         med->med_idmap = NULL;
5066         spin_lock(&exp->exp_lock);
5067         exp->exp_connecting = 1;
5068         spin_unlock(&exp->exp_lock);
5069
5070         /* self-export doesn't need client data and ldlm initialization */
5071         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5072                                      &exp->exp_client_uuid)))
5073                 RETURN(0);
5074
5075         rc = tgt_client_alloc(exp);
5076         if (rc)
5077                 GOTO(err, rc);
5078
5079         rc = ldlm_init_export(exp);
5080         if (rc)
5081                 GOTO(err_free, rc);
5082
5083         RETURN(rc);
5084
5085 err_free:
5086         tgt_client_free(exp);
5087 err:
5088         CERROR("%s: Failed to initialize export: rc = %d\n",
5089                exp->exp_obd->obd_name, rc);
5090         return rc;
5091 }
5092
5093 static int mdt_destroy_export(struct obd_export *exp)
5094 {
5095         ENTRY;
5096
5097         if (exp_connect_rmtclient(exp))
5098                 mdt_cleanup_idmap(&exp->exp_mdt_data);
5099
5100         target_destroy_export(exp);
5101         /* destroy can be called from failed obd_setup, so
5102          * checking uuid is safer than obd_self_export */
5103         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5104                                      &exp->exp_client_uuid)))
5105                 RETURN(0);
5106
5107         ldlm_destroy_export(exp);
5108         tgt_client_free(exp);
5109
5110         LASSERT(cfs_list_empty(&exp->exp_outstanding_replies));
5111         LASSERT(cfs_list_empty(&exp->exp_mdt_data.med_open_head));
5112
5113         RETURN(0);
5114 }
5115
5116 /** The maximum depth that fid2path() will search.
5117  * This is limited only because we want to store the fids for
5118  * historical path lookup purposes.
5119  */
5120 #define MAX_PATH_DEPTH 100
5121
5122 /** mdt_path() lookup structure. */
5123 struct path_lookup_info {
5124         __u64                   pli_recno;      /**< history point */
5125         __u64                   pli_currec;     /**< current record */
5126         struct lu_fid           pli_fid;
5127         struct lu_fid           pli_fids[MAX_PATH_DEPTH]; /**< path, in fids */
5128         struct mdt_object       *pli_mdt_obj;
5129         char                    *pli_path;      /**< full path */
5130         int                     pli_pathlen;
5131         int                     pli_linkno;     /**< which hardlink to follow */
5132         int                     pli_fidcount;   /**< number of \a pli_fids */
5133 };
5134
5135 static int mdt_links_read(struct mdt_thread_info *info,
5136                           struct mdt_object *mdt_obj, struct linkea_data *ldata)
5137 {
5138         int rc;
5139
5140         LASSERT(ldata->ld_buf->lb_buf != NULL);
5141
5142         if (!mdt_object_exists(mdt_obj))
5143                 return -ENODATA;
5144
5145         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5146                           ldata->ld_buf, XATTR_NAME_LINK);
5147         if (rc == -ERANGE) {
5148                 /* Buf was too small, figure out what we need. */
5149                 lu_buf_free(ldata->ld_buf);
5150                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5151                                   ldata->ld_buf, XATTR_NAME_LINK);
5152                 if (rc < 0)
5153                         return rc;
5154                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
5155                 if (ldata->ld_buf->lb_buf == NULL)
5156                         return -ENOMEM;
5157                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5158                                   ldata->ld_buf, XATTR_NAME_LINK);
5159         }
5160         if (rc < 0)
5161                 return rc;
5162
5163         return linkea_init(ldata);
5164 }
5165
5166 static int mdt_path_current(struct mdt_thread_info *info,
5167                             struct path_lookup_info *pli)
5168 {
5169         struct mdt_device       *mdt = info->mti_mdt;
5170         struct mdt_object       *mdt_obj;
5171         struct link_ea_header   *leh;
5172         struct link_ea_entry    *lee;
5173         struct lu_name          *tmpname = &info->mti_name;
5174         struct lu_fid           *tmpfid = &info->mti_tmp_fid1;
5175         struct lu_buf           *buf = &info->mti_big_buf;
5176         char                    *ptr;
5177         int                     reclen;
5178         struct linkea_data      ldata = { 0 };
5179         int                     rc = 0;
5180         ENTRY;
5181
5182         /* temp buffer for path element, the buffer will be finally freed
