lustre/mdt/mdt_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2010, 2013, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/mdt/mdt_handler.c
  37  *
  38  * Lustre Metadata Target (mdt) request handler
  39  *
  40  * Author: Peter Braam <braam@clusterfs.com>
  41  * Author: Andreas Dilger <adilger@clusterfs.com>
  42  * Author: Phil Schwan <phil@clusterfs.com>
  43  * Author: Mike Shaver <shaver@clusterfs.com>
  44  * Author: Nikita Danilov <nikita@clusterfs.com>
  45  * Author: Huang Hua <huanghua@clusterfs.com>
  46  * Author: Yury Umanets <umka@clusterfs.com>
  47  */
  48
  49 #define DEBUG_SUBSYSTEM S_MDS
  50
  51 #include <linux/module.h>
  52 /*
  53  * struct OBD_{ALLOC,FREE}*()
  54  */
  55 #include <obd_support.h>
  56 #include <lustre_ioctl.h>
  57 /* struct ptlrpc_request */
  58 #include <lustre_net.h>
  59 /* struct obd_export */
  60 #include <lustre_export.h>
  61 /* struct obd_device */
  62 #include <obd.h>
  63 /* lu2dt_dev() */
  64 #include <dt_object.h>
  65 #include <lustre_mds.h>
  66 #include <lustre_log.h>
  67 #include "mdt_internal.h"
  68 #include <lustre_acl.h>
  69 #include <lustre_param.h>
  70 #include <lustre_quota.h>
  71 #include <lustre_lfsck.h>
  72 #include <lustre_nodemap.h>
  73
  74 mdl_mode_t mdt_mdl_lock_modes[] = {
  75         [LCK_MINMODE] = MDL_MINMODE,
  76         [LCK_EX]      = MDL_EX,
  77         [LCK_PW]      = MDL_PW,
  78         [LCK_PR]      = MDL_PR,
  79         [LCK_CW]      = MDL_CW,
  80         [LCK_CR]      = MDL_CR,
  81         [LCK_NL]      = MDL_NL,
  82         [LCK_GROUP]   = MDL_GROUP
  83 };
  84
  85 ldlm_mode_t mdt_dlm_lock_modes[] = {
  86         [MDL_MINMODE] = LCK_MINMODE,
  87         [MDL_EX]      = LCK_EX,
  88         [MDL_PW]      = LCK_PW,
  89         [MDL_PR]      = LCK_PR,
  90         [MDL_CW]      = LCK_CW,
  91         [MDL_CR]      = LCK_CR,
  92         [MDL_NL]      = LCK_NL,
  93         [MDL_GROUP]   = LCK_GROUP
  94 };
  95
  96 static struct mdt_device *mdt_dev(struct lu_device *d);
  97 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
  98
  99 static const struct lu_object_operations mdt_obj_ops;
 100
 101 /* Slab for MDT object allocation */
 102 static struct kmem_cache *mdt_object_kmem;
 103
 104 /* For HSM restore handles */
 105 struct kmem_cache *mdt_hsm_cdt_kmem;
 106
 107 /* For HSM request handles */
 108 struct kmem_cache *mdt_hsm_car_kmem;
 109
 110 static struct lu_kmem_descr mdt_caches[] = {
 111         {
 112                 .ckd_cache = &mdt_object_kmem,
 113                 .ckd_name  = "mdt_obj",
 114                 .ckd_size  = sizeof(struct mdt_object)
 115         },
 116         {
 117                 .ckd_cache      = &mdt_hsm_cdt_kmem,
 118                 .ckd_name       = "mdt_cdt_restore_handle",
 119                 .ckd_size       = sizeof(struct cdt_restore_handle)
 120         },
 121         {
 122                 .ckd_cache      = &mdt_hsm_car_kmem,
 123                 .ckd_name       = "mdt_cdt_agent_req",
 124                 .ckd_size       = sizeof(struct cdt_agent_req)
 125         },
 126         {
 127                 .ckd_cache = NULL
 128         }
 129 };
 130
 131 int mdt_get_disposition(struct ldlm_reply *rep, int flag)
 132 {
 133         if (!rep)
 134                 return 0;
 135         return (rep->lock_policy_res1 & flag);
 136 }
 137
 138 void mdt_clear_disposition(struct mdt_thread_info *info,
 139                            struct ldlm_reply *rep, int flag)
 140 {
 141         if (info) {
 142                 info->mti_opdata &= ~flag;
 143                 tgt_opdata_clear(info->mti_env, flag);
 144         }
 145         if (rep)
 146                 rep->lock_policy_res1 &= ~flag;
 147 }
 148
 149 void mdt_set_disposition(struct mdt_thread_info *info,
 150                          struct ldlm_reply *rep, int flag)
 151 {
 152         if (info) {
 153                 info->mti_opdata |= flag;
 154                 tgt_opdata_set(info->mti_env, flag);
 155         }
 156         if (rep)
 157                 rep->lock_policy_res1 |= flag;
 158 }
 159
 160 void mdt_lock_reg_init(struct mdt_lock_handle *lh, ldlm_mode_t lm)
 161 {
 162         lh->mlh_pdo_hash = 0;
 163         lh->mlh_reg_mode = lm;
 164         lh->mlh_rreg_mode = lm;
 165         lh->mlh_type = MDT_REG_LOCK;
 166 }
 167
 168 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, ldlm_mode_t lock_mode,
 169                        const struct lu_name *lname)
 170 {
 171         lh->mlh_reg_mode = lock_mode;
 172         lh->mlh_rreg_mode = lock_mode;
 173         lh->mlh_type = MDT_PDO_LOCK;
 174
 175         if (lu_name_is_valid(lname)) {
 176                 lh->mlh_pdo_hash = full_name_hash(lname->ln_name,
 177                                                   lname->ln_namelen);
 178                 /* XXX Workaround for LU-2856
 179                  *
 180                  * Zero is a valid return value of full_name_hash, but
 181                  * several users of mlh_pdo_hash assume a non-zero
 182                  * hash value. We therefore map zero onto an
 183                  * arbitrary, but consistent value (1) to avoid
 184                  * problems further down the road. */
 185                 if (unlikely(lh->mlh_pdo_hash == 0))
 186                         lh->mlh_pdo_hash = 1;
 187         } else {
 188                 lh->mlh_pdo_hash = 0;
 189         }
 190 }
 191
 192 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 193                               struct mdt_lock_handle *lh)
 194 {
 195         mdl_mode_t mode;
 196         ENTRY;
 197
 198         /*
 199          * Any dir access needs couple of locks:
 200          *
 201          * 1) on part of dir we gonna take lookup/modify;
 202          *
 203          * 2) on whole dir to protect it from concurrent splitting and/or to
 204          * flush client's cache for readdir().
 205          *
 206          * so, for a given mode and object this routine decides what lock mode
 207          * to use for lock #2:
 208          *
 209          * 1) if caller's gonna lookup in dir then we need to protect dir from
 210          * being splitted only - LCK_CR
 211          *
 212          * 2) if caller's gonna modify dir then we need to protect dir from
 213          * being splitted and to flush cache - LCK_CW
 214          *
 215          * 3) if caller's gonna modify dir and that dir seems ready for
 216          * splitting then we need to protect it from any type of access
 217          * (lookup/modify/split) - LCK_EX --bzzz
 218          */
 219
 220         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 221         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 222
 223         /*
 224          * Ask underlaying level its opinion about preferable PDO lock mode
 225          * having access type passed as regular lock mode:
 226          *
 227          * - MDL_MINMODE means that lower layer does not want to specify lock
 228          * mode;
 229          *
 230          * - MDL_NL means that no PDO lock should be taken. This is used in some
 231          * cases. Say, for non-splittable directories no need to use PDO locks
 232          * at all.
 233          */
 234         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 235                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 236
 237         if (mode != MDL_MINMODE) {
 238                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 239         } else {
 240                 /*
 241                  * Lower layer does not want to specify locking mode. We do it
 242                  * our selves. No special protection is needed, just flush
 243                  * client's cache on modification and allow concurrent
 244                  * mondification.
 245                  */
 246                 switch (lh->mlh_reg_mode) {
 247                 case LCK_EX:
 248                         lh->mlh_pdo_mode = LCK_EX;
 249                         break;
 250                 case LCK_PR:
 251                         lh->mlh_pdo_mode = LCK_CR;
 252                         break;
 253                 case LCK_PW:
 254                         lh->mlh_pdo_mode = LCK_CW;
 255                         break;
 256                 default:
 257                         CERROR("Not expected lock type (0x%x)\n",
 258                                (int)lh->mlh_reg_mode);
 259                         LBUG();
 260                 }
 261         }
 262
 263         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 264         EXIT;
 265 }
 266
 267 static int mdt_getstatus(struct tgt_session_info *tsi)
 268 {
 269         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
 270         struct mdt_device       *mdt = info->mti_mdt;
 271         struct mdt_body         *repbody;
 272         int                      rc;
 273
 274         ENTRY;
 275
 276         rc = mdt_check_ucred(info);
 277         if (rc)
 278                 GOTO(out, rc = err_serious(rc));
 279
 280         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
 281                 GOTO(out, rc = err_serious(-ENOMEM));
 282
 283         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 284         repbody->mbo_fid1 = mdt->mdt_md_root_fid;
 285         repbody->mbo_valid |= OBD_MD_FLID;
 286
 287         if (tsi->tsi_tgt->lut_mds_capa &&
 288             exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) {
 289                 struct mdt_object       *root;
 290                 struct lustre_capa      *capa;
 291
 292                 root = mdt_object_find(info->mti_env, mdt, &repbody->mbo_fid1);
 293                 if (IS_ERR(root))
 294                         GOTO(out, rc = PTR_ERR(root));
 295
 296                 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
 297                 LASSERT(capa);
 298                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
 299                 rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa,
 300                                  0);
 301                 mdt_object_put(info->mti_env, root);
 302                 if (rc == 0)
 303                         repbody->mbo_valid |= OBD_MD_FLMDSCAPA;
 304         }
 305         EXIT;
 306 out:
 307         mdt_thread_info_fini(info);
 308         return rc;
 309 }
 310
 311 static int mdt_statfs(struct tgt_session_info *tsi)
 312 {
 313         struct ptlrpc_request           *req = tgt_ses_req(tsi);
 314         struct mdt_thread_info          *info = tsi2mdt_info(tsi);
 315         struct md_device                *next = info->mti_mdt->mdt_child;
 316         struct ptlrpc_service_part      *svcpt;
 317         struct obd_statfs               *osfs;
 318         int                             rc;
 319
 320         ENTRY;
 321
 322         svcpt = req->rq_rqbd->rqbd_svcpt;
 323
 324         /* This will trigger a watchdog timeout */
 325         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 326                          (MDT_SERVICE_WATCHDOG_FACTOR *
 327                           at_get(&svcpt->scp_at_estimate)) + 1);
 328
 329         rc = mdt_check_ucred(info);
 330         if (rc)
 331                 GOTO(out, rc = err_serious(rc));
 332
 333         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
 334                 GOTO(out, rc = err_serious(-ENOMEM));
 335
 336         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 337         if (!osfs)
 338                 GOTO(out, rc = -EPROTO);
 339
 340         /** statfs information are cached in the mdt_device */
 341         if (cfs_time_before_64(info->mti_mdt->mdt_osfs_age,
 342                                cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS))) {
 343                 /** statfs data is too old, get up-to-date one */
 344                 rc = next->md_ops->mdo_statfs(info->mti_env, next, osfs);
 345                 if (rc)
 346                         GOTO(out, rc);
 347                 spin_lock(&info->mti_mdt->mdt_osfs_lock);
 348                 info->mti_mdt->mdt_osfs = *osfs;
 349                 info->mti_mdt->mdt_osfs_age = cfs_time_current_64();
 350                 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
 351         } else {
 352                 /** use cached statfs data */
 353                 spin_lock(&info->mti_mdt->mdt_osfs_lock);
 354                 *osfs = info->mti_mdt->mdt_osfs;
 355                 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
 356         }
 357
 358         if (rc == 0)
 359                 mdt_counter_incr(req, LPROC_MDT_STATFS);
 360 out:
 361         mdt_thread_info_fini(info);
 362         RETURN(rc);
 363 }
 364
 365 /**
 366  * Pack SOM attributes into the reply.
 367  * Call under a DLM UPDATE lock.
 368  */
 369 static void mdt_pack_size2body(struct mdt_thread_info *info,
 370                                struct mdt_object *mo)
 371 {
 372         struct mdt_body *b;
 373         struct md_attr *ma = &info->mti_attr;
 374
 375         LASSERT(ma->ma_attr.la_valid & LA_MODE);
 376         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 377
 378         /* Check if Size-on-MDS is supported, if this is a regular file,
 379          * if SOM is enabled on the object and if SOM cache exists and valid.
 380          * Otherwise do not pack Size-on-MDS attributes to the reply. */
 381         if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) ||
 382             !S_ISREG(ma->ma_attr.la_mode) ||
 383             !mdt_object_is_som_enabled(mo) ||
 384             !(ma->ma_valid & MA_SOM))
 385                 return;
 386
 387         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 388         b->mbo_size = ma->ma_som->msd_size;
 389         b->mbo_blocks = ma->ma_som->msd_blocks;
 390 }
 391
 392 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 393                         const struct lu_attr *attr, const struct lu_fid *fid)
 394 {
 395         struct md_attr          *ma = &info->mti_attr;
 396         struct obd_export       *exp = info->mti_exp;
 397         struct lu_nodemap       *nodemap = exp->exp_target_data.ted_nodemap;
 398
 399         LASSERT(ma->ma_valid & MA_INODE);
 400
 401         b->mbo_atime      = attr->la_atime;
 402         b->mbo_mtime      = attr->la_mtime;
 403         b->mbo_ctime      = attr->la_ctime;
 404         b->mbo_mode       = attr->la_mode;
 405         b->mbo_size       = attr->la_size;
 406         b->mbo_blocks     = attr->la_blocks;
 407         b->mbo_uid        = nodemap_map_id(nodemap, NODEMAP_UID,
 408                                            NODEMAP_FS_TO_CLIENT,
 409                                            attr->la_uid);
 410         b->mbo_gid        = nodemap_map_id(nodemap, NODEMAP_GID,
 411                                            NODEMAP_FS_TO_CLIENT,
 412                                            attr->la_gid);
 413         b->mbo_flags      = attr->la_flags;
 414         b->mbo_nlink      = attr->la_nlink;
 415         b->mbo_rdev       = attr->la_rdev;
 416
 417         /* XXX should pack the reply body according to lu_valid */
 418         b->mbo_valid |= OBD_MD_FLCTIME | OBD_MD_FLUID   |
 419                         OBD_MD_FLGID   | OBD_MD_FLTYPE  |
 420                         OBD_MD_FLMODE  | OBD_MD_FLNLINK | OBD_MD_FLFLAGS |
 421                         OBD_MD_FLATIME | OBD_MD_FLMTIME ;
 422
 423         if (!S_ISREG(attr->la_mode)) {
 424                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 425         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
 426                 /* means no objects are allocated on osts. */
 427                 LASSERT(!(ma->ma_valid & MA_LOV));
 428                 /* just ignore blocks occupied by extend attributes on MDS */
 429                 b->mbo_blocks = 0;
 430                 /* if no object is allocated on osts, the size on mds is valid.
 431                  * b=22272 */
 432                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 433         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL &&
 434                    ma->ma_lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) {
 435                 /* A released file stores its size on MDS. */
 436                 /* But return 1 block for released file, unless tools like tar
 437                  * will consider it fully sparse. (LU-3864)
 438                  */
 439                 if (unlikely(b->mbo_size == 0))
 440                         b->mbo_blocks = 0;
 441                 else
 442                         b->mbo_blocks = 1;
 443                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 444         }
 445
 446         if (fid) {
 447                 b->mbo_fid1 = *fid;
 448                 b->mbo_valid |= OBD_MD_FLID;
 449                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n",
 450                                 PFID(fid), b->mbo_nlink,
 451                        b->mbo_mode, b->mbo_size);
 452         }
 453
 454         if (info)
 455                 mdt_body_reverse_idmap(info, b);
 456
 457         if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE))
 458                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 459                        PFID(fid), (unsigned long long)b->mbo_size);
 460 }
 461
 462 static inline int mdt_body_has_lov(const struct lu_attr *la,
 463                                    const struct mdt_body *body)
 464 {
 465         return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
 466                (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
 467 }
 468
 469 void mdt_client_compatibility(struct mdt_thread_info *info)
 470 {
 471         struct mdt_body       *body;
 472         struct ptlrpc_request *req = mdt_info_req(info);
 473         struct obd_export     *exp = req->rq_export;
 474         struct md_attr        *ma = &info->mti_attr;
 475         struct lu_attr        *la = &ma->ma_attr;
 476         ENTRY;
 477
 478         if (exp_connect_layout(exp))
 479                 /* the client can deal with 16-bit lmm_stripe_count */
 480                 RETURN_EXIT;
 481
 482         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 483
 484         if (!mdt_body_has_lov(la, body))
 485                 RETURN_EXIT;
 486
 487         /* now we have a reply with a lov for a client not compatible with the
 488          * layout lock so we have to clean the layout generation number */
 489         if (S_ISREG(la->la_mode))
 490                 ma->ma_lmm->lmm_layout_gen = 0;
 491         EXIT;
 492 }
 493
 494 int mdt_attr_get_eabuf_size(struct mdt_thread_info *info, struct mdt_object *o)
 495 {
 496         const struct lu_env *env = info->mti_env;
 497         int rc, rc2;
 498
 499         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 500                           XATTR_NAME_LOV);
 501
 502         if (rc == -ENODATA)
 503                 rc = 0;
 504
 505         if (rc < 0)
 506                 goto out;
 507
 508         /* Is it a directory? Let's check for the LMV as well */
 509         if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
 510                 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 511                                    XATTR_NAME_LMV);
 512
 513                 if (rc2 == -ENODATA)
 514                         rc2 = mo_xattr_get(env, mdt_object_child(o),
 515                                            &LU_BUF_NULL,
 516                                            XATTR_NAME_DEFAULT_LMV);
 517
 518                 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
 519                         rc = rc2;
 520         }
 521
 522 out:
 523         return rc;
 524 }
 525
 526 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
 527                       const char *name)
 528 {
 529         const struct lu_env *env = info->mti_env;
 530         int rc;
 531         ENTRY;
 532
 533         LASSERT(info->mti_big_lmm_used == 0);
 534         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
 535         if (rc < 0)
 536                 RETURN(rc);
 537
 538         /* big_lmm may need to be grown */
 539         if (info->mti_big_lmmsize < rc) {
 540                 int size = size_roundup_power2(rc);
 541
 542                 if (info->mti_big_lmmsize > 0) {
 543                         /* free old buffer */
 544                         LASSERT(info->mti_big_lmm);
 545                         OBD_FREE_LARGE(info->mti_big_lmm,
 546                                        info->mti_big_lmmsize);
 547                         info->mti_big_lmm = NULL;
 548                         info->mti_big_lmmsize = 0;
 549                 }
 550
 551                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
 552                 if (info->mti_big_lmm == NULL)
 553                         RETURN(-ENOMEM);
 554                 info->mti_big_lmmsize = size;
 555         }
 556         LASSERT(info->mti_big_lmmsize >= rc);
 557
 558         info->mti_buf.lb_buf = info->mti_big_lmm;
 559         info->mti_buf.lb_len = info->mti_big_lmmsize;
 560         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
 561
 562         RETURN(rc);
 563 }
 564
 565 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
 566                    struct md_attr *ma, const char *name)
 567 {
 568         struct md_object *next = mdt_object_child(o);
 569         struct lu_buf    *buf = &info->mti_buf;
 570         int rc;
 571
 572         if (strcmp(name, XATTR_NAME_LOV) == 0) {
 573                 buf->lb_buf = ma->ma_lmm;
 574                 buf->lb_len = ma->ma_lmm_size;
 575                 LASSERT(!(ma->ma_valid & MA_LOV));
 576         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 577                 buf->lb_buf = ma->ma_lmv;
 578                 buf->lb_len = ma->ma_lmv_size;
 579                 LASSERT(!(ma->ma_valid & MA_LMV));
 580         } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 581                 buf->lb_buf = ma->ma_lmv;
 582                 buf->lb_len = ma->ma_lmv_size;
 583                 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
 584         } else {
 585                 return -EINVAL;
 586         }
 587
 588         rc = mo_xattr_get(info->mti_env, next, buf, name);
 589         if (rc > 0) {
 590
 591 got:
 592                 if (strcmp(name, XATTR_NAME_LOV) == 0) {
 593                         if (info->mti_big_lmm_used)
 594                                 ma->ma_lmm = info->mti_big_lmm;
 595
 596                         /* NOT return LOV EA with hole to old client. */
 597                         if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
 598                                      LOV_PATTERN_F_HOLE) &&
 599                             !(exp_connect_flags(info->mti_exp) &
 600                               OBD_CONNECT_LFSCK)) {
 601                                 return -EIO;
 602                         } else {
 603                                 ma->ma_lmm_size = rc;
 604                                 ma->ma_valid |= MA_LOV;
 605                         }
 606                 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 607                         if (info->mti_big_lmm_used)
 608                                 ma->ma_lmv = info->mti_big_lmm;
 609
 610                         ma->ma_lmv_size = rc;
 611                         ma->ma_valid |= MA_LMV;
 612                 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 613                         ma->ma_lmv_size = rc;
 614                         ma->ma_valid |= MA_LMV_DEF;
 615                 }
 616
 617                 /* Update mdt_max_mdsize so all clients will be aware that */
 618                 if (info->mti_mdt->mdt_max_mdsize < rc)
 619                         info->mti_mdt->mdt_max_mdsize = rc;
 620
 621                 rc = 0;
 622         } else if (rc == -ENODATA) {
 623                 /* no LOV EA */
 624                 rc = 0;
 625         } else if (rc == -ERANGE) {
 626                 /* Default LMV has fixed size, so it must be able to fit
 627                  * in the original buffer */
 628                 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
 629                         return rc;
 630                 rc = mdt_big_xattr_get(info, o, name);
 631                 if (rc > 0) {
 632                         info->mti_big_lmm_used = 1;
 633                         goto got;
 634                 }
 635         }
 636
 637         return rc;
 638 }
 639
 640 static int mdt_attr_get_pfid(struct mdt_thread_info *info,
 641                              struct mdt_object *o, struct lu_fid *pfid)
 642 {
 643         struct lu_buf           *buf = &info->mti_buf;
 644         struct link_ea_header   *leh;
 645         struct link_ea_entry    *lee;
 646         int                      rc;
 647         ENTRY;
 648
 649         buf->lb_buf = info->mti_big_lmm;
 650         buf->lb_len = info->mti_big_lmmsize;
 651         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
 652                           buf, XATTR_NAME_LINK);
 653         /* ignore errors, MA_PFID won't be set and it is
 654          * up to the caller to treat this as an error */
 655         if (rc == -ERANGE || buf->lb_len == 0) {
 656                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
 657                 buf->lb_buf = info->mti_big_lmm;
 658                 buf->lb_len = info->mti_big_lmmsize;
 659         }
 660
 661         if (rc < 0)
 662                 RETURN(rc);
 663         if (rc < sizeof(*leh)) {
 664                 CERROR("short LinkEA on "DFID": rc = %d\n",
 665                        PFID(mdt_object_fid(o)), rc);
 666                 RETURN(-ENODATA);
 667         }
 668
 669         leh = (struct link_ea_header *) buf->lb_buf;
 670         lee = (struct link_ea_entry *)(leh + 1);
 671         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
 672                 leh->leh_magic = LINK_EA_MAGIC;
 673                 leh->leh_reccount = __swab32(leh->leh_reccount);
 674                 leh->leh_len = __swab64(leh->leh_len);
 675         }
 676         if (leh->leh_magic != LINK_EA_MAGIC)
 677                 RETURN(-EINVAL);
 678         if (leh->leh_reccount == 0)
 679                 RETURN(-ENODATA);
 680
 681         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
 682         fid_be_to_cpu(pfid, pfid);
 683
 684         RETURN(0);
 685 }
 686
 687 int mdt_attr_get_complex(struct mdt_thread_info *info,
 688                          struct mdt_object *o, struct md_attr *ma)
 689 {
 690         const struct lu_env *env = info->mti_env;
 691         struct md_object    *next = mdt_object_child(o);
 692         struct lu_buf       *buf = &info->mti_buf;
 693         int                  need = ma->ma_need;
 694         int                  rc = 0, rc2;
 695         u32                  mode;
 696         ENTRY;
 697
 698         ma->ma_valid = 0;
 699
 700         if (mdt_object_exists(o) == 0)
 701                 GOTO(out, rc = -ENOENT);
 702         mode = lu_object_attr(&next->mo_lu);
 703
 704         if (need & MA_INODE) {
 705                 ma->ma_need = MA_INODE;
 706                 rc = mo_attr_get(env, next, ma);
 707                 if (rc)
 708                         GOTO(out, rc);
 709                 ma->ma_valid |= MA_INODE;
 710         }
 711
 712         if (need & MA_PFID) {
 713                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
 714                 if (rc == 0)
 715                         ma->ma_valid |= MA_PFID;
 716                 /* ignore this error, parent fid is not mandatory */
 717                 rc = 0;
 718         }
 719
 720         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
 721                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
 722                 if (rc)
 723                         GOTO(out, rc);
 724         }
 725
 726         if (need & MA_LMV && S_ISDIR(mode)) {
 727                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
 728                 if (rc != 0)
 729                         GOTO(out, rc);
 730         }
 731
 732         if (need & MA_LMV_DEF && S_ISDIR(mode)) {
 733                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
 734                 if (rc != 0)
 735                         GOTO(out, rc);
 736         }
 737
 738         if (need & MA_SOM && S_ISREG(mode)) {
 739                 buf->lb_buf = info->mti_xattr_buf;
 740                 buf->lb_len = sizeof(info->mti_xattr_buf);
 741                 CLASSERT(sizeof(struct som_attrs) <=
 742                          sizeof(info->mti_xattr_buf));
 743                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_SOM);
 744                 rc2 = lustre_buf2som(info->mti_xattr_buf, rc2, ma->ma_som);
 745                 if (rc2 == 0)
 746                         ma->ma_valid |= MA_SOM;
 747                 else if (rc2 < 0 && rc2 != -ENODATA)
 748                         GOTO(out, rc = rc2);
 749         }
 750
 751         if (need & MA_HSM && S_ISREG(mode)) {
 752                 buf->lb_buf = info->mti_xattr_buf;
