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, 2015, 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 #include <dt_object.h>
  54 #include <lustre_acl.h>
  55 #include <lustre_export.h>
  56 #include <lustre_ioctl.h>
  57 #include <lustre_lfsck.h>
  58 #include <lustre_log.h>
  59 #include <lustre_net.h>
  60 #include <lustre_nodemap.h>
  61 #include <lustre_mds.h>
  62 #include <lustre_param.h>
  63 #include <lustre_quota.h>
  64 #include <lustre_swab.h>
  65 #include <obd.h>
  66 #include <obd_support.h>
  67
  68 #include "mdt_internal.h"
  69
  70
  71 static unsigned int max_mod_rpcs_per_client = 8;
  72 CFS_MODULE_PARM(max_mod_rpcs_per_client, "i", uint, 0644,
  73                 "maximum number of modify RPCs in flight allowed per client");
  74
  75 mdl_mode_t mdt_mdl_lock_modes[] = {
  76         [LCK_MINMODE] = MDL_MINMODE,
  77         [LCK_EX]      = MDL_EX,
  78         [LCK_PW]      = MDL_PW,
  79         [LCK_PR]      = MDL_PR,
  80         [LCK_CW]      = MDL_CW,
  81         [LCK_CR]      = MDL_CR,
  82         [LCK_NL]      = MDL_NL,
  83         [LCK_GROUP]   = MDL_GROUP
  84 };
  85
  86 enum ldlm_mode mdt_dlm_lock_modes[] = {
  87         [MDL_MINMODE]   = LCK_MINMODE,
  88         [MDL_EX]        = LCK_EX,
  89         [MDL_PW]        = LCK_PW,
  90         [MDL_PR]        = LCK_PR,
  91         [MDL_CW]        = LCK_CW,
  92         [MDL_CR]        = LCK_CR,
  93         [MDL_NL]        = LCK_NL,
  94         [MDL_GROUP]     = LCK_GROUP
  95 };
  96
  97 static struct mdt_device *mdt_dev(struct lu_device *d);
  98 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
  99
 100 static const struct lu_object_operations mdt_obj_ops;
 101
 102 /* Slab for MDT object allocation */
 103 static struct kmem_cache *mdt_object_kmem;
 104
 105 /* For HSM restore handles */
 106 struct kmem_cache *mdt_hsm_cdt_kmem;
 107
 108 /* For HSM request handles */
 109 struct kmem_cache *mdt_hsm_car_kmem;
 110
 111 static struct lu_kmem_descr mdt_caches[] = {
 112         {
 113                 .ckd_cache = &mdt_object_kmem,
 114                 .ckd_name  = "mdt_obj",
 115                 .ckd_size  = sizeof(struct mdt_object)
 116         },
 117         {
 118                 .ckd_cache      = &mdt_hsm_cdt_kmem,
 119                 .ckd_name       = "mdt_cdt_restore_handle",
 120                 .ckd_size       = sizeof(struct cdt_restore_handle)
 121         },
 122         {
 123                 .ckd_cache      = &mdt_hsm_car_kmem,
 124                 .ckd_name       = "mdt_cdt_agent_req",
 125                 .ckd_size       = sizeof(struct cdt_agent_req)
 126         },
 127         {
 128                 .ckd_cache = NULL
 129         }
 130 };
 131
 132 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag)
 133 {
 134         if (!rep)
 135                 return 0;
 136         return rep->lock_policy_res1 & op_flag;
 137 }
 138
 139 void mdt_clear_disposition(struct mdt_thread_info *info,
 140                            struct ldlm_reply *rep, __u64 op_flag)
 141 {
 142         if (info) {
 143                 info->mti_opdata &= ~op_flag;
 144                 tgt_opdata_clear(info->mti_env, op_flag);
 145         }
 146         if (rep)
 147                 rep->lock_policy_res1 &= ~op_flag;
 148 }
 149
 150 void mdt_set_disposition(struct mdt_thread_info *info,
 151                          struct ldlm_reply *rep, __u64 op_flag)
 152 {
 153         if (info) {
 154                 info->mti_opdata |= op_flag;
 155                 tgt_opdata_set(info->mti_env, op_flag);
 156         }
 157         if (rep)
 158                 rep->lock_policy_res1 |= op_flag;
 159 }
 160
 161 void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm)
 162 {
 163         lh->mlh_pdo_hash = 0;
 164         lh->mlh_reg_mode = lm;
 165         lh->mlh_rreg_mode = lm;
 166         lh->mlh_type = MDT_REG_LOCK;
 167 }
 168
 169 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode,
 170                        const struct lu_name *lname)
 171 {
 172         lh->mlh_reg_mode = lock_mode;
 173         lh->mlh_pdo_mode = LCK_MINMODE;
 174         lh->mlh_rreg_mode = lock_mode;
 175         lh->mlh_type = MDT_PDO_LOCK;
 176
 177         if (lu_name_is_valid(lname)) {
 178                 lh->mlh_pdo_hash = full_name_hash(lname->ln_name,
 179                                                   lname->ln_namelen);
 180                 /* XXX Workaround for LU-2856
 181                  *
 182                  * Zero is a valid return value of full_name_hash, but
 183                  * several users of mlh_pdo_hash assume a non-zero
 184                  * hash value. We therefore map zero onto an
 185                  * arbitrary, but consistent value (1) to avoid
 186                  * problems further down the road. */
 187                 if (unlikely(lh->mlh_pdo_hash == 0))
 188                         lh->mlh_pdo_hash = 1;
 189         } else {
 190                 lh->mlh_pdo_hash = 0;
 191         }
 192 }
 193
 194 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 195                               struct mdt_lock_handle *lh)
 196 {
 197         mdl_mode_t mode;
 198         ENTRY;
 199
 200         /*
 201          * Any dir access needs couple of locks:
 202          *
 203          * 1) on part of dir we gonna take lookup/modify;
 204          *
 205          * 2) on whole dir to protect it from concurrent splitting and/or to
 206          * flush client's cache for readdir().
 207          *
 208          * so, for a given mode and object this routine decides what lock mode
 209          * to use for lock #2:
 210          *
 211          * 1) if caller's gonna lookup in dir then we need to protect dir from
 212          * being splitted only - LCK_CR
 213          *
 214          * 2) if caller's gonna modify dir then we need to protect dir from
 215          * being splitted and to flush cache - LCK_CW
 216          *
 217          * 3) if caller's gonna modify dir and that dir seems ready for
 218          * splitting then we need to protect it from any type of access
 219          * (lookup/modify/split) - LCK_EX --bzzz
 220          */
 221
 222         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 223         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 224
 225         /*
 226          * Ask underlaying level its opinion about preferable PDO lock mode
 227          * having access type passed as regular lock mode:
 228          *
 229          * - MDL_MINMODE means that lower layer does not want to specify lock
 230          * mode;
 231          *
 232          * - MDL_NL means that no PDO lock should be taken. This is used in some
 233          * cases. Say, for non-splittable directories no need to use PDO locks
 234          * at all.
 235          */
 236         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 237                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 238
 239         if (mode != MDL_MINMODE) {
 240                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 241         } else {
 242                 /*
 243                  * Lower layer does not want to specify locking mode. We do it
 244                  * our selves. No special protection is needed, just flush
 245                  * client's cache on modification and allow concurrent
 246                  * mondification.
 247                  */
 248                 switch (lh->mlh_reg_mode) {
 249                 case LCK_EX:
 250                         lh->mlh_pdo_mode = LCK_EX;
 251                         break;
 252                 case LCK_PR:
 253                         lh->mlh_pdo_mode = LCK_CR;
 254                         break;
 255                 case LCK_PW:
 256                         lh->mlh_pdo_mode = LCK_CW;
 257                         break;
 258                 default:
 259                         CERROR("Not expected lock type (0x%x)\n",
 260                                (int)lh->mlh_reg_mode);
 261                         LBUG();
 262                 }
 263         }
 264
 265         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 266         EXIT;
 267 }
 268
 269 static int mdt_getstatus(struct tgt_session_info *tsi)
 270 {
 271         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
 272         struct mdt_device       *mdt = info->mti_mdt;
 273         struct mdt_body         *repbody;
 274         int                      rc;
 275
 276         ENTRY;
 277
 278         rc = mdt_check_ucred(info);
 279         if (rc)
 280                 GOTO(out, rc = err_serious(rc));
 281
 282         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK))
 283                 GOTO(out, rc = err_serious(-ENOMEM));
 284
 285         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 286         repbody->mbo_fid1 = mdt->mdt_md_root_fid;
 287         repbody->mbo_valid |= OBD_MD_FLID;
 288
 289         EXIT;
 290 out:
 291         mdt_thread_info_fini(info);
 292         return rc;
 293 }
 294
 295 static int mdt_statfs(struct tgt_session_info *tsi)
 296 {
 297         struct ptlrpc_request           *req = tgt_ses_req(tsi);
 298         struct mdt_thread_info          *info = tsi2mdt_info(tsi);
 299         struct md_device                *next = info->mti_mdt->mdt_child;
 300         struct ptlrpc_service_part      *svcpt;
 301         struct obd_statfs               *osfs;
 302         int                             rc;
 303
 304         ENTRY;
 305
 306         svcpt = req->rq_rqbd->rqbd_svcpt;
 307
 308         /* This will trigger a watchdog timeout */
 309         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 310                          (MDT_SERVICE_WATCHDOG_FACTOR *
 311                           at_get(&svcpt->scp_at_estimate)) + 1);
 312
 313         rc = mdt_check_ucred(info);
 314         if (rc)
 315                 GOTO(out, rc = err_serious(rc));
 316
 317         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
 318                 GOTO(out, rc = err_serious(-ENOMEM));
 319
 320         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 321         if (!osfs)
 322                 GOTO(out, rc = -EPROTO);
 323
 324         /** statfs information are cached in the mdt_device */
 325         if (cfs_time_before_64(info->mti_mdt->mdt_osfs_age,
 326                                cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS))) {
 327                 /** statfs data is too old, get up-to-date one */
 328                 rc = next->md_ops->mdo_statfs(info->mti_env, next, osfs);
 329                 if (rc)
 330                         GOTO(out, rc);
 331                 spin_lock(&info->mti_mdt->mdt_osfs_lock);
 332                 info->mti_mdt->mdt_osfs = *osfs;
 333                 info->mti_mdt->mdt_osfs_age = cfs_time_current_64();
 334                 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
 335         } else {
 336                 /** use cached statfs data */
 337                 spin_lock(&info->mti_mdt->mdt_osfs_lock);
 338                 *osfs = info->mti_mdt->mdt_osfs;
 339                 spin_unlock(&info->mti_mdt->mdt_osfs_lock);
 340         }
 341
 342         if (rc == 0)
 343                 mdt_counter_incr(req, LPROC_MDT_STATFS);
 344 out:
 345         mdt_thread_info_fini(info);
 346         RETURN(rc);
 347 }
 348
 349 #ifdef CONFIG_FS_POSIX_ACL
 350 /*
 351  * Pack ACL data into the reply. UIDs/GIDs are mapped and filtered by nodemap.