5183          * in mdt_thread_info_fini */
5184         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
5185         if (buf->lb_buf == NULL)
5186                 RETURN(-ENOMEM);
5187
5188         ldata.ld_buf = buf;
5189         ptr = pli->pli_path + pli->pli_pathlen - 1;
5190         *ptr = 0;
5191         --ptr;
5192         pli->pli_fidcount = 0;
5193         pli->pli_fids[0] = *(struct lu_fid *)mdt_object_fid(pli->pli_mdt_obj);
5194
5195         /* root FID only exists on MDT0, and fid2path should also ends at MDT0,
5196          * so checking root_fid can only happen on MDT0. */
5197         while (!lu_fid_eq(&mdt->mdt_md_root_fid,
5198                           &pli->pli_fids[pli->pli_fidcount])) {
5199                 mdt_obj = mdt_object_find(info->mti_env, mdt,
5200                                           &pli->pli_fids[pli->pli_fidcount]);
5201                 if (IS_ERR(mdt_obj))
5202                         GOTO(out, rc = PTR_ERR(mdt_obj));
5203                 if (mdt_object_remote(mdt_obj)) {
5204                         mdt_object_put(info->mti_env, mdt_obj);
5205                         GOTO(remote_out, rc = -EREMOTE);
5206                 }
5207                 if (!mdt_object_exists(mdt_obj)) {
5208                         mdt_object_put(info->mti_env, mdt_obj);
5209                         GOTO(out, rc = -ENOENT);
5210                 }
5211
5212                 rc = mdt_links_read(info, mdt_obj, &ldata);
5213                 mdt_object_put(info->mti_env, mdt_obj);
5214                 if (rc != 0)
5215                         GOTO(out, rc);
5216
5217                 leh = buf->lb_buf;
5218                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
5219                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
5220                 /* If set, use link #linkno for path lookup, otherwise use
5221                    link #0.  Only do this for the final path element. */
5222                 if (pli->pli_fidcount == 0 &&
5223                     pli->pli_linkno < leh->leh_reccount) {
5224                         int count;
5225                         for (count = 0; count < pli->pli_linkno; count++) {
5226                                 lee = (struct link_ea_entry *)
5227                                      ((char *)lee + reclen);
5228                                 linkea_entry_unpack(lee, &reclen, tmpname,
5229                                                     tmpfid);
5230                         }
5231                         if (pli->pli_linkno < leh->leh_reccount - 1)
5232                                 /* indicate to user there are more links */
5233                                 pli->pli_linkno++;
5234                 }
5235
5236                 /* Pack the name in the end of the buffer */
5237                 ptr -= tmpname->ln_namelen;
5238                 if (ptr - 1 <= pli->pli_path)
5239                         GOTO(out, rc = -EOVERFLOW);
5240                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
5241                 *(--ptr) = '/';
5242
5243                 /* Store the parent fid for historic lookup */
5244                 if (++pli->pli_fidcount >= MAX_PATH_DEPTH)
5245                         GOTO(out, rc = -EOVERFLOW);
5246                 pli->pli_fids[pli->pli_fidcount] = *tmpfid;
5247         }
5248
5249 remote_out:
5250         ptr++; /* skip leading / */
5251         memmove(pli->pli_path, ptr, pli->pli_path + pli->pli_pathlen - ptr);
5252
5253         EXIT;
5254 out:
5255         return rc;
5256 }
5257
5258 /* Returns the full path to this fid, as of changelog record recno. */
5259 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
5260                     char *path, int pathlen, __u64 *recno, int *linkno,
5261                     struct lu_fid *fid)
5262 {
5263         struct mdt_device       *mdt = info->mti_mdt;
5264         struct path_lookup_info *pli;
5265         int                     tries = 3;
5266         int                     rc = -EAGAIN;
5267         ENTRY;
5268
5269         if (pathlen < 3)
5270                 RETURN(-EOVERFLOW);
5271
5272         if (lu_fid_eq(&mdt->mdt_md_root_fid, mdt_object_fid(obj))) {
5273                 path[0] = '\0';
5274                 RETURN(0);
5275         }
5276
5277         OBD_ALLOC_PTR(pli);
5278         if (pli == NULL)
5279                 RETURN(-ENOMEM);
5280
5281         pli->pli_mdt_obj = obj;
5282         pli->pli_recno = *recno;
5283         pli->pli_path = path;
5284         pli->pli_pathlen = pathlen;
5285         pli->pli_linkno = *linkno;
5286
5287         /* Retry multiple times in case file is being moved */
5288         while (tries-- && rc == -EAGAIN)