 753                 buf->lb_len = sizeof(info->mti_xattr_buf);
 754                 CLASSERT(sizeof(struct hsm_attrs) <=
 755                          sizeof(info->mti_xattr_buf));
 756                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
 757                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
 758                 if (rc2 == 0)
 759                         ma->ma_valid |= MA_HSM;
 760                 else if (rc2 < 0 && rc2 != -ENODATA)
 761                         GOTO(out, rc = rc2);
 762         }
 763
 764 #ifdef CONFIG_FS_POSIX_ACL
 765         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
 766                 buf->lb_buf = ma->ma_acl;
 767                 buf->lb_len = ma->ma_acl_size;
 768                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
 769                 if (rc2 > 0) {
 770                         ma->ma_acl_size = rc2;
 771                         ma->ma_valid |= MA_ACL_DEF;
 772                 } else if (rc2 == -ENODATA) {
 773                         /* no ACLs */
 774                         ma->ma_acl_size = 0;
 775                 } else
 776                         GOTO(out, rc = rc2);
 777         }
 778 #endif
 779 out:
 780         ma->ma_need = need;
 781         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
 782                rc, ma->ma_valid, ma->ma_lmm);
 783         RETURN(rc);
 784 }
 785
 786 static int mdt_getattr_internal(struct mdt_thread_info *info,
 787                                 struct mdt_object *o, int ma_need)
 788 {
 789         struct md_object        *next = mdt_object_child(o);
 790         const struct mdt_body   *reqbody = info->mti_body;
 791         struct ptlrpc_request   *req = mdt_info_req(info);
 792         struct md_attr          *ma = &info->mti_attr;
 793         struct lu_attr          *la = &ma->ma_attr;
 794         struct req_capsule      *pill = info->mti_pill;
 795         const struct lu_env     *env = info->mti_env;
 796         struct mdt_body         *repbody;
 797         struct lu_buf           *buffer = &info->mti_buf;
 798         struct obd_export       *exp = info->mti_exp;
 799         struct lu_nodemap       *nodemap = exp->exp_target_data.ted_nodemap;
 800         int                      rc;
 801         int                      is_root;
 802         ENTRY;
 803
 804         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
 805                 RETURN(err_serious(-ENOMEM));
 806
 807         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 808
 809         ma->ma_valid = 0;
 810
 811         if (mdt_object_remote(o)) {
 812                 /* This object is located on remote node.*/
 813                 /* Return -ENOTSUPP for old client */
 814                 if (!mdt_is_dne_client(req->rq_export))
 815                         GOTO(out, rc = -ENOTSUPP);
 816
 817                 repbody->mbo_fid1 = *mdt_object_fid(o);
 818                 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
 819                 GOTO(out, rc = 0);
 820         }
 821
 822         if (reqbody->mbo_eadatasize > 0) {
 823                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
 824                 if (buffer->lb_buf == NULL)
 825                         GOTO(out, rc = -EPROTO);
 826                 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
 827                                                       RCL_SERVER);
 828         } else {
 829                 buffer->lb_buf = NULL;
 830                 buffer->lb_len = 0;
 831                 ma_need &= ~(MA_LOV | MA_LMV);
 832                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
 833                        mdt_obd_name(info->mti_mdt),
 834                        req->rq_export->exp_client_uuid.uuid);
 835         }
 836
 837         /* If it is dir object and client require MEA, then we got MEA */
 838         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 839             (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
 840                 /* Assumption: MDT_MD size is enough for lmv size. */
 841                 ma->ma_lmv = buffer->lb_buf;
 842                 ma->ma_lmv_size = buffer->lb_len;
 843                 ma->ma_need = MA_INODE;
 844                 if (ma->ma_lmv_size > 0) {
 845                         if (reqbody->mbo_valid & OBD_MD_MEA)
 846                                 ma->ma_need |= MA_LMV;
 847                         else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA)
 848                                 ma->ma_need |= MA_LMV_DEF;
 849                 }
 850         } else {
 851                 ma->ma_lmm = buffer->lb_buf;
 852                 ma->ma_lmm_size = buffer->lb_len;
 853                 ma->ma_need = MA_INODE | MA_HSM;
 854                 if (ma->ma_lmm_size > 0)
 855                         ma->ma_need |= MA_LOV;
 856         }
 857
 858         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 859             reqbody->mbo_valid & OBD_MD_FLDIREA  &&
 860             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
 861                 /* get default stripe info for this dir. */
 862                 ma->ma_need |= MA_LOV_DEF;
 863         }
 864         ma->ma_need |= ma_need;
 865         if (ma->ma_need & MA_SOM)
 866                 ma->ma_som = &info->mti_u.som.data;
 867
 868         rc = mdt_attr_get_complex(info, o, ma);
 869         if (unlikely(rc)) {
 870                 CERROR("%s: getattr error for "DFID": rc = %d\n",
 871                        mdt_obd_name(info->mti_mdt),
 872                        PFID(mdt_object_fid(o)), rc);
 873                 RETURN(rc);
 874         }
 875
 876         /* if file is released, check if a restore is running */
 877         if ((ma->ma_valid & MA_HSM) && (ma->ma_hsm.mh_flags & HS_RELEASED) &&
 878             mdt_hsm_restore_is_running(info, mdt_object_fid(o))) {
 879                 repbody->mbo_t_state = MS_RESTORE;
 880                 repbody->mbo_valid |= OBD_MD_TSTATE;
 881         }
 882
 883         is_root = lu_fid_eq(mdt_object_fid(o), &info->mti_mdt->mdt_md_root_fid);
 884
 885         /* the Lustre protocol supposes to return default striping
 886          * on the user-visible root if explicitly requested */
 887         if ((ma->ma_valid & MA_LOV) == 0 && S_ISDIR(la->la_mode) &&
 888             (ma->ma_need & MA_LOV_DEF && is_root) && ma->ma_need & MA_LOV) {
 889                 struct lu_fid      rootfid;
 890                 struct mdt_object *root;
 891                 struct mdt_device *mdt = info->mti_mdt;
 892
 893                 rc = dt_root_get(env, mdt->mdt_bottom, &rootfid);
 894                 if (rc)
 895                         RETURN(rc);
 896                 root = mdt_object_find(env, mdt, &rootfid);
 897                 if (IS_ERR(root))
 898                         RETURN(PTR_ERR(root));
 899                 rc = mdt_stripe_get(info, root, ma, XATTR_NAME_LOV);
 900                 mdt_object_put(info->mti_env, root);
 901                 if (unlikely(rc)) {
 902                         CERROR("%s: getattr error for "DFID": rc = %d\n",
 903                                mdt_obd_name(info->mti_mdt),
 904                                PFID(mdt_object_fid(o)), rc);
 905                         RETURN(rc);
 906                 }
 907         }
 908
 909         if (likely(ma->ma_valid & MA_INODE))
 910                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
 911         else
 912                 RETURN(-EFAULT);
 913
 914         if (mdt_body_has_lov(la, reqbody)) {
 915                 if (ma->ma_valid & MA_LOV) {
 916                         LASSERT(ma->ma_lmm_size);
 917                         repbody->mbo_eadatasize = ma->ma_lmm_size;
 918                         if (S_ISDIR(la->la_mode))
 919                                 repbody->mbo_valid |= OBD_MD_FLDIREA;
 920                         else
 921                                 repbody->mbo_valid |= OBD_MD_FLEASIZE;
 922                         mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
 923                 }
 924                 if (ma->ma_valid & MA_LMV) {
 925                         /* Return -ENOTSUPP for old client */
 926                         if (!mdt_is_striped_client(req->rq_export))
 927                                 RETURN(-ENOTSUPP);
 928
 929                         LASSERT(S_ISDIR(la->la_mode));
 930                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
 931                         repbody->mbo_eadatasize = ma->ma_lmv_size;
 932                         repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
 933                 }
 934                 if (ma->ma_valid & MA_LMV_DEF) {
 935                         /* Return -ENOTSUPP for old client */
 936                         if (!mdt_is_striped_client(req->rq_export))
 937                                 RETURN(-ENOTSUPP);
 938                         LASSERT(S_ISDIR(la->la_mode));
 939                         repbody->mbo_eadatasize = ma->ma_lmv_size;
 940                         repbody->mbo_valid |= (OBD_MD_FLDIREA |
 941                                                OBD_MD_DEFAULT_MEA);
 942                 }
 943         } else if (S_ISLNK(la->la_mode) &&
 944                    reqbody->mbo_valid & OBD_MD_LINKNAME) {
 945                 buffer->lb_buf = ma->ma_lmm;
 946                 /* eadatasize from client includes NULL-terminator, so
 947                  * there is no need to read it */
 948                 buffer->lb_len = reqbody->mbo_eadatasize - 1;
 949                 rc = mo_readlink(env, next, buffer);
 950                 if (unlikely(rc <= 0)) {
 951                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
 952                                mdt_obd_name(info->mti_mdt),
 953                                PFID(mdt_object_fid(o)), rc);
 954                         rc = -EFAULT;
 955                 } else {
 956                         int print_limit = min_t(int, PAGE_CACHE_SIZE - 128, rc);
 957
 958                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
 959                                 rc -= 2;
 960                         repbody->mbo_valid |= OBD_MD_LINKNAME;
 961                         /* we need to report back size with NULL-terminator
 962                          * because client expects that */
 963                         repbody->mbo_eadatasize = rc + 1;
 964                         if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
 965                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
 966                                        "on "DFID ", expected %d\n",
 967                                        mdt_obd_name(info->mti_mdt),
 968                                        rc, PFID(mdt_object_fid(o)),
 969                                        reqbody->mbo_eadatasize - 1);
 970                         /* NULL terminate */
 971                         ((char *)ma->ma_lmm)[rc] = 0;
 972
 973                         /* If the total CDEBUG() size is larger than a page, it
 974                          * will print a warning to the console, avoid this by
 975                          * printing just the last part of the symlink. */
 976                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
 977                                print_limit < rc ? "..." : "", print_limit,
 978                                (char *)ma->ma_lmm + rc - print_limit, rc);
 979                         rc = 0;
 980                 }
 981         }
 982
 983         if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
 984                 repbody->mbo_max_cookiesize = 0;
 985                 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
 986                 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
 987                 CDEBUG(D_INODE, "I am going to change the MAX_MD_SIZE & "
 988                        "MAX_COOKIE to : %d:%d\n", repbody->mbo_max_mdsize,
 989                        repbody->mbo_max_cookiesize);
 990         }
 991
 992         if (exp_connect_rmtclient(info->mti_exp) &&
 993             reqbody->mbo_valid & OBD_MD_FLRMTPERM) {
 994                 void *buf = req_capsule_server_get(pill, &RMF_ACL);
 995
 996                 /* mdt_getattr_lock only */
 997                 rc = mdt_pack_remote_perm(info, o, buf);
 998                 if (rc) {
 999                         repbody->mbo_valid &= ~OBD_MD_FLRMTPERM;
1000                         repbody->mbo_aclsize = 0;
1001                         RETURN(rc);
1002                 } else {
1003                         repbody->mbo_valid |= OBD_MD_FLRMTPERM;
1004                         repbody->mbo_aclsize = sizeof(struct mdt_remote_perm);
1005                 }
1006         }
1007 #ifdef CONFIG_FS_POSIX_ACL
1008         else if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1009                  (reqbody->mbo_valid & OBD_MD_FLACL)) {
1010                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
1011                 buffer->lb_len = req_capsule_get_size(pill,
1012                                                       &RMF_ACL, RCL_SERVER);
1013                 if (buffer->lb_len > 0) {
1014                         rc = mo_xattr_get(env, next, buffer,
1015                                           XATTR_NAME_ACL_ACCESS);
1016                         if (rc < 0) {
1017                                 if (rc == -ENODATA) {
1018                                         repbody->mbo_aclsize = 0;
1019                                         repbody->mbo_valid |= OBD_MD_FLACL;
1020                                         rc = 0;
1021                                 } else if (rc == -EOPNOTSUPP) {
1022                                         rc = 0;
1023                                 } else {
1024                                         CERROR("%s: unable to read "DFID
1025                                                " ACL: rc = %d\n",
1026                                                mdt_obd_name(info->mti_mdt),
1027                                                PFID(mdt_object_fid(o)), rc);
1028                                 }
1029                         } else {
1030                                 rc = nodemap_map_acl(nodemap, buffer->lb_buf,
1031                                                      rc, NODEMAP_FS_TO_CLIENT);
1032                                 /* if all ACLs mapped out, rc is still >= 0 */
1033                                 if (rc < 0) {
1034                                         CERROR("%s: nodemap_map_acl unable to"
1035                                                " parse "DFID" ACL: rc = %d\n",
1036                                                mdt_obd_name(info->mti_mdt),
1037                                                PFID(mdt_object_fid(o)), rc);
1038                                 } else {
1039                                         repbody->mbo_aclsize = rc;
1040                                         repbody->mbo_valid |= OBD_MD_FLACL;
1041                                         rc = 0;
1042                                 }
1043                         }
1044                 }
1045         }
1046 #endif
1047
1048         if (reqbody->mbo_valid & OBD_MD_FLMDSCAPA &&
1049             info->mti_mdt->mdt_lut.lut_mds_capa &&
1050             exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) {
1051                 struct lustre_capa *capa;
1052
1053                 capa = req_capsule_server_get(pill, &RMF_CAPA1);
1054                 LASSERT(capa);
1055                 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
1056                 rc = mo_capa_get(env, next, capa, 0);
1057                 if (rc)
1058                         RETURN(rc);
1059                 repbody->mbo_valid |= OBD_MD_FLMDSCAPA;
1060         }
1061
1062 out:
1063         if (rc == 0)
1064                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
1065
1066         RETURN(rc);
1067 }
1068
1069 static int mdt_renew_capa(struct mdt_thread_info *info)
1070 {
1071         struct mdt_object  *obj = info->mti_object;
1072         struct mdt_body    *body;
1073         struct lustre_capa *capa, *c;
1074         int rc;
1075         ENTRY;
1076
1077         /* if object doesn't exist, or server has disabled capability,
1078          * return directly, client will find body->valid OBD_MD_FLOSSCAPA
1079          * flag not set.
1080          */
1081         if (!obj || !info->mti_mdt->mdt_lut.lut_oss_capa ||
1082             !(exp_connect_flags(info->mti_exp) & OBD_CONNECT_OSS_CAPA))
1083                 RETURN(0);
1084
1085         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1086         LASSERT(body != NULL);
1087
1088         c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1);
1089         LASSERT(c);
1090
1091         capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
1092         LASSERT(capa);
1093
1094         *capa = *c;
1095         rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1);
1096         if (rc == 0)
1097                 body->mbo_valid |= OBD_MD_FLOSSCAPA;
1098         RETURN(rc);
1099 }
1100
1101 static int mdt_getattr(struct tgt_session_info *tsi)
1102 {
1103         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1104         struct mdt_object       *obj = info->mti_object;
1105         struct req_capsule      *pill = info->mti_pill;
1106         struct mdt_body         *reqbody;
1107         struct mdt_body         *repbody;
1108         int rc, rc2;
1109         ENTRY;
1110
1111         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1112         LASSERT(reqbody);
1113
1114         if (reqbody->mbo_valid & OBD_MD_FLOSSCAPA) {
1115                 rc = req_capsule_server_pack(pill);
1116                 if (unlikely(rc))
1117                         RETURN(err_serious(rc));
1118                 rc = mdt_renew_capa(info);
1119                 GOTO(out_shrink, rc);
1120         }
1121
1122         LASSERT(obj != NULL);
1123         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1124
1125         /* Unlike intent case where we need to pre-fill out buffers early on
1126          * in intent policy for ldlm reasons, here we can have a much better
1127          * guess at EA size by just reading it from disk.
1128          * Exceptions are readdir and (missing) directory striping */
1129         /* Readlink */
1130         if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1131                 /* No easy way to know how long is the symlink, but it cannot
1132                  * be more than PATH_MAX, so we allocate +1 */
1133                 rc = PATH_MAX + 1;
1134
1135         /* A special case for fs ROOT: getattr there might fetch
1136          * default EA for entire fs, not just for this dir!
1137          */
1138         } else if (lu_fid_eq(mdt_object_fid(obj),
1139                              &info->mti_mdt->mdt_md_root_fid) &&
1140                    (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1141                    (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1142                                                                  MDS_GETATTR)) {
1143                 /* Should the default strping be bigger, mdt_fix_reply
1144                  * will reallocate */
1145                 rc = DEF_REP_MD_SIZE;
1146         } else {
1147                 /* Read the actual EA size from disk */
1148                 rc = mdt_attr_get_eabuf_size(info, obj);
1149         }
1150
1151         if (rc < 0)
1152                 GOTO(out, rc = err_serious(rc));
1153
1154         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1155
1156         rc = req_capsule_server_pack(pill);
1157         if (unlikely(rc != 0))
1158                 GOTO(out, rc = err_serious(rc));
1159
1160         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1161         LASSERT(repbody != NULL);
1162         repbody->mbo_eadatasize = 0;
1163         repbody->mbo_aclsize = 0;
1164
1165         if (reqbody->mbo_valid & OBD_MD_FLRMTPERM)
1166                 rc = mdt_init_ucred(info, reqbody);
1167         else
1168                 rc = mdt_check_ucred(info);
1169         if (unlikely(rc))
1170                 GOTO(out_shrink, rc);
1171
1172         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1173
1174         /*
1175          * Don't check capability at all, because rename might getattr for
1176          * remote obj, and at that time no capability is available.
1177          */
1178         mdt_set_capainfo(info, 1, &reqbody->mbo_fid1, BYPASS_CAPA);
1179         rc = mdt_getattr_internal(info, obj, 0);
1180         if (reqbody->mbo_valid & OBD_MD_FLRMTPERM)
1181                 mdt_exit_ucred(info);
1182         EXIT;
1183 out_shrink:
1184         mdt_client_compatibility(info);
1185         rc2 = mdt_fix_reply(info);
1186         if (rc == 0)
1187                 rc = rc2;
1188 out:
1189         mdt_thread_info_fini(info);
1190         return rc;
1191 }
1192
1193 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1194 {
1195         struct mdt_thread_info  *info;
1196         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1197         struct obd_export       *exp = req->rq_export;
1198         struct mdt_object       *o1, *o2, *o;
1199         struct mdt_lock_handle  *lh1, *lh2;
1200         struct mdc_swap_layouts *msl;
1201         int                      rc;
1202         ENTRY;
1203
1204         /* client does not support layout lock, so layout swaping
1205          * is disabled.
1206          * FIXME: there is a problem for old clients which don't support
1207          * layout lock yet. If those clients have already opened the file
1208          * they won't be notified at all so that old layout may still be
1209          * used to do IO. This can be fixed after file release is landed by
1210          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1211         if (!exp_connect_layout(exp))
1212                 RETURN(-EOPNOTSUPP);
1213
1214         info = tsi2mdt_info(tsi);
1215
1216         if (info->mti_dlm_req != NULL)
1217                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1218
1219         if (req_capsule_get_size(info->mti_pill, &RMF_CAPA1, RCL_CLIENT))
1220                 mdt_set_capainfo(info, 0, &info->mti_body->mbo_fid1,
1221                                  req_capsule_client_get(info->mti_pill,
1222                                                         &RMF_CAPA1));
1223
1224         if (req_capsule_get_size(info->mti_pill, &RMF_CAPA2, RCL_CLIENT))
1225                 mdt_set_capainfo(info, 1, &info->mti_body->mbo_fid2,
1226                                  req_capsule_client_get(info->mti_pill,
1227                                                         &RMF_CAPA2));
1228
1229         o1 = info->mti_object;
1230         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1231                                 &info->mti_body->mbo_fid2);
1232         if (IS_ERR(o))
1233                 GOTO(out, rc = PTR_ERR(o));
1234
1235         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1236                 GOTO(put, rc = -ENOENT);
1237
1238         rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1239         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1240                 GOTO(put, rc);
1241
1242         if (rc < 0)
1243                 swap(o1, o2);
1244
1245         /* permission check. Make sure the calling process having permission
1246          * to write both files. */
1247         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1248                                 MAY_WRITE);
1249         if (rc < 0)
1250                 GOTO(put, rc);
1251
1252         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1253                                 MAY_WRITE);
1254         if (rc < 0)
1255                 GOTO(put, rc);
1256
1257         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1258         if (msl == NULL)
1259                 GOTO(put, rc = -EPROTO);
1260
1261         lh1 = &info->mti_lh[MDT_LH_NEW];
1262         mdt_lock_reg_init(lh1, LCK_EX);
1263         lh2 = &info->mti_lh[MDT_LH_OLD];
1264         mdt_lock_reg_init(lh2, LCK_EX);
1265
1266         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1267                              MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK);
1268         if (rc < 0)
1269                 GOTO(put, rc);
1270
1271         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1272                              MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK);
1273         if (rc < 0)
1274                 GOTO(unlock1, rc);
1275
1276         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1277                              mdt_object_child(o2), msl->msl_flags);
1278         GOTO(unlock2, rc);
1279 unlock2:
1280         mdt_object_unlock(info, o2, lh2, rc);
1281 unlock1:
1282         mdt_object_unlock(info, o1, lh1, rc);
1283 put:
1284         mdt_object_put(info->mti_env, o);
1285 out:
1286         mdt_thread_info_fini(info);
1287         RETURN(rc);
1288 }
1289
1290 static int mdt_raw_lookup(struct mdt_thread_info *info,
1291                           struct mdt_object *parent,
1292                           const struct lu_name *lname,
1293                           struct ldlm_reply *ldlm_rep)
1294 {
1295         struct md_object *next = mdt_object_child(info->mti_object);
1296         const struct mdt_body *reqbody = info->mti_body;
1297         struct lu_fid *child_fid = &info->mti_tmp_fid1;
1298         struct mdt_body *repbody;
1299         int rc;
1300         ENTRY;
1301
1302         if (reqbody->mbo_valid != OBD_MD_FLID)
1303                 RETURN(0);
1304
1305         LASSERT(!info->mti_cross_ref);
1306
1307         /* Only got the fid of this obj by name */
1308         fid_zero(child_fid);
1309         rc = mdo_lookup(info->mti_env, next, lname, child_fid,
1310                         &info->mti_spec);
1311 #if 0
1312         /* XXX is raw_lookup possible as intent operation? */
1313         if (rc != 0) {
1314                 if (rc == -ENOENT)
1315                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1316                 RETURN(rc);
1317         } else
1318                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1319
1320         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1321 #endif
1322         if (rc == 0) {
1323                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1324                 repbody->mbo_fid1 = *child_fid;
1325                 repbody->mbo_valid = OBD_MD_FLID;
1326         }
1327         RETURN(1);
1328 }
1329
1330 /*
1331  * UPDATE lock should be taken against parent, and be release before exit;
1332  * child_bits lock should be taken against child, and be returned back:
1333  *            (1)normal request should release the child lock;
1334  *            (2)intent request will grant the lock to client.
1335  */
1336 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1337                                  struct mdt_lock_handle *lhc,
1338                                  __u64 child_bits,
1339                                  struct ldlm_reply *ldlm_rep)
1340 {
1341         struct ptlrpc_request  *req       = mdt_info_req(info);
1342         struct mdt_body        *reqbody   = NULL;
1343         struct mdt_object      *parent    = info->mti_object;
1344         struct mdt_object      *child;
1345         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1346         struct lu_name         *lname     = NULL;
1347         struct mdt_lock_handle *lhp       = NULL;
1348         struct ldlm_lock       *lock;
1349         bool                    is_resent;
1350         bool                    try_layout;
1351         int                     ma_need = 0;
1352         int                     rc;
1353         ENTRY;
1354
1355         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1356         LASSERT(ergo(is_resent,
1357                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1358
1359         if (parent == NULL)
1360                 RETURN(-ENOENT);
1361
1362         if (info->mti_cross_ref) {
1363                 /* Only getattr on the child. Parent is on another node. */
1364                 mdt_set_disposition(info, ldlm_rep,
1365                                     DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1366                 child = parent;
1367                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1368                        "ldlm_rep = %p\n",
1369                        PFID(mdt_object_fid(child)), ldlm_rep);
1370
1371                 rc = mdt_check_resent_lock(info, child, lhc);
1372                 if (rc < 0) {
1373                         RETURN(rc);
1374                 } else if (rc > 0) {
1375                         mdt_lock_handle_init(lhc);
1376                         mdt_lock_reg_init(lhc, LCK_PR);
1377
1378                         /*
1379                          * Object's name is on another MDS, no lookup or layout
1380                          * lock is needed here but update lock is.