 352  *
 353  * \param       info    thread info object
 354  * \param       repbody reply to pack ACLs into
 355  * \param       o       mdt object of file to examine
 356  * \param       nodemap nodemap of client to reply to
 357  * \retval      0       success
 358  * \retval      -errno  error getting or parsing ACL from disk
 359  */
 360 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
 361                       struct mdt_object *o, struct lu_nodemap *nodemap)
 362 {
 363         const struct lu_env     *env = info->mti_env;
 364         struct md_object        *next = mdt_object_child(o);
 365         struct lu_buf           *buf = &info->mti_buf;
 366         int rc;
 367
 368         buf->lb_buf = req_capsule_server_get(info->mti_pill, &RMF_ACL);
 369         buf->lb_len = req_capsule_get_size(info->mti_pill, &RMF_ACL,
 370                                            RCL_SERVER);
 371         if (buf->lb_len == 0)
 372                 return 0;
 373
 374         rc = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_ACCESS);
 375         if (rc < 0) {
 376                 if (rc == -ENODATA) {
 377                         repbody->mbo_aclsize = 0;
 378                         repbody->mbo_valid |= OBD_MD_FLACL;
 379                         rc = 0;
 380                 } else if (rc == -EOPNOTSUPP) {
 381                         rc = 0;
 382                 } else {
 383                         CERROR("%s: unable to read "DFID" ACL: rc = %d\n",
 384                                mdt_obd_name(info->mti_mdt),
 385                                PFID(mdt_object_fid(o)), rc);
 386                 }
 387         } else {
 388                 rc = nodemap_map_acl(nodemap, buf->lb_buf,
 389                                      rc, NODEMAP_FS_TO_CLIENT);
 390                 /* if all ACLs mapped out, rc is still >= 0 */
 391                 if (rc < 0) {
 392                         CERROR("%s: nodemap_map_acl unable to parse "DFID
 393                                " ACL: rc = %d\n", mdt_obd_name(info->mti_mdt),
 394                                PFID(mdt_object_fid(o)), rc);
 395                 } else {
 396                         repbody->mbo_aclsize = rc;
 397                         repbody->mbo_valid |= OBD_MD_FLACL;
 398                         rc = 0;
 399                 }
 400         }
 401         return rc;
 402 }
 403 #endif
 404
 405 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 406                         const struct lu_attr *attr, const struct lu_fid *fid)
 407 {
 408         struct md_attr          *ma = &info->mti_attr;
 409         struct obd_export       *exp = info->mti_exp;
 410         struct lu_nodemap       *nodemap = exp->exp_target_data.ted_nodemap;
 411
 412         LASSERT(ma->ma_valid & MA_INODE);
 413
 414         if (attr->la_valid & LA_ATIME) {
 415                 b->mbo_atime = attr->la_atime;
 416                 b->mbo_valid |= OBD_MD_FLATIME;
 417         }
 418         if (attr->la_valid & LA_MTIME) {
 419                 b->mbo_mtime = attr->la_mtime;
 420                 b->mbo_valid |= OBD_MD_FLMTIME;
 421         }
 422         if (attr->la_valid & LA_CTIME) {
 423                 b->mbo_ctime = attr->la_ctime;
 424                 b->mbo_valid |= OBD_MD_FLCTIME;
 425         }
 426         if (attr->la_valid & LA_FLAGS) {
 427                 b->mbo_flags = attr->la_flags;
 428                 b->mbo_valid |= OBD_MD_FLFLAGS;
 429         }
 430         if (attr->la_valid & LA_NLINK) {
 431                 b->mbo_nlink = attr->la_nlink;
 432                 b->mbo_valid |= OBD_MD_FLNLINK;
 433         }
 434         if (attr->la_valid & LA_UID) {
 435                 b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 436                                             NODEMAP_FS_TO_CLIENT,
 437                                             attr->la_uid);
 438                 b->mbo_valid |= OBD_MD_FLUID;
 439         }
 440         if (attr->la_valid & LA_GID) {
 441                 b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 442                                             NODEMAP_FS_TO_CLIENT,
 443                                             attr->la_gid);
 444                 b->mbo_valid |= OBD_MD_FLGID;
 445         }
 446         b->mbo_mode = attr->la_mode;
 447         if (attr->la_valid & LA_MODE)
 448                 b->mbo_valid |= OBD_MD_FLMODE;
 449         if (attr->la_valid & LA_TYPE)
 450                 b->mbo_valid |= OBD_MD_FLTYPE;
 451
 452         if (fid != NULL) {
 453                 b->mbo_fid1 = *fid;
 454                 b->mbo_valid |= OBD_MD_FLID;
 455                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid="LPX64"\n",
 456                        PFID(fid), b->mbo_nlink, b->mbo_mode, b->mbo_valid);
 457         }
 458
 459         if (info != NULL)
 460                 mdt_body_reverse_idmap(info, b);
 461
 462         if (!(attr->la_valid & LA_TYPE))
 463                 return;
 464
 465         b->mbo_rdev   = attr->la_rdev;
 466         b->mbo_size   = attr->la_size;
 467         b->mbo_blocks = attr->la_blocks;
 468
 469         if (!S_ISREG(attr->la_mode)) {
 470                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 471         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
 472                 /* means no objects are allocated on osts. */
 473                 LASSERT(!(ma->ma_valid & MA_LOV));
 474                 /* just ignore blocks occupied by extend attributes on MDS */
 475                 b->mbo_blocks = 0;
 476                 /* if no object is allocated on osts, the size on mds is valid.
 477                  * b=22272 */
 478                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 479         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL &&
 480                    ma->ma_lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) {
 481                 /* A released file stores its size on MDS. */
 482                 /* But return 1 block for released file, unless tools like tar
 483                  * will consider it fully sparse. (LU-3864)
 484                  */
 485                 if (unlikely(b->mbo_size == 0))
 486                         b->mbo_blocks = 0;
 487                 else
 488                         b->mbo_blocks = 1;
 489                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 490         }
 491
 492         if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE))
 493                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 494                        PFID(fid), (unsigned long long)b->mbo_size);
 495 }
 496
 497 static inline int mdt_body_has_lov(const struct lu_attr *la,
 498                                    const struct mdt_body *body)
 499 {
 500         return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
 501                (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
 502 }
 503
 504 void mdt_client_compatibility(struct mdt_thread_info *info)
 505 {
 506         struct mdt_body       *body;
 507         struct ptlrpc_request *req = mdt_info_req(info);
 508         struct obd_export     *exp = req->rq_export;
 509         struct md_attr        *ma = &info->mti_attr;
 510         struct lu_attr        *la = &ma->ma_attr;
 511         ENTRY;
 512
 513         if (exp_connect_layout(exp))
 514                 /* the client can deal with 16-bit lmm_stripe_count */
 515                 RETURN_EXIT;
 516
 517         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 518
 519         if (!mdt_body_has_lov(la, body))
 520                 RETURN_EXIT;
 521
 522         /* now we have a reply with a lov for a client not compatible with the
 523          * layout lock so we have to clean the layout generation number */
 524         if (S_ISREG(la->la_mode))
 525                 ma->ma_lmm->lmm_layout_gen = 0;
 526         EXIT;
 527 }
 528
 529 static int mdt_attr_get_eabuf_size(struct mdt_thread_info *info,
 530                                    struct mdt_object *o)
 531 {
 532         const struct lu_env *env = info->mti_env;
 533         int rc, rc2;
 534
 535         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 536                           XATTR_NAME_LOV);
 537
 538         if (rc == -ENODATA)
 539                 rc = 0;
 540
 541         if (rc < 0)
 542                 goto out;
 543
 544         /* Is it a directory? Let's check for the LMV as well */
 545         if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
 546                 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 547                                    XATTR_NAME_LMV);
 548
 549                 if (rc2 == -ENODATA)
 550                         rc2 = mo_xattr_get(env, mdt_object_child(o),
 551                                            &LU_BUF_NULL,
 552                                            XATTR_NAME_DEFAULT_LMV);
 553
 554                 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
 555                         rc = rc2;
 556         }
 557
 558 out:
 559         return rc;
 560 }
 561
 562 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
 563                       const char *name)
 564 {
 565         const struct lu_env *env = info->mti_env;
 566         int rc;
 567         ENTRY;
 568
 569         LASSERT(info->mti_big_lmm_used == 0);
 570         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
 571         if (rc < 0)
 572                 RETURN(rc);
 573
 574         /* big_lmm may need to be grown */
 575         if (info->mti_big_lmmsize < rc) {
 576                 int size = size_roundup_power2(rc);
 577
 578                 if (info->mti_big_lmmsize > 0) {
 579                         /* free old buffer */
 580                         LASSERT(info->mti_big_lmm);
 581                         OBD_FREE_LARGE(info->mti_big_lmm,
 582                                        info->mti_big_lmmsize);
 583                         info->mti_big_lmm = NULL;
 584                         info->mti_big_lmmsize = 0;
 585                 }
 586
 587                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
 588                 if (info->mti_big_lmm == NULL)
 589                         RETURN(-ENOMEM);
 590                 info->mti_big_lmmsize = size;
 591         }
 592         LASSERT(info->mti_big_lmmsize >= rc);
 593
 594         info->mti_buf.lb_buf = info->mti_big_lmm;
 595         info->mti_buf.lb_len = info->mti_big_lmmsize;
 596         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
 597
 598         RETURN(rc);
 599 }
 600
 601 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
 602                    struct md_attr *ma, const char *name)
 603 {
 604         struct md_object *next = mdt_object_child(o);
 605         struct lu_buf    *buf = &info->mti_buf;
 606         int rc;
 607
 608         if (strcmp(name, XATTR_NAME_LOV) == 0) {
 609                 buf->lb_buf = ma->ma_lmm;
 610                 buf->lb_len = ma->ma_lmm_size;
 611                 LASSERT(!(ma->ma_valid & MA_LOV));
 612         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 613                 buf->lb_buf = ma->ma_lmv;
 614                 buf->lb_len = ma->ma_lmv_size;
 615                 LASSERT(!(ma->ma_valid & MA_LMV));
 616         } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 617                 buf->lb_buf = ma->ma_lmv;
 618                 buf->lb_len = ma->ma_lmv_size;
 619                 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
 620         } else {
 621                 return -EINVAL;
 622         }
 623
 624         rc = mo_xattr_get(info->mti_env, next, buf, name);
 625         if (rc > 0) {
 626
 627 got:
 628                 if (strcmp(name, XATTR_NAME_LOV) == 0) {
 629                         if (info->mti_big_lmm_used)
 630                                 ma->ma_lmm = info->mti_big_lmm;
 631
 632                         /* NOT return LOV EA with hole to old client. */
 633                         if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
 634                                      LOV_PATTERN_F_HOLE) &&
 635                             !(exp_connect_flags(info->mti_exp) &
 636                               OBD_CONNECT_LFSCK)) {
 637                                 return -EIO;
 638                         } else {
 639                                 ma->ma_lmm_size = rc;
 640                                 ma->ma_valid |= MA_LOV;
 641                         }
 642                 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 643                         if (info->mti_big_lmm_used)
 644                                 ma->ma_lmv = info->mti_big_lmm;
 645
 646                         ma->ma_lmv_size = rc;
 647                         ma->ma_valid |= MA_LMV;
 648                 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 649                         ma->ma_lmv_size = rc;
 650                         ma->ma_valid |= MA_LMV_DEF;
 651                 }
 652
 653                 /* Update mdt_max_mdsize so all clients will be aware that */
 654                 if (info->mti_mdt->mdt_max_mdsize < rc)
 655                         info->mti_mdt->mdt_max_mdsize = rc;
 656
 657                 rc = 0;
 658         } else if (rc == -ENODATA) {
 659                 /* no LOV EA */
 660                 rc = 0;