5289                 rc = mdt_path_current(info, pli);
5290
5291         /* return the last resolved fids to the client, so the client will
5292          * build the left path on another MDT for remote object */
5293         *fid = pli->pli_fids[pli->pli_fidcount];
5294
5295         *recno = pli->pli_currec;
5296         /* Return next link index to caller */
5297         *linkno = pli->pli_linkno;
5298
5299         OBD_FREE_PTR(pli);
5300
5301         RETURN(rc);
5302 }
5303
5304 static int mdt_fid2path(struct mdt_thread_info *info,
5305                         struct getinfo_fid2path *fp)
5306 {
5307         struct mdt_device *mdt = info->mti_mdt;
5308         struct mdt_object *obj;
5309         int    rc;
5310         ENTRY;
5311
5312         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
5313                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
5314
5315         if (!fid_is_sane(&fp->gf_fid))
5316                 RETURN(-EINVAL);
5317
5318         if (!fid_is_namespace_visible(&fp->gf_fid)) {
5319                 CWARN("%s: "DFID" is invalid, sequence should be "
5320                       ">= "LPX64"\n", mdt_obd_name(mdt),
5321                       PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
5322                 RETURN(-EINVAL);
5323         }
5324
5325         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
5326         if (obj == NULL || IS_ERR(obj)) {
5327                 CDEBUG(D_IOCTL, "no object "DFID": %ld\n", PFID(&fp->gf_fid),
5328                        PTR_ERR(obj));
5329                 RETURN(-EINVAL);
5330         }
5331
5332         if (mdt_object_remote(obj))
5333                 rc = -EREMOTE;
5334         else if (!mdt_object_exists(obj))
5335                 rc = -ENOENT;
5336         else
5337                 rc = 0;
5338
5339         if (rc < 0) {
5340                 mdt_object_put(info->mti_env, obj);
5341                 CDEBUG(D_IOCTL, "nonlocal object "DFID": %d\n",
5342                        PFID(&fp->gf_fid), rc);
5343                 RETURN(rc);
5344         }
5345
5346         rc = mdt_path(info, obj, fp->gf_path, fp->gf_pathlen, &fp->gf_recno,
5347                       &fp->gf_linkno, &fp->gf_fid);
5348
5349         CDEBUG(D_INFO, "fid "DFID", path %s recno "LPX64" linkno %u\n",
5350                PFID(&fp->gf_fid), fp->gf_path, fp->gf_recno, fp->gf_linkno);
5351
5352         mdt_object_put(info->mti_env, obj);
5353
5354         RETURN(rc);
5355 }
5356
5357 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key,
5358                             void *val, int vallen)
5359 {
5360         struct getinfo_fid2path *fpout, *fpin;
5361         int rc = 0;
5362
5363         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
5364         fpout = val;
5365
5366         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5367                 lustre_swab_fid2path(fpin);
5368
5369         memcpy(fpout, fpin, sizeof(*fpin));
5370         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
5371                 RETURN(-EINVAL);
5372
5373         rc = mdt_fid2path(info, fpout);
5374         RETURN(rc);
5375 }
5376
5377 int mdt_get_info(struct tgt_session_info *tsi)
5378 {
5379         char    *key;
5380         int      keylen;
5381         __u32   *vallen;
5382         void    *valout;
5383         int      rc;
5384
5385         ENTRY;
5386
5387         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
5388         if (key == NULL) {
5389                 CDEBUG(D_IOCTL, "No GETINFO key");
5390                 RETURN(err_serious(-EFAULT));
5391         }
5392         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
5393                                       RCL_CLIENT);
5394
5395         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
5396         if (vallen == NULL) {
5397                 CDEBUG(D_IOCTL, "Unable to get RMF_GETINFO_VALLEN buffer");
5398                 RETURN(err_serious(-EFAULT));
5399         }
5400
5401         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
5402                              *vallen);
5403         rc = req_capsule_server_pack(tsi->tsi_pill);
5404         if (rc)
5405                 RETURN(err_serious(rc));
5406
5407         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
5408         if (valout == NULL) {
5409                 CDEBUG(D_IOCTL, "Unable to get get-info RPC out buffer");
5410                 RETURN(err_serious(-EFAULT));
5411         }
5412
5413         if (KEY_IS(KEY_FID2PATH)) {
5414                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
5415
5416                 rc = mdt_rpc_fid2path(info, key, valout, *vallen);