1381                          */
1382                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
1383                                         MDS_INODELOCK_LAYOUT);
1384                         child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1385
1386                         rc = mdt_object_lock(info, child, lhc, child_bits,
1387                                              MDT_LOCAL_LOCK);
1388                         if (rc < 0)
1389                                 RETURN(rc);
1390                 }
1391
1392                 /* Finally, we can get attr for child. */
1393                 if (!mdt_object_exists(child)) {
1394                         LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1395                                         &child->mot_obj,
1396                                         "remote object doesn't exist.\n");
1397                         mdt_object_unlock(info, child, lhc, 1);
1398                         RETURN(-ENOENT);
1399                 }
1400
1401                 mdt_set_capainfo(info, 0, mdt_object_fid(child), BYPASS_CAPA);
1402                 rc = mdt_getattr_internal(info, child, 0);
1403                 if (unlikely(rc != 0))
1404                         mdt_object_unlock(info, child, lhc, 1);
1405
1406                 RETURN(rc);
1407         }
1408
1409         lname = &info->mti_name;
1410         mdt_name_unpack(info->mti_pill, &RMF_NAME, lname, MNF_FIX_ANON);
1411
1412         if (lu_name_is_valid(lname)) {
1413                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1414                        "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1415                        PNAME(lname), ldlm_rep);
1416         } else {
1417                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1418                 if (unlikely(reqbody == NULL))
1419                         RETURN(err_serious(-EPROTO));
1420
1421                 *child_fid = reqbody->mbo_fid2;
1422
1423                 if (unlikely(!fid_is_sane(child_fid)))
1424                         RETURN(err_serious(-EINVAL));
1425
1426                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1427                        "ldlm_rep = %p\n",
1428                        PFID(mdt_object_fid(parent)),
1429                        PFID(&reqbody->mbo_fid2), ldlm_rep);
1430         }
1431
1432         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1433
1434         if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1435                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1436                                 &parent->mot_obj,
1437                                 "Parent doesn't exist!\n");
1438                 RETURN(-ESTALE);
1439         }
1440
1441         if (mdt_object_remote(parent)) {
1442                 CERROR("%s: parent "DFID" is on remote target\n",
1443                        mdt_obd_name(info->mti_mdt),
1444                        PFID(mdt_object_fid(parent)));
1445                 RETURN(-EIO);
1446         }
1447
1448         if (lu_name_is_valid(lname)) {
1449                 rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1450                 if (rc != 0) {
1451                         if (rc > 0)
1452                                 rc = 0;
1453                         RETURN(rc);
1454                 }
1455
1456                 /* step 1: lock parent only if parent is a directory */
1457                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1458                         lhp = &info->mti_lh[MDT_LH_PARENT];
1459                         mdt_lock_pdo_init(lhp, LCK_PR, lname);
1460                         rc = mdt_object_lock(info, parent, lhp,
1461                                              MDS_INODELOCK_UPDATE,
1462                                              MDT_LOCAL_LOCK);
1463                         if (unlikely(rc != 0))
1464                                 RETURN(rc);
1465                 }
1466
1467                 /* step 2: lookup child's fid by name */
1468                 fid_zero(child_fid);
1469                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
1470                                 child_fid, &info->mti_spec);
1471                 if (rc == -ENOENT)
1472                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1473
1474                 if (rc != 0)
1475                         GOTO(out_parent, rc);
1476         }
1477
1478         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1479
1480         /*
1481          *step 3: find the child object by fid & lock it.
1482          *        regardless if it is local or remote.
1483          *
1484          *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1485          *      set parent dir fid the same as child fid in getattr by fid case
1486          *      we should not lu_object_find() the object again, could lead
1487          *      to hung if there is a concurrent unlink destroyed the object.
1488          */
1489         if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1490                 mdt_object_get(info->mti_env, parent);
1491                 child = parent;
1492         } else {
1493                 child = mdt_object_find(info->mti_env, info->mti_mdt,
1494                                         child_fid);
1495         }
1496
1497         if (unlikely(IS_ERR(child)))
1498                 GOTO(out_parent, rc = PTR_ERR(child));
1499
1500         rc = mdt_check_resent_lock(info, child, lhc);
1501         if (rc < 0) {
1502                 GOTO(out_child, rc);
1503         } else if (rc > 0) {
1504                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout*2);
1505                 mdt_lock_handle_init(lhc);
1506                 mdt_lock_reg_init(lhc, LCK_PR);
1507                 try_layout = false;
1508
1509                 if (!mdt_object_exists(child)) {
1510                         LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1511                                         &child->mot_obj,
1512                                         "Object doesn't exist!\n");
1513                         GOTO(out_child, rc = -ENOENT);
1514                 }
1515
1516                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1517                       mdt_object_exists(child) && !mdt_object_remote(child)) {
1518                         struct md_attr *ma = &info->mti_attr;
1519
1520                         ma->ma_valid = 0;
1521                         ma->ma_need = MA_INODE;
1522                         rc = mdt_attr_get_complex(info, child, ma);
1523                         if (unlikely(rc != 0))
1524                                 GOTO(out_child, rc);
1525
1526                         /* If the file has not been changed for some time, we
1527                          * return not only a LOOKUP lock, but also an UPDATE
1528                          * lock and this might save us RPC on later STAT. For
1529                          * directories, it also let negative dentry cache start
1530                          * working for this dir. */
1531                         if (ma->ma_valid & MA_INODE &&
1532                             ma->ma_attr.la_valid & LA_CTIME &&
1533                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1534                                 ma->ma_attr.la_ctime < cfs_time_current_sec())
1535                                 child_bits |= MDS_INODELOCK_UPDATE;
1536                 }
1537
1538                 /* layout lock must be granted in a best-effort way
1539                  * for IT operations */
1540                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1541                 if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1542                     exp_connect_layout(info->mti_exp) &&
1543                     S_ISREG(lu_object_attr(&child->mot_obj)) &&
1544                     !mdt_object_remote(child) && ldlm_rep != NULL) {
1545                         /* try to grant layout lock for regular file. */
1546                         try_layout = true;
1547                 }
1548
1549                 rc = 0;
1550                 if (try_layout) {
1551                         child_bits |= MDS_INODELOCK_LAYOUT;
1552                         /* try layout lock, it may fail to be granted due to
1553                          * contention at LOOKUP or UPDATE */
1554                         if (!mdt_object_lock_try(info, child, lhc, child_bits,
1555                                                  MDT_CROSS_LOCK)) {
1556                                 child_bits &= ~MDS_INODELOCK_LAYOUT;
1557                                 LASSERT(child_bits != 0);
1558                                 rc = mdt_object_lock(info, child, lhc,
1559                                                 child_bits, MDT_CROSS_LOCK);
1560                         } else {
1561                                 ma_need |= MA_LOV;
1562                         }
1563                 } else {
1564                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1565                          * client will enqueue the lock to the remote MDT */
1566                         if (mdt_object_remote(child))
1567                                 child_bits &= ~MDS_INODELOCK_UPDATE;
1568                         rc = mdt_object_lock(info, child, lhc, child_bits,
1569                                                 MDT_CROSS_LOCK);
1570                 }
1571                 if (unlikely(rc != 0))
1572                         GOTO(out_child, rc);
1573         }
1574
1575         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1576         /* Get MA_SOM attributes if update lock is given. */
1577         if (lock &&
1578             lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_UPDATE &&
1579             S_ISREG(lu_object_attr(&mdt_object_child(child)->mo_lu)))
1580                 ma_need |= MA_SOM;
1581
1582         /* finally, we can get attr for child. */
1583         mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
1584         rc = mdt_getattr_internal(info, child, ma_need);
1585         if (unlikely(rc != 0)) {
1586                 mdt_object_unlock(info, child, lhc, 1);
1587         } else if (lock) {
1588                 /* Debugging code. */
1589                 LDLM_DEBUG(lock, "Returning lock to client");
1590                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1591                                          &lock->l_resource->lr_name),
1592                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1593                          PLDLMRES(lock->l_resource),
1594                          PFID(mdt_object_fid(child)));
1595                 if (mdt_object_exists(child) && !mdt_object_remote(child))
1596                         mdt_pack_size2body(info, child);
1597         }
1598         if (lock)
1599                 LDLM_LOCK_PUT(lock);
1600
1601         EXIT;
1602 out_child:
1603         mdt_object_put(info->mti_env, child);
1604 out_parent:
1605         if (lhp)
1606                 mdt_object_unlock(info, parent, lhp, 1);
1607         return rc;
1608 }
1609
1610 /* normal handler: should release the child lock */
1611 static int mdt_getattr_name(struct tgt_session_info *tsi)
1612 {
1613         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1614         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1615         struct mdt_body        *reqbody;
1616         struct mdt_body        *repbody;
1617         int rc, rc2;
1618         ENTRY;
1619
1620         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1621         LASSERT(reqbody != NULL);
1622         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1623         LASSERT(repbody != NULL);
1624
1625         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1626         repbody->mbo_eadatasize = 0;
1627         repbody->mbo_aclsize = 0;
1628
1629         rc = mdt_init_ucred(info, reqbody);
1630         if (unlikely(rc))
1631                 GOTO(out_shrink, rc);
1632
1633         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1634         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1635                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1636                 lhc->mlh_reg_lh.cookie = 0;
1637         }
1638         mdt_exit_ucred(info);
1639         EXIT;
1640 out_shrink:
1641         mdt_client_compatibility(info);
1642         rc2 = mdt_fix_reply(info);
1643         if (rc == 0)
1644                 rc = rc2;
1645         mdt_thread_info_fini(info);
1646         return rc;
1647 }
1648
1649 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1650                          void *karg, void *uarg);
1651
1652 static int mdt_set_info(struct tgt_session_info *tsi)
1653 {
1654         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1655         char                    *key;
1656         void                    *val;
1657         int                      keylen, vallen, rc = 0;
1658
1659         ENTRY;
1660
1661         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
1662         if (key == NULL) {
1663                 DEBUG_REQ(D_HA, req, "no set_info key");
1664                 RETURN(err_serious(-EFAULT));
1665         }
1666
1667         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
1668                                       RCL_CLIENT);
1669
1670         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
1671         if (val == NULL) {
1672                 DEBUG_REQ(D_HA, req, "no set_info val");
1673                 RETURN(err_serious(-EFAULT));
1674         }
1675
1676         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
1677                                       RCL_CLIENT);
1678
1679         /* Swab any part of val you need to here */
1680         if (KEY_IS(KEY_READ_ONLY)) {
1681                 spin_lock(&req->rq_export->exp_lock);
1682                 if (*(__u32 *)val)
1683                         *exp_connect_flags_ptr(req->rq_export) |=
1684                                 OBD_CONNECT_RDONLY;
1685                 else
1686                         *exp_connect_flags_ptr(req->rq_export) &=
1687                                 ~OBD_CONNECT_RDONLY;
1688                 spin_unlock(&req->rq_export->exp_lock);
1689         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1690                 struct changelog_setinfo *cs = val;
1691
1692                 if (vallen != sizeof(*cs)) {
1693                         CERROR("%s: bad changelog_clear setinfo size %d\n",
1694                                tgt_name(tsi->tsi_tgt), vallen);
1695                         RETURN(-EINVAL);
1696                 }
1697                 if (ptlrpc_req_need_swab(req)) {
1698                         __swab64s(&cs->cs_recno);
1699                         __swab32s(&cs->cs_id);
1700                 }
1701
1702                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
1703                                    vallen, val, NULL);
1704         } else {
1705                 RETURN(-EINVAL);
1706         }
1707         RETURN(rc);
1708 }
1709
1710 static int mdt_readpage(struct tgt_session_info *tsi)
1711 {
1712         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
1713         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
1714         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
1715         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
1716         struct mdt_body         *repbody;
1717         int                      rc;
1718         int                      i;
1719
1720         ENTRY;
1721
1722         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1723                 RETURN(err_serious(-ENOMEM));
1724
1725         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
1726         if (repbody == NULL || reqbody == NULL)
1727                 RETURN(err_serious(-EFAULT));
1728
1729         /*
1730          * prepare @rdpg before calling lower layers and transfer itself. Here
1731          * reqbody->size contains offset of where to start to read and
1732          * reqbody->nlink contains number bytes to read.
1733          */
1734         rdpg->rp_hash = reqbody->mbo_size;
1735         if (rdpg->rp_hash != reqbody->mbo_size) {
1736                 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1737                        rdpg->rp_hash, reqbody->mbo_size);
1738                 RETURN(-EFAULT);
1739         }
1740
1741         rdpg->rp_attrs = reqbody->mbo_mode;
1742         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
1743                 rdpg->rp_attrs |= LUDA_64BITHASH;
1744         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
1745                                 exp_max_brw_size(tsi->tsi_exp));
1746         rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE - 1) >>
1747                           PAGE_CACHE_SHIFT;
1748         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1749         if (rdpg->rp_pages == NULL)
1750                 RETURN(-ENOMEM);
1751
1752         for (i = 0; i < rdpg->rp_npages; ++i) {
1753                 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1754                 if (rdpg->rp_pages[i] == NULL)
1755                         GOTO(free_rdpg, rc = -ENOMEM);
1756         }
1757
1758         /* call lower layers to fill allocated pages with directory data */
1759         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
1760         if (rc < 0)
1761                 GOTO(free_rdpg, rc);
1762
1763         /* send pages to client */
1764         rc = tgt_sendpage(tsi, rdpg, rc);
1765
1766         EXIT;
1767 free_rdpg:
1768
1769         for (i = 0; i < rdpg->rp_npages; i++)
1770                 if (rdpg->rp_pages[i] != NULL)
1771                         __free_page(rdpg->rp_pages[i]);
1772         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1773
1774         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1775                 RETURN(0);
1776
1777         return rc;
1778 }
1779
1780 static int mdt_reint_internal(struct mdt_thread_info *info,
1781                               struct mdt_lock_handle *lhc,
1782                               __u32 op)
1783 {
1784         struct req_capsule      *pill = info->mti_pill;
1785         struct mdt_body         *repbody;
1786         int                      rc = 0, rc2;
1787
1788         ENTRY;
1789
1790         rc = mdt_reint_unpack(info, op);
1791         if (rc != 0) {
1792                 CERROR("Can't unpack reint, rc %d\n", rc);
1793                 RETURN(err_serious(rc));
1794         }
1795
1796         /* for replay (no_create) lmm is not needed, client has it already */
1797         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1798                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1799                                      DEF_REP_MD_SIZE);
1800
1801         /* llog cookies are always 0, the field is kept for compatibility */
1802         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1803                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
1804
1805         rc = req_capsule_server_pack(pill);
1806         if (rc != 0) {
1807                 CERROR("Can't pack response, rc %d\n", rc);
1808                 RETURN(err_serious(rc));
1809         }
1810
1811         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1812                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1813                 LASSERT(repbody);
1814                 repbody->mbo_eadatasize = 0;
1815                 repbody->mbo_aclsize = 0;
1816         }
1817
1818         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1819
1820         /* for replay no cookkie / lmm need, because client have this already */
1821         if (info->mti_spec.no_create)
1822                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1823                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1824
1825         rc = mdt_init_ucred_reint(info);
1826         if (rc)
1827                 GOTO(out_shrink, rc);
1828
1829         rc = mdt_fix_attr_ucred(info, op);
1830         if (rc != 0)
1831                 GOTO(out_ucred, rc = err_serious(rc));
1832
1833         if (mdt_check_resent(info, mdt_reconstruct, lhc)) {
1834                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
1835                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1836                 GOTO(out_ucred, rc);
1837         }
1838         rc = mdt_reint_rec(info, lhc);
1839         EXIT;
1840 out_ucred:
1841         mdt_exit_ucred(info);
1842 out_shrink:
1843         mdt_client_compatibility(info);
1844         rc2 = mdt_fix_reply(info);
1845         if (rc == 0)
1846                 rc = rc2;
1847         return rc;
1848 }
1849
1850 static long mdt_reint_opcode(struct ptlrpc_request *req,
1851                              const struct req_format **fmt)
1852 {
1853         struct mdt_device       *mdt;
1854         struct mdt_rec_reint    *rec;
1855         long                     opc;
1856
1857         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
1858         if (rec != NULL) {
1859                 opc = rec->rr_opcode;
1860                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
1861                 if (opc < REINT_MAX && fmt[opc] != NULL)
1862                         req_capsule_extend(&req->rq_pill, fmt[opc]);
1863                 else {
1864                         mdt = mdt_exp2dev(req->rq_export);
1865                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
1866                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
1867                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
1868                         opc = err_serious(-EFAULT);
1869                 }
1870         } else {
1871                 opc = err_serious(-EFAULT);
1872         }
1873         return opc;
1874 }
1875
1876 static int mdt_reint(struct tgt_session_info *tsi)
1877 {
1878         long opc;
1879         int  rc;
1880         static const struct req_format *reint_fmts[REINT_MAX] = {
1881                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
1882                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
1883                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
1884                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
1885                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
1886                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
1887                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
1888                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
1889                 [REINT_MIGRATE]  = &RQF_MDS_REINT_RENAME
1890         };
1891
1892         ENTRY;
1893
1894         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
1895         if (opc >= 0) {
1896                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1897                 /*
1898                  * No lock possible here from client to pass it to reint code
1899                  * path.
1900                  */
1901                 rc = mdt_reint_internal(info, NULL, opc);
1902                 mdt_thread_info_fini(info);
1903         } else {
1904                 rc = opc;
1905         }
1906
1907         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1908         RETURN(rc);
1909 }
1910
1911 /* this should sync the whole device */
1912 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1913 {
1914         struct dt_device *dt = mdt->mdt_bottom;
1915         int rc;
1916         ENTRY;
1917
1918         rc = dt->dd_ops->dt_sync(env, dt);
1919         RETURN(rc);
1920 }
1921
1922 /* this should sync this object */
1923 static int mdt_object_sync(struct mdt_thread_info *info)
1924 {
1925         struct md_object *next;
1926         int rc;
1927         ENTRY;
1928
1929         if (!mdt_object_exists(info->mti_object)) {
1930                 CWARN("%s: non existing object "DFID": rc = %d\n",
1931                       mdt_obd_name(info->mti_mdt),
1932                       PFID(mdt_object_fid(info->mti_object)), -ESTALE);
1933                 RETURN(-ESTALE);
1934         }
1935         next = mdt_object_child(info->mti_object);
1936         rc = mo_object_sync(info->mti_env, next);
1937
1938         RETURN(rc);
1939 }
1940
1941 static int mdt_sync(struct tgt_session_info *tsi)
1942 {
1943         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1944         struct req_capsule      *pill = tsi->tsi_pill;
1945         struct mdt_body         *body;
1946         int                      rc;
1947
1948         ENTRY;
1949
1950         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1951                 RETURN(err_serious(-ENOMEM));
1952
1953         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
1954                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
1955         } else {
1956                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1957
1958                 /* sync an object */
1959                 rc = mdt_object_sync(info);
1960                 if (rc == 0) {
1961                         const struct lu_fid *fid;
1962                         struct lu_attr *la = &info->mti_attr.ma_attr;
1963
1964                         info->mti_attr.ma_need = MA_INODE;
1965                         info->mti_attr.ma_valid = 0;
1966                         rc = mdt_attr_get_complex(info, info->mti_object,
1967                                                   &info->mti_attr);
1968                         if (rc == 0) {
1969                                 body = req_capsule_server_get(pill,
1970                                                               &RMF_MDT_BODY);
1971                                 fid = mdt_object_fid(info->mti_object);
1972                                 mdt_pack_attr2body(info, body, la, fid);
1973                         }
1974                 }
1975                 mdt_thread_info_fini(info);
1976         }
1977         if (rc == 0)
1978                 mdt_counter_incr(req, LPROC_MDT_SYNC);
1979
1980         RETURN(rc);
1981 }
1982
1983 /*
1984  * Handle quota control requests to consult current usage/limit, but also
1985  * to configure quota enforcement
1986  */
1987 static int mdt_quotactl(struct tgt_session_info *tsi)
1988 {
1989         struct obd_export       *exp  = tsi->tsi_exp;
1990         struct req_capsule      *pill = tsi->tsi_pill;
1991         struct obd_quotactl     *oqctl, *repoqc;
1992         int                      id, rc;
1993         struct mdt_device       *mdt = mdt_exp2dev(exp);
1994         struct lu_device        *qmt = mdt->mdt_qmt_dev;
1995         struct lu_nodemap       *nodemap = exp->exp_target_data.ted_nodemap;
1996         ENTRY;
1997
1998         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1999         if (oqctl == NULL)
2000                 RETURN(err_serious(-EPROTO));
2001
2002         rc = req_capsule_server_pack(pill);
2003         if (rc)
2004                 RETURN(err_serious(rc));
2005
2006         switch (oqctl->qc_cmd) {
2007         case Q_QUOTACHECK:
2008         case LUSTRE_Q_INVALIDATE:
2009         case LUSTRE_Q_FINVALIDATE:
2010         case Q_QUOTAON:
2011         case Q_QUOTAOFF:
2012         case Q_INITQUOTA:
2013                 /* deprecated, not used any more */
2014                 RETURN(-EOPNOTSUPP);
2015                 /* master quotactl */
2016         case Q_SETINFO:
2017         case Q_SETQUOTA:
2018                 if (!nodemap_can_setquota(nodemap))
2019                         RETURN(-EPERM);
2020         case Q_GETINFO:
2021         case Q_GETQUOTA:
2022                 if (qmt == NULL)
2023                         RETURN(-EOPNOTSUPP);
2024                 /* slave quotactl */
2025         case Q_GETOINFO:
2026         case Q_GETOQUOTA:
2027                 break;
2028         default:
2029                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2030                 RETURN(-EFAULT);
2031         }
2032
2033         /* map uid/gid for remote client */
2034         id = oqctl->qc_id;
2035         if (exp_connect_rmtclient(exp)) {
2036                 struct lustre_idmap_table *idmap;
2037
2038                 idmap = exp->exp_mdt_data.med_idmap;
2039
2040                 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
2041                              oqctl->qc_cmd != Q_GETINFO))
2042                         RETURN(-EPERM);
2043
2044                 if (oqctl->qc_type == USRQUOTA)
2045                         id = lustre_idmap_lookup_uid(NULL, idmap, 0,
2046                                                      oqctl->qc_id);
2047                 else if (oqctl->qc_type == GRPQUOTA)
2048                         id = lustre_idmap_lookup_gid(NULL, idmap, 0,
2049                                                      oqctl->qc_id);
2050                 else
2051                         RETURN(-EINVAL);
2052
2053                 if (id == CFS_IDMAP_NOTFOUND) {
2054                         CDEBUG(D_QUOTA, "no mapping for id %u\n", oqctl->qc_id);
2055                         RETURN(-EACCES);
2056                 }
2057         }
2058
2059         if (oqctl->qc_type == USRQUOTA)
2060                 id = nodemap_map_id(nodemap, NODEMAP_UID,
2061                                     NODEMAP_CLIENT_TO_FS, id);
2062         else if (oqctl->qc_type == GRPQUOTA)
2063                 id = nodemap_map_id(nodemap, NODEMAP_UID,
2064                                     NODEMAP_CLIENT_TO_FS, id);
2065
2066         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2067         if (repoqc == NULL)
2068                 RETURN(err_serious(-EFAULT));
2069
2070         if (oqctl->qc_id != id)
2071                 swap(oqctl->qc_id, id);
2072
2073         switch (oqctl->qc_cmd) {
2074
2075         case Q_GETINFO:
2076         case Q_SETINFO:
2077         case Q_SETQUOTA:
2078         case Q_GETQUOTA:
2079                 /* forward quotactl request to QMT */
2080                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2081                 break;
2082
2083         case Q_GETOINFO:
2084         case Q_GETOQUOTA:
2085                 /* slave quotactl */
2086                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2087                                    oqctl);
2088                 break;
2089
2090         default:
2091                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2092                 RETURN(-EFAULT);
2093         }
2094
2095         if (oqctl->qc_id != id)
2096                 swap(oqctl->qc_id, id);
2097
2098         *repoqc = *oqctl;
2099         RETURN(rc);
2100 }
2101
2102 /** clone llog ctxt from child (mdd)
2103  * This allows remote llog (replicator) access.
2104  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2105  * context was originally set up, or we can handle them directly.
2106  * I choose the latter, but that means I need any llog
2107  * contexts set up by child to be accessable by the mdt.  So we clone the
2108  * context into our context list here.
2109  */
2110 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2111                                int idx)
2112 {
2113         struct md_device  *next = mdt->mdt_child;
2114         struct llog_ctxt *ctxt;
2115         int rc;
2116
2117         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2118                 return 0;
2119
2120         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2121         if (rc || ctxt == NULL) {
2122                 return 0;
2123         }
2124
2125         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2126         if (rc)
2127                 CERROR("Can't set mdt ctxt %d\n", rc);
2128
2129         return rc;
2130 }
2131
2132 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2133                                  struct mdt_device *mdt, int idx)
2134 {
2135         struct llog_ctxt *ctxt;
2136
2137         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2138         if (ctxt == NULL)
2139                 return 0;
2140         /* Put once for the get we just did, and once for the clone */
2141         llog_ctxt_put(ctxt);
2142         llog_ctxt_put(ctxt);
2143         return 0;
2144 }
2145
2146 /*
2147  * sec context handlers
2148  */
2149 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2150 {
2151         int rc;
2152
2153         rc = mdt_handle_idmap(tsi);
2154         if (unlikely(rc)) {
2155                 struct ptlrpc_request   *req = tgt_ses_req(tsi);
2156                 __u32                    opc;
2157
2158                 opc = lustre_msg_get_opc(req->rq_reqmsg);
2159                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2160                         sptlrpc_svc_ctx_invalidate(req);
2161         }
2162
2163         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2164
2165         return rc;
2166 }
2167
2168 /*
2169  * quota request handlers
2170  */
2171 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2172 {
2173         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
2174         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2175         int                      rc;
2176         ENTRY;
2177
2178         if (qmt == NULL)
2179                 RETURN(err_serious(-EOPNOTSUPP));
2180
2181         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2182         RETURN(rc);
2183 }
2184
2185 struct mdt_object *mdt_object_new(const struct lu_env *env,
2186                                   struct mdt_device *d,
2187                                   const struct lu_fid *f)
2188 {
2189         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2190         struct lu_object *o;
2191         struct mdt_object *m;
2192         ENTRY;
2193
2194         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2195         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2196         if (unlikely(IS_ERR(o)))
2197                 m = (struct mdt_object *)o;
2198         else
2199                 m = mdt_obj(o);
2200         RETURN(m);
2201 }
2202
2203 struct mdt_object *mdt_object_find(const struct lu_env *env,
2204                                    struct mdt_device *d,
2205                                    const struct lu_fid *f)
2206 {
2207         struct lu_object *o;
2208         struct mdt_object *m;
2209         ENTRY;
2210
2211         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2212         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2213         if (unlikely(IS_ERR(o)))
2214                 m = (struct mdt_object *)o;
2215         else
2216                 m = mdt_obj(o);
2217
2218         RETURN(m);
2219 }
2220
2221 /**
2222  * Asyncronous commit for mdt device.