 661         } else if (rc == -ERANGE) {
 662                 /* Default LMV has fixed size, so it must be able to fit
 663                  * in the original buffer */
 664                 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
 665                         return rc;
 666                 rc = mdt_big_xattr_get(info, o, name);
 667                 if (rc > 0) {
 668                         info->mti_big_lmm_used = 1;
 669                         goto got;
 670                 }
 671         }
 672
 673         return rc;
 674 }
 675
 676 static int mdt_attr_get_pfid(struct mdt_thread_info *info,
 677                              struct mdt_object *o, struct lu_fid *pfid)
 678 {
 679         struct lu_buf           *buf = &info->mti_buf;
 680         struct link_ea_header   *leh;
 681         struct link_ea_entry    *lee;
 682         int                      rc;
 683         ENTRY;
 684
 685         buf->lb_buf = info->mti_big_lmm;
 686         buf->lb_len = info->mti_big_lmmsize;
 687         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
 688                           buf, XATTR_NAME_LINK);
 689         /* ignore errors, MA_PFID won't be set and it is
 690          * up to the caller to treat this as an error */
 691         if (rc == -ERANGE || buf->lb_len == 0) {
 692                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
 693                 buf->lb_buf = info->mti_big_lmm;
 694                 buf->lb_len = info->mti_big_lmmsize;
 695         }
 696
 697         if (rc < 0)
 698                 RETURN(rc);
 699         if (rc < sizeof(*leh)) {
 700                 CERROR("short LinkEA on "DFID": rc = %d\n",
 701                        PFID(mdt_object_fid(o)), rc);
 702                 RETURN(-ENODATA);
 703         }
 704
 705         leh = (struct link_ea_header *) buf->lb_buf;
 706         lee = (struct link_ea_entry *)(leh + 1);
 707         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
 708                 leh->leh_magic = LINK_EA_MAGIC;
 709                 leh->leh_reccount = __swab32(leh->leh_reccount);
 710                 leh->leh_len = __swab64(leh->leh_len);
 711         }
 712         if (leh->leh_magic != LINK_EA_MAGIC)
 713                 RETURN(-EINVAL);
 714         if (leh->leh_reccount == 0)
 715                 RETURN(-ENODATA);
 716
 717         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
 718         fid_be_to_cpu(pfid, pfid);
 719
 720         RETURN(0);
 721 }
 722
 723 int mdt_attr_get_complex(struct mdt_thread_info *info,
 724                          struct mdt_object *o, struct md_attr *ma)
 725 {
 726         const struct lu_env *env = info->mti_env;
 727         struct md_object    *next = mdt_object_child(o);
 728         struct lu_buf       *buf = &info->mti_buf;
 729         int                  need = ma->ma_need;
 730         int                  rc = 0, rc2;
 731         u32                  mode;
 732         ENTRY;
 733
 734         ma->ma_valid = 0;
 735
 736         if (mdt_object_exists(o) == 0)
 737                 GOTO(out, rc = -ENOENT);
 738         mode = lu_object_attr(&next->mo_lu);
 739
 740         if (need & MA_INODE) {
 741                 ma->ma_need = MA_INODE;
 742                 rc = mo_attr_get(env, next, ma);
 743                 if (rc)
 744                         GOTO(out, rc);
 745                 ma->ma_valid |= MA_INODE;
 746         }
 747
 748         if (need & MA_PFID) {
 749                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
 750                 if (rc == 0)
 751                         ma->ma_valid |= MA_PFID;
 752                 /* ignore this error, parent fid is not mandatory */
 753                 rc = 0;
 754         }
 755
 756         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
 757                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
 758                 if (rc)
 759                         GOTO(out, rc);
 760         }
 761
 762         if (need & MA_LMV && S_ISDIR(mode)) {
 763                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
 764                 if (rc != 0)
 765                         GOTO(out, rc);
 766         }
 767
 768         if (need & MA_LMV_DEF && S_ISDIR(mode)) {
 769                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
 770                 if (rc != 0)
 771                         GOTO(out, rc);
 772         }
 773
 774         if (need & MA_HSM && S_ISREG(mode)) {
 775                 buf->lb_buf = info->mti_xattr_buf;
 776                 buf->lb_len = sizeof(info->mti_xattr_buf);
 777                 CLASSERT(sizeof(struct hsm_attrs) <=
 778                          sizeof(info->mti_xattr_buf));
 779                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
 780                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
 781                 if (rc2 == 0)
 782                         ma->ma_valid |= MA_HSM;
 783                 else if (rc2 < 0 && rc2 != -ENODATA)
 784                         GOTO(out, rc = rc2);
 785         }
 786
 787 #ifdef CONFIG_FS_POSIX_ACL
 788         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
 789                 buf->lb_buf = ma->ma_acl;
 790                 buf->lb_len = ma->ma_acl_size;
 791                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
 792                 if (rc2 > 0) {
 793                         ma->ma_acl_size = rc2;
 794                         ma->ma_valid |= MA_ACL_DEF;
 795                 } else if (rc2 == -ENODATA) {
 796                         /* no ACLs */
 797                         ma->ma_acl_size = 0;
 798                 } else
 799                         GOTO(out, rc = rc2);
 800         }
 801 #endif
 802 out:
 803         ma->ma_need = need;
 804         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64" ma_lmm=%p\n",
 805                rc, ma->ma_valid, ma->ma_lmm);
 806         RETURN(rc);
 807 }
 808
 809 static int mdt_getattr_internal(struct mdt_thread_info *info,
 810                                 struct mdt_object *o, int ma_need)
 811 {
 812         struct md_object        *next = mdt_object_child(o);
 813         const struct mdt_body   *reqbody = info->mti_body;
 814         struct ptlrpc_request   *req = mdt_info_req(info);
 815         struct md_attr          *ma = &info->mti_attr;
 816         struct lu_attr          *la = &ma->ma_attr;
 817         struct req_capsule      *pill = info->mti_pill;
 818         const struct lu_env     *env = info->mti_env;
 819         struct mdt_body         *repbody;
 820         struct lu_buf           *buffer = &info->mti_buf;
 821         struct obd_export       *exp = info->mti_exp;
 822         struct lu_nodemap       *nodemap = exp->exp_target_data.ted_nodemap;
 823         int                      rc;
 824         int                      is_root;
 825         ENTRY;
 826
 827         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
 828                 RETURN(err_serious(-ENOMEM));
 829
 830         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 831
 832         ma->ma_valid = 0;
 833
 834         if (mdt_object_remote(o)) {
 835                 /* This object is located on remote node.*/
 836                 /* Return -ENOTSUPP for old client */
 837                 if (!mdt_is_dne_client(req->rq_export))
 838                         GOTO(out, rc = -ENOTSUPP);
 839
 840                 repbody->mbo_fid1 = *mdt_object_fid(o);
 841                 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
 842                 GOTO(out, rc = 0);
 843         }
 844
 845         if (reqbody->mbo_eadatasize > 0) {
 846                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
 847                 if (buffer->lb_buf == NULL)
 848                         GOTO(out, rc = -EPROTO);
 849                 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
 850                                                       RCL_SERVER);
 851         } else {
 852                 buffer->lb_buf = NULL;
 853                 buffer->lb_len = 0;
 854                 ma_need &= ~(MA_LOV | MA_LMV);
 855                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
 856                        mdt_obd_name(info->mti_mdt),
 857                        req->rq_export->exp_client_uuid.uuid);
 858         }
 859
 860         /* If it is dir object and client require MEA, then we got MEA */
 861         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 862             (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
 863                 /* Assumption: MDT_MD size is enough for lmv size. */
 864                 ma->ma_lmv = buffer->lb_buf;
 865                 ma->ma_lmv_size = buffer->lb_len;
 866                 ma->ma_need = MA_INODE;
 867                 if (ma->ma_lmv_size > 0) {
 868                         if (reqbody->mbo_valid & OBD_MD_MEA)
 869                                 ma->ma_need |= MA_LMV;
 870                         else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA)
 871                                 ma->ma_need |= MA_LMV_DEF;
 872                 }
 873         } else {
 874                 ma->ma_lmm = buffer->lb_buf;
 875                 ma->ma_lmm_size = buffer->lb_len;
 876                 ma->ma_need = MA_INODE | MA_HSM;
 877                 if (ma->ma_lmm_size > 0)
 878                         ma->ma_need |= MA_LOV;
 879         }
 880
 881         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
 882             reqbody->mbo_valid & OBD_MD_FLDIREA  &&
 883             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
 884                 /* get default stripe info for this dir. */
 885                 ma->ma_need |= MA_LOV_DEF;
 886         }
 887         ma->ma_need |= ma_need;
 888
 889         rc = mdt_attr_get_complex(info, o, ma);
 890         if (unlikely(rc)) {
 891                 CERROR("%s: getattr error for "DFID": rc = %d\n",
 892                        mdt_obd_name(info->mti_mdt),
 893                        PFID(mdt_object_fid(o)), rc);
 894                 RETURN(rc);
 895         }
 896
 897         /* if file is released, check if a restore is running */
 898         if (ma->ma_valid & MA_HSM) {
 899                 repbody->mbo_valid |= OBD_MD_TSTATE;
 900                 if ((ma->ma_hsm.mh_flags & HS_RELEASED) &&
 901                     mdt_hsm_restore_is_running(info, mdt_object_fid(o)))
 902                         repbody->mbo_t_state = MS_RESTORE;
 903         }
 904
 905         is_root = lu_fid_eq(mdt_object_fid(o), &info->mti_mdt->mdt_md_root_fid);
 906
 907         /* the Lustre protocol supposes to return default striping
 908          * on the user-visible root if explicitly requested */
 909         if ((ma->ma_valid & MA_LOV) == 0 && S_ISDIR(la->la_mode) &&
 910             (ma->ma_need & MA_LOV_DEF && is_root) && ma->ma_need & MA_LOV) {
 911                 struct lu_fid      rootfid;
 912                 struct mdt_object *root;
 913                 struct mdt_device *mdt = info->mti_mdt;
 914
 915                 rc = dt_root_get(env, mdt->mdt_bottom, &rootfid);
 916                 if (rc)
 917                         RETURN(rc);
 918                 root = mdt_object_find(env, mdt, &rootfid);
 919                 if (IS_ERR(root))
 920                         RETURN(PTR_ERR(root));
 921                 rc = mdt_stripe_get(info, root, ma, XATTR_NAME_LOV);
 922                 mdt_object_put(info->mti_env, root);
 923                 if (unlikely(rc)) {
 924                         CERROR("%s: getattr error for "DFID": rc = %d\n",
 925                                mdt_obd_name(info->mti_mdt),
 926                                PFID(mdt_object_fid(o)), rc);
 927                         RETURN(rc);
 928                 }
 929         }
 930
 931         if (likely(ma->ma_valid & MA_INODE))
 932                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
 933         else
 934                 RETURN(-EFAULT);
 935
 936         if (mdt_body_has_lov(la, reqbody)) {
 937                 if (ma->ma_valid & MA_LOV) {
 938                         LASSERT(ma->ma_lmm_size);
 939                         repbody->mbo_eadatasize = ma->ma_lmm_size;
 940                         if (S_ISDIR(la->la_mode))
 941                                 repbody->mbo_valid |= OBD_MD_FLDIREA;
 942                         else
 943                                 repbody->mbo_valid |= OBD_MD_FLEASIZE;
 944                         mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
 945                 }
 946                 if (ma->ma_valid & MA_LMV) {
 947                         /* Return -ENOTSUPP for old client */
 948                         if (!mdt_is_striped_client(req->rq_export))
 949                                 RETURN(-ENOTSUPP);
 950
 951                         LASSERT(S_ISDIR(la->la_mode));
 952                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
 953                         repbody->mbo_eadatasize = ma->ma_lmv_size;
 954                         repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
 955                 }
 956                 if (ma->ma_valid & MA_LMV_DEF) {
 957                         /* Return -ENOTSUPP for old client */
 958                         if (!mdt_is_striped_client(req->rq_export))