5417                 mdt_thread_info_fini(info);
5418         } else {
5419                 rc = -EINVAL;
5420         }
5421         RETURN(rc);
5422 }
5423
5424 /* Pass the ioc down */
5425 static int mdt_ioc_child(struct lu_env *env, struct mdt_device *mdt,
5426                          unsigned int cmd, int len, void *data)
5427 {
5428         struct lu_context ioctl_session;
5429         struct md_device *next = mdt->mdt_child;
5430         int rc;
5431         ENTRY;
5432
5433         rc = lu_context_init(&ioctl_session, LCT_SERVER_SESSION);
5434         if (rc)
5435                 RETURN(rc);
5436         ioctl_session.lc_thread = (struct ptlrpc_thread *)current;
5437         lu_context_enter(&ioctl_session);
5438         env->le_ses = &ioctl_session;
5439
5440         LASSERT(next->md_ops->mdo_iocontrol);
5441         rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data);
5442
5443         lu_context_exit(&ioctl_session);
5444         lu_context_fini(&ioctl_session);
5445         RETURN(rc);
5446 }
5447
5448 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
5449 {
5450         struct obd_ioctl_data *data = karg;
5451         struct lu_fid *fid;
5452         __u64 version;
5453         struct mdt_object *obj;
5454         struct mdt_lock_handle  *lh;
5455         int rc;
5456         ENTRY;
5457
5458         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
5459             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
5460                 RETURN(-EINVAL);
5461
5462         fid = (struct lu_fid *)data->ioc_inlbuf1;
5463
5464         if (!fid_is_sane(fid))
5465                 RETURN(-EINVAL);
5466
5467         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
5468
5469         lh = &mti->mti_lh[MDT_LH_PARENT];
5470         mdt_lock_reg_init(lh, LCK_CR);
5471
5472         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
5473         if (IS_ERR(obj))
5474                 RETURN(PTR_ERR(obj));
5475
5476         if (mdt_object_remote(obj)) {
5477                 rc = -EREMOTE;
5478                 /**
5479                  * before calling version get the correct MDS should be
5480                  * fid, this is error to find remote object here
5481                  */
5482                 CERROR("nonlocal object "DFID"\n", PFID(fid));
5483         } else if (!mdt_object_exists(obj)) {
5484                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
5485                 rc = -ENOENT;
5486         } else {
5487                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
5488                *(__u64 *)data->ioc_inlbuf2 = version;
5489                 rc = 0;
5490         }
5491         mdt_object_unlock_put(mti, obj, lh, 1);
5492         RETURN(rc);
5493 }
5494
5495 /* ioctls on obd dev */
5496 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
5497                          void *karg, void *uarg)
5498 {
5499         struct lu_env      env;
5500         struct obd_device *obd = exp->exp_obd;
5501         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
5502         struct dt_device  *dt = mdt->mdt_bottom;
5503         int rc;
5504
5505         ENTRY;
5506         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
5507         rc = lu_env_init(&env, LCT_MD_THREAD);
5508         if (rc)
5509                 RETURN(rc);
5510
5511         switch (cmd) {
5512         case OBD_IOC_SYNC:
5513                 rc = mdt_device_sync(&env, mdt);
5514                 break;
5515         case OBD_IOC_SET_READONLY:
5516                 rc = dt->dd_ops->dt_ro(&env, dt);
5517                 break;
5518         case OBD_IOC_ABORT_RECOVERY:
5519                 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
5520                 target_stop_recovery_thread(obd);
5521                 rc = 0;
5522                 break;
5523         case OBD_IOC_CHANGELOG_REG:
5524         case OBD_IOC_CHANGELOG_DEREG:
5525         case OBD_IOC_CHANGELOG_CLEAR:
5526                 rc = mdt_ioc_child(&env, mdt, cmd, len, karg);
5527                 break;
5528         case OBD_IOC_START_LFSCK: {
5529                 struct md_device *next = mdt->mdt_child;
5530                 struct obd_ioctl_data *data = karg;
5531                 struct lfsck_start_param lsp;
5532
5533                 if (unlikely(data == NULL)) {
5534                         rc = -EINVAL;
5535                         break;
5536                 }
5537
5538                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
5539                 lsp.lsp_namespace = mdt->mdt_namespace;
5540                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
5541                 break;