2223  *
2224  * Pass asynchonous commit call down the MDS stack.
2225  *
2226  * \param env environment
2227  * \param mdt the mdt device
2228  */
2229 static void mdt_device_commit_async(const struct lu_env *env,
2230                                     struct mdt_device *mdt)
2231 {
2232         struct dt_device *dt = mdt->mdt_bottom;
2233         int rc;
2234
2235         rc = dt->dd_ops->dt_commit_async(env, dt);
2236         if (unlikely(rc != 0))
2237                 CWARN("%s: async commit start failed: rc = %d\n",
2238                       mdt_obd_name(mdt), rc);
2239 }
2240
2241 /**
2242  * Mark the lock as "synchonous".
2243  *
2244  * Mark the lock to deffer transaction commit to the unlock time.
2245  *
2246  * \param lock the lock to mark as "synchonous"
2247  *
2248  * \see mdt_is_lock_sync
2249  * \see mdt_save_lock
2250  */
2251 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2252 {
2253         lock->l_ast_data = (void*)1;
2254 }
2255
2256 /**
2257  * Check whehter the lock "synchonous" or not.
2258  *
2259  * \param lock the lock to check
2260  * \retval 1 the lock is "synchonous"
2261  * \retval 0 the lock isn't "synchronous"
2262  *
2263  * \see mdt_set_lock_sync
2264  * \see mdt_save_lock
2265  */
2266 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2267 {
2268         return lock->l_ast_data != NULL;
2269 }
2270
2271 /**
2272  * Blocking AST for mdt locks.
2273  *
2274  * Starts transaction commit if in case of COS lock conflict or
2275  * deffers such a commit to the mdt_save_lock.
2276  *
2277  * \param lock the lock which blocks a request or cancelling lock
2278  * \param desc unused
2279  * \param data unused
2280  * \param flag indicates whether this cancelling or blocking callback
2281  * \retval 0
2282  * \see ldlm_blocking_ast_nocheck
2283  */
2284 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2285                      void *data, int flag)
2286 {
2287         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2288         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2289         int rc;
2290         ENTRY;
2291
2292         if (flag == LDLM_CB_CANCELING)
2293                 RETURN(0);
2294         lock_res_and_lock(lock);
2295         if (lock->l_blocking_ast != mdt_blocking_ast) {
2296                 unlock_res_and_lock(lock);
2297                 RETURN(0);
2298         }
2299         if (mdt_cos_is_enabled(mdt) &&
2300             lock->l_req_mode & (LCK_PW | LCK_EX) &&
2301             lock->l_blocking_lock != NULL &&
2302             lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) {
2303                 mdt_set_lock_sync(lock);
2304         }
2305         rc = ldlm_blocking_ast_nocheck(lock);
2306
2307         /* There is no lock conflict if l_blocking_lock == NULL,
2308          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2309          * when the last reference to a local lock was released */
2310         if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2311                 struct lu_env env;
2312
2313                 rc = lu_env_init(&env, LCT_LOCAL);
2314                 if (unlikely(rc != 0))
2315                         CWARN("%s: lu_env initialization failed, cannot "
2316                               "start asynchronous commit: rc = %d\n",
2317                               obd->obd_name, rc);
2318                 else
2319                         mdt_device_commit_async(&env, mdt);
2320                 lu_env_fini(&env);
2321         }
2322         RETURN(rc);
2323 }
2324
2325 /* Used for cross-MDT lock */
2326 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2327                             void *data, int flag)
2328 {
2329         struct lustre_handle lockh;
2330         int               rc;
2331
2332         switch (flag) {
2333         case LDLM_CB_BLOCKING:
2334                 ldlm_lock2handle(lock, &lockh);
2335                 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
2336                 if (rc < 0) {
2337                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2338                         RETURN(rc);
2339                 }
2340                 break;
2341         case LDLM_CB_CANCELING:
2342                 LDLM_DEBUG(lock, "Revoke remote lock\n");
2343                 break;
2344         default:
2345                 LBUG();
2346         }
2347         RETURN(0);
2348 }
2349
2350 int mdt_check_resent_lock(struct mdt_thread_info *info,
2351                           struct mdt_object *mo,
2352                           struct mdt_lock_handle *lhc)
2353 {
2354         /* the lock might already be gotten in ldlm_handle_enqueue() */
2355         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
2356                 struct ptlrpc_request *req = mdt_info_req(info);
2357                 struct ldlm_lock      *lock;
2358
2359                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2360                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
2361                 if (lock == NULL) {
2362                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
2363                          * a resend case, however, it could be already destroyed
2364                          * due to client eviction or a raced cancel RPC. */
2365                         LDLM_DEBUG_NOLOCK("Invalid lock handle "LPX64"\n",
2366                                           lhc->mlh_reg_lh.cookie);
2367                         RETURN(-ESTALE);
2368                 }
2369
2370                 if (!fid_res_name_eq(mdt_object_fid(mo),
2371                                      &lock->l_resource->lr_name)) {
2372                         CWARN("%s: Although resent, but still not "
2373                               "get child lock:"DFID"\n",
2374                               info->mti_exp->exp_obd->obd_name,
2375                               PFID(mdt_object_fid(mo)));
2376                         LDLM_LOCK_PUT(lock);
2377                         RETURN(-EPROTO);
2378                 }
2379                 LDLM_LOCK_PUT(lock);
2380                 return 0;
2381         }
2382         return 1;
2383 }
2384
2385 int mdt_remote_object_lock(struct mdt_thread_info *mti,
2386                            struct mdt_object *o, const struct lu_fid *fid,
2387                            struct lustre_handle *lh, ldlm_mode_t mode,
2388                            __u64 ibits)
2389 {
2390         struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
2391         ldlm_policy_data_t *policy = &mti->mti_policy;
2392         struct ldlm_res_id *res_id = &mti->mti_res_id;
2393         int rc = 0;
2394         ENTRY;
2395
2396         LASSERT(mdt_object_remote(o));
2397
2398         LASSERT(ibits == MDS_INODELOCK_UPDATE);
2399
2400         fid_build_reg_res_name(fid, res_id);
2401
2402         memset(einfo, 0, sizeof(*einfo));
2403         einfo->ei_type = LDLM_IBITS;
2404         einfo->ei_mode = mode;
2405         einfo->ei_cb_bl = mdt_remote_blocking_ast;
2406         einfo->ei_cb_cp = ldlm_completion_ast;
2407         einfo->ei_enq_slave = 0;
2408         einfo->ei_res_id = res_id;
2409
2410         memset(policy, 0, sizeof(*policy));
2411         policy->l_inodebits.bits = ibits;
2412
2413         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2414                             policy);
2415         RETURN(rc);
2416 }
2417
2418 static int mdt_object_local_lock(struct mdt_thread_info *info,
2419                                  struct mdt_object *o,
2420                                  struct mdt_lock_handle *lh, __u64 ibits,
2421                                  bool nonblock, int locality)
2422 {
2423         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2424         ldlm_policy_data_t *policy = &info->mti_policy;
2425         struct ldlm_res_id *res_id = &info->mti_res_id;
2426         __u64 dlmflags;
2427         int rc;
2428         ENTRY;
2429
2430         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2431         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2432         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2433         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2434
2435         /* Only enqueue LOOKUP lock for remote object */
2436         if (mdt_object_remote(o))
2437                 LASSERT(ibits == MDS_INODELOCK_LOOKUP);
2438
2439         if (lh->mlh_type == MDT_PDO_LOCK) {
2440                 /* check for exists after object is locked */
2441                 if (mdt_object_exists(o) == 0) {
2442                         /* Non-existent object shouldn't have PDO lock */
2443                         RETURN(-ESTALE);
2444                 } else {
2445                         /* Non-dir object shouldn't have PDO lock */
2446                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2447                                 RETURN(-ENOTDIR);
2448                 }
2449         }
2450
2451         memset(policy, 0, sizeof(*policy));
2452         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2453
2454         dlmflags = LDLM_FL_ATOMIC_CB;
2455         if (nonblock)
2456                 dlmflags |= LDLM_FL_BLOCK_NOWAIT;
2457
2458         /*
2459          * Take PDO lock on whole directory and build correct @res_id for lock
2460          * on part of directory.
2461          */
2462         if (lh->mlh_pdo_hash != 0) {
2463                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2464                 mdt_lock_pdo_mode(info, o, lh);
2465                 if (lh->mlh_pdo_mode != LCK_NL) {
2466                         /*
2467                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2468                          * is never going to be sent to client and we do not
2469                          * want it slowed down due to possible cancels.
2470                          */
2471                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2472                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2473                                           policy, res_id, dlmflags,
2474                                           info->mti_exp == NULL ? NULL :
2475                                           &info->mti_exp->exp_handle.h_cookie);
2476                         if (unlikely(rc != 0))
2477                                 GOTO(out_unlock, rc);
2478                 }
2479
2480                 /*
2481                  * Finish res_id initializing by name hash marking part of
2482                  * directory which is taking modification.
2483                  */
2484                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2485         }
2486
2487         policy->l_inodebits.bits = ibits;
2488
2489         /*
2490          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2491          * going to be sent to client. If it is - mdt_intent_policy() path will
2492          * fix it up and turn FL_LOCAL flag off.
2493          */
2494         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2495                           res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2496                           info->mti_exp == NULL ? NULL :
2497                           &info->mti_exp->exp_handle.h_cookie);
2498 out_unlock:
2499         if (rc != 0)
2500                 mdt_object_unlock(info, o, lh, 1);
2501         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2502                    lh->mlh_pdo_hash != 0 &&
2503                    (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
2504                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2505
2506         RETURN(rc);
2507 }
2508
2509 static int
2510 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
2511                          struct mdt_lock_handle *lh, __u64 ibits,
2512                          bool nonblock, int locality)
2513 {
2514         int rc;
2515         ENTRY;
2516
2517         if (!mdt_object_remote(o))
2518                 return mdt_object_local_lock(info, o, lh, ibits, nonblock,
2519                                              locality);
2520
2521         if (locality == MDT_LOCAL_LOCK) {
2522                 CERROR("%s: try to get local lock for remote object"
2523                        DFID".\n", mdt_obd_name(info->mti_mdt),
2524                        PFID(mdt_object_fid(o)));
2525                 RETURN(-EPROTO);
2526         }
2527
2528         /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
2529         ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
2530                    MDS_INODELOCK_XATTR);
2531         if (ibits & MDS_INODELOCK_UPDATE) {
2532                 /* Sigh, PDO needs to enqueue 2 locks right now, but
2533                  * enqueue RPC can only request 1 lock, to avoid extra
2534                  * RPC, so it will instead enqueue EX lock for remote
2535                  * object anyway XXX*/
2536                 if (lh->mlh_type == MDT_PDO_LOCK &&
2537                     lh->mlh_pdo_hash != 0) {
2538                         CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
2539                                "EX lock.\n", mdt_obd_name(info->mti_mdt),
2540                                PFID(mdt_object_fid(o)));
2541                         lh->mlh_pdo_hash = 0;
2542                         lh->mlh_rreg_mode = LCK_EX;
2543                         lh->mlh_type = MDT_REG_LOCK;
2544                 }
2545                 rc = mdt_remote_object_lock(info, o, mdt_object_fid(o),
2546                                             &lh->mlh_rreg_lh,
2547                                             lh->mlh_rreg_mode,
2548                                             MDS_INODELOCK_UPDATE);
2549                 if (rc != ELDLM_OK)
2550                         RETURN(rc);
2551         }
2552
2553         /* Only enqueue LOOKUP lock for remote object */
2554         if (ibits & MDS_INODELOCK_LOOKUP) {
2555                 rc = mdt_object_local_lock(info, o, lh,
2556                                            MDS_INODELOCK_LOOKUP,
2557                                            nonblock, locality);
2558                 if (rc != ELDLM_OK)
2559                         RETURN(rc);
2560         }
2561
2562         RETURN(0);
2563 }
2564
2565 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2566                     struct mdt_lock_handle *lh, __u64 ibits, int locality)
2567 {
2568         return mdt_object_lock_internal(info, o, lh, ibits, false, locality);
2569 }
2570
2571 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
2572                         struct mdt_lock_handle *lh, __u64 ibits, int locality)
2573 {
2574         struct mdt_lock_handle tmp = *lh;
2575         int rc;
2576
2577         rc = mdt_object_lock_internal(info, o, &tmp, ibits, true, locality);
2578         if (rc == 0)
2579                 *lh = tmp;
2580
2581         return rc == 0;
2582 }
2583
2584 /**
2585  * Save a lock within request object.
2586  *
2587  * Keep the lock referenced until whether client ACK or transaction
2588  * commit happens or release the lock immediately depending on input
2589  * parameters. If COS is ON, a write lock is converted to COS lock
2590  * before saving.
2591  *
2592  * \param info thead info object
2593  * \param h lock handle
2594  * \param mode lock mode
2595  * \param decref force immediate lock releasing
2596  */
2597 static
2598 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2599                    ldlm_mode_t mode, int decref)
2600 {
2601         ENTRY;
2602
2603         if (lustre_handle_is_used(h)) {
2604                 if (decref || !info->mti_has_trans ||
2605                     !(mode & (LCK_PW | LCK_EX))){
2606                         mdt_fid_unlock(h, mode);
2607                 } else {
2608                         struct mdt_device *mdt = info->mti_mdt;
2609                         struct ldlm_lock *lock = ldlm_handle2lock(h);
2610                         struct ptlrpc_request *req = mdt_info_req(info);
2611                         int no_ack = 0;
2612
2613                         LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2614                                  h->cookie);
2615                         /* there is no request if mdt_object_unlock() is called
2616                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
2617                         if (likely(req != NULL)) {
2618                                 CDEBUG(D_HA, "request = %p reply state = %p"
2619                                        " transno = "LPD64"\n", req,
2620                                        req->rq_reply_state, req->rq_transno);
2621                                 if (mdt_cos_is_enabled(mdt)) {
2622                                         no_ack = 1;
2623                                         ldlm_lock_downgrade(lock, LCK_COS);
2624                                         mode = LCK_COS;
2625                                 }
2626                                 ptlrpc_save_lock(req, h, mode, no_ack);
2627                         } else {
2628                                 ldlm_lock_decref(h, mode);
2629                         }
2630                         if (mdt_is_lock_sync(lock)) {
2631                                 CDEBUG(D_HA, "found sync-lock,"
2632                                        " async commit started\n");
2633                                 mdt_device_commit_async(info->mti_env,
2634                                                         mdt);
2635                         }
2636                         LDLM_LOCK_PUT(lock);
2637                 }
2638                 h->cookie = 0ull;
2639         }
2640
2641         EXIT;
2642 }
2643
2644 /**
2645  * Unlock mdt object.
2646  *
2647  * Immeditely release the regular lock and the PDO lock or save the
2648  * lock in reqeuest and keep them referenced until client ACK or
2649  * transaction commit.
2650  *
2651  * \param info thread info object
2652  * \param o mdt object
2653  * \param lh mdt lock handle referencing regular and PDO locks
2654  * \param decref force immediate lock releasing
2655  */
2656 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2657                        struct mdt_lock_handle *lh, int decref)
2658 {
2659         ENTRY;
2660
2661         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2662         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2663
2664         if (lustre_handle_is_used(&lh->mlh_rreg_lh))
2665                 ldlm_lock_decref(&lh->mlh_rreg_lh, lh->mlh_rreg_mode);
2666
2667         EXIT;
2668 }
2669
2670 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2671                                         const struct lu_fid *f,
2672                                         struct mdt_lock_handle *lh,
2673                                         __u64 ibits)
2674 {
2675         struct mdt_object *o;
2676
2677         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2678         if (!IS_ERR(o)) {
2679                 int rc;
2680
2681                 rc = mdt_object_lock(info, o, lh, ibits,
2682                                      MDT_LOCAL_LOCK);
2683                 if (rc != 0) {
2684                         mdt_object_put(info->mti_env, o);
2685                         o = ERR_PTR(rc);
2686                 }
2687         }
2688         return o;
2689 }
2690
2691 void mdt_object_unlock_put(struct mdt_thread_info * info,
2692                            struct mdt_object * o,
2693                            struct mdt_lock_handle *lh,
2694                            int decref)
2695 {
2696         mdt_object_unlock(info, o, lh, decref);
2697         mdt_object_put(info->mti_env, o);
2698 }
2699
2700 /*
2701  * Generic code handling requests that have struct mdt_body passed in:
2702  *
2703  *  - extract mdt_body from request and save it in @info, if present;
2704  *
2705  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
2706  *  @info;
2707  *
2708  *  - if HABEO_CORPUS flag is set for this request type check whether object
2709  *  actually exists on storage (lu_object_exists()).
2710  *
2711  */
2712 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2713 {
2714         const struct mdt_body    *body;
2715         struct mdt_object        *obj;
2716         const struct lu_env      *env;
2717         struct req_capsule       *pill;
2718         int                       rc;
2719         ENTRY;
2720
2721         env = info->mti_env;
2722         pill = info->mti_pill;
2723
2724         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2725         if (body == NULL)
2726                 RETURN(-EFAULT);
2727
2728         if (!(body->mbo_valid & OBD_MD_FLID))
2729                 RETURN(0);
2730
2731         if (!fid_is_sane(&body->mbo_fid1)) {
2732                 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
2733                 RETURN(-EINVAL);
2734         }
2735
2736         /*
2737          * Do not get size or any capa fields before we check that request
2738          * contains capa actually. There are some requests which do not, for
2739          * instance MDS_IS_SUBDIR.
2740          */
2741         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2742             req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT))
2743                 mdt_set_capainfo(info, 0, &body->mbo_fid1,
2744                                  req_capsule_client_get(pill, &RMF_CAPA1));
2745
2746         obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
2747         if (!IS_ERR(obj)) {
2748                 if ((flags & HABEO_CORPUS) &&
2749                     !mdt_object_exists(obj)) {
2750                         mdt_object_put(env, obj);
2751                         /* for capability renew ENOENT will be handled in
2752                          * mdt_renew_capa */
2753                         if (body->mbo_valid & OBD_MD_FLOSSCAPA)
2754                                 rc = 0;
2755                         else
2756                                 rc = -ENOENT;
2757                 } else {
2758                         info->mti_object = obj;
2759                         rc = 0;
2760                 }
2761         } else
2762                 rc = PTR_ERR(obj);
2763
2764         RETURN(rc);
2765 }
2766
2767 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2768 {
2769         struct req_capsule *pill = info->mti_pill;
2770         int rc;
2771         ENTRY;
2772
2773         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2774                 rc = mdt_body_unpack(info, flags);
2775         else
2776                 rc = 0;
2777
2778         if (rc == 0 && (flags & HABEO_REFERO)) {
2779                 /* Pack reply. */
2780                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2781                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2782                                              DEF_REP_MD_SIZE);
2783                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2784                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
2785                                              RCL_SERVER, 0);
2786
2787                 rc = req_capsule_server_pack(pill);
2788         }
2789         RETURN(rc);
2790 }
2791
2792 static int mdt_init_capa_ctxt(const struct lu_env *env, struct mdt_device *m)
2793 {
2794         struct md_device *next = m->mdt_child;
2795
2796         return next->md_ops->mdo_init_capa_ctxt(env, next,
2797                                                 m->mdt_lut.lut_mds_capa,
2798                                                 m->mdt_capa_timeout,
2799                                                 m->mdt_capa_alg,
2800                                                 m->mdt_capa_keys);
2801 }
2802
2803 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2804 {
2805         lh->mlh_type = MDT_NUL_LOCK;
2806         lh->mlh_reg_lh.cookie = 0ull;
2807         lh->mlh_reg_mode = LCK_MINMODE;
2808         lh->mlh_pdo_lh.cookie = 0ull;
2809         lh->mlh_pdo_mode = LCK_MINMODE;
2810         lh->mlh_rreg_lh.cookie = 0ull;
2811         lh->mlh_rreg_mode = LCK_MINMODE;
2812 }
2813
2814 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2815 {
2816         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2817         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2818 }
2819
2820 /*
2821  * Initialize fields of struct mdt_thread_info. Other fields are left in
2822  * uninitialized state, because it's too expensive to zero out whole
2823  * mdt_thread_info (> 1K) on each request arrival.
2824  */
2825 void mdt_thread_info_init(struct ptlrpc_request *req,
2826                           struct mdt_thread_info *info)
2827 {
2828         int i;
2829
2830         info->mti_pill = &req->rq_pill;
2831
2832         /* lock handle */
2833         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2834                 mdt_lock_handle_init(&info->mti_lh[i]);
2835
2836         /* mdt device: it can be NULL while CONNECT */
2837         if (req->rq_export) {
2838                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2839                 info->mti_exp = req->rq_export;
2840         } else
2841                 info->mti_mdt = NULL;
2842         info->mti_env = req->rq_svc_thread->t_env;
2843         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2844
2845         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2846         info->mti_big_buf = LU_BUF_NULL;
2847         info->mti_body = NULL;
2848         info->mti_object = NULL;
2849         info->mti_dlm_req = NULL;
2850         info->mti_has_trans = 0;
2851         info->mti_cross_ref = 0;
2852         info->mti_opdata = 0;
2853         info->mti_big_lmm_used = 0;
2854
2855         info->mti_spec.no_create = 0;
2856         info->mti_spec.sp_rm_entry = 0;
2857
2858         info->mti_spec.u.sp_ea.eadata = NULL;
2859         info->mti_spec.u.sp_ea.eadatalen = 0;
2860 }
2861
2862 void mdt_thread_info_fini(struct mdt_thread_info *info)
2863 {
2864         int i;
2865
2866         if (info->mti_object != NULL) {
2867                 mdt_object_put(info->mti_env, info->mti_object);
2868                 info->mti_object = NULL;
2869         }
2870
2871         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2872                 mdt_lock_handle_fini(&info->mti_lh[i]);
2873         info->mti_env = NULL;
2874         info->mti_pill = NULL;
2875         info->mti_exp = NULL;
2876
2877         if (unlikely(info->mti_big_buf.lb_buf != NULL))
2878                 lu_buf_free(&info->mti_big_buf);
2879 }
2880
2881 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
2882 {
2883         struct mdt_thread_info  *mti;
2884         struct lustre_capa      *lc;
2885
2886         mti = mdt_th_info(tsi->tsi_env);
2887         LASSERT(mti != NULL);
2888
2889         mdt_thread_info_init(tgt_ses_req(tsi), mti);
2890         if (tsi->tsi_corpus != NULL) {
2891                 struct req_capsule *pill = tsi->tsi_pill;
2892
2893                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
2894                 lu_object_get(tsi->tsi_corpus);
2895
2896                 /*
2897                  * XXX: must be part of tgt_mdt_body_unpack but moved here
2898                  * due to mdt_set_capainfo().
2899                  */
2900                 if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_CLIENT) &&
2901                     req_capsule_get_size(pill, &RMF_CAPA1, RCL_CLIENT) > 0) {
2902                         lc = req_capsule_client_get(pill, &RMF_CAPA1);
2903                         mdt_set_capainfo(mti, 0, &tsi->tsi_mdt_body->mbo_fid1,
2904                                          lc);
2905                 }
2906         }
2907         mti->mti_body = tsi->tsi_mdt_body;
2908         mti->mti_dlm_req = tsi->tsi_dlm_req;
2909
2910         return mti;
2911 }
2912
2913 static int mdt_tgt_connect(struct tgt_session_info *tsi)
2914 {
2915         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2916         int                      rc;
2917
2918         ENTRY;
2919
2920         if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
2921             cfs_fail_val ==
2922             tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id)
2923                 schedule_timeout_and_set_state(TASK_UNINTERRUPTIBLE, HZ * 3);
2924
2925         rc = tgt_connect(tsi);
2926         if (rc != 0)
2927                 RETURN(rc);
2928
2929         rc = mdt_init_idmap(tsi);
2930         if (rc != 0)
2931                 GOTO(err, rc);
2932         RETURN(0);
2933 err:
2934         obd_disconnect(class_export_get(req->rq_export));
2935         return rc;
2936 }
2937
2938 enum mdt_it_code {
2939         MDT_IT_OPEN,
2940         MDT_IT_OCREAT,
2941         MDT_IT_CREATE,
2942         MDT_IT_GETATTR,
2943         MDT_IT_READDIR,
2944         MDT_IT_LOOKUP,
2945         MDT_IT_UNLINK,
2946         MDT_IT_TRUNC,
2947         MDT_IT_GETXATTR,
2948         MDT_IT_LAYOUT,
2949         MDT_IT_QUOTA,
2950         MDT_IT_NR
2951 };
2952
2953 static int mdt_intent_getattr(enum mdt_it_code opcode,
2954                               struct mdt_thread_info *info,
2955                               struct ldlm_lock **,
2956                               __u64);
2957
2958 static int mdt_intent_getxattr(enum mdt_it_code opcode,
2959                                 struct mdt_thread_info *info,
2960                                 struct ldlm_lock **lockp,
2961                                 __u64 flags);
2962
2963 static int mdt_intent_layout(enum mdt_it_code opcode,
2964                              struct mdt_thread_info *info,
2965                              struct ldlm_lock **,
2966                              __u64);
2967 static int mdt_intent_reint(enum mdt_it_code opcode,
2968                             struct mdt_thread_info *info,
2969                             struct ldlm_lock **,
2970                             __u64);
2971
2972 static struct mdt_it_flavor {
2973         const struct req_format *it_fmt;
2974         __u32                    it_flags;
2975         int                    (*it_act)(enum mdt_it_code ,
2976                                          struct mdt_thread_info *,
2977                                          struct ldlm_lock **,
2978                                          __u64);
2979         long                     it_reint;
2980 } mdt_it_flavor[] = {
2981         [MDT_IT_OPEN]     = {
2982                 .it_fmt   = &RQF_LDLM_INTENT,
2983                 /*.it_flags = HABEO_REFERO,*/
2984                 .it_flags = 0,
2985                 .it_act   = mdt_intent_reint,
2986                 .it_reint = REINT_OPEN
2987         },
2988         [MDT_IT_OCREAT]   = {
2989                 .it_fmt   = &RQF_LDLM_INTENT,
2990                 /*
2991                  * OCREAT is not a MUTABOR request as if the file
2992                  * already exists.