 959                                 RETURN(-ENOTSUPP);
 960                         LASSERT(S_ISDIR(la->la_mode));
 961                         repbody->mbo_eadatasize = ma->ma_lmv_size;
 962                         repbody->mbo_valid |= (OBD_MD_FLDIREA |
 963                                                OBD_MD_DEFAULT_MEA);
 964                 }
 965         } else if (S_ISLNK(la->la_mode) &&
 966                    reqbody->mbo_valid & OBD_MD_LINKNAME) {
 967                 buffer->lb_buf = ma->ma_lmm;
 968                 /* eadatasize from client includes NULL-terminator, so
 969                  * there is no need to read it */
 970                 buffer->lb_len = reqbody->mbo_eadatasize - 1;
 971                 rc = mo_readlink(env, next, buffer);
 972                 if (unlikely(rc <= 0)) {
 973                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
 974                                mdt_obd_name(info->mti_mdt),
 975                                PFID(mdt_object_fid(o)), rc);
 976                         rc = -EFAULT;
 977                 } else {
 978                         int print_limit = min_t(int, PAGE_CACHE_SIZE - 128, rc);
 979
 980                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
 981                                 rc -= 2;
 982                         repbody->mbo_valid |= OBD_MD_LINKNAME;
 983                         /* we need to report back size with NULL-terminator
 984                          * because client expects that */
 985                         repbody->mbo_eadatasize = rc + 1;
 986                         if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
 987                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
 988                                        "on "DFID ", expected %d\n",
 989                                        mdt_obd_name(info->mti_mdt),
 990                                        rc, PFID(mdt_object_fid(o)),
 991                                        reqbody->mbo_eadatasize - 1);
 992                         /* NULL terminate */
 993                         ((char *)ma->ma_lmm)[rc] = 0;
 994
 995                         /* If the total CDEBUG() size is larger than a page, it
 996                          * will print a warning to the console, avoid this by
 997                          * printing just the last part of the symlink. */
 998                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
 999                                print_limit < rc ? "..." : "", print_limit,
1000                                (char *)ma->ma_lmm + rc - print_limit, rc);
1001                         rc = 0;
1002                 }
1003         }
1004
1005         if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
1006                 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1007                 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1008                 CDEBUG(D_INODE, "changing the max MD size to %u\n",
1009                        repbody->mbo_max_mdsize);
1010         }
1011
1012         if (exp_connect_rmtclient(info->mti_exp) &&
1013             reqbody->mbo_valid & OBD_MD_FLRMTPERM) {
1014                 void *buf = req_capsule_server_get(pill, &RMF_ACL);
1015
1016                 /* mdt_getattr_lock only */
1017                 rc = mdt_pack_remote_perm(info, o, buf);
1018                 if (rc) {
1019                         repbody->mbo_valid &= ~OBD_MD_FLRMTPERM;
1020                         repbody->mbo_aclsize = 0;
1021                         RETURN(rc);
1022                 } else {
1023                         repbody->mbo_valid |= OBD_MD_FLRMTPERM;
1024                         repbody->mbo_aclsize = sizeof(struct mdt_remote_perm);
1025                 }
1026         }
1027 #ifdef CONFIG_FS_POSIX_ACL
1028         else if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1029                  (reqbody->mbo_valid & OBD_MD_FLACL))
1030                 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
1031 #endif
1032
1033 out:
1034         if (rc == 0)
1035                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
1036
1037         RETURN(rc);
1038 }
1039
1040 static int mdt_getattr(struct tgt_session_info *tsi)
1041 {
1042         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1043         struct mdt_object       *obj = info->mti_object;
1044         struct req_capsule      *pill = info->mti_pill;
1045         struct mdt_body         *reqbody;
1046         struct mdt_body         *repbody;
1047         int rc, rc2;
1048         ENTRY;
1049
1050         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1051         LASSERT(reqbody);
1052         LASSERT(obj != NULL);
1053         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1054
1055         /* Unlike intent case where we need to pre-fill out buffers early on
1056          * in intent policy for ldlm reasons, here we can have a much better
1057          * guess at EA size by just reading it from disk.
1058          * Exceptions are readdir and (missing) directory striping */
1059         /* Readlink */
1060         if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1061                 /* No easy way to know how long is the symlink, but it cannot
1062                  * be more than PATH_MAX, so we allocate +1 */
1063                 rc = PATH_MAX + 1;
1064
1065         /* A special case for fs ROOT: getattr there might fetch
1066          * default EA for entire fs, not just for this dir!
1067          */
1068         } else if (lu_fid_eq(mdt_object_fid(obj),
1069                              &info->mti_mdt->mdt_md_root_fid) &&
1070                    (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1071                    (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1072                                                                  MDS_GETATTR)) {
1073                 /* Should the default strping be bigger, mdt_fix_reply
1074                  * will reallocate */
1075                 rc = DEF_REP_MD_SIZE;
1076         } else {
1077                 /* Read the actual EA size from disk */
1078                 rc = mdt_attr_get_eabuf_size(info, obj);
1079         }
1080
1081         if (rc < 0)
1082                 GOTO(out, rc = err_serious(rc));
1083
1084         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1085
1086         rc = req_capsule_server_pack(pill);
1087         if (unlikely(rc != 0))
1088                 GOTO(out, rc = err_serious(rc));
1089
1090         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1091         LASSERT(repbody != NULL);
1092         repbody->mbo_eadatasize = 0;
1093         repbody->mbo_aclsize = 0;
1094
1095         if (reqbody->mbo_valid & OBD_MD_FLRMTPERM)
1096                 rc = mdt_init_ucred(info, reqbody);
1097         else
1098                 rc = mdt_check_ucred(info);
1099         if (unlikely(rc))
1100                 GOTO(out_shrink, rc);
1101
1102         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1103
1104         rc = mdt_getattr_internal(info, obj, 0);
1105         if (reqbody->mbo_valid & OBD_MD_FLRMTPERM)
1106                 mdt_exit_ucred(info);
1107         EXIT;
1108 out_shrink:
1109         mdt_client_compatibility(info);
1110         rc2 = mdt_fix_reply(info);
1111         if (rc == 0)
1112                 rc = rc2;
1113 out:
1114         mdt_thread_info_fini(info);
1115         return rc;
1116 }
1117
1118 /**
1119  * Exchange MOF_LOV_CREATED flags between two objects after a
1120  * layout swap. No assumption is made on whether o1 or o2 have
1121  * created objects or not.
1122  *
1123  * \param[in,out] o1    First swap layout object
1124  * \param[in,out] o2    Second swap layout object
1125  */
1126 static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2)
1127 {
1128         __u64   o1_flags;
1129
1130         mutex_lock(&o1->mot_lov_mutex);
1131         mutex_lock(&o2->mot_lov_mutex);
1132
1133         o1_flags = o1->mot_flags;
1134         o1->mot_flags = (o1->mot_flags & ~MOF_LOV_CREATED) |
1135                         (o2->mot_flags & MOF_LOV_CREATED);
1136
1137         o2->mot_flags = (o2->mot_flags & ~MOF_LOV_CREATED) |
1138                         (o1_flags & MOF_LOV_CREATED);
1139
1140         mutex_unlock(&o2->mot_lov_mutex);
1141         mutex_unlock(&o1->mot_lov_mutex);
1142 }
1143
1144 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1145 {
1146         struct mdt_thread_info  *info;
1147         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1148         struct obd_export       *exp = req->rq_export;
1149         struct mdt_object       *o1, *o2, *o;
1150         struct mdt_lock_handle  *lh1, *lh2;
1151         struct mdc_swap_layouts *msl;
1152         int                      rc;
1153         ENTRY;
1154
1155         /* client does not support layout lock, so layout swaping
1156          * is disabled.
1157          * FIXME: there is a problem for old clients which don't support
1158          * layout lock yet. If those clients have already opened the file
1159          * they won't be notified at all so that old layout may still be
1160          * used to do IO. This can be fixed after file release is landed by
1161          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1162         if (!exp_connect_layout(exp))
1163                 RETURN(-EOPNOTSUPP);
1164
1165         info = tsi2mdt_info(tsi);
1166
1167         if (info->mti_dlm_req != NULL)
1168                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1169
1170         o1 = info->mti_object;
1171         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1172                                 &info->mti_body->mbo_fid2);
1173         if (IS_ERR(o))
1174                 GOTO(out, rc = PTR_ERR(o));
1175
1176         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1177                 GOTO(put, rc = -ENOENT);
1178
1179         rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1180         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1181                 GOTO(put, rc);
1182
1183         if (rc < 0)
1184                 swap(o1, o2);
1185
1186         /* permission check. Make sure the calling process having permission
1187          * to write both files. */
1188         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1189                                 MAY_WRITE);
1190         if (rc < 0)
1191                 GOTO(put, rc);
1192
1193         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1194                                 MAY_WRITE);
1195         if (rc < 0)
1196                 GOTO(put, rc);
1197
1198         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1199         if (msl == NULL)
1200                 GOTO(put, rc = -EPROTO);
1201
1202         lh1 = &info->mti_lh[MDT_LH_NEW];
1203         mdt_lock_reg_init(lh1, LCK_EX);
1204         lh2 = &info->mti_lh[MDT_LH_OLD];
1205         mdt_lock_reg_init(lh2, LCK_EX);
1206
1207         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1208                              MDS_INODELOCK_XATTR);
1209         if (rc < 0)
1210                 GOTO(put, rc);
1211
1212         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1213                              MDS_INODELOCK_XATTR);
1214         if (rc < 0)
1215                 GOTO(unlock1, rc);
1216
1217         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1218                              mdt_object_child(o2), msl->msl_flags);
1219         if (rc < 0)
1220                 GOTO(unlock2, rc);
1221
1222         mdt_swap_lov_flag(o1, o2);
1223
1224 unlock2:
1225         mdt_object_unlock(info, o2, lh2, rc);
1226 unlock1:
1227         mdt_object_unlock(info, o1, lh1, rc);
1228 put:
1229         mdt_object_put(info->mti_env, o);
1230 out:
1231         mdt_thread_info_fini(info);
1232         RETURN(rc);
1233 }
1234
1235 static int mdt_raw_lookup(struct mdt_thread_info *info,
1236                           struct mdt_object *parent,
1237                           const struct lu_name *lname,
1238                           struct ldlm_reply *ldlm_rep)
1239 {
1240         struct lu_fid   *child_fid = &info->mti_tmp_fid1;
1241         int              rc;
1242         ENTRY;
1243
1244         LASSERT(!info->mti_cross_ref);
1245
1246         /* Only got the fid of this obj by name */
1247         fid_zero(child_fid);
1248         rc = mdo_lookup(info->mti_env, mdt_object_child(info->mti_object),
1249                         lname, child_fid, &info->mti_spec);
1250         if (rc == 0) {
1251                 struct mdt_body *repbody;
1252
1253                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1254                 repbody->mbo_fid1 = *child_fid;
1255                 repbody->mbo_valid = OBD_MD_FLID;
1256                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1257         } else if (rc == -ENOENT) {
1258                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1259         }
1260
1261         RETURN(rc);
1262 }
1263
1264 /*
1265  * UPDATE lock should be taken against parent, and be released before exit;
1266  * child_bits lock should be taken against child, and be returned back:
1267  *            (1)normal request should release the child lock;
1268  *            (2)intent request will grant the lock to client.