5542         }
5543         case OBD_IOC_STOP_LFSCK: {
5544                 struct md_device *next = mdt->mdt_child;
5545
5546                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, NULL);
5547                 break;
5548         }
5549         case OBD_IOC_GET_OBJ_VERSION: {
5550                 struct mdt_thread_info *mti;
5551                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5552                 memset(mti, 0, sizeof *mti);
5553                 mti->mti_env = &env;
5554                 mti->mti_mdt = mdt;
5555                 mti->mti_exp = exp;
5556
5557                 rc = mdt_ioc_version_get(mti, karg);
5558                 break;
5559         }
5560         case OBD_IOC_CATLOGLIST:
5561                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg);
5562                 break;
5563         default:
5564                 rc = -EOPNOTSUPP;
5565                 CERROR("%s: Not supported cmd = %d, rc = %d\n",
5566                         mdt_obd_name(mdt), cmd, rc);
5567         }
5568
5569         lu_env_fini(&env);
5570         RETURN(rc);
5571 }
5572
5573 int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
5574 {
5575         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
5576         int rc;
5577         ENTRY;
5578
5579         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
5580         RETURN(rc);
5581 }
5582
5583 int mdt_obd_postrecov(struct obd_device *obd)
5584 {
5585         struct lu_env env;
5586         int rc;
5587
5588         rc = lu_env_init(&env, LCT_MD_THREAD);
5589         if (rc)
5590                 RETURN(rc);
5591         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
5592         lu_env_fini(&env);
5593         return rc;
5594 }
5595
5596 static struct obd_ops mdt_obd_device_ops = {
5597         .o_owner          = THIS_MODULE,
5598         .o_set_info_async = mdt_obd_set_info_async,
5599         .o_connect        = mdt_obd_connect,
5600         .o_reconnect      = mdt_obd_reconnect,
5601         .o_disconnect     = mdt_obd_disconnect,
5602         .o_init_export    = mdt_init_export,
5603         .o_destroy_export = mdt_destroy_export,
5604         .o_iocontrol      = mdt_iocontrol,
5605         .o_postrecov      = mdt_obd_postrecov,
5606 };
5607
5608 static struct lu_device* mdt_device_fini(const struct lu_env *env,
5609                                          struct lu_device *d)
5610 {
5611         struct mdt_device *m = mdt_dev(d);
5612         ENTRY;
5613
5614         mdt_fini(env, m);
5615         RETURN(NULL);
5616 }
5617
5618 static struct lu_device *mdt_device_free(const struct lu_env *env,
5619                                          struct lu_device *d)
5620 {
5621         struct mdt_device *m = mdt_dev(d);
5622         ENTRY;
5623
5624         lu_device_fini(&m->mdt_lu_dev);
5625         OBD_FREE_PTR(m);
5626
5627         RETURN(NULL);
5628 }
5629
5630 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
5631                                           struct lu_device_type *t,
5632                                           struct lustre_cfg *cfg)
5633 {
5634         struct lu_device  *l;
5635         struct mdt_device *m;
5636
5637         OBD_ALLOC_PTR(m);
5638         if (m != NULL) {
5639                 int rc;
5640
5641                 l = &m->mdt_lu_dev;
5642                 rc = mdt_init0(env, m, t, cfg);
5643                 if (rc != 0) {
5644                         mdt_device_free(env, l);
5645                         l = ERR_PTR(rc);
5646                         return l;
5647                 }
5648         } else
5649                 l = ERR_PTR(-ENOMEM);
5650         return l;
5651 }
5652
5653 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
5654 LU_KEY_INIT(mdt, struct mdt_thread_info);
5655
5656 static void mdt_key_fini(const struct lu_context *ctx,
5657                          struct lu_context_key *key, void* data)
5658 {
5659         struct mdt_thread_info *info = data;
5660
5661         if (info->mti_big_lmm) {
5662                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
5663                 info->mti_big_lmm = NULL;
5664                 info->mti_big_lmmsize = 0;
5665         }
5666         OBD_FREE_PTR(info);
5667 }
5668
5669 /* context key: mdt_thread_key */
5670 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
5671
5672 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
5673 {
5674         return lu_ucred(info->mti_env);
5675 }
5676
5677 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
5678 {
5679         return lu_ucred_check(info->mti_env);
5680 }
5681
5682 /**
5683  * Enable/disable COS (Commit On Sharing).