2993                  * We do the extra check of OBD_CONNECT_RDONLY in
2994                  * mdt_reint_open() when we really need to create
2995                  * the object.
2996                  */
2997                 .it_flags = 0,
2998                 .it_act   = mdt_intent_reint,
2999                 .it_reint = REINT_OPEN
3000         },
3001         [MDT_IT_CREATE]   = {
3002                 .it_fmt   = &RQF_LDLM_INTENT,
3003                 .it_flags = MUTABOR,
3004                 .it_act   = mdt_intent_reint,
3005                 .it_reint = REINT_CREATE
3006         },
3007         [MDT_IT_GETATTR]  = {
3008                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3009                 .it_flags = HABEO_REFERO,
3010                 .it_act   = mdt_intent_getattr
3011         },
3012         [MDT_IT_READDIR]  = {
3013                 .it_fmt   = NULL,
3014                 .it_flags = 0,
3015                 .it_act   = NULL
3016         },
3017         [MDT_IT_LOOKUP]   = {
3018                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3019                 .it_flags = HABEO_REFERO,
3020                 .it_act   = mdt_intent_getattr
3021         },
3022         [MDT_IT_UNLINK]   = {
3023                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
3024                 .it_flags = MUTABOR,
3025                 .it_act   = NULL,
3026                 .it_reint = REINT_UNLINK
3027         },
3028         [MDT_IT_TRUNC]    = {
3029                 .it_fmt   = NULL,
3030                 .it_flags = MUTABOR,
3031                 .it_act   = NULL
3032         },
3033         [MDT_IT_GETXATTR] = {
3034                 .it_fmt   = &RQF_LDLM_INTENT_GETXATTR,
3035                 .it_flags = HABEO_CORPUS,
3036                 .it_act   = mdt_intent_getxattr
3037         },
3038         [MDT_IT_LAYOUT] = {
3039                 .it_fmt   = &RQF_LDLM_INTENT_LAYOUT,
3040                 .it_flags = 0,
3041                 .it_act   = mdt_intent_layout
3042         }
3043 };
3044
3045 static int
3046 mdt_intent_lock_replace(struct mdt_thread_info *info,
3047                         struct ldlm_lock **lockp,
3048                         struct mdt_lock_handle *lh,
3049                         __u64 flags)
3050 {
3051         struct ptlrpc_request  *req = mdt_info_req(info);
3052         struct ldlm_lock       *lock = *lockp;
3053         struct ldlm_lock       *new_lock;
3054
3055         /* If possible resent found a lock, @lh is set to its handle */
3056         new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3057
3058         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3059                 lh->mlh_reg_lh.cookie = 0;
3060                 RETURN(0);
3061         }
3062
3063         LASSERTF(new_lock != NULL,
3064                  "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
3065
3066         /*
3067          * If we've already given this lock to a client once, then we should
3068          * have no readers or writers.  Otherwise, we should have one reader
3069          * _or_ writer ref (which will be zeroed below) before returning the
3070          * lock to a client.
3071          */
3072         if (new_lock->l_export == req->rq_export) {
3073                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3074         } else {
3075                 LASSERT(new_lock->l_export == NULL);
3076                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3077         }
3078
3079         *lockp = new_lock;
3080
3081         if (new_lock->l_export == req->rq_export) {
3082                 /*
3083                  * Already gave this to the client, which means that we
3084                  * reconstructed a reply.
3085                  */
3086                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3087                         MSG_RESENT);
3088
3089                 LDLM_LOCK_RELEASE(new_lock);
3090                 lh->mlh_reg_lh.cookie = 0;
3091                 RETURN(ELDLM_LOCK_REPLACED);
3092         }
3093
3094         /*
3095          * Fixup the lock to be given to the client.
3096          */
3097         lock_res_and_lock(new_lock);
3098         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3099          * possible blocking AST. */
3100         while (new_lock->l_readers > 0) {
3101                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3102                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3103                 new_lock->l_readers--;
3104         }
3105         while (new_lock->l_writers > 0) {
3106                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3107                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3108                 new_lock->l_writers--;
3109         }
3110
3111         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3112         new_lock->l_blocking_ast = lock->l_blocking_ast;
3113         new_lock->l_completion_ast = lock->l_completion_ast;
3114         new_lock->l_remote_handle = lock->l_remote_handle;
3115         new_lock->l_flags &= ~LDLM_FL_LOCAL;
3116
3117         unlock_res_and_lock(new_lock);
3118
3119         cfs_hash_add(new_lock->l_export->exp_lock_hash,
3120                      &new_lock->l_remote_handle,
3121                      &new_lock->l_exp_hash);
3122
3123         LDLM_LOCK_RELEASE(new_lock);
3124         lh->mlh_reg_lh.cookie = 0;
3125
3126         RETURN(ELDLM_LOCK_REPLACED);
3127 }
3128
3129 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3130                                     struct ldlm_lock *new_lock,
3131                                     struct mdt_lock_handle *lh,
3132                                     __u64 flags)
3133 {
3134         struct ptlrpc_request  *req = mdt_info_req(info);
3135         struct ldlm_request    *dlmreq;
3136
3137         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3138                 return;
3139
3140         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3141
3142         /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
3143          * lock was found by ldlm_handle_enqueue(); if so @lh must be
3144          * initialized. */
3145         if (flags & LDLM_FL_RESENT) {
3146                 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
3147                 lh->mlh_reg_mode = new_lock->l_granted_mode;
3148
3149                 LDLM_DEBUG(new_lock, "Restoring lock cookie");
3150                 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie "LPX64,
3151                           lh->mlh_reg_lh.cookie);
3152                 return;
3153         }
3154
3155         /*
3156          * If the xid matches, then we know this is a resent request, and allow
3157          * it. (It's probably an OPEN, for which we don't send a lock.
3158          */
3159         if (req_xid_is_last(req))
3160                 return;
3161
3162         /*
3163          * This remote handle isn't enqueued, so we never received or processed
3164          * this request.  Clear MSG_RESENT, because it can be handled like any
3165          * normal request now.
3166          */
3167         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3168
3169         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
3170                   dlmreq->lock_handle[0].cookie);
3171 }
3172
3173 static int mdt_intent_getxattr(enum mdt_it_code opcode,
3174                                 struct mdt_thread_info *info,
3175                                 struct ldlm_lock **lockp,
3176                                 __u64 flags)
3177 {
3178         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3179         struct ldlm_reply      *ldlm_rep = NULL;
3180         int rc, grc;
3181
3182         /*
3183          * Initialize lhc->mlh_reg_lh either from a previously granted lock
3184          * (for the resend case) or a new lock. Below we will use it to
3185          * replace the original lock.
3186          */
3187         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3188         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3189                 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3190                 rc = mdt_object_lock(info, info->mti_object, lhc,
3191                                         MDS_INODELOCK_XATTR,
3192                                         MDT_LOCAL_LOCK);
3193                 if (rc)
3194                         return rc;
3195         }
3196
3197         grc = mdt_getxattr(info);
3198
3199         rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3200
3201         if (mdt_info_req(info)->rq_repmsg != NULL)
3202                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3203         if (ldlm_rep == NULL)
3204                 RETURN(err_serious(-EFAULT));
3205
3206         ldlm_rep->lock_policy_res2 = grc;
3207
3208         return rc;
3209 }
3210
3211 static int mdt_intent_getattr(enum mdt_it_code opcode,
3212                               struct mdt_thread_info *info,
3213                               struct ldlm_lock **lockp,
3214                               __u64 flags)
3215 {
3216         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3217         __u64                   child_bits;
3218         struct ldlm_reply      *ldlm_rep;
3219         struct mdt_body        *reqbody;
3220         struct mdt_body        *repbody;
3221         int                     rc, rc2;
3222         ENTRY;
3223
3224         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3225         LASSERT(reqbody);
3226
3227         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3228         LASSERT(repbody);
3229
3230         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
3231         repbody->mbo_eadatasize = 0;
3232         repbody->mbo_aclsize = 0;
3233
3234         switch (opcode) {
3235         case MDT_IT_LOOKUP:
3236                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3237                 break;
3238         case MDT_IT_GETATTR:
3239                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3240                              MDS_INODELOCK_PERM;
3241                 break;
3242         default:
3243                 CERROR("Unsupported intent (%d)\n", opcode);
3244                 GOTO(out_shrink, rc = -EINVAL);
3245         }
3246
3247         rc = mdt_init_ucred(info, reqbody);
3248         if (rc)
3249                 GOTO(out_shrink, rc);
3250
3251         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3252         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3253
3254         /* Get lock from request for possible resent case. */
3255         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3256
3257         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3258         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3259
3260         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3261                 ldlm_rep->lock_policy_res2 = 0;
3262         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3263             ldlm_rep->lock_policy_res2) {
3264                 lhc->mlh_reg_lh.cookie = 0ull;
3265                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3266         }
3267
3268         rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3269         EXIT;
3270 out_ucred:
3271         mdt_exit_ucred(info);
3272 out_shrink:
3273         mdt_client_compatibility(info);
3274         rc2 = mdt_fix_reply(info);
3275         if (rc == 0)
3276                 rc = rc2;
3277         return rc;
3278 }
3279
3280 static int mdt_intent_layout(enum mdt_it_code opcode,
3281                              struct mdt_thread_info *info,
3282                              struct ldlm_lock **lockp,
3283                              __u64 flags)
3284 {
3285         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_LAYOUT];
3286         struct layout_intent *layout;
3287         struct lu_fid *fid;
3288         struct mdt_object *obj = NULL;
3289         int rc = 0;
3290         ENTRY;
3291
3292         if (opcode != MDT_IT_LAYOUT) {
3293                 CERROR("%s: Unknown intent (%d)\n", mdt_obd_name(info->mti_mdt),
3294                         opcode);
3295                 RETURN(-EINVAL);
3296         }
3297
3298         fid = &info->mti_tmp_fid2;
3299         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3300
3301         /* Get lock from request for possible resent case. */
3302         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3303
3304         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3305         if (IS_ERR(obj))
3306                 RETURN(PTR_ERR(obj));
3307
3308         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3309                 /* get the length of lsm */
3310                 rc = mdt_attr_get_eabuf_size(info, obj);
3311                 if (rc < 0) {
3312                         mdt_object_put(info->mti_env, obj);
3313                         RETURN(rc);
3314                 }
3315
3316                 if (rc > info->mti_mdt->mdt_max_mdsize)
3317                         info->mti_mdt->mdt_max_mdsize = rc;
3318         }
3319
3320         mdt_object_put(info->mti_env, obj);
3321
3322         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3323         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER, rc);
3324         rc = req_capsule_server_pack(info->mti_pill);
3325         if (rc != 0)
3326                 RETURN(-EINVAL);
3327
3328         if (lustre_handle_is_used(&lhc->mlh_reg_lh))
3329                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3330
3331         layout = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3332         LASSERT(layout != NULL);
3333         if (layout->li_opc == LAYOUT_INTENT_ACCESS)
3334                 /* return to normal/resent ldlm handling */
3335                 RETURN(rc);
3336
3337         CERROR("%s: Unsupported layout intent (%d)\n",
3338                 mdt_obd_name(info->mti_mdt), layout->li_opc);
3339         RETURN(-EINVAL);
3340 }
3341
3342 static int mdt_intent_reint(enum mdt_it_code opcode,
3343                             struct mdt_thread_info *info,
3344                             struct ldlm_lock **lockp,
3345                             __u64 flags)
3346 {
3347         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3348         struct ldlm_reply      *rep = NULL;
3349         long                    opc;
3350         int                     rc;
3351
3352         static const struct req_format *intent_fmts[REINT_MAX] = {
3353                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3354                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3355         };
3356
3357         ENTRY;
3358
3359         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
3360         if (opc < 0)
3361                 RETURN(opc);
3362
3363         if (mdt_it_flavor[opcode].it_reint != opc) {
3364                 CERROR("Reint code %ld doesn't match intent: %d\n",
3365                        opc, opcode);
3366                 RETURN(err_serious(-EPROTO));
3367         }
3368
3369         /* Get lock from request for possible resent case. */
3370         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3371
3372         rc = mdt_reint_internal(info, lhc, opc);
3373
3374         /* Check whether the reply has been packed successfully. */
3375         if (mdt_info_req(info)->rq_repmsg != NULL)
3376                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3377         if (rep == NULL)
3378                 RETURN(err_serious(-EFAULT));
3379
3380         /* MDC expects this in any case */
3381         if (rc != 0)
3382                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3383
3384         /* the open lock or the lock for cross-ref object should be
3385          * returned to the client */
3386         if (rc == -MDT_EREMOTE_OPEN || mdt_get_disposition(rep, DISP_OPEN_LOCK)) {
3387                 LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh));
3388                 rep->lock_policy_res2 = 0;
3389                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3390                 RETURN(rc);
3391         }
3392
3393         rep->lock_policy_res2 = clear_serious(rc);
3394
3395         if (rep->lock_policy_res2 == -ENOENT &&
3396             mdt_get_disposition(rep, DISP_LOOKUP_NEG))
3397                 rep->lock_policy_res2 = 0;
3398
3399         if (rc == -ENOTCONN || rc == -ENODEV ||
3400             rc == -EOVERFLOW) { /**< if VBR failure then return error */
3401                 /*
3402                  * If it is the disconnect error (ENODEV & ENOCONN), the error
3403                  * will be returned by rq_status, and client at ptlrpc layer
3404                  * will detect this, then disconnect, reconnect the import
3405                  * immediately, instead of impacting the following the rpc.
3406                  */
3407                 lhc->mlh_reg_lh.cookie = 0ull;
3408                 RETURN(rc);
3409         } else {
3410                 /*
3411                  * For other cases, the error will be returned by intent.
3412                  * and client will retrieve the result from intent.
3413                  */
3414                  /*
3415                   * FIXME: when open lock is finished, that should be
3416                   * checked here.
3417                   */
3418                 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3419                         LASSERTF(rc == 0, "Error occurred but lock handle "
3420                                  "is still in use, rc = %d\n", rc);
3421                         rep->lock_policy_res2 = 0;
3422                         rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3423                         RETURN(rc);
3424                 } else {
3425                         lhc->mlh_reg_lh.cookie = 0ull;
3426                         RETURN(ELDLM_LOCK_ABORTED);
3427                 }
3428         }
3429 }
3430
3431 static int mdt_intent_code(long itcode)
3432 {
3433         int rc;
3434
3435         switch(itcode) {
3436         case IT_OPEN:
3437                 rc = MDT_IT_OPEN;
3438                 break;
3439         case IT_OPEN|IT_CREAT:
3440                 rc = MDT_IT_OCREAT;
3441                 break;
3442         case IT_CREAT:
3443                 rc = MDT_IT_CREATE;
3444                 break;
3445         case IT_READDIR:
3446                 rc = MDT_IT_READDIR;
3447                 break;
3448         case IT_GETATTR:
3449                 rc = MDT_IT_GETATTR;
3450                 break;
3451         case IT_LOOKUP:
3452                 rc = MDT_IT_LOOKUP;
3453                 break;
3454         case IT_UNLINK:
3455                 rc = MDT_IT_UNLINK;
3456                 break;
3457         case IT_TRUNC:
3458                 rc = MDT_IT_TRUNC;
3459                 break;
3460         case IT_GETXATTR:
3461                 rc = MDT_IT_GETXATTR;
3462                 break;
3463         case IT_LAYOUT:
3464                 rc = MDT_IT_LAYOUT;
3465                 break;
3466         case IT_QUOTA_DQACQ:
3467         case IT_QUOTA_CONN:
3468                 rc = MDT_IT_QUOTA;
3469                 break;
3470         default:
3471                 CERROR("Unknown intent opcode: %ld\n", itcode);
3472                 rc = -EINVAL;
3473                 break;
3474         }
3475         return rc;
3476 }
3477
3478 static int mdt_intent_opc(long itopc, struct mdt_thread_info *info,
3479                           struct ldlm_lock **lockp, __u64 flags)
3480 {
3481         struct req_capsule      *pill = info->mti_pill;
3482         struct ptlrpc_request   *req = mdt_info_req(info);
3483         struct mdt_it_flavor    *flv;
3484         int opc;
3485         int rc;
3486         ENTRY;
3487
3488         opc = mdt_intent_code(itopc);
3489         if (opc < 0)
3490                 RETURN(-EINVAL);
3491
3492         if (opc == MDT_IT_QUOTA) {
3493                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
3494
3495                 if (qmt == NULL)
3496                         RETURN(-EOPNOTSUPP);
3497
3498                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
3499                 /* pass the request to quota master */
3500                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
3501                                                  mdt_info_req(info), lockp,
3502                                                  flags);
3503                 RETURN(rc);
3504         }
3505
3506         flv = &mdt_it_flavor[opc];
3507         if (flv->it_fmt != NULL)
3508                 req_capsule_extend(pill, flv->it_fmt);
3509
3510         rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3511         if (rc < 0)
3512                 RETURN(rc);
3513
3514         if (flv->it_flags & MUTABOR &&
3515             exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
3516                 RETURN(-EROFS);
3517
3518         if (flv->it_act != NULL) {
3519                 struct ldlm_reply *rep;
3520
3521                 /* execute policy */
3522                 rc = flv->it_act(opc, info, lockp, flags);
3523
3524                 /* Check whether the reply has been packed successfully. */
3525                 if (req->rq_repmsg != NULL) {
3526                         rep = req_capsule_server_get(info->mti_pill,
3527                                                      &RMF_DLM_REP);
3528                         rep->lock_policy_res2 =
3529                                 ptlrpc_status_hton(rep->lock_policy_res2);
3530                 }
3531         }
3532
3533         RETURN(rc);
3534 }
3535
3536 static int mdt_intent_policy(struct ldlm_namespace *ns,
3537                              struct ldlm_lock **lockp, void *req_cookie,
3538                              ldlm_mode_t mode, __u64 flags, void *data)
3539 {
3540         struct tgt_session_info *tsi;
3541         struct mdt_thread_info  *info;
3542         struct ptlrpc_request   *req  =  req_cookie;
3543         struct ldlm_intent      *it;
3544         struct req_capsule      *pill;
3545         int rc;
3546
3547         ENTRY;
3548
3549         LASSERT(req != NULL);
3550
3551         tsi = tgt_ses_info(req->rq_svc_thread->t_env);
3552
3553         info = tsi2mdt_info(tsi);
3554         LASSERT(info != NULL);
3555         pill = info->mti_pill;
3556         LASSERT(pill->rc_req == req);
3557
3558         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3559                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
3560                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3561                 if (it != NULL) {
3562                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
3563                         if (rc == 0)
3564                                 rc = ELDLM_OK;
3565
3566                         /* Lock without inodebits makes no sense and will oops
3567                          * later in ldlm. Let's check it now to see if we have
3568                          * ibits corrupted somewhere in mdt_intent_opc().
3569                          * The case for client miss to set ibits has been
3570                          * processed by others. */
3571                         LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
3572                                         lr_type == LDLM_IBITS,
3573                                      info->mti_dlm_req->lock_desc.\
3574                                         l_policy_data.l_inodebits.bits != 0));
3575                 } else
3576                         rc = err_serious(-EFAULT);
3577         } else {
3578                 /* No intent was provided */
3579                 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3580                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
3581                 rc = req_capsule_server_pack(pill);
3582                 if (rc)
3583                         rc = err_serious(rc);
3584         }
3585         mdt_thread_info_fini(info);
3586         RETURN(rc);
3587 }
3588
3589 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
3590 {
3591         struct seq_server_site  *ss = mdt_seq_site(mdt);
3592
3593         if (ss->ss_node_id == 0)
3594                 return;
3595
3596         if (ss->ss_client_seq != NULL) {
3597                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3598                 ss->ss_client_seq->lcs_exp = NULL;
3599         }
3600
3601         if (ss->ss_server_fld != NULL) {
3602                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
3603                 ss->ss_server_fld->lsf_control_exp = NULL;
3604         }
3605 }
3606
3607 static void mdt_seq_fini_cli(struct mdt_device *mdt)
3608 {
3609         struct seq_server_site *ss = mdt_seq_site(mdt);
3610
3611         if (ss == NULL)
3612                 return;
3613
3614         if (ss->ss_server_seq != NULL)
3615                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
3616 }
3617
3618 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
3619 {
3620         mdt_seq_fini_cli(mdt);
3621         mdt_deregister_seq_exp(mdt);
3622
3623         return seq_site_fini(env, mdt_seq_site(mdt));
3624 }
3625
3626 /**
3627  * It will retrieve its FLDB entries from MDT0, and it only happens
3628  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
3629  * and it needs to refresh FLDB from the MDT0.
3630  **/
3631 static int mdt_register_lwp_callback(void *data)
3632 {
3633         struct lu_env           env;
3634         struct mdt_device       *mdt = data;
3635         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
3636         int                     rc;
3637         ENTRY;
3638
3639         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
3640
3641         if (!likely(fld->lsf_new))
3642                 RETURN(0);
3643
3644         rc = lu_env_init(&env, LCT_MD_THREAD);
3645         if (rc) {
3646                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
3647                 RETURN(rc);
3648         }
3649
3650         rc = fld_update_from_controller(&env, fld);
3651         if (rc != 0) {
3652                 CERROR("%s: cannot update controller: rc = %d\n",
3653                        mdt_obd_name(mdt), rc);
3654                 GOTO(out, rc);
3655         }
3656 out:
3657         lu_env_fini(&env);
3658         RETURN(rc);
3659 }
3660
3661 static int mdt_register_seq_exp(struct mdt_device *mdt)
3662 {
3663         struct seq_server_site  *ss = mdt_seq_site(mdt);
3664         char                    *lwp_name = NULL;
3665         int                     rc;
3666
3667         if (ss->ss_node_id == 0)
3668                 return 0;
3669
3670         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
3671         if (lwp_name == NULL)
3672                 GOTO(out_free, rc = -ENOMEM);
3673
3674         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
3675         if (rc != 0)
3676                 GOTO(out_free, rc);
3677
3678         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
3679                                       NULL, NULL);
3680         if (rc != 0)
3681                 GOTO(out_free, rc);
3682
3683         rc = lustre_register_lwp_item(lwp_name,
3684                                       &ss->ss_server_fld->lsf_control_exp,
3685                                       mdt_register_lwp_callback, mdt);
3686         if (rc != 0) {
3687                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3688                 ss->ss_client_seq->lcs_exp = NULL;
3689                 GOTO(out_free, rc);
3690         }
3691 out_free:
3692         if (lwp_name != NULL)
3693                 OBD_FREE(lwp_name, MAX_OBD_NAME);
3694
3695         return rc;
3696 }
3697
3698 /*
3699  * Init client sequence manager which is used by local MDS to talk to sequence
3700  * controller on remote node.