1269  */
1270 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1271                                  struct mdt_lock_handle *lhc,
1272                                  __u64 child_bits,
1273                                  struct ldlm_reply *ldlm_rep)
1274 {
1275         struct ptlrpc_request  *req       = mdt_info_req(info);
1276         struct mdt_body        *reqbody   = NULL;
1277         struct mdt_object      *parent    = info->mti_object;
1278         struct mdt_object      *child;
1279         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1280         struct lu_name         *lname     = NULL;
1281         struct mdt_lock_handle *lhp       = NULL;
1282         struct ldlm_lock       *lock;
1283         bool                    is_resent;
1284         bool                    try_layout;
1285         int                     ma_need = 0;
1286         int                     rc;
1287         ENTRY;
1288
1289         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1290         LASSERT(ergo(is_resent,
1291                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1292
1293         if (parent == NULL)
1294                 RETURN(-ENOENT);
1295
1296         if (info->mti_cross_ref) {
1297                 /* Only getattr on the child. Parent is on another node. */
1298                 mdt_set_disposition(info, ldlm_rep,
1299                                     DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1300                 child = parent;
1301                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1302                        "ldlm_rep = %p\n",
1303                        PFID(mdt_object_fid(child)), ldlm_rep);
1304
1305                 rc = mdt_check_resent_lock(info, child, lhc);
1306                 if (rc < 0) {
1307                         RETURN(rc);
1308                 } else if (rc > 0) {
1309                         mdt_lock_handle_init(lhc);
1310                         mdt_lock_reg_init(lhc, LCK_PR);
1311
1312                         /*
1313                          * Object's name is on another MDS, no lookup or layout
1314                          * lock is needed here but update lock is.
1315                          */
1316                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
1317                                         MDS_INODELOCK_LAYOUT);
1318                         child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1319
1320                         rc = mdt_object_lock(info, child, lhc, child_bits);
1321                         if (rc < 0)
1322                                 RETURN(rc);
1323                 }
1324
1325                 /* Finally, we can get attr for child. */
1326                 if (!mdt_object_exists(child)) {
1327                         LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1328                                         &child->mot_obj,
1329                                         "remote object doesn't exist.\n");
1330                         mdt_object_unlock(info, child, lhc, 1);
1331                         RETURN(-ENOENT);
1332                 }
1333
1334                 rc = mdt_getattr_internal(info, child, 0);
1335                 if (unlikely(rc != 0))
1336                         mdt_object_unlock(info, child, lhc, 1);
1337
1338                 RETURN(rc);
1339         }
1340
1341         lname = &info->mti_name;
1342         mdt_name_unpack(info->mti_pill, &RMF_NAME, lname, MNF_FIX_ANON);
1343
1344         if (lu_name_is_valid(lname)) {
1345                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1346                        "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1347                        PNAME(lname), ldlm_rep);
1348         } else {
1349                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1350                 if (unlikely(reqbody == NULL))
1351                         RETURN(err_serious(-EPROTO));
1352
1353                 *child_fid = reqbody->mbo_fid2;
1354
1355                 if (unlikely(!fid_is_sane(child_fid)))
1356                         RETURN(err_serious(-EINVAL));
1357
1358                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1359                        "ldlm_rep = %p\n",
1360                        PFID(mdt_object_fid(parent)),
1361                        PFID(&reqbody->mbo_fid2), ldlm_rep);
1362         }
1363
1364         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1365
1366         if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1367                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1368                                 &parent->mot_obj,
1369                                 "Parent doesn't exist!\n");
1370                 RETURN(-ESTALE);
1371         }
1372
1373         if (mdt_object_remote(parent)) {
1374                 CERROR("%s: parent "DFID" is on remote target\n",
1375                        mdt_obd_name(info->mti_mdt),
1376                        PFID(mdt_object_fid(parent)));
1377                 RETURN(-EIO);
1378         }
1379
1380         if (lu_name_is_valid(lname)) {
1381                 /* Always allow to lookup ".." */
1382                 if (unlikely(lname->ln_namelen == 2 &&
1383                              lname->ln_name[0] == '.' &&
1384                              lname->ln_name[1] == '.'))
1385                         info->mti_spec.sp_permitted = 1;
1386
1387                 if (info->mti_body->mbo_valid == OBD_MD_FLID) {
1388                         rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1389
1390                         RETURN(rc);
1391                 }
1392
1393                 /* step 1: lock parent only if parent is a directory */
1394                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1395                         lhp = &info->mti_lh[MDT_LH_PARENT];
1396                         mdt_lock_pdo_init(lhp, LCK_PR, lname);
1397                         rc = mdt_object_lock(info, parent, lhp,
1398                                              MDS_INODELOCK_UPDATE);
1399                         if (unlikely(rc != 0))
1400                                 RETURN(rc);
1401                 }
1402
1403                 /* step 2: lookup child's fid by name */
1404                 fid_zero(child_fid);
1405                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
1406                                 child_fid, &info->mti_spec);
1407                 if (rc == -ENOENT)
1408                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1409
1410                 if (rc != 0)
1411                         GOTO(out_parent, rc);
1412         }
1413
1414         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1415
1416         /*
1417          *step 3: find the child object by fid & lock it.
1418          *        regardless if it is local or remote.
1419          *
1420          *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1421          *      set parent dir fid the same as child fid in getattr by fid case
1422          *      we should not lu_object_find() the object again, could lead
1423          *      to hung if there is a concurrent unlink destroyed the object.
1424          */
1425         if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1426                 mdt_object_get(info->mti_env, parent);
1427                 child = parent;
1428         } else {
1429                 child = mdt_object_find(info->mti_env, info->mti_mdt,
1430                                         child_fid);
1431         }
1432
1433         if (unlikely(IS_ERR(child)))
1434                 GOTO(out_parent, rc = PTR_ERR(child));
1435
1436         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
1437         if (!mdt_object_exists(child)) {
1438                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1439                                 &child->mot_obj,
1440                                 "Object doesn't exist!\n");
1441                 GOTO(out_child, rc = -ENOENT);
1442         }
1443
1444         rc = mdt_check_resent_lock(info, child, lhc);
1445         if (rc < 0) {
1446                 GOTO(out_child, rc);
1447         } else if (rc > 0) {
1448                 mdt_lock_handle_init(lhc);
1449                 mdt_lock_reg_init(lhc, LCK_PR);
1450                 try_layout = false;
1451
1452                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1453                       mdt_object_exists(child) && !mdt_object_remote(child)) {
1454                         struct md_attr *ma = &info->mti_attr;
1455
1456                         ma->ma_valid = 0;
1457                         ma->ma_need = MA_INODE;
1458                         rc = mdt_attr_get_complex(info, child, ma);
1459                         if (unlikely(rc != 0))
1460                                 GOTO(out_child, rc);
1461
1462                         /* If the file has not been changed for some time, we
1463                          * return not only a LOOKUP lock, but also an UPDATE
1464                          * lock and this might save us RPC on later STAT. For
1465                          * directories, it also let negative dentry cache start
1466                          * working for this dir. */
1467                         if (ma->ma_valid & MA_INODE &&
1468                             ma->ma_attr.la_valid & LA_CTIME &&
1469                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1470                                 ma->ma_attr.la_ctime < cfs_time_current_sec())
1471                                 child_bits |= MDS_INODELOCK_UPDATE;
1472                 }
1473
1474                 /* layout lock must be granted in a best-effort way
1475                  * for IT operations */
1476                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1477                 if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1478                     exp_connect_layout(info->mti_exp) &&
1479                     S_ISREG(lu_object_attr(&child->mot_obj)) &&
1480                     !mdt_object_remote(child) && ldlm_rep != NULL) {
1481                         /* try to grant layout lock for regular file. */
1482                         try_layout = true;
1483                 }
1484
1485                 rc = 0;
1486                 if (try_layout) {
1487                         child_bits |= MDS_INODELOCK_LAYOUT;
1488                         /* try layout lock, it may fail to be granted due to
1489                          * contention at LOOKUP or UPDATE */
1490                         if (!mdt_object_lock_try(info, child, lhc,
1491                                                  child_bits)) {
1492                                 child_bits &= ~MDS_INODELOCK_LAYOUT;
1493                                 LASSERT(child_bits != 0);
1494                                 rc = mdt_object_lock(info, child, lhc,
1495                                                      child_bits);
1496                         } else {
1497                                 ma_need |= MA_LOV;
1498                         }
1499                 } else {
1500                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1501                          * client will enqueue the lock to the remote MDT */
1502                         if (mdt_object_remote(child))
1503                                 child_bits &= ~MDS_INODELOCK_UPDATE;
1504                         rc = mdt_object_lock(info, child, lhc, child_bits);
1505                 }
1506                 if (unlikely(rc != 0))
1507                         GOTO(out_child, rc);
1508         }
1509
1510         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1511
1512         /* finally, we can get attr for child. */
1513         rc = mdt_getattr_internal(info, child, ma_need);
1514         if (unlikely(rc != 0)) {
1515                 mdt_object_unlock(info, child, lhc, 1);
1516         } else if (lock) {
1517                 /* Debugging code. */
1518                 LDLM_DEBUG(lock, "Returning lock to client");
1519                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1520                                          &lock->l_resource->lr_name),
1521                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1522                          PLDLMRES(lock->l_resource),
1523                          PFID(mdt_object_fid(child)));
1524         }
1525         if (lock)
1526                 LDLM_LOCK_PUT(lock);
1527