5684  *
5685  * Set/Clear the COS flag in mdt options.
5686  *
5687  * \param mdt mdt device
5688  * \param val 0 disables COS, other values enable COS
5689  */
5690 void mdt_enable_cos(struct mdt_device *mdt, int val)
5691 {
5692         struct lu_env env;
5693         int rc;
5694
5695         mdt->mdt_opts.mo_cos = !!val;
5696         rc = lu_env_init(&env, LCT_LOCAL);
5697         if (unlikely(rc != 0)) {
5698                 CWARN("lu_env initialization failed with rc = %d,"
5699                       "cannot sync\n", rc);
5700                 return;
5701         }
5702         mdt_device_sync(&env, mdt);
5703         lu_env_fini(&env);
5704 }
5705
5706 /**
5707  * Check COS (Commit On Sharing) status.
5708  *
5709  * Return COS flag status.
5710  *
5711  * \param mdt mdt device
5712  */
5713 int mdt_cos_is_enabled(struct mdt_device *mdt)
5714 {
5715         return mdt->mdt_opts.mo_cos != 0;
5716 }
5717
5718 static struct lu_device_type_operations mdt_device_type_ops = {
5719         .ldto_device_alloc = mdt_device_alloc,
5720         .ldto_device_free  = mdt_device_free,
5721         .ldto_device_fini  = mdt_device_fini
5722 };
5723
5724 static struct lu_device_type mdt_device_type = {
5725         .ldt_tags     = LU_DEVICE_MD,
5726         .ldt_name     = LUSTRE_MDT_NAME,
5727         .ldt_ops      = &mdt_device_type_ops,
5728         .ldt_ctx_tags = LCT_MD_THREAD
5729 };
5730
5731 static int __init mdt_mod_init(void)
5732 {
5733         struct lprocfs_static_vars lvars;
5734         int rc;
5735
5736         CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") ==
5737                  FID_NOBRACE_LEN + 1);
5738         CLASSERT(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") ==
5739                  FID_LEN + 1);
5740         rc = lu_kmem_init(mdt_caches);
5741         if (rc)
5742                 return rc;
5743
5744         rc = mds_mod_init();
5745         if (rc)
5746                 GOTO(lu_fini, rc);
5747
5748         lprocfs_mdt_init_vars(&lvars);
5749         rc = class_register_type(&mdt_obd_device_ops, NULL, NULL,
5750 #ifndef HAVE_ONLY_PROCFS_SEQ
5751                                 lvars.module_vars,
5752 #endif
5753                                 LUSTRE_MDT_NAME, &mdt_device_type);
5754         if (rc)
5755                 GOTO(mds_fini, rc);
5756 lu_fini:
5757         if (rc)
5758                 lu_kmem_fini(mdt_caches);
5759 mds_fini:
5760         if (rc)
5761                 mds_mod_exit();
5762         return rc;
5763 }
5764
5765 static void __exit mdt_mod_exit(void)
5766 {
5767         class_unregister_type(LUSTRE_MDT_NAME);
5768         mds_mod_exit();
5769         lu_kmem_fini(mdt_caches);
5770 }
5771
5772 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
5773 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
5774 MODULE_LICENSE("GPL");
5775
5776 cfs_module(mdt, LUSTRE_VERSION_STRING, mdt_mod_init, mdt_mod_exit);