3701  */
3702 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
3703 {
3704         struct seq_server_site  *ss = mdt_seq_site(mdt);
3705         int                     rc;
3706         char                    *prefix;
3707         ENTRY;
3708
3709         /* check if this is adding the first MDC and controller is not yet
3710          * initialized. */
3711         OBD_ALLOC_PTR(ss->ss_client_seq);
3712         if (ss->ss_client_seq == NULL)
3713                 RETURN(-ENOMEM);
3714
3715         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3716         if (prefix == NULL) {
3717                 OBD_FREE_PTR(ss->ss_client_seq);
3718                 ss->ss_client_seq = NULL;
3719                 RETURN(-ENOMEM);
3720         }
3721
3722         /* Note: seq_client_fini will be called in seq_site_fini */
3723         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
3724         rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
3725                              prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
3726                                                             NULL);
3727         OBD_FREE(prefix, MAX_OBD_NAME + 5);
3728         if (rc != 0) {
3729                 OBD_FREE_PTR(ss->ss_client_seq);
3730                 ss->ss_client_seq = NULL;
3731                 RETURN(rc);
3732         }
3733
3734         rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
3735
3736         RETURN(rc);
3737 }
3738
3739 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
3740 {
3741         struct seq_server_site  *ss;
3742         int                     rc;
3743         ENTRY;
3744
3745         ss = mdt_seq_site(mdt);
3746         /* init sequence controller server(MDT0) */
3747         if (ss->ss_node_id == 0) {
3748                 OBD_ALLOC_PTR(ss->ss_control_seq);
3749                 if (ss->ss_control_seq == NULL)
3750                         RETURN(-ENOMEM);
3751
3752                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
3753                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
3754                                      ss);
3755                 if (rc)
3756                         GOTO(out_seq_fini, rc);
3757         }
3758
3759         /* Init normal sequence server */
3760         OBD_ALLOC_PTR(ss->ss_server_seq);
3761         if (ss->ss_server_seq == NULL)
3762                 GOTO(out_seq_fini, rc = -ENOMEM);
3763
3764         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
3765                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
3766         if (rc)
3767                 GOTO(out_seq_fini, rc);
3768
3769         /* init seq client for seq server to talk to seq controller(MDT0) */
3770         rc = mdt_seq_init_cli(env, mdt);
3771         if (rc != 0)
3772                 GOTO(out_seq_fini, rc);
3773
3774         if (ss->ss_node_id != 0)
3775                 /* register controler export through lwp */
3776                 rc = mdt_register_seq_exp(mdt);
3777
3778         EXIT;
3779 out_seq_fini:
3780         if (rc)
3781                 mdt_seq_fini(env, mdt);
3782
3783         return rc;
3784 }
3785
3786 /*
3787  * FLD wrappers
3788  */
3789 static int mdt_fld_fini(const struct lu_env *env,
3790                         struct mdt_device *m)
3791 {
3792         struct seq_server_site *ss = mdt_seq_site(m);
3793         ENTRY;
3794
3795         if (ss && ss->ss_server_fld) {
3796                 fld_server_fini(env, ss->ss_server_fld);
3797                 OBD_FREE_PTR(ss->ss_server_fld);
3798                 ss->ss_server_fld = NULL;
3799         }
3800
3801         RETURN(0);
3802 }
3803
3804 static int mdt_fld_init(const struct lu_env *env,
3805                         const char *uuid,
3806                         struct mdt_device *m)
3807 {
3808         struct seq_server_site *ss;
3809         int rc;
3810         ENTRY;
3811
3812         ss = mdt_seq_site(m);
3813
3814         OBD_ALLOC_PTR(ss->ss_server_fld);
3815         if (ss->ss_server_fld == NULL)
3816                 RETURN(rc = -ENOMEM);
3817
3818         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
3819                              LU_SEQ_RANGE_MDT);
3820         if (rc) {
3821                 OBD_FREE_PTR(ss->ss_server_fld);
3822                 ss->ss_server_fld = NULL;
3823                 RETURN(rc);
3824         }
3825
3826         RETURN(0);
3827 }
3828
3829 static void mdt_stack_pre_fini(const struct lu_env *env,
3830                            struct mdt_device *m, struct lu_device *top)
3831 {
3832         struct lustre_cfg_bufs  *bufs;
3833         struct lustre_cfg       *lcfg;
3834         struct mdt_thread_info  *info;
3835         ENTRY;
3836
3837         LASSERT(top);
3838
3839         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3840         LASSERT(info != NULL);
3841
3842         bufs = &info->mti_u.bufs;
3843
3844         LASSERT(m->mdt_child_exp);
3845         LASSERT(m->mdt_child_exp->exp_obd);
3846
3847         /* process cleanup, pass mdt obd name to get obd umount flags */
3848         /* XXX: this is needed because all layers are referenced by
3849          * objects (some of them are pinned by osd, for example *
3850          * the proper solution should be a model where object used
3851          * by osd only doesn't have mdt/mdd slices -bzzz */
3852         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
3853         lustre_cfg_bufs_set_string(bufs, 1, NULL);
3854         lcfg = lustre_cfg_new(LCFG_PRE_CLEANUP, bufs);
3855         if (!lcfg) {
3856                 CERROR("%s: cannot alloc lcfg\n", mdt_obd_name(m));
3857                 return;
3858         }
3859         top->ld_ops->ldo_process_config(env, top, lcfg);
3860         lustre_cfg_free(lcfg);
3861         EXIT;
3862 }
3863
3864 static void mdt_stack_fini(const struct lu_env *env,
3865                            struct mdt_device *m, struct lu_device *top)
3866 {
3867         struct obd_device       *obd = mdt2obd_dev(m);
3868         struct lustre_cfg_bufs  *bufs;
3869         struct lustre_cfg       *lcfg;
3870         struct mdt_thread_info  *info;
3871         char                     flags[3] = "";
3872         ENTRY;
3873
3874         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3875         LASSERT(info != NULL);
3876
3877         lu_dev_del_linkage(top->ld_site, top);
3878
3879         lu_site_purge(env, top->ld_site, -1);
3880
3881         bufs = &info->mti_u.bufs;
3882         /* process cleanup, pass mdt obd name to get obd umount flags */
3883         /* another purpose is to let all layers to release their objects */
3884         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
3885         if (obd->obd_force)
3886                 strcat(flags, "F");
3887         if (obd->obd_fail)
3888                 strcat(flags, "A");
3889         lustre_cfg_bufs_set_string(bufs, 1, flags);
3890         lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
3891         if (!lcfg) {
3892                 CERROR("Cannot alloc lcfg!\n");
3893                 return;
3894         }
3895         LASSERT(top);
3896         top->ld_ops->ldo_process_config(env, top, lcfg);
3897         lustre_cfg_free(lcfg);
3898
3899         lu_site_purge(env, top->ld_site, -1);
3900
3901         m->mdt_child = NULL;
3902         m->mdt_bottom = NULL;
3903
3904         obd_disconnect(m->mdt_child_exp);
3905         m->mdt_child_exp = NULL;
3906
3907         obd_disconnect(m->mdt_bottom_exp);
3908         m->mdt_child_exp = NULL;
3909 }
3910
3911 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
3912                                const char *next, struct obd_export **exp)
3913 {
3914         struct obd_connect_data *data = NULL;
3915         struct obd_device       *obd;
3916         int                      rc;
3917         ENTRY;
3918
3919         OBD_ALLOC_PTR(data);
3920         if (data == NULL)
3921                 GOTO(out, rc = -ENOMEM);
3922
3923         obd = class_name2obd(next);
3924         if (obd == NULL) {
3925                 CERROR("%s: can't locate next device: %s\n",
3926                        mdt_obd_name(m), next);
3927                 GOTO(out, rc = -ENOTCONN);
3928         }
3929
3930         data->ocd_connect_flags = OBD_CONNECT_VERSION;
3931         data->ocd_version = LUSTRE_VERSION_CODE;
3932
3933         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
3934         if (rc) {
3935                 CERROR("%s: cannot connect to next dev %s (%d)\n",
3936                        mdt_obd_name(m), next, rc);
3937                 GOTO(out, rc);
3938         }
3939
3940 out:
3941         if (data)
3942                 OBD_FREE_PTR(data);
3943         RETURN(rc);
3944 }
3945
3946 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
3947                           struct lustre_cfg *cfg)
3948 {
3949         char                   *dev = lustre_cfg_string(cfg, 0);
3950         int                     rc, name_size, uuid_size;
3951         char                   *name, *uuid, *p;
3952         struct lustre_cfg_bufs *bufs;
3953         struct lustre_cfg      *lcfg;
3954         struct obd_device      *obd;
3955         struct lustre_profile  *lprof;
3956         struct lu_site         *site;
3957         ENTRY;
3958
3959         /* in 1.8 we had the only device in the stack - MDS.
3960          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
3961          * in 2.3 OSD is instantiated by obd_mount.c, so we need
3962          * to generate names and setup MDT, MDD. MDT will be using
3963          * generated name to connect to MDD. for MDD the next device
3964          * will be LOD with name taken from so called "profile" which
3965          * is generated by mount_option line
3966          *
3967          * 1.8 MGS generates config. commands like this:
3968          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
3969          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
3970          * 2.0 MGS generates config. commands like this:
3971          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
3972          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
3973          *                    3:lustre-MDT0000-mdtlov  4:f
3974          *
3975          * we generate MDD name from MDT one, just replacing T with D
3976          *
3977          * after all the preparations, the logical equivalent will be
3978          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
3979          *                    3:lustre-MDT0000-mdtlov  4:f
3980          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
3981          *                    3:lustre-MDD0000  4:f
3982          *
3983          *  notice we build the stack from down to top: MDD first, then MDT */
3984
3985         name_size = MAX_OBD_NAME;
3986         uuid_size = MAX_OBD_NAME;
3987
3988         OBD_ALLOC(name, name_size);
3989         OBD_ALLOC(uuid, uuid_size);
3990         if (name == NULL || uuid == NULL)
3991                 GOTO(cleanup_mem, rc = -ENOMEM);
3992
3993         OBD_ALLOC_PTR(bufs);
3994         if (!bufs)
3995                 GOTO(cleanup_mem, rc = -ENOMEM);
3996
3997         strcpy(name, dev);
3998         p = strstr(name, "-MDT");
3999         if (p == NULL)
4000                 GOTO(free_bufs, rc = -ENOMEM);
4001         p[3] = 'D';
4002
4003         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
4004
4005         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4006         if (lprof == NULL || lprof->lp_dt == NULL) {
4007                 CERROR("can't find the profile: %s\n",
4008                        lustre_cfg_string(cfg, 0));
4009                 GOTO(free_bufs, rc = -EINVAL);
4010         }
4011
4012         lustre_cfg_bufs_reset(bufs, name);
4013         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
4014         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4015         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4016
4017         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4018         if (!lcfg)
4019                 GOTO(free_bufs, rc = -ENOMEM);
4020
4021         rc = class_attach(lcfg);
4022         if (rc)
4023                 GOTO(lcfg_cleanup, rc);
4024
4025         obd = class_name2obd(name);
4026         if (!obd) {
4027                 CERROR("Can not find obd %s (%s in config)\n",
4028                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
4029                 GOTO(lcfg_cleanup, rc = -EINVAL);
4030         }
4031
4032         lustre_cfg_free(lcfg);
4033
4034         lustre_cfg_bufs_reset(bufs, name);
4035         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4036         lustre_cfg_bufs_set_string(bufs, 2, dev);
4037         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4038
4039         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4040
4041         rc = class_setup(obd, lcfg);
4042         if (rc)
4043                 GOTO(class_detach, rc);
4044
4045         /* connect to MDD we just setup */
4046         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
4047         if (rc)
4048                 GOTO(class_detach, rc);
4049
4050         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
4051         LASSERT(site);
4052         LASSERT(mdt_lu_site(mdt) == NULL);
4053         mdt->mdt_lu_dev.ld_site = site;
4054         site->ls_top_dev = &mdt->mdt_lu_dev;
4055         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
4056
4057         /* now connect to bottom OSD */
4058         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
4059         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
4060         if (rc)
4061                 GOTO(class_detach, rc);
4062         mdt->mdt_bottom =
4063                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
4064
4065         rc = lu_env_refill((struct lu_env *)env);
4066         if (rc != 0)
4067                 CERROR("Failure to refill session: '%d'\n", rc);
4068
4069         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
4070
4071         EXIT;
4072 class_detach:
4073         if (rc)
4074                 class_detach(obd, lcfg);
4075 lcfg_cleanup:
4076         lustre_cfg_free(lcfg);
4077 free_bufs:
4078         OBD_FREE_PTR(bufs);
4079 cleanup_mem:
4080         if (name)
4081                 OBD_FREE(name, name_size);
4082         if (uuid)
4083                 OBD_FREE(uuid, uuid_size);
4084         RETURN(rc);
4085 }
4086
4087 /* setup quota master target on MDT0 */
4088 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
4089                           struct lustre_cfg *cfg)
4090 {
4091         struct obd_device       *obd;
4092         char                    *dev = lustre_cfg_string(cfg, 0);
4093         char                    *qmtname, *uuid, *p;
4094         struct lustre_cfg_bufs  *bufs;
4095         struct lustre_cfg       *lcfg;
4096         struct lustre_profile   *lprof;
4097         struct obd_connect_data *data;
4098         int                      rc;
4099         ENTRY;
4100
4101         LASSERT(mdt->mdt_qmt_exp == NULL);
4102         LASSERT(mdt->mdt_qmt_dev == NULL);
4103
4104         /* quota master is on MDT0 only for now */
4105         if (mdt->mdt_seq_site.ss_node_id != 0)
4106                 RETURN(0);
4107
4108         /* MGS generates config commands which look as follows:
4109          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4110          *                    3:lustre-MDT0000-mdtlov  4:f
4111          *
4112          * We generate the QMT name from the MDT one, just replacing MD with QM
4113          * after all the preparations, the logical equivalent will be:
4114          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
4115          *                    3:lustre-MDT0000-osd  4:f */
4116         OBD_ALLOC(qmtname, MAX_OBD_NAME);
4117         OBD_ALLOC(uuid, UUID_MAX);
4118         OBD_ALLOC_PTR(bufs);
4119         OBD_ALLOC_PTR(data);
4120         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
4121                 GOTO(cleanup_mem, rc = -ENOMEM);
4122
4123         strcpy(qmtname, dev);
4124         p = strstr(qmtname, "-MDT");
4125         if (p == NULL)
4126                 GOTO(cleanup_mem, rc = -ENOMEM);
4127         /* replace MD with QM */
4128         p[1] = 'Q';
4129         p[2] = 'M';
4130
4131         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
4132
4133         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4134         if (lprof == NULL || lprof->lp_dt == NULL) {
4135                 CERROR("can't find profile for %s\n",
4136                        lustre_cfg_string(cfg, 0));
4137                 GOTO(cleanup_mem, rc = -EINVAL);
4138         }
4139
4140         lustre_cfg_bufs_reset(bufs, qmtname);
4141         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
4142         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4143         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4144
4145         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4146         if (!lcfg)
4147                 GOTO(cleanup_mem, rc = -ENOMEM);
4148
4149         rc = class_attach(lcfg);
4150         if (rc)
4151                 GOTO(lcfg_cleanup, rc);
4152
4153         obd = class_name2obd(qmtname);
4154         if (!obd) {
4155                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
4156                        lustre_cfg_string(cfg, 0));
4157                 GOTO(lcfg_cleanup, rc = -EINVAL);
4158         }
4159
4160         lustre_cfg_free(lcfg);
4161
4162         lustre_cfg_bufs_reset(bufs, qmtname);
4163         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4164         lustre_cfg_bufs_set_string(bufs, 2, dev);
4165
4166         /* for quota, the next device should be the OSD device */
4167         lustre_cfg_bufs_set_string(bufs, 3,
4168                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
4169
4170         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4171
4172         rc = class_setup(obd, lcfg);
4173         if (rc)
4174                 GOTO(class_detach, rc);
4175
4176         mdt->mdt_qmt_dev = obd->obd_lu_dev;
4177
4178         /* configure local quota objects */
4179         rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
4180                                                    &mdt->mdt_lu_dev,
4181                                                    mdt->mdt_qmt_dev);
4182         if (rc)
4183                 GOTO(class_cleanup, rc);
4184
4185         /* connect to quota master target */
4186         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4187         data->ocd_version = LUSTRE_VERSION_CODE;
4188         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
4189                          data, NULL);
4190         if (rc) {
4191                 CERROR("cannot connect to quota master device %s (%d)\n",
4192                        qmtname, rc);
4193                 GOTO(class_cleanup, rc);
4194         }
4195
4196         EXIT;
4197 class_cleanup:
4198         if (rc) {
4199                 class_manual_cleanup(obd);
4200                 mdt->mdt_qmt_dev = NULL;
4201         }
4202 class_detach:
4203         if (rc)
4204                 class_detach(obd, lcfg);
4205 lcfg_cleanup:
4206         lustre_cfg_free(lcfg);
4207 cleanup_mem:
4208         if (bufs)
4209                 OBD_FREE_PTR(bufs);
4210         if (qmtname)
4211                 OBD_FREE(qmtname, MAX_OBD_NAME);
4212         if (uuid)
4213                 OBD_FREE(uuid, UUID_MAX);
4214         if (data)
4215                 OBD_FREE_PTR(data);
4216         return rc;
4217 }
4218
4219 /* Shutdown quota master target associated with mdt */
4220 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
4221 {
4222         ENTRY;
4223
4224         if (mdt->mdt_qmt_exp == NULL)
4225                 RETURN_EXIT;
4226         LASSERT(mdt->mdt_qmt_dev != NULL);
4227
4228         /* the qmt automatically shuts down when the mdt disconnects */
4229         obd_disconnect(mdt->mdt_qmt_exp);
4230         mdt->mdt_qmt_exp = NULL;
4231         mdt->mdt_qmt_dev = NULL;
4232         EXIT;
4233 }
4234
4235 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
4236  * for now. This will be removed along with converting rest of MDT code
4237  * to use tgt_session_info */
4238 int mdt_tgt_getxattr(struct tgt_session_info *tsi)
4239 {
4240         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
4241         int                      rc;
4242
4243         rc = mdt_getxattr(info);
4244
4245         mdt_thread_info_fini(info);
4246         return rc;
4247 }
4248
4249 static struct tgt_handler mdt_tgt_handlers[] = {
4250 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4251                 0,                      MDS_CONNECT,    mdt_tgt_connect,
4252                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
4253 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4254                 0,                      MDS_DISCONNECT, tgt_disconnect,
4255                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
4256 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4257                 HABEO_REFERO,           MDS_SET_INFO,   mdt_set_info,
4258                 &RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION),
4259 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
4260 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_GETSTATUS,  mdt_getstatus),
4261 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETATTR,    mdt_getattr),
4262 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_GETATTR_NAME,
4263                                                         mdt_getattr_name),
4264 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETXATTR,   mdt_tgt_getxattr),
4265 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_STATFS,     mdt_statfs),
4266 TGT_MDT_HDL(0           | MUTABOR,      MDS_REINT,      mdt_reint),
4267 TGT_MDT_HDL(HABEO_CORPUS,               MDS_CLOSE,      mdt_close),
4268 TGT_MDT_HDL(HABEO_CORPUS,               MDS_DONE_WRITING,
4269                                                         mdt_done_writing),
4270 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_READPAGE,   mdt_readpage),
4271 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_SYNC,       mdt_sync),
4272 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
4273 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_PROGRESS,
4274                                                         mdt_hsm_progress),
4275 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_REGISTER,
4276                                                         mdt_hsm_ct_register),
4277 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_UNREGISTER,
4278                                                         mdt_hsm_ct_unregister),
4279 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_STATE_GET,
4280                                                         mdt_hsm_state_get),
4281 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_STATE_SET,
4282                                                         mdt_hsm_state_set),
4283 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_ACTION, mdt_hsm_action),
4284 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_REQUEST,
4285                                                         mdt_hsm_request),
4286 TGT_MDT_HDL(HABEO_CLAVIS | HABEO_CORPUS | HABEO_REFERO | MUTABOR,
4287             MDS_SWAP_LAYOUTS,
4288             mdt_swap_layouts),
4289 };
4290
4291 static struct tgt_handler mdt_sec_ctx_ops[] = {
4292 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
4293 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
4294 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
4295 };
4296
4297 static struct tgt_handler mdt_quota_ops[] = {
4298 TGT_QUOTA_HDL(HABEO_REFERO,             QUOTA_DQACQ,      mdt_quota_dqacq),
4299 };
4300
4301 static struct tgt_opc_slice mdt_common_slice[] = {
4302         {
4303                 .tos_opc_start  = MDS_FIRST_OPC,
4304                 .tos_opc_end    = MDS_LAST_OPC,
4305                 .tos_hs         = mdt_tgt_handlers
4306         },
4307         {
4308                 .tos_opc_start  = OBD_FIRST_OPC,
4309                 .tos_opc_end    = OBD_LAST_OPC,
4310                 .tos_hs         = tgt_obd_handlers
4311         },
4312         {
4313                 .tos_opc_start  = LDLM_FIRST_OPC,
4314                 .tos_opc_end    = LDLM_LAST_OPC,
4315                 .tos_hs         = tgt_dlm_handlers
4316         },
4317         {
4318                 .tos_opc_start  = SEC_FIRST_OPC,
4319                 .tos_opc_end    = SEC_LAST_OPC,
4320                 .tos_hs         = mdt_sec_ctx_ops
4321         },
4322         {
4323                 .tos_opc_start  = OUT_UPDATE_FIRST_OPC,
4324                 .tos_opc_end    = OUT_UPDATE_LAST_OPC,
4325                 .tos_hs         = tgt_out_handlers
4326         },
4327         {
4328                 .tos_opc_start  = FLD_FIRST_OPC,
4329                 .tos_opc_end    = FLD_LAST_OPC,
4330                 .tos_hs         = fld_handlers
4331         },
4332         {
4333                 .tos_opc_start  = SEQ_FIRST_OPC,
4334                 .tos_opc_end    = SEQ_LAST_OPC,
4335                 .tos_hs         = seq_handlers
4336         },
4337         {
4338                 .tos_opc_start  = QUOTA_DQACQ,
4339                 .tos_opc_end    = QUOTA_LAST_OPC,
4340                 .tos_hs         = mdt_quota_ops
4341         },
4342         {
4343                 .tos_opc_start  = LLOG_FIRST_OPC,
4344                 .tos_opc_end    = LLOG_LAST_OPC,
4345                 .tos_hs         = tgt_llog_handlers
4346         },
4347         {
4348                 .tos_opc_start  = LFSCK_FIRST_OPC,
4349                 .tos_opc_end    = LFSCK_LAST_OPC,
4350                 .tos_hs         = tgt_lfsck_handlers
4351         },
4352
4353         {
4354                 .tos_hs         = NULL
4355         }
4356 };
4357
4358 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4359 {
4360         struct md_device        *next = m->mdt_child;
4361         struct lu_device        *d    = &m->mdt_lu_dev;
4362         struct obd_device       *obd  = mdt2obd_dev(m);
4363         struct lfsck_stop        stop;
4364         ENTRY;
4365
4366         stop.ls_status = LS_PAUSED;
4367         stop.ls_flags = 0;
4368         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
4369
4370         target_recovery_fini(obd);
4371         ping_evictor_stop();
4372         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
4373
4374         if (m->mdt_opts.mo_coordinator)
4375                 mdt_hsm_cdt_stop(m);
4376
4377         mdt_hsm_cdt_fini(m);
4378
4379         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
4380         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4381         obd_exports_barrier(obd);
4382         obd_zombie_barrier();
4383
4384         mdt_procfs_fini(m);
4385
4386         tgt_fini(env, &m->mdt_lut);
4387         mdt_fs_cleanup(env, m);
4388         upcall_cache_cleanup(m->mdt_identity_cache);
4389         m->mdt_identity_cache = NULL;
4390
4391         if (m->mdt_namespace != NULL) {
4392                 ldlm_namespace_free(m->mdt_namespace, NULL,
4393                                     d->ld_obd->obd_force);
4394                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4395         }
4396
4397         mdt_quota_fini(env, m);
4398
4399         cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
4400
4401         mdt_seq_fini(env, m);
4402         mdt_fld_fini(env, m);
4403
4404         next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL);
4405         cfs_timer_disarm(&m->mdt_ck_timer);
4406         mdt_ck_thread_stop(m);
4407
4408         /*
4409          * Finish the stack
4410          */
4411         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4412
4413         LASSERT(atomic_read(&d->ld_ref) == 0);
4414
4415         server_put_mount(mdt_obd_name(m), true);
4416
4417         EXIT;
4418 }
4419
4420 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4421
4422 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4423                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
4424 {
4425         struct mdt_thread_info    *info;
4426         struct obd_device         *obd;
4427         const char                *dev = lustre_cfg_string(cfg, 0);
4428         const char                *num = lustre_cfg_string(cfg, 2);
4429         struct lustre_mount_info  *lmi = NULL;
4430         struct lustre_sb_info     *lsi;
4431         struct lu_site            *s;
4432         struct seq_server_site    *ss_site;
4433         const char                *identity_upcall = "NONE";
4434         struct md_device          *next;
4435         int                        rc;
4436         long                       node_id;
4437         mntopt_t                   mntopts;
4438         ENTRY;
4439
4440         lu_device_init(&m->mdt_lu_dev, ldt);
4441         /*
4442          * Environment (env) might be missing mdt_thread_key values at that
4443          * point, if device is allocated when mdt_thread_key is in QUIESCENT
4444          * mode.
4445          *
4446          * Usually device allocation path doesn't use module key values, but
4447          * mdt has to do a lot of work here, so allocate key value.
4448          */
4449         rc = lu_env_refill((struct lu_env *)env);
4450         if (rc != 0)
4451                 RETURN(rc);
4452
4453         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4454         LASSERT(info != NULL);
4455
4456         obd = class_name2obd(dev);
4457         LASSERT(obd != NULL);
4458
4459         m->mdt_max_mdsize = MAX_MD_SIZE; /* 4 stripes */
4460
4461         m->mdt_som_conf = 0;
4462
4463         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4464
4465         /* default is coordinator off, it is started through conf_param
4466          * or /proc */
4467         m->mdt_opts.mo_coordinator = 0;
4468
4469         lmi = server_get_mount(dev);
4470         if (lmi == NULL) {
4471                 CERROR("Cannot get mount info for %s!\n", dev);
4472                 RETURN(-EFAULT);
4473         } else {
4474                 lsi = s2lsi(lmi->lmi_sb);
4475                 /* CMD is supported only in IAM mode */
4476                 LASSERT(num);
4477                 node_id = simple_strtol(num, NULL, 10);
4478                 obd->u.obt.obt_magic = OBT_MAGIC;
4479         }
4480
4481         spin_lock_init(&m->mdt_ioepoch_lock);
4482         m->mdt_capa_timeout = CAPA_TIMEOUT;
4483         m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1;
4484         m->mdt_ck_timeout = CAPA_KEY_TIMEOUT;
4485         m->mdt_squash.rsi_uid = 0;
4486         m->mdt_squash.rsi_gid = 0;
4487         INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
4488         init_rwsem(&m->mdt_squash.rsi_sem);
4489         spin_lock_init(&m->mdt_osfs_lock);
4490         m->mdt_osfs_age = cfs_time_shift_64(-1000);
4491         m->mdt_enable_remote_dir = 0;
4492         m->mdt_enable_remote_dir_gid = 0;
4493
4494         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
4495         m->mdt_lu_dev.ld_obd = obd;
4496         /* Set this lu_device to obd for error handling purposes. */
4497         obd->obd_lu_dev = &m->mdt_lu_dev;
4498
4499         /* init the stack */
4500         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
4501         if (rc) {
4502                 CERROR("%s: Can't init device stack, rc %d\n",
4503                        mdt_obd_name(m), rc);
4504                 GOTO(err_lmi, rc);
4505         }
4506
4507         s = mdt_lu_site(m);
4508         ss_site = mdt_seq_site(m);
4509         s->ld_seq_site = ss_site;
4510         ss_site->ss_lu = s;
4511
4512         /* set server index */
4513         ss_site->ss_node_id = node_id;
4514
4515         /* failover is the default
4516          * FIXME: we do not failout mds0/mgs, which may cause some problems.