1528         EXIT;
1529 out_child:
1530         mdt_object_put(info->mti_env, child);
1531 out_parent:
1532         if (lhp)
1533                 mdt_object_unlock(info, parent, lhp, 1);
1534         return rc;
1535 }
1536
1537 /* normal handler: should release the child lock */
1538 static int mdt_getattr_name(struct tgt_session_info *tsi)
1539 {
1540         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1541         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1542         struct mdt_body        *reqbody;
1543         struct mdt_body        *repbody;
1544         int rc, rc2;
1545         ENTRY;
1546
1547         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1548         LASSERT(reqbody != NULL);
1549         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1550         LASSERT(repbody != NULL);
1551
1552         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1553         repbody->mbo_eadatasize = 0;
1554         repbody->mbo_aclsize = 0;
1555
1556         rc = mdt_init_ucred_intent_getattr(info, reqbody);
1557         if (unlikely(rc))
1558                 GOTO(out_shrink, rc);
1559
1560         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1561         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1562                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1563                 lhc->mlh_reg_lh.cookie = 0;
1564         }
1565         mdt_exit_ucred(info);
1566         EXIT;
1567 out_shrink:
1568         mdt_client_compatibility(info);
1569         rc2 = mdt_fix_reply(info);
1570         if (rc == 0)
1571                 rc = rc2;
1572         mdt_thread_info_fini(info);
1573         return rc;
1574 }
1575
1576 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1577                          void *karg, void *uarg);
1578
1579 static int mdt_set_info(struct tgt_session_info *tsi)
1580 {
1581         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1582         char                    *key;
1583         void                    *val;
1584         int                      keylen, vallen, rc = 0;
1585
1586         ENTRY;
1587
1588         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
1589         if (key == NULL) {
1590                 DEBUG_REQ(D_HA, req, "no set_info key");
1591                 RETURN(err_serious(-EFAULT));
1592         }
1593
1594         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
1595                                       RCL_CLIENT);
1596
1597         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
1598         if (val == NULL) {
1599                 DEBUG_REQ(D_HA, req, "no set_info val");
1600                 RETURN(err_serious(-EFAULT));
1601         }
1602
1603         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
1604                                       RCL_CLIENT);
1605
1606         /* Swab any part of val you need to here */
1607         if (KEY_IS(KEY_READ_ONLY)) {
1608                 spin_lock(&req->rq_export->exp_lock);
1609                 if (*(__u32 *)val)
1610                         *exp_connect_flags_ptr(req->rq_export) |=
1611                                 OBD_CONNECT_RDONLY;
1612                 else
1613                         *exp_connect_flags_ptr(req->rq_export) &=
1614                                 ~OBD_CONNECT_RDONLY;
1615                 spin_unlock(&req->rq_export->exp_lock);
1616         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1617                 struct changelog_setinfo *cs = val;
1618
1619                 if (vallen != sizeof(*cs)) {
1620                         CERROR("%s: bad changelog_clear setinfo size %d\n",
1621                                tgt_name(tsi->tsi_tgt), vallen);
1622                         RETURN(-EINVAL);
1623                 }
1624                 if (ptlrpc_req_need_swab(req)) {
1625                         __swab64s(&cs->cs_recno);
1626                         __swab32s(&cs->cs_id);
1627                 }
1628
1629                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
1630                                    vallen, val, NULL);
1631         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
1632                 if (vallen > 0)
1633                         obd_export_evict_by_nid(req->rq_export->exp_obd, val);
1634         } else {
1635                 RETURN(-EINVAL);
1636         }
1637         RETURN(rc);
1638 }
1639
1640 static int mdt_readpage(struct tgt_session_info *tsi)
1641 {
1642         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
1643         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
1644         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
1645         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
1646         struct mdt_body         *repbody;
1647         int                      rc;
1648         int                      i;
1649
1650         ENTRY;
1651
1652         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1653                 RETURN(err_serious(-ENOMEM));
1654
1655         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
1656         if (repbody == NULL || reqbody == NULL)
1657                 RETURN(err_serious(-EFAULT));
1658
1659         /*
1660          * prepare @rdpg before calling lower layers and transfer itself. Here
1661          * reqbody->size contains offset of where to start to read and
1662          * reqbody->nlink contains number bytes to read.
1663          */
1664         rdpg->rp_hash = reqbody->mbo_size;
1665         if (rdpg->rp_hash != reqbody->mbo_size) {
1666                 CERROR("Invalid hash: "LPX64" != "LPX64"\n",
1667                        rdpg->rp_hash, reqbody->mbo_size);
1668                 RETURN(-EFAULT);
1669         }
1670
1671         rdpg->rp_attrs = reqbody->mbo_mode;
1672         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
1673                 rdpg->rp_attrs |= LUDA_64BITHASH;
1674         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
1675                                 exp_max_brw_size(tsi->tsi_exp));
1676         rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE - 1) >>
1677                           PAGE_CACHE_SHIFT;
1678         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1679         if (rdpg->rp_pages == NULL)
1680                 RETURN(-ENOMEM);
1681
1682         for (i = 0; i < rdpg->rp_npages; ++i) {
1683                 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1684                 if (rdpg->rp_pages[i] == NULL)
1685                         GOTO(free_rdpg, rc = -ENOMEM);
1686         }
1687
1688         /* call lower layers to fill allocated pages with directory data */
1689         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
1690         if (rc < 0)
1691                 GOTO(free_rdpg, rc);
1692
1693         /* send pages to client */
1694         rc = tgt_sendpage(tsi, rdpg, rc);
1695
1696         EXIT;
1697 free_rdpg:
1698
1699         for (i = 0; i < rdpg->rp_npages; i++)
1700                 if (rdpg->rp_pages[i] != NULL)
1701                         __free_page(rdpg->rp_pages[i]);
1702         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1703
1704         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1705                 RETURN(0);
1706
1707         return rc;
1708 }
1709
1710 static int mdt_reint_internal(struct mdt_thread_info *info,
1711                               struct mdt_lock_handle *lhc,
1712                               __u32 op)
1713 {
1714         struct req_capsule      *pill = info->mti_pill;
1715         struct mdt_body         *repbody;
1716         int                      rc = 0, rc2;
1717
1718         ENTRY;
1719
1720         rc = mdt_reint_unpack(info, op);
1721         if (rc != 0) {
1722                 CERROR("Can't unpack reint, rc %d\n", rc);
1723                 RETURN(err_serious(rc));
1724         }
1725
1726         /* for replay (no_create) lmm is not needed, client has it already */
1727         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1728                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1729                                      DEF_REP_MD_SIZE);
1730
1731         /* llog cookies are always 0, the field is kept for compatibility */
1732         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1733                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
1734
1735         rc = req_capsule_server_pack(pill);
1736         if (rc != 0) {
1737                 CERROR("Can't pack response, rc %d\n", rc);
1738                 RETURN(err_serious(rc));
1739         }
1740
1741         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1742                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1743                 LASSERT(repbody);
1744                 repbody->mbo_eadatasize = 0;
1745                 repbody->mbo_aclsize = 0;
1746         }
1747
1748         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1749
1750         /* for replay no cookkie / lmm need, because client have this already */
1751         if (info->mti_spec.no_create)
1752                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1753                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1754
1755         rc = mdt_init_ucred_reint(info);
1756         if (rc)
1757                 GOTO(out_shrink, rc);
1758
1759         rc = mdt_fix_attr_ucred(info, op);
1760         if (rc != 0)
1761                 GOTO(out_ucred, rc = err_serious(rc));
1762
1763         rc = mdt_check_resent(info, mdt_reconstruct, lhc);
1764         if (rc < 0) {
1765                 GOTO(out_ucred, rc);
1766         } else if (rc == 1) {
1767                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
1768                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1769                 GOTO(out_ucred, rc);
1770         }
1771         rc = mdt_reint_rec(info, lhc);
1772         EXIT;
1773 out_ucred:
1774         mdt_exit_ucred(info);
1775 out_shrink:
1776         mdt_client_compatibility(info);
1777         rc2 = mdt_fix_reply(info);
1778         if (rc == 0)
1779                 rc = rc2;
1780         return rc;
1781 }
1782
1783 static long mdt_reint_opcode(struct ptlrpc_request *req,
1784                              const struct req_format **fmt)
1785 {
1786         struct mdt_device       *mdt;
1787         struct mdt_rec_reint    *rec;
1788         long                     opc;
1789
1790         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
1791         if (rec != NULL) {
1792                 opc = rec->rr_opcode;
1793                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
1794                 if (opc < REINT_MAX && fmt[opc] != NULL)
1795                         req_capsule_extend(&req->rq_pill, fmt[opc]);
1796                 else {
1797                         mdt = mdt_exp2dev(req->rq_export);
1798                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
1799                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
1800                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
1801                         opc = err_serious(-EFAULT);
1802                 }
1803         } else {
1804                 opc = err_serious(-EFAULT);
1805         }
1806         return opc;
1807 }
1808
1809 static int mdt_reint(struct tgt_session_info *tsi)
1810 {
1811         long opc;
1812         int  rc;
1813         static const struct req_format *reint_fmts[REINT_MAX] = {
1814                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
1815                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
1816                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
1817                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
1818                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
1819                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
1820                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
1821                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
1822                 [REINT_MIGRATE]  = &RQF_MDS_REINT_RENAME
1823         };
1824
1825         ENTRY;
1826
1827         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
1828         if (opc >= 0) {
1829                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1830                 /*
1831                  * No lock possible here from client to pass it to reint code
1832                  * path.