4517          * assumed whose ss_node_id == 0 XXX
4518          * */
4519         obd->obd_replayable = 1;
4520         /* No connection accepted until configurations will finish */
4521         obd->obd_no_conn = 1;
4522
4523         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4524                 char *str = lustre_cfg_string(cfg, 4);
4525                 if (strchr(str, 'n')) {
4526                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
4527                         obd->obd_replayable = 0;
4528                 }
4529         }
4530
4531         rc = mdt_fld_init(env, mdt_obd_name(m), m);
4532         if (rc)
4533                 GOTO(err_fini_stack, rc);
4534
4535         rc = mdt_seq_init(env, m);
4536         if (rc)
4537                 GOTO(err_fini_fld, rc);
4538
4539         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
4540                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
4541         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4542                                               LDLM_NAMESPACE_SERVER,
4543                                               LDLM_NAMESPACE_GREEDY,
4544                                               LDLM_NS_TYPE_MDT);
4545         if (m->mdt_namespace == NULL)
4546                 GOTO(err_fini_seq, rc = -ENOMEM);
4547
4548         m->mdt_namespace->ns_lvbp = m;
4549         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
4550
4551         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4552         /* set obd_namespace for compatibility with old code */
4553         obd->obd_namespace = m->mdt_namespace;
4554
4555         cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m);
4556
4557         rc = mdt_hsm_cdt_init(m);
4558         if (rc != 0) {
4559                 CERROR("%s: error initializing coordinator, rc %d\n",
4560                        mdt_obd_name(m), rc);
4561                 GOTO(err_free_ns, rc);
4562         }
4563
4564         rc = mdt_ck_thread_start(m);
4565         if (rc)
4566                 GOTO(err_free_hsm, rc);
4567
4568         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
4569                       OBD_FAIL_MDS_ALL_REQUEST_NET,
4570                       OBD_FAIL_MDS_ALL_REPLY_NET);
4571         if (rc)
4572                 GOTO(err_capa, rc);
4573
4574         rc = mdt_fs_setup(env, m, obd, lsi);
4575         if (rc)
4576                 GOTO(err_tgt, rc);
4577
4578         tgt_adapt_sptlrpc_conf(&m->mdt_lut, 1);
4579
4580         next = m->mdt_child;
4581         rc = next->md_ops->mdo_iocontrol(env, next, OBD_IOC_GET_MNTOPT, 0,
4582                                          &mntopts);
4583         if (rc)
4584                 GOTO(err_fs_cleanup, rc);
4585
4586         if (mntopts & MNTOPT_USERXATTR)
4587                 m->mdt_opts.mo_user_xattr = 1;
4588         else
4589                 m->mdt_opts.mo_user_xattr = 0;
4590
4591         rc = next->md_ops->mdo_maxeasize_get(env, next, &m->mdt_max_ea_size);
4592         if (rc)
4593                 GOTO(err_fs_cleanup, rc);
4594
4595         if (mntopts & MNTOPT_ACL)
4596                 m->mdt_opts.mo_acl = 1;
4597         else
4598                 m->mdt_opts.mo_acl = 0;
4599
4600         /* XXX: to support suppgid for ACL, we enable identity_upcall
4601          * by default, otherwise, maybe got unexpected -EACCESS. */
4602         if (m->mdt_opts.mo_acl)
4603                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4604
4605         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
4606                                                 identity_upcall,
4607                                                 &mdt_identity_upcall_cache_ops);
4608         if (IS_ERR(m->mdt_identity_cache)) {
4609                 rc = PTR_ERR(m->mdt_identity_cache);
4610                 m->mdt_identity_cache = NULL;
4611                 GOTO(err_fs_cleanup, rc);
4612         }
4613
4614         rc = mdt_procfs_init(m, dev);
4615         if (rc) {
4616                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4617                 GOTO(err_recovery, rc);
4618         }
4619
4620         rc = mdt_quota_init(env, m, cfg);
4621         if (rc)
4622                 GOTO(err_procfs, rc);
4623
4624         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
4625         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
4626                            "mdt_ldlm_client", m->mdt_ldlm_client);
4627
4628         ping_evictor_start();
4629
4630         /* recovery will be started upon mdt_prepare()
4631          * when the whole stack is complete and ready
4632          * to serve the requests */
4633
4634         mdt_init_capa_ctxt(env, m);
4635
4636         /* Reduce the initial timeout on an MDS because it doesn't need such
4637          * a long timeout as an OST does. Adaptive timeouts will adjust this
4638          * value appropriately. */
4639         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4640                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4641
4642         RETURN(0);
4643 err_procfs:
4644         mdt_procfs_fini(m);
4645 err_recovery:
4646         target_recovery_fini(obd);
4647         upcall_cache_cleanup(m->mdt_identity_cache);
4648         m->mdt_identity_cache = NULL;
4649 err_fs_cleanup:
4650         mdt_fs_cleanup(env, m);
4651 err_tgt:
4652         tgt_fini(env, &m->mdt_lut);
4653 err_capa:
4654         cfs_timer_disarm(&m->mdt_ck_timer);
4655         mdt_ck_thread_stop(m);
4656 err_free_hsm:
4657         mdt_hsm_cdt_fini(m);
4658 err_free_ns:
4659         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4660         obd->obd_namespace = m->mdt_namespace = NULL;
4661 err_fini_seq:
4662         mdt_seq_fini(env, m);
4663 err_fini_fld:
4664         mdt_fld_fini(env, m);
4665 err_fini_stack:
4666         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4667 err_lmi:
4668         if (lmi)
4669                 server_put_mount(dev, true);
4670         return(rc);
4671 }
4672
4673 /* For interoperability, the left element is old parameter, the right one
4674  * is the new version of the parameter, if some parameter is deprecated,
4675  * the new version should be set as NULL. */
4676 static struct cfg_interop_param mdt_interop_param[] = {
4677         { "mdt.group_upcall",   NULL },
4678         { "mdt.quota_type",     NULL },
4679         { "mdd.quota_type",     NULL },
4680         { "mdt.rootsquash",     "mdt.root_squash" },
4681         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
4682         { NULL }
4683 };
4684
4685 /* used by MGS to process specific configurations */
4686 static int mdt_process_config(const struct lu_env *env,
4687                               struct lu_device *d, struct lustre_cfg *cfg)
4688 {
4689         struct mdt_device *m = mdt_dev(d);
4690         struct md_device *md_next = m->mdt_child;
4691         struct lu_device *next = md2lu_dev(md_next);
4692         int rc;
4693         ENTRY;
4694
4695         switch (cfg->lcfg_command) {
4696         case LCFG_PARAM: {
4697                 struct obd_device          *obd = d->ld_obd;
4698
4699                 /* For interoperability */
4700                 struct cfg_interop_param   *ptr = NULL;
4701                 struct lustre_cfg          *old_cfg = NULL;
4702                 char                       *param = NULL;
4703
4704                 param = lustre_cfg_string(cfg, 1);
4705                 if (param == NULL) {
4706                         CERROR("param is empty\n");
4707                         rc = -EINVAL;
4708                         break;
4709                 }
4710
4711                 ptr = class_find_old_param(param, mdt_interop_param);
4712                 if (ptr != NULL) {
4713                         if (ptr->new_param == NULL) {
4714                                 rc = 0;
4715                                 CWARN("For interoperability, skip this %s."
4716                                       " It is obsolete.\n", ptr->old_param);
4717                                 break;
4718                         }
4719
4720                         CWARN("Found old param %s, changed it to %s.\n",
4721                               ptr->old_param, ptr->new_param);
4722
4723                         old_cfg = cfg;
4724                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
4725                         if (IS_ERR(cfg)) {
4726                                 rc = PTR_ERR(cfg);
4727                                 break;
4728                         }
4729                 }
4730
4731                 rc = class_process_proc_param(PARAM_MDT, obd->obd_vars,
4732                                               cfg, obd);
4733                 if (rc > 0 || rc == -ENOSYS) {
4734                         /* is it an HSM var ? */
4735                         rc = class_process_proc_param(PARAM_HSM,
4736                                                       hsm_cdt_get_proc_vars(),
4737                                                       cfg, obd);
4738                         if (rc > 0 || rc == -ENOSYS)
4739                                 /* we don't understand; pass it on */
4740                                 rc = next->ld_ops->ldo_process_config(env, next,
4741                                                                       cfg);
4742                 }
4743
4744                 if (old_cfg != NULL)
4745                         lustre_cfg_free(cfg);
4746
4747                 break;
4748         }
4749         default:
4750                 /* others are passed further */
4751                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4752                 break;
4753         }
4754         RETURN(rc);
4755 }
4756
4757 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4758                                           const struct lu_object_header *hdr,
4759                                           struct lu_device *d)
4760 {
4761         struct mdt_object *mo;
4762
4763         ENTRY;
4764
4765         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
4766         if (mo != NULL) {
4767                 struct lu_object *o;
4768                 struct lu_object_header *h;
4769
4770                 o = &mo->mot_obj;
4771                 h = &mo->mot_header;
4772                 lu_object_header_init(h);
4773                 lu_object_init(o, h, d);
4774                 lu_object_add_top(h, o);
4775                 o->lo_ops = &mdt_obj_ops;
4776                 mutex_init(&mo->mot_ioepoch_mutex);
4777                 mutex_init(&mo->mot_lov_mutex);
4778                 init_rwsem(&mo->mot_open_sem);
4779                 RETURN(o);
4780         }
4781         RETURN(NULL);
4782 }
4783
4784 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
4785                            const struct lu_object_conf *unused)
4786 {
4787         struct mdt_device *d = mdt_dev(o->lo_dev);
4788         struct lu_device  *under;
4789         struct lu_object  *below;
4790         int                rc = 0;
4791         ENTRY;
4792
4793         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4794                PFID(lu_object_fid(o)));
4795
4796         under = &d->mdt_child->md_lu_dev;
4797         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4798         if (below != NULL) {
4799                 lu_object_add(o, below);
4800         } else
4801                 rc = -ENOMEM;
4802
4803         RETURN(rc);
4804 }
4805
4806 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4807 {
4808         struct mdt_object *mo = mdt_obj(o);
4809         struct lu_object_header *h;
4810         ENTRY;
4811
4812         h = o->lo_header;
4813         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4814                PFID(lu_object_fid(o)));
4815
4816         LASSERT(atomic_read(&mo->mot_open_count) == 0);
4817         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
4818
4819         lu_object_fini(o);
4820         lu_object_header_fini(h);
4821         OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
4822
4823         EXIT;
4824 }
4825
4826 static int mdt_object_print(const struct lu_env *env, void *cookie,
4827                             lu_printer_t p, const struct lu_object *o)
4828 {
4829         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
4830         return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p(ioepoch="LPU64" "
4831                     "flags="LPX64", epochcount=%d, writecount=%d)",
4832                     mdto, mdto->mot_ioepoch, mdto->mot_flags,
4833                     mdto->mot_ioepoch_count, mdto->mot_writecount);
4834 }
4835
4836 static int mdt_prepare(const struct lu_env *env,
4837                 struct lu_device *pdev,
4838                 struct lu_device *cdev)
4839 {
4840         struct mdt_device *mdt = mdt_dev(cdev);
4841         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
4842         struct obd_device *obd = cdev->ld_obd;
4843         int rc;
4844
4845         ENTRY;
4846
4847         LASSERT(obd);
4848
4849         rc = next->ld_ops->ldo_prepare(env, cdev, next);
4850         if (rc)
4851                 RETURN(rc);
4852
4853         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
4854         if (rc)
4855                 RETURN(rc);
4856
4857         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
4858         if (rc)
4859                 RETURN(rc);
4860
4861         rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
4862         /* The LFSCK instance is registered just now, so it must be there when
4863          * register the namespace to such instance. */
4864         LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
4865
4866         if (mdt->mdt_seq_site.ss_node_id == 0) {
4867                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
4868                                                          &mdt->mdt_md_root_fid);
4869                 if (rc)
4870                         RETURN(rc);
4871         }
4872
4873         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
4874         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
4875         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
4876         LASSERT(obd->obd_no_conn);
4877         spin_lock(&obd->obd_dev_lock);
4878         obd->obd_no_conn = 0;
4879         spin_unlock(&obd->obd_dev_lock);
4880
4881         if (obd->obd_recovering == 0)
4882                 mdt_postrecov(env, mdt);
4883
4884         RETURN(rc);
4885 }
4886
4887 const struct lu_device_operations mdt_lu_ops = {
4888         .ldo_object_alloc   = mdt_object_alloc,
4889         .ldo_process_config = mdt_process_config,
4890         .ldo_prepare        = mdt_prepare,
4891 };
4892
4893 static const struct lu_object_operations mdt_obj_ops = {
4894         .loo_object_init    = mdt_object_init,
4895         .loo_object_free    = mdt_object_free,
4896         .loo_object_print   = mdt_object_print
4897 };
4898
4899 static int mdt_obd_set_info_async(const struct lu_env *env,
4900                                   struct obd_export *exp,
4901                                   __u32 keylen, void *key,
4902                                   __u32 vallen, void *val,
4903                                   struct ptlrpc_request_set *set)
4904 {
4905         int rc;
4906
4907         ENTRY;
4908
4909         if (KEY_IS(KEY_SPTLRPC_CONF)) {
4910                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp), 0);
4911                 RETURN(rc);
4912         }
4913
4914         RETURN(0);
4915 }
4916
4917 /**
4918  * Match client and server connection feature flags.
4919  *
4920  * Compute the compatibility flags for a connection request based on
4921  * features mutually supported by client and server.
4922  *
4923  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
4924  * must not be updated here, otherwise a partially initialized value may
4925  * be exposed. After the connection request is successfully processed,
4926  * the top-level MDT connect request handler atomically updates the export
4927  * connect flags from the obd_connect_data::ocd_connect_flags field of the
4928  * reply. \see mdt_connect().
4929  *
4930  * \param exp   the obd_export associated with this client/target pair
4931  * \param mdt   the target device for the connection
4932  * \param data  stores data for this connect request
4933  *
4934  * \retval 0       success
4935  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
4936  * \retval -EBADE  client and server feature requirements are incompatible
4937  */
4938 static int mdt_connect_internal(struct obd_export *exp,
4939                                 struct mdt_device *mdt,
4940                                 struct obd_connect_data *data)
4941 {
4942         LASSERT(data != NULL);
4943
4944         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
4945         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
4946
4947         if (!(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
4948             !(data->ocd_connect_flags & OBD_CONNECT_IBITS)) {
4949                 CWARN("%s: client %s does not support ibits lock, either "
4950                       "very old or an invalid client: flags "LPX64"\n",
4951                       mdt_obd_name(mdt), exp->exp_client_uuid.uuid,
4952                       data->ocd_connect_flags);
4953                 return -EBADE;
4954         }
4955
4956         if (!mdt->mdt_opts.mo_acl)
4957                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
4958
4959         if (!mdt->mdt_opts.mo_user_xattr)
4960                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
4961
4962         if (!mdt->mdt_som_conf)
4963                 data->ocd_connect_flags &= ~OBD_CONNECT_SOM;
4964
4965         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
4966                 data->ocd_brw_size = min(data->ocd_brw_size,
4967                                          (__u32)MD_MAX_BRW_SIZE);
4968                 if (data->ocd_brw_size == 0) {
4969                         CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
4970                                " ocd_version: %x ocd_grant: %d "
4971                                "ocd_index: %u ocd_brw_size is "
4972                                "unexpectedly zero, network data "
4973                                "corruption? Refusing connection of this"
4974                                " client\n",
4975                                mdt_obd_name(mdt),
4976                                exp->exp_client_uuid.uuid,
4977                                exp, data->ocd_connect_flags, data->ocd_version,
4978                                data->ocd_grant, data->ocd_index);
4979                         return -EPROTO;
4980                 }
4981         }
4982
4983         /* NB: Disregard the rule against updating
4984          * exp_connect_data.ocd_connect_flags in this case, since
4985          * tgt_client_new() needs to know if this is a lightweight
4986          * connection, and it is safe to expose this flag before
4987          * connection processing completes. */
4988         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
4989                 spin_lock(&exp->exp_lock);
4990                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
4991                 spin_unlock(&exp->exp_lock);
4992         }
4993
4994         data->ocd_version = LUSTRE_VERSION_CODE;
4995
4996         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
4997                 CWARN("%s: MDS requires FID support, but client not\n",
4998                       mdt_obd_name(mdt));
4999                 return -EBADE;
5000         }
5001
5002         if (mdt->mdt_som_conf &&
5003             !(data->ocd_connect_flags & (OBD_CONNECT_LIGHTWEIGHT |
5004                                          OBD_CONNECT_MDS_MDS |
5005                                          OBD_CONNECT_SOM))) {
5006                 CWARN("%s: MDS has SOM enabled, but client does not support "
5007                       "it\n", mdt_obd_name(mdt));
5008                 return -EBADE;
5009         }
5010
5011         if (OCD_HAS_FLAG(data, PINGLESS)) {
5012                 if (ptlrpc_pinger_suppress_pings()) {
5013                         spin_lock(&exp->exp_obd->obd_dev_lock);
5014                         list_del_init(&exp->exp_obd_chain_timed);
5015                         spin_unlock(&exp->exp_obd->obd_dev_lock);
5016                 } else {
5017                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
5018                 }
5019         }
5020
5021         data->ocd_max_easize = mdt->mdt_max_ea_size;
5022
5023         return 0;
5024 }
5025
5026 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
5027                                    struct mdt_thread_info *info,
5028                                    struct mdt_file_data *mfd)
5029 {
5030         struct lu_context ses;
5031         int rc;
5032         ENTRY;
5033
5034         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
5035         if (rc)
5036                 RETURN(rc);
5037
5038         env->le_ses = &ses;
5039         lu_context_enter(&ses);
5040
5041         mdt_ucred(info)->uc_valid = UCRED_OLD;
5042         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
5043
5044         lu_context_exit(&ses);
5045         lu_context_fini(&ses);
5046         env->le_ses = NULL;
5047
5048         RETURN(rc);
5049 }
5050
5051 static int mdt_export_cleanup(struct obd_export *exp)
5052 {
5053         struct list_head         closing_list;
5054         struct mdt_export_data  *med = &exp->exp_mdt_data;
5055         struct obd_device       *obd = exp->exp_obd;
5056         struct mdt_device       *mdt;
5057         struct mdt_thread_info  *info;
5058         struct lu_env            env;
5059         struct mdt_file_data    *mfd, *n;
5060         int rc = 0;
5061         ENTRY;
5062
5063         INIT_LIST_HEAD(&closing_list);
5064         spin_lock(&med->med_open_lock);
5065         while (!list_empty(&med->med_open_head)) {
5066                 struct list_head *tmp = med->med_open_head.next;
5067                 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
5068
5069                 /* Remove mfd handle so it can't be found again.
5070                  * We are consuming the mfd_list reference here. */
5071                 class_handle_unhash(&mfd->mfd_handle);
5072                 list_move_tail(&mfd->mfd_list, &closing_list);
5073         }
5074         spin_unlock(&med->med_open_lock);
5075         mdt = mdt_dev(obd->obd_lu_dev);
5076         LASSERT(mdt != NULL);
5077
5078         rc = lu_env_init(&env, LCT_MD_THREAD);
5079         if (rc)
5080                 RETURN(rc);
5081
5082         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5083         LASSERT(info != NULL);
5084         memset(info, 0, sizeof *info);
5085         info->mti_env = &env;
5086         info->mti_mdt = mdt;
5087         info->mti_exp = exp;
5088
5089         if (!list_empty(&closing_list)) {
5090                 struct md_attr *ma = &info->mti_attr;
5091
5092                 /* Close any open files (which may also cause orphan
5093                  * unlinking). */
5094                 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5095                         list_del_init(&mfd->mfd_list);
5096                         ma->ma_need = ma->ma_valid = 0;
5097
5098                         /* This file is being closed due to an eviction, it
5099                          * could have been modified and now dirty regarding to
5100                          * HSM archive, check this!
5101                          * The logic here is to mark a file dirty if there's a
5102                          * chance it was dirtied before the client was evicted,
5103                          * so that we don't have to wait for a release attempt
5104                          * before finding out the file was actually dirty and
5105                          * fail the release. Aggressively marking it dirty here
5106                          * will cause the policy engine to attempt to
5107                          * re-archive it; when rearchiving, we can compare the
5108                          * current version to the HSM data_version and make the
5109                          * archive request into a noop if it's not actually
5110                          * dirty.
5111                          */
5112                         if (mfd->mfd_mode & (FMODE_WRITE|MDS_FMODE_TRUNC))
5113                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
5114
5115                         /* Don't unlink orphan on failover umount, LU-184 */
5116                         if (exp->exp_flags & OBD_OPT_FAILOVER) {
5117                                 ma->ma_valid = MA_FLAGS;
5118                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5119                         }
5120                         mdt_mfd_close(info, mfd);
5121                 }
5122         }
5123         info->mti_mdt = NULL;
5124         /* cleanup client slot early */
5125         /* Do not erase record for recoverable client. */
5126         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5127                 tgt_client_del(&env, exp);
5128         lu_env_fini(&env);
5129
5130         RETURN(rc);
5131 }
5132
5133 static int mdt_obd_disconnect(struct obd_export *exp)
5134 {
5135         int rc;
5136         ENTRY;
5137
5138         LASSERT(exp);
5139         class_export_get(exp);
5140
5141         nodemap_del_member(exp);
5142         rc = server_disconnect_export(exp);
5143         if (rc != 0)
5144                 CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
5145
5146         rc = mdt_export_cleanup(exp);
5147         class_export_put(exp);
5148         RETURN(rc);
5149 }
5150
5151 /* mds_connect copy */
5152 static int mdt_obd_connect(const struct lu_env *env,
5153                            struct obd_export **exp, struct obd_device *obd,
5154                            struct obd_uuid *cluuid,
5155                            struct obd_connect_data *data,
5156                            void *localdata)
5157 {
5158         struct obd_export       *lexp;
5159         struct lustre_handle    conn = { 0 };
5160         struct mdt_device       *mdt;
5161         int                      rc;
5162         lnet_nid_t              *client_nid = localdata;
5163         ENTRY;
5164
5165         LASSERT(env != NULL);
5166         if (!exp || !obd || !cluuid)
5167                 RETURN(-EINVAL);
5168
5169         mdt = mdt_dev(obd->obd_lu_dev);
5170
5171         /*
5172          * first, check whether the stack is ready to handle requests
5173          * XXX: probably not very appropriate method is used now
5174          *      at some point we should find a better one
5175          */
5176         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) && data != NULL &&
5177             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
5178                 rc = obd_get_info(env, mdt->mdt_child_exp,
5179                                   sizeof(KEY_OSP_CONNECTED),
5180                                   KEY_OSP_CONNECTED, NULL, NULL, NULL);
5181                 if (rc)
5182                         RETURN(-EAGAIN);
5183                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
5184         }
5185
5186         rc = class_connect(&conn, obd, cluuid);
5187         if (rc)
5188                 RETURN(rc);
5189
5190         lexp = class_conn2export(&conn);
5191         LASSERT(lexp != NULL);
5192
5193         rc = mdt_connect_internal(lexp, mdt, data);
5194         if (rc == 0) {
5195                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5196
5197                 LASSERT(lcd);
5198                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5199                 rc = tgt_client_new(env, lexp);
5200                 if (rc == 0) {
5201                         rc = nodemap_add_member(*client_nid, lexp);
5202                         if (rc != 0 && rc != -EEXIST)
5203                                 goto out;
5204
5205                         mdt_export_stats_init(obd, lexp, localdata);
5206                 }
5207
5208                 /* For phase I, sync for cross-ref operation. */
5209                 spin_lock(&lexp->exp_lock);
5210                 lexp->exp_keep_sync = 1;
5211                 spin_unlock(&lexp->exp_lock);
5212         }
5213 out:
5214         if (rc != 0) {
5215                 class_disconnect(lexp);
5216                 *exp = NULL;
5217         } else {
5218                 *exp = lexp;
5219         }
5220
5221         RETURN(rc);
5222 }
5223
5224 static int mdt_obd_reconnect(const struct lu_env *env,
5225                              struct obd_export *exp, struct obd_device *obd,
5226                              struct obd_uuid *cluuid,
5227                              struct obd_connect_data *data,
5228                              void *localdata)
5229 {
5230         lnet_nid_t             *client_nid = localdata;
5231         int                     rc;
5232         ENTRY;
5233
5234         if (exp == NULL || obd == NULL || cluuid == NULL)
5235                 RETURN(-EINVAL);
5236
5237         rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
5238         if (rc == 0) {
5239                 rc = nodemap_add_member(*client_nid, exp);
5240                 if (rc == 0 || rc == -EEXIST)
5241                         mdt_export_stats_init(obd, exp, localdata);
5242         }
5243
5244         RETURN(rc);
5245 }
5246
5247 /* FIXME: Can we avoid using these two interfaces? */
5248 static int mdt_init_export(struct obd_export *exp)
5249 {
5250         struct mdt_export_data *med = &exp->exp_mdt_data;
5251         int                     rc;
5252         ENTRY;
5253
5254         INIT_LIST_HEAD(&med->med_open_head);
5255         spin_lock_init(&med->med_open_lock);
5256         mutex_init(&med->med_idmap_mutex);
5257         med->med_idmap = NULL;
5258         spin_lock(&exp->exp_lock);
5259         exp->exp_connecting = 1;
5260         spin_unlock(&exp->exp_lock);
5261
5262         /* self-export doesn't need client data and ldlm initialization */
5263         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5264                                      &exp->exp_client_uuid)))
5265                 RETURN(0);
5266
5267         rc = tgt_client_alloc(exp);
5268         if (rc)
5269                 GOTO(err, rc);
5270
5271         rc = ldlm_init_export(exp);
5272         if (rc)
5273                 GOTO(err_free, rc);
5274
5275         RETURN(rc);
5276
5277 err_free:
5278         tgt_client_free(exp);
5279 err:
5280         CERROR("%s: Failed to initialize export: rc = %d\n",
5281                exp->exp_obd->obd_name, rc);
5282         return rc;
5283 }
5284
5285 static int mdt_destroy_export(struct obd_export *exp)
5286 {
5287         ENTRY;
5288
5289         if (exp_connect_rmtclient(exp))
5290                 mdt_cleanup_idmap(&exp->exp_mdt_data);
5291
5292         target_destroy_export(exp);
5293         /* destroy can be called from failed obd_setup, so
5294          * checking uuid is safer than obd_self_export */
5295         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5296                                      &exp->exp_client_uuid)))
5297                 RETURN(0);
5298
5299         ldlm_destroy_export(exp);
5300         tgt_client_free(exp);
5301
5302         LASSERT(list_empty(&exp->exp_outstanding_replies));
5303         LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
5304
5305         RETURN(0);
5306 }
5307
5308 /** The maximum depth that fid2path() will search.
5309  * This is limited only because we want to store the fids for
5310  * historical path lookup purposes.
5311  */
5312 #define MAX_PATH_DEPTH 100
5313
5314 /** mdt_path() lookup structure. */
5315 struct path_lookup_info {
5316         __u64                   pli_recno;      /**< history point */
5317         __u64                   pli_currec;     /**< current record */
5318         struct lu_fid           pli_fid;
5319         struct lu_fid           pli_fids[MAX_PATH_DEPTH]; /**< path, in fids */
5320         struct mdt_object       *pli_mdt_obj;
5321         char                    *pli_path;      /**< full path */
5322         int                     pli_pathlen;
5323         int                     pli_linkno;     /**< which hardlink to follow */
5324         int                     pli_fidcount;   /**< number of \a pli_fids */
5325 };
5326
5327 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
5328                    struct linkea_data *ldata)
5329 {
5330         int rc;
5331
5332         LASSERT(ldata->ld_buf->lb_buf != NULL);
5333
5334         if (!mdt_object_exists(mdt_obj))
5335                 return -ENODATA;
5336
5337         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5338                           ldata->ld_buf, XATTR_NAME_LINK);
5339         if (rc == -ERANGE) {
5340                 /* Buf was too small, figure out what we need. */
5341                 lu_buf_free(ldata->ld_buf);
5342                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5343                                   ldata->ld_buf, XATTR_NAME_LINK);
5344                 if (rc < 0)
5345                         return rc;
5346                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
5347                 if (ldata->ld_buf->lb_buf == NULL)
5348                         return -ENOMEM;
5349                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5350                                   ldata->ld_buf, XATTR_NAME_LINK);
5351         }
5352         if (rc < 0)
5353                 return rc;
5354
5355         return linkea_init(ldata);
5356 }
5357
5358 static int mdt_path_current(struct mdt_thread_info *info,
5359                             struct path_lookup_info *pli)
5360 {
5361         struct mdt_device       *mdt = info->mti_mdt;
5362         struct mdt_object       *mdt_obj;
5363         struct link_ea_header   *leh;
5364         struct link_ea_entry    *lee;
5365         struct lu_name          *tmpname = &info->mti_name;
5366         struct lu_fid           *tmpfid = &info->mti_tmp_fid1;
5367         struct lu_buf           *buf = &info->mti_big_buf;
5368         char                    *ptr;
5369         int                     reclen;
5370         struct linkea_data      ldata = { 0 };
5371         int                     rc = 0;
5372         ENTRY;
5373
5374         /* temp buffer for path element, the buffer will be finally freed
5375          * in mdt_thread_info_fini */
5376         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
5377         if (buf->lb_buf == NULL)
5378                 RETURN(-ENOMEM);
5379
5380         ldata.ld_buf = buf;
5381         ptr = pli->pli_path + pli->pli_pathlen - 1;
5382         *ptr = 0;
5383         --ptr;
5384         pli->pli_fidcount = 0;
5385         pli->pli_fids[0] = *(struct lu_fid *)mdt_object_fid(pli->pli_mdt_obj);
5386         *tmpfid = pli->pli_fids[0];
5387         /* root FID only exists on MDT0, and fid2path should also ends at MDT0,
5388          * so checking root_fid can only happen on MDT0. */
5389         while (!lu_fid_eq(&mdt->mdt_md_root_fid,
5390                           &pli->pli_fids[pli->pli_fidcount])) {
5391                 struct lu_buf           lmv_buf;
5392
5393                 mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
5394                 if (IS_ERR(mdt_obj))
5395                         GOTO(out, rc = PTR_ERR(mdt_obj));
5396
5397                 if (!mdt_object_exists(mdt_obj)) {
5398                         mdt_object_put(info->mti_env, mdt_obj);
5399                         GOTO(out, rc = -ENOENT);
5400                 }
5401
5402                 if (mdt_object_remote(mdt_obj)) {
5403                         mdt_object_put(info->mti_env, mdt_obj);
5404                         GOTO(remote_out, rc = -EREMOTE);
5405                 }
5406
5407                 rc = mdt_links_read(info, mdt_obj, &ldata);
5408                 if (rc != 0) {
5409                         mdt_object_put(info->mti_env, mdt_obj);
5410                         GOTO(out, rc);
5411                 }
5412
5413                 leh = buf->lb_buf;
5414                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
5415                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
5416                 /* If set, use link #linkno for path lookup, otherwise use
5417                    link #0.  Only do this for the final path element. */
5418                 if (pli->pli_fidcount == 0 &&
5419                     pli->pli_linkno < leh->leh_reccount) {
5420                         int count;
5421                         for (count = 0; count < pli->pli_linkno; count++) {
5422                                 lee = (struct link_ea_entry *)
5423                                      ((char *)lee + reclen);
5424                                 linkea_entry_unpack(lee, &reclen, tmpname,
5425                                                     tmpfid);
5426                         }
5427                         if (pli->pli_linkno < leh->leh_reccount - 1)
5428                                 /* indicate to user there are more links */
5429                                 pli->pli_linkno++;
5430                 }
5431
5432                 lmv_buf.lb_buf = info->mti_xattr_buf;
5433                 lmv_buf.lb_len = sizeof(info->mti_xattr_buf);
5434                 /* Check if it is slave stripes */
5435                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5436                                   &lmv_buf, XATTR_NAME_LMV);
5437                 mdt_object_put(info->mti_env, mdt_obj);
5438                 if (rc > 0) {
5439                         union lmv_mds_md *lmm = lmv_buf.lb_buf;
5440
5441                         /* For slave stripes, get its master */
5442                         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE) {
5443                                 pli->pli_fids[pli->pli_fidcount] = *tmpfid;
5444                                 continue;
5445                         }
5446                 } else if (rc < 0 && rc != -ENODATA) {
5447                         GOTO(out, rc);
5448                 }
5449
5450                 rc = 0;
5451
5452                 /* Pack the name in the end of the buffer */
5453                 ptr -= tmpname->ln_namelen;
5454                 if (ptr - 1 <= pli->pli_path)
5455                         GOTO(out, rc = -EOVERFLOW);
5456                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
5457                 *(--ptr) = '/';
5458
5459                 /* Store the parent fid for historic lookup */
5460                 if (++pli->pli_fidcount >= MAX_PATH_DEPTH)
5461                         GOTO(out, rc = -EOVERFLOW);
5462                 pli->pli_fids[pli->pli_fidcount] = *tmpfid;
5463         }
5464
5465 remote_out:
5466         ptr++; /* skip leading / */
5467         memmove(pli->pli_path, ptr, pli->pli_path + pli->pli_pathlen - ptr);
5468
5469         EXIT;
5470 out:
5471         return rc;
5472 }
5473
5474 /* Returns the full path to this fid, as of changelog record recno. */
5475 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
5476                     char *path, int pathlen, __u64 *recno, int *linkno,
5477                     struct lu_fid *fid)
5478 {
5479         struct mdt_device       *mdt = info->mti_mdt;
5480         struct path_lookup_info *pli;
5481         int                     tries = 3;
5482         int                     rc = -EAGAIN;
5483         ENTRY;
5484
5485         if (pathlen < 3)
5486                 RETURN(-EOVERFLOW);
5487
5488         if (lu_fid_eq(&mdt->mdt_md_root_fid, mdt_object_fid(obj))) {
5489                 path[0] = '\0';
5490                 RETURN(0);
5491         }
5492
5493         OBD_ALLOC_PTR(pli);
5494         if (pli == NULL)
5495                 RETURN(-ENOMEM);
5496
5497         pli->pli_mdt_obj = obj;
5498         pli->pli_recno = *recno;
5499         pli->pli_path = path;
5500         pli->pli_pathlen = pathlen;
5501         pli->pli_linkno = *linkno;
5502
5503         /* Retry multiple times in case file is being moved */
5504         while (tries-- && rc == -EAGAIN)
5505                 rc = mdt_path_current(info, pli);
5506
5507         /* return the last resolved fids to the client, so the client will
5508          * build the left path on another MDT for remote object */
5509         *fid = pli->pli_fids[pli->pli_fidcount];
5510
5511         *recno = pli->pli_currec;
5512         /* Return next link index to caller */
5513         *linkno = pli->pli_linkno;
5514
5515         OBD_FREE_PTR(pli);
5516
5517         RETURN(rc);
5518 }
5519
5520 static int mdt_fid2path(struct mdt_thread_info *info,
5521                         struct getinfo_fid2path *fp)
5522 {
5523         struct mdt_device *mdt = info->mti_mdt;
5524         struct mdt_object *obj;
5525         int    rc;
5526         ENTRY;
5527
5528         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
5529                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
5530
5531         if (!fid_is_sane(&fp->gf_fid))
5532                 RETURN(-EINVAL);
5533
5534         if (!fid_is_namespace_visible(&fp->gf_fid)) {
5535                 CWARN("%s: "DFID" is invalid, sequence should be "
5536                       ">= "LPX64"\n", mdt_obd_name(mdt),
5537                       PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
5538                 RETURN(-EINVAL);
5539         }
5540
5541         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
5542         if (obj == NULL || IS_ERR(obj)) {
5543                 CDEBUG(D_IOCTL, "no object "DFID": %ld\n", PFID(&fp->gf_fid),
5544                        PTR_ERR(obj));
5545                 RETURN(-EINVAL);
5546         }
5547
5548         if (mdt_object_remote(obj))
5549                 rc = -EREMOTE;
5550         else if (!mdt_object_exists(obj))
5551                 rc = -ENOENT;
5552         else
5553                 rc = 0;
5554
5555         if (rc < 0) {
5556                 mdt_object_put(info->mti_env, obj);
5557                 CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n",
5558                        PFID(&fp->gf_fid), rc);
5559                 RETURN(rc);
5560         }
5561
5562         rc = mdt_path(info, obj, fp->gf_path, fp->gf_pathlen, &fp->gf_recno,
5563                       &fp->gf_linkno, &fp->gf_fid);
5564
5565         CDEBUG(D_INFO, "fid "DFID", path %s recno "LPX64" linkno %u\n",
5566                PFID(&fp->gf_fid), fp->gf_path, fp->gf_recno, fp->gf_linkno);
5567
5568         mdt_object_put(info->mti_env, obj);
5569
5570         RETURN(rc);
5571 }
5572
5573 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key,
5574                             void *val, int vallen)
5575 {
5576         struct getinfo_fid2path *fpout, *fpin;
5577         int rc = 0;
5578
5579         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
5580         fpout = val;
5581
5582         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5583                 lustre_swab_fid2path(fpin);
5584
5585         memcpy(fpout, fpin, sizeof(*fpin));
5586         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
5587                 RETURN(-EINVAL);
5588
5589         rc = mdt_fid2path(info, fpout);
5590         RETURN(rc);
5591 }
5592
5593 int mdt_get_info(struct tgt_session_info *tsi)
5594 {
5595         char    *key;
5596         int      keylen;
5597         __u32   *vallen;
5598         void    *valout;
5599         int      rc;
5600
5601         ENTRY;
5602
5603         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
5604         if (key == NULL) {
5605                 CDEBUG(D_IOCTL, "No GETINFO key\n");
5606                 RETURN(err_serious(-EFAULT));
5607         }
5608         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
5609                                       RCL_CLIENT);
5610
5611         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
5612         if (vallen == NULL) {
5613                 CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n",
5614                                 tgt_name(tsi->tsi_tgt));
5615                 RETURN(err_serious(-EFAULT));
5616         }
5617
5618         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
5619                              *vallen);
5620         rc = req_capsule_server_pack(tsi->tsi_pill);
5621         if (rc)
5622                 RETURN(err_serious(rc));
5623
5624         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
5625         if (valout == NULL) {
5626                 CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n",
5627                                 tgt_name(tsi->tsi_tgt));
5628                 RETURN(err_serious(-EFAULT));
5629         }
5630
5631         if (KEY_IS(KEY_FID2PATH)) {
5632                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
5633
5634                 rc = mdt_rpc_fid2path(info, key, valout, *vallen);
5635                 mdt_thread_info_fini(info);
5636         } else {
5637                 rc = -EINVAL;
5638         }
5639         RETURN(rc);
5640 }
5641
5642 /* Pass the ioc down */
5643 static int mdt_ioc_child(struct lu_env *env, struct mdt_device *mdt,
5644                          unsigned int cmd, int len, void *data)
5645 {
5646         struct lu_context ioctl_session;
5647         struct md_device *next = mdt->mdt_child;
5648         int rc;
5649         ENTRY;
5650
5651         rc = lu_context_init(&ioctl_session, LCT_SERVER_SESSION);
5652         if (rc)
5653                 RETURN(rc);
5654         ioctl_session.lc_thread = (struct ptlrpc_thread *)current;
5655         lu_context_enter(&ioctl_session);
5656         env->le_ses = &ioctl_session;
5657
5658         LASSERT(next->md_ops->mdo_iocontrol);
5659         rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data);
5660
5661         lu_context_exit(&ioctl_session);
5662         lu_context_fini(&ioctl_session);
5663         RETURN(rc);
5664 }
5665
5666 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
5667 {
5668         struct obd_ioctl_data *data = karg;
5669         struct lu_fid *fid;
5670         __u64 version;
5671         struct mdt_object *obj;
5672         struct mdt_lock_handle  *lh;
5673         int rc;
5674         ENTRY;
5675
5676         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
5677             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
5678                 RETURN(-EINVAL);
5679
5680         fid = (struct lu_fid *)data->ioc_inlbuf1;
5681
5682         if (!fid_is_sane(fid))
5683                 RETURN(-EINVAL);
5684
5685         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
5686
5687         lh = &mti->mti_lh[MDT_LH_PARENT];
5688         mdt_lock_reg_init(lh, LCK_CR);
5689
5690         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
5691         if (IS_ERR(obj))
5692                 RETURN(PTR_ERR(obj));
5693
5694         if (mdt_object_remote(obj)) {
5695                 rc = -EREMOTE;
5696                 /**
5697                  * before calling version get the correct MDS should be
5698                  * fid, this is error to find remote object here
5699                  */
5700                 CERROR("nonlocal object "DFID"\n", PFID(fid));
5701         } else if (!mdt_object_exists(obj)) {
5702                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
5703                 rc = -ENOENT;
5704         } else {
5705                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
5706                *(__u64 *)data->ioc_inlbuf2 = version;
5707                 rc = 0;
5708         }
5709         mdt_object_unlock_put(mti, obj, lh, 1);
5710         RETURN(rc);
5711 }
5712
5713 /* ioctls on obd dev */
5714 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
5715                          void *karg, void *uarg)
5716 {
5717         struct lu_env      env;
5718         struct obd_device *obd = exp->exp_obd;
5719         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
5720         struct dt_device  *dt = mdt->mdt_bottom;
5721         int rc;
5722
5723         ENTRY;
5724         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
5725         rc = lu_env_init(&env, LCT_MD_THREAD);
5726         if (rc)
5727                 RETURN(rc);
5728
5729         switch (cmd) {
5730         case OBD_IOC_SYNC:
5731                 rc = mdt_device_sync(&env, mdt);
5732                 break;
5733         case OBD_IOC_SET_READONLY:
5734                 rc = dt->dd_ops->dt_ro(&env, dt);
5735                 break;
5736         case OBD_IOC_ABORT_RECOVERY:
5737                 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
5738                 target_stop_recovery_thread(obd);
5739                 rc = 0;
5740                 break;
5741         case OBD_IOC_CHANGELOG_REG:
5742         case OBD_IOC_CHANGELOG_DEREG:
5743         case OBD_IOC_CHANGELOG_CLEAR:
5744                 rc = mdt_ioc_child(&env, mdt, cmd, len, karg);
5745                 break;
5746         case OBD_IOC_START_LFSCK: {
5747                 struct md_device *next = mdt->mdt_child;
5748                 struct obd_ioctl_data *data = karg;
5749                 struct lfsck_start_param lsp;
5750
5751                 if (unlikely(data == NULL)) {
5752                         rc = -EINVAL;
5753                         break;
5754                 }
5755
5756                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
5757                 lsp.lsp_index_valid = 0;
5758                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
5759                 break;
5760         }
5761         case OBD_IOC_STOP_LFSCK: {
5762                 struct md_device        *next = mdt->mdt_child;
5763                 struct obd_ioctl_data   *data = karg;
5764                 struct lfsck_stop        stop;
5765
5766                 stop.ls_status = LS_STOPPED;
5767                 /* Old lfsck utils may pass NULL @stop. */
5768                 if (data->ioc_inlbuf1 == NULL)
5769                         stop.ls_flags = 0;
5770                 else
5771                         stop.ls_flags =
5772                         ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
5773
5774                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
5775                 break;
5776         }
5777         case OBD_IOC_GET_OBJ_VERSION: {
5778                 struct mdt_thread_info *mti;
5779                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5780                 memset(mti, 0, sizeof *mti);
5781                 mti->mti_env = &env;
5782                 mti->mti_mdt = mdt;
5783                 mti->mti_exp = exp;
5784
5785                 rc = mdt_ioc_version_get(mti, karg);
5786                 break;
5787         }
5788         case OBD_IOC_CATLOGLIST: {
5789                 struct mdt_thread_info *mti;
5790
5791                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5792                 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
5793                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
5794                                        &mti->mti_tmp_fid1);
5795                 break;
5796          }
5797         default:
5798                 rc = -EOPNOTSUPP;
5799                 CERROR("%s: Not supported cmd = %d, rc = %d\n",
5800                         mdt_obd_name(mdt), cmd, rc);
5801         }
5802
5803         lu_env_fini(&env);
5804         RETURN(rc);
5805 }
5806
5807 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
5808 {
5809         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
5810         struct lfsck_start_param lsp;
5811         int rc;
5812         ENTRY;
5813
5814         lsp.lsp_start = NULL;
5815         lsp.lsp_index_valid = 0;
5816         rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
5817                                                    OBD_IOC_START_LFSCK,
5818                                                    0, &lsp);
5819         if (rc != 0 && rc != -EALREADY)
5820                 CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
5821                       mdt_obd_name(mdt), rc);
5822
5823         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
5824         RETURN(rc);
5825 }
5826
5827 static int mdt_obd_postrecov(struct obd_device *obd)
5828 {
5829         struct lu_env env;
5830         int rc;
5831
5832         rc = lu_env_init(&env, LCT_MD_THREAD);
5833         if (rc)
5834                 RETURN(rc);
5835         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
5836         lu_env_fini(&env);
5837         return rc;
5838 }
5839
5840 static struct obd_ops mdt_obd_device_ops = {
5841         .o_owner          = THIS_MODULE,
5842         .o_set_info_async = mdt_obd_set_info_async,
5843         .o_connect        = mdt_obd_connect,
5844         .o_reconnect      = mdt_obd_reconnect,
5845         .o_disconnect     = mdt_obd_disconnect,
5846         .o_init_export    = mdt_init_export,
5847         .o_destroy_export = mdt_destroy_export,
5848         .o_iocontrol      = mdt_iocontrol,
5849         .o_postrecov      = mdt_obd_postrecov,
5850 };
5851
5852 static struct lu_device* mdt_device_fini(const struct lu_env *env,
5853                                          struct lu_device *d)
5854 {
5855         struct mdt_device *m = mdt_dev(d);
5856         ENTRY;
5857
5858         mdt_fini(env, m);
5859         RETURN(NULL);
5860 }
5861
5862 static struct lu_device *mdt_device_free(const struct lu_env *env,
5863                                          struct lu_device *d)
5864 {
5865         struct mdt_device *m = mdt_dev(d);
5866         ENTRY;
5867
5868         lu_device_fini(&m->mdt_lu_dev);
5869         OBD_FREE_PTR(m);
5870
5871         RETURN(NULL);
5872 }
5873
5874 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
5875                                           struct lu_device_type *t,
5876                                           struct lustre_cfg *cfg)
5877 {
5878         struct lu_device  *l;
5879         struct mdt_device *m;
5880
5881         OBD_ALLOC_PTR(m);
5882         if (m != NULL) {
5883                 int rc;
5884
5885                 l = &m->mdt_lu_dev;
5886                 rc = mdt_init0(env, m, t, cfg);
5887                 if (rc != 0) {
5888                         mdt_device_free(env, l);
5889                         l = ERR_PTR(rc);
5890                         return l;
5891                 }
5892         } else
5893                 l = ERR_PTR(-ENOMEM);
5894         return l;
5895 }
5896
5897 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
5898 LU_KEY_INIT(mdt, struct mdt_thread_info);
5899
5900 static void mdt_key_fini(const struct lu_context *ctx,
5901                          struct lu_context_key *key, void* data)
5902 {
5903         struct mdt_thread_info *info = data;
5904
5905         if (info->mti_big_lmm) {
5906                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
5907                 info->mti_big_lmm = NULL;
5908                 info->mti_big_lmmsize = 0;
5909         }
5910         OBD_FREE_PTR(info);
5911 }
5912
5913 /* context key: mdt_thread_key */
5914 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
5915
5916 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
5917 {
5918         return lu_ucred(info->mti_env);
5919 }
5920
5921 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
5922 {
5923         return lu_ucred_check(info->mti_env);
5924 }
5925
5926 /**
5927  * Enable/disable COS (Commit On Sharing).
5928  *
5929  * Set/Clear the COS flag in mdt options.
5930  *
5931  * \param mdt mdt device
5932  * \param val 0 disables COS, other values enable COS
5933  */
5934 void mdt_enable_cos(struct mdt_device *mdt, int val)
5935 {
5936         struct lu_env env;
5937         int rc;
5938
5939         mdt->mdt_opts.mo_cos = !!val;
5940         rc = lu_env_init(&env, LCT_LOCAL);
5941         if (unlikely(rc != 0)) {
5942                 CWARN("%s: lu_env initialization failed, cannot "
5943                       "sync: rc = %d\n", mdt_obd_name(mdt), rc);
5944                 return;
5945         }
5946         mdt_device_sync(&env, mdt);
5947         lu_env_fini(&env);
5948 }
5949
5950 /**
5951  * Check COS (Commit On Sharing) status.
5952  *
5953  * Return COS flag status.
5954  *
5955  * \param mdt mdt device
5956  */
5957 int mdt_cos_is_enabled(struct mdt_device *mdt)
5958 {
5959         return mdt->mdt_opts.mo_cos != 0;
5960 }
5961
5962 static struct lu_device_type_operations mdt_device_type_ops = {
5963         .ldto_device_alloc = mdt_device_alloc,
5964         .ldto_device_free  = mdt_device_free,
5965         .ldto_device_fini  = mdt_device_fini
5966 };
5967
5968 static struct lu_device_type mdt_device_type = {
5969         .ldt_tags     = LU_DEVICE_MD,
5970         .ldt_name     = LUSTRE_MDT_NAME,
5971         .ldt_ops      = &mdt_device_type_ops,
5972         .ldt_ctx_tags = LCT_MD_THREAD
5973 };
5974
5975 static int __init mdt_mod_init(void)
5976 {
5977         int rc;
5978
5979         CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") ==
5980                  FID_NOBRACE_LEN + 1);
5981         CLASSERT(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") ==
5982                  FID_LEN + 1);
5983         rc = lu_kmem_init(mdt_caches);
5984         if (rc)
5985                 return rc;
5986
5987         rc = mds_mod_init();
5988         if (rc)
5989                 GOTO(lu_fini, rc);
5990
5991         rc = class_register_type(&mdt_obd_device_ops, NULL, true, NULL,
5992                                  LUSTRE_MDT_NAME, &mdt_device_type);
5993         if (rc)
5994                 GOTO(mds_fini, rc);
5995 lu_fini:
5996         if (rc)
5997                 lu_kmem_fini(mdt_caches);
5998 mds_fini:
5999         if (rc)
6000                 mds_mod_exit();
6001         return rc;
6002 }
6003
6004 static void __exit mdt_mod_exit(void)
6005 {
6006         class_unregister_type(LUSTRE_MDT_NAME);
6007         mds_mod_exit();
6008         lu_kmem_fini(mdt_caches);
6009 }
6010
6011 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
6012 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
6013 MODULE_LICENSE("GPL");
6014
6015 cfs_module(mdt, LUSTRE_VERSION_STRING, mdt_mod_init, mdt_mod_exit);