1833                  */
1834                 rc = mdt_reint_internal(info, NULL, opc);
1835                 mdt_thread_info_fini(info);
1836         } else {
1837                 rc = opc;
1838         }
1839
1840         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
1841         RETURN(rc);
1842 }
1843
1844 /* this should sync the whole device */
1845 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
1846 {
1847         struct dt_device *dt = mdt->mdt_bottom;
1848         int rc;
1849         ENTRY;
1850
1851         rc = dt->dd_ops->dt_sync(env, dt);
1852         RETURN(rc);
1853 }
1854
1855 /* this should sync this object */
1856 static int mdt_object_sync(struct mdt_thread_info *info)
1857 {
1858         struct md_object *next;
1859         int rc;
1860         ENTRY;
1861
1862         if (!mdt_object_exists(info->mti_object)) {
1863                 CWARN("%s: non existing object "DFID": rc = %d\n",
1864                       mdt_obd_name(info->mti_mdt),
1865                       PFID(mdt_object_fid(info->mti_object)), -ESTALE);
1866                 RETURN(-ESTALE);
1867         }
1868         next = mdt_object_child(info->mti_object);
1869         rc = mo_object_sync(info->mti_env, next);
1870
1871         RETURN(rc);
1872 }
1873
1874 static int mdt_sync(struct tgt_session_info *tsi)
1875 {
1876         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1877         struct req_capsule      *pill = tsi->tsi_pill;
1878         struct mdt_body         *body;
1879         int                      rc;
1880
1881         ENTRY;
1882
1883         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
1884                 RETURN(err_serious(-ENOMEM));
1885
1886         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
1887                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
1888         } else {
1889                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
1890
1891                 /* sync an object */
1892                 rc = mdt_object_sync(info);
1893                 if (rc == 0) {
1894                         const struct lu_fid *fid;
1895                         struct lu_attr *la = &info->mti_attr.ma_attr;
1896
1897                         info->mti_attr.ma_need = MA_INODE;
1898                         info->mti_attr.ma_valid = 0;
1899                         rc = mdt_attr_get_complex(info, info->mti_object,
1900                                                   &info->mti_attr);
1901                         if (rc == 0) {
1902                                 body = req_capsule_server_get(pill,
1903                                                               &RMF_MDT_BODY);
1904                                 fid = mdt_object_fid(info->mti_object);
1905                                 mdt_pack_attr2body(info, body, la, fid);
1906                         }
1907                 }
1908                 mdt_thread_info_fini(info);
1909         }
1910         if (rc == 0)
1911                 mdt_counter_incr(req, LPROC_MDT_SYNC);
1912
1913         RETURN(rc);
1914 }
1915
1916 /*
1917  * Handle quota control requests to consult current usage/limit, but also
1918  * to configure quota enforcement
1919  */
1920 static int mdt_quotactl(struct tgt_session_info *tsi)
1921 {
1922         struct obd_export       *exp  = tsi->tsi_exp;
1923         struct req_capsule      *pill = tsi->tsi_pill;
1924         struct obd_quotactl     *oqctl, *repoqc;
1925         int                      id, rc;
1926         struct mdt_device       *mdt = mdt_exp2dev(exp);
1927         struct lu_device        *qmt = mdt->mdt_qmt_dev;
1928         struct lu_nodemap       *nodemap = exp->exp_target_data.ted_nodemap;
1929         ENTRY;
1930
1931         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
1932         if (oqctl == NULL)
1933                 RETURN(err_serious(-EPROTO));
1934
1935         rc = req_capsule_server_pack(pill);
1936         if (rc)
1937                 RETURN(err_serious(rc));
1938
1939         switch (oqctl->qc_cmd) {
1940                 /* master quotactl */
1941         case Q_SETINFO:
1942         case Q_SETQUOTA:
1943                 if (!nodemap_can_setquota(nodemap))
1944                         RETURN(-EPERM);
1945         case Q_GETINFO:
1946         case Q_GETQUOTA:
1947                 if (qmt == NULL)
1948                         RETURN(-EOPNOTSUPP);
1949                 /* slave quotactl */
1950         case Q_GETOINFO:
1951         case Q_GETOQUOTA:
1952                 break;
1953         default:
1954                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
1955                 RETURN(-EFAULT);
1956         }
1957
1958         /* map uid/gid for remote client */
1959         id = oqctl->qc_id;
1960         if (exp_connect_rmtclient(exp)) {
1961                 struct lustre_idmap_table *idmap;
1962
1963                 idmap = exp->exp_mdt_data.med_idmap;
1964
1965                 if (unlikely(oqctl->qc_cmd != Q_GETQUOTA &&
1966                              oqctl->qc_cmd != Q_GETINFO))
1967                         RETURN(-EPERM);
1968
1969                 if (oqctl->qc_type == USRQUOTA)
1970                         id = lustre_idmap_lookup_uid(NULL, idmap, 0,
1971                                                      oqctl->qc_id);
1972                 else if (oqctl->qc_type == GRPQUOTA)
1973                         id = lustre_idmap_lookup_gid(NULL, idmap, 0,
1974                                                      oqctl->qc_id);
1975                 else
1976                         RETURN(-EINVAL);
1977
1978                 if (id == CFS_IDMAP_NOTFOUND) {
1979                         CDEBUG(D_QUOTA, "no mapping for id %u\n", oqctl->qc_id);
1980                         RETURN(-EACCES);
1981                 }
1982         }
1983
1984         if (oqctl->qc_type == USRQUOTA)
1985                 id = nodemap_map_id(nodemap, NODEMAP_UID,
1986                                     NODEMAP_CLIENT_TO_FS, id);
1987         else if (oqctl->qc_type == GRPQUOTA)
1988                 id = nodemap_map_id(nodemap, NODEMAP_UID,
1989                                     NODEMAP_CLIENT_TO_FS, id);
1990
1991         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
1992         if (repoqc == NULL)
1993                 RETURN(err_serious(-EFAULT));
1994
1995         if (oqctl->qc_id != id)
1996                 swap(oqctl->qc_id, id);
1997
1998         switch (oqctl->qc_cmd) {
1999
2000         case Q_GETINFO:
2001         case Q_SETINFO:
2002         case Q_SETQUOTA:
2003         case Q_GETQUOTA:
2004                 /* forward quotactl request to QMT */
2005                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2006                 break;
2007
2008         case Q_GETOINFO:
2009         case Q_GETOQUOTA:
2010                 /* slave quotactl */
2011                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2012                                    oqctl);
2013                 break;
2014
2015         default:
2016                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2017                 RETURN(-EFAULT);
2018         }
2019
2020         if (oqctl->qc_id != id)
2021                 swap(oqctl->qc_id, id);
2022
2023         *repoqc = *oqctl;
2024         RETURN(rc);
2025 }
2026
2027 /** clone llog ctxt from child (mdd)
2028  * This allows remote llog (replicator) access.
2029  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2030  * context was originally set up, or we can handle them directly.
2031  * I choose the latter, but that means I need any llog
2032  * contexts set up by child to be accessable by the mdt.  So we clone the
2033  * context into our context list here.
2034  */
2035 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2036                                int idx)
2037 {
2038         struct md_device  *next = mdt->mdt_child;
2039         struct llog_ctxt *ctxt;
2040         int rc;
2041
2042         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2043                 return 0;
2044
2045         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2046         if (rc || ctxt == NULL) {
2047                 return 0;
2048         }
2049
2050         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2051         if (rc)
2052                 CERROR("Can't set mdt ctxt %d\n", rc);
2053
2054         return rc;
2055 }
2056
2057 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2058                                  struct mdt_device *mdt, int idx)
2059 {
2060         struct llog_ctxt *ctxt;
2061
2062         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2063         if (ctxt == NULL)
2064                 return 0;
2065         /* Put once for the get we just did, and once for the clone */
2066         llog_ctxt_put(ctxt);
2067         llog_ctxt_put(ctxt);
2068         return 0;
2069 }
2070
2071 /*
2072  * sec context handlers
2073  */
2074 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2075 {
2076         int rc;
2077
2078         rc = mdt_handle_idmap(tsi);
2079         if (unlikely(rc)) {
2080                 struct ptlrpc_request   *req = tgt_ses_req(tsi);
2081                 __u32                    opc;
2082
2083                 opc = lustre_msg_get_opc(req->rq_reqmsg);
2084                 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT)
2085                         sptlrpc_svc_ctx_invalidate(req);
2086         }
2087
2088         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2089
2090         return rc;
2091 }
2092
2093 /*
2094  * quota request handlers
2095  */
2096 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2097 {
2098         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
2099         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2100         int                      rc;
2101         ENTRY;
2102
2103         if (qmt == NULL)
2104                 RETURN(err_serious(-EOPNOTSUPP));
2105
2106         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2107         RETURN(rc);
2108 }
2109
2110 struct mdt_object *mdt_object_new(const struct lu_env *env,
2111                                   struct mdt_device *d,
2112                                   const struct lu_fid *f)
2113 {
2114         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2115         struct lu_object *o;
2116         struct mdt_object *m;
2117         ENTRY;
2118
2119         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2120         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2121         if (unlikely(IS_ERR(o)))
2122                 m = (struct mdt_object *)o;
2123         else
2124                 m = mdt_obj(o);
2125         RETURN(m);
2126 }
2127
2128 struct mdt_object *mdt_object_find(const struct lu_env *env,
2129                                    struct mdt_device *d,
2130                                    const struct lu_fid *f)
2131 {
2132         struct lu_object *o;
2133         struct mdt_object *m;
2134         ENTRY;
2135
2136         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2137         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2138         if (unlikely(IS_ERR(o)))
2139                 m = (struct mdt_object *)o;
2140         else
2141                 m = mdt_obj(o);
2142
2143         RETURN(m);
2144 }
2145
2146 /**
2147  * Asyncronous commit for mdt device.
2148  *
2149  * Pass asynchonous commit call down the MDS stack.
2150  *
2151  * \param env environment
2152  * \param mdt the mdt device
2153  */
2154 static void mdt_device_commit_async(const struct lu_env *env,
2155                                     struct mdt_device *mdt)
2156 {
2157         struct dt_device *dt = mdt->mdt_bottom;
2158         int rc;
2159         ENTRY;
2160
2161         rc = dt->dd_ops->dt_commit_async(env, dt);
2162         if (unlikely(rc != 0))
2163                 CWARN("%s: async commit start failed: rc = %d\n",
2164                       mdt_obd_name(mdt), rc);
2165         atomic_inc(&mdt->mdt_async_commit_count);
2166         EXIT;
2167 }
2168
2169 /**
2170  * Mark the lock as "synchonous".
2171  *
2172  * Mark the lock to deffer transaction commit to the unlock time.
2173  *
2174  * \param lock the lock to mark as "synchonous"
2175  *
2176  * \see mdt_is_lock_sync
2177  * \see mdt_save_lock
2178  */
2179 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2180 {
2181         lock->l_ast_data = (void*)1;
2182 }
2183
2184 /**
2185  * Check whehter the lock "synchonous" or not.
2186  *
2187  * \param lock the lock to check
2188  * \retval 1 the lock is "synchonous"
2189  * \retval 0 the lock isn't "synchronous"
2190  *
2191  * \see mdt_set_lock_sync
2192  * \see mdt_save_lock
2193  */
2194 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2195 {
2196         return lock->l_ast_data != NULL;
2197 }
2198
2199 /**
2200  * Blocking AST for mdt locks.
2201  *
2202  * Starts transaction commit if in case of COS lock conflict or
2203  * deffers such a commit to the mdt_save_lock.
2204  *
2205  * \param lock the lock which blocks a request or cancelling lock
2206  * \param desc unused
2207  * \param data unused
2208  * \param flag indicates whether this cancelling or blocking callback
2209  * \retval 0
2210  * \see ldlm_blocking_ast_nocheck
2211  */
2212 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2213                      void *data, int flag)
2214 {
2215         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2216         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2217         int rc;
2218         ENTRY;
2219
2220         if (flag == LDLM_CB_CANCELING)
2221                 RETURN(0);
2222
2223         lock_res_and_lock(lock);
2224         if (lock->l_blocking_ast != mdt_blocking_ast) {
2225                 unlock_res_and_lock(lock);
2226                 RETURN(0);
2227         }
2228         if (lock->l_req_mode & (LCK_PW | LCK_EX) &&
2229             lock->l_blocking_lock != NULL) {
2230                 if (mdt_cos_is_enabled(mdt) &&
2231                     lock->l_client_cookie !=
2232                     lock->l_blocking_lock->l_client_cookie)
2233                         mdt_set_lock_sync(lock);
2234                 else if (mdt_slc_is_enabled(mdt) &&
2235                          ldlm_is_cos_incompat(lock->l_blocking_lock))
2236                         mdt_set_lock_sync(lock);
2237         }
2238         rc = ldlm_blocking_ast_nocheck(lock);
2239
2240         /* There is no lock conflict if l_blocking_lock == NULL,
2241          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2242          * when the last reference to a local lock was released */
2243         if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2244                 struct lu_env env;
2245
2246                 rc = lu_env_init(&env, LCT_LOCAL);
2247                 if (unlikely(rc != 0))
2248                         CWARN("%s: lu_env initialization failed, cannot "
2249                               "start asynchronous commit: rc = %d\n",
2250                               obd->obd_name, rc);
2251                 else
2252                         mdt_device_commit_async(&env, mdt);
2253                 lu_env_fini(&env);
2254         }
2255         RETURN(rc);
2256 }
2257
2258 /*
2259  * Blocking AST for cross-MDT lock
2260  *
2261  * Discard lock from uncommitted_slc_locks and cancel it.
2262  *
2263  * \param lock  the lock which blocks a request or cancelling lock
2264  * \param desc  unused
2265  * \param data  unused
2266  * \param flag  indicates whether this cancelling or blocking callback
2267  * \retval      0 on success
2268  * \retval      negative number on error
2269  */
2270 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2271                             void *data, int flag)
2272 {
2273         struct lustre_handle lockh;
2274         int               rc;
2275         ENTRY;
2276
2277         switch (flag) {
2278         case LDLM_CB_BLOCKING:
2279                 ldlm_lock2handle(lock, &lockh);
2280                 rc = ldlm_cli_cancel(&lockh,
2281                         ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
2282                 if (rc < 0) {
2283                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2284                         RETURN(rc);
2285                 }
2286                 break;
2287         case LDLM_CB_CANCELING:
2288                 LDLM_DEBUG(lock, "Revoke remote lock\n");
2289                 /* discard slc lock here so that it can be cleaned anytime,
2290                  * especially for cleanup_resource() */
2291                 tgt_discard_slc_lock(lock);
2292                 break;
2293         default:
2294                 LBUG();
2295         }
2296
2297         RETURN(0);
2298 }
2299
2300 int mdt_check_resent_lock(struct mdt_thread_info *info,
2301                           struct mdt_object *mo,
2302                           struct mdt_lock_handle *lhc)
2303 {
2304         /* the lock might already be gotten in ldlm_handle_enqueue() */
2305         if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
2306                 struct ptlrpc_request *req = mdt_info_req(info);
2307                 struct ldlm_lock      *lock;
2308
2309                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2310                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
2311                 if (lock == NULL) {
2312                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
2313                          * a resend case, however, it could be already destroyed
2314                          * due to client eviction or a raced cancel RPC. */
2315                         LDLM_DEBUG_NOLOCK("Invalid lock handle "LPX64"\n",
2316                                           lhc->mlh_reg_lh.cookie);
2317                         RETURN(-ESTALE);
2318                 }
2319
2320                 if (!fid_res_name_eq(mdt_object_fid(mo),
2321                                      &lock->l_resource->lr_name)) {
2322                         CWARN("%s: Although resent, but still not "
2323                               "get child lock:"DFID"\n",
2324                               info->mti_exp->exp_obd->obd_name,
2325                               PFID(mdt_object_fid(mo)));
2326                         LDLM_LOCK_PUT(lock);
2327                         RETURN(-EPROTO);
2328                 }
2329                 LDLM_LOCK_PUT(lock);
2330                 return 0;
2331         }
2332         return 1;
2333 }
2334
2335 int mdt_remote_object_lock(struct mdt_thread_info *mti, struct mdt_object *o,
2336                            const struct lu_fid *fid, struct lustre_handle *lh,
2337                            enum ldlm_mode mode, __u64 ibits, bool nonblock)
2338 {
2339         struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
2340         union ldlm_policy_data *policy = &mti->mti_policy;
2341         struct ldlm_res_id *res_id = &mti->mti_res_id;
2342         int rc = 0;
2343         ENTRY;
2344
2345         LASSERT(mdt_object_remote(o));
2346
2347         fid_build_reg_res_name(fid, res_id);
2348
2349         memset(einfo, 0, sizeof(*einfo));
2350         einfo->ei_type = LDLM_IBITS;
2351         einfo->ei_mode = mode;
2352         einfo->ei_cb_bl = mdt_remote_blocking_ast;
2353         einfo->ei_cb_cp = ldlm_completion_ast;
2354         einfo->ei_enq_slave = 0;
2355         einfo->ei_res_id = res_id;
2356         if (nonblock)
2357                 einfo->ei_nonblock = 1;
2358
2359         memset(policy, 0, sizeof(*policy));
2360         policy->l_inodebits.bits = ibits;
2361
2362         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2363                             policy);
2364         RETURN(rc);
2365 }
2366
2367 static int mdt_object_local_lock(struct mdt_thread_info *info,
2368                                  struct mdt_object *o,
2369                                  struct mdt_lock_handle *lh, __u64 ibits,
2370                                  bool nonblock, bool cos_incompat)
2371 {
2372         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2373         union ldlm_policy_data *policy = &info->mti_policy;
2374         struct ldlm_res_id *res_id = &info->mti_res_id;
2375         __u64 dlmflags = 0;
2376         int rc;
2377         ENTRY;
2378
2379         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2380         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2381         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2382         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2383
2384         if (cos_incompat) {
2385                 LASSERT(lh->mlh_reg_mode == LCK_PW ||
2386                         lh->mlh_reg_mode == LCK_EX);
2387                 dlmflags |= LDLM_FL_COS_INCOMPAT;
2388         } else if (mdt_cos_is_enabled(info->mti_mdt)) {
2389                 dlmflags |= LDLM_FL_COS_ENABLED;
2390         }
2391
2392         /* Only enqueue LOOKUP lock for remote object */
2393         if (mdt_object_remote(o))
2394                 LASSERT(ibits == MDS_INODELOCK_LOOKUP);
2395
2396         if (lh->mlh_type == MDT_PDO_LOCK) {
2397                 /* check for exists after object is locked */
2398                 if (mdt_object_exists(o) == 0) {
2399                         /* Non-existent object shouldn't have PDO lock */
2400                         RETURN(-ESTALE);
2401                 } else {
2402                         /* Non-dir object shouldn't have PDO lock */
2403                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2404                                 RETURN(-ENOTDIR);
2405                 }
2406         }
2407
2408         memset(policy, 0, sizeof(*policy));
2409         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2410
2411         dlmflags |= LDLM_FL_ATOMIC_CB;
2412         if (nonblock)
2413                 dlmflags |= LDLM_FL_BLOCK_NOWAIT;
2414
2415         /*
2416          * Take PDO lock on whole directory and build correct @res_id for lock
2417          * on part of directory.
2418          */
2419         if (lh->mlh_pdo_hash != 0) {
2420                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2421                 mdt_lock_pdo_mode(info, o, lh);
2422                 if (lh->mlh_pdo_mode != LCK_NL) {
2423                         /*
2424                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2425                          * is never going to be sent to client and we do not
2426                          * want it slowed down due to possible cancels.
2427                          */
2428                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2429                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2430                                           policy, res_id, dlmflags,
2431                                           info->mti_exp == NULL ? NULL :
2432                                           &info->mti_exp->exp_handle.h_cookie);
2433                         if (unlikely(rc != 0))
2434                                 GOTO(out_unlock, rc);
2435                 }
2436
2437                 /*
2438                  * Finish res_id initializing by name hash marking part of
2439                  * directory which is taking modification.
2440                  */
2441                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2442         }
2443
2444         policy->l_inodebits.bits = ibits;
2445
2446         /*
2447          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2448          * going to be sent to client. If it is - mdt_intent_policy() path will
2449          * fix it up and turn FL_LOCAL flag off.
2450          */
2451         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2452                           res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2453                           info->mti_exp == NULL ? NULL :
2454                           &info->mti_exp->exp_handle.h_cookie);
2455 out_unlock:
2456         if (rc != 0)
2457                 mdt_object_unlock(info, o, lh, 1);
2458         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2459                    lh->mlh_pdo_hash != 0 &&
2460                    (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
2461                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2462
2463         RETURN(rc);
2464 }
2465
2466 static int
2467 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
2468                          struct mdt_lock_handle *lh, __u64 ibits, bool nonblock,
2469                          bool cos_incompat)
2470 {
2471         struct mdt_lock_handle *local_lh = NULL;
2472         int rc;
2473         ENTRY;
2474
2475         if (!mdt_object_remote(o)) {
2476                 rc = mdt_object_local_lock(info, o, lh, ibits, nonblock,
2477                                            cos_incompat);
2478                 RETURN(rc);
2479         }
2480
2481         /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
2482         ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
2483                    MDS_INODELOCK_XATTR);
2484
2485         /* Only enqueue LOOKUP lock for remote object */
2486         if (ibits & MDS_INODELOCK_LOOKUP) {
2487                 rc = mdt_object_local_lock(info, o, lh, MDS_INODELOCK_LOOKUP,
2488                                            nonblock, cos_incompat);
2489                 if (rc != ELDLM_OK)
2490                         RETURN(rc);
2491
2492                 local_lh = lh;
2493         }
2494
2495         if (ibits & MDS_INODELOCK_UPDATE) {
2496                 /* Sigh, PDO needs to enqueue 2 locks right now, but
2497                  * enqueue RPC can only request 1 lock, to avoid extra
2498                  * RPC, so it will instead enqueue EX lock for remote
2499                  * object anyway XXX*/
2500                 if (lh->mlh_type == MDT_PDO_LOCK &&
2501                     lh->mlh_pdo_hash != 0) {
2502                         CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
2503                                "EX lock.\n", mdt_obd_name(info->mti_mdt),
2504                                PFID(mdt_object_fid(o)));
2505                         lh->mlh_pdo_hash = 0;
2506                         lh->mlh_rreg_mode = LCK_EX;
2507                         lh->mlh_type = MDT_REG_LOCK;
2508                 }
2509                 rc = mdt_remote_object_lock(info, o, mdt_object_fid(o),
2510                                             &lh->mlh_rreg_lh,
2511                                             lh->mlh_rreg_mode,
2512                                             MDS_INODELOCK_UPDATE, nonblock);
2513                 if (rc != ELDLM_OK) {
2514                         if (local_lh != NULL)
2515                                 mdt_object_unlock(info, o, local_lh, rc);
2516                         RETURN(rc);
2517                 }
2518         }
2519
2520         RETURN(0);
2521 }
2522
2523 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2524                     struct mdt_lock_handle *lh, __u64 ibits)
2525 {
2526         return mdt_object_lock_internal(info, o, lh, ibits, false, false);
2527 }
2528
2529 int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2530                           struct mdt_lock_handle *lh, __u64 ibits,
2531                           bool cos_incompat)
2532 {
2533         LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
2534         return mdt_object_lock_internal(info, o, lh, ibits, false,
2535                                         cos_incompat);
2536 }
2537
2538 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
2539                         struct mdt_lock_handle *lh, __u64 ibits)
2540 {
2541         struct mdt_lock_handle tmp = *lh;
2542         int rc;
2543
2544         rc = mdt_object_lock_internal(info, o, &tmp, ibits, true, false);
2545         if (rc == 0)
2546                 *lh = tmp;
2547
2548         return rc == 0;
2549 }
2550
2551 int mdt_reint_object_lock_try(struct mdt_thread_info *info,
2552                               struct mdt_object *o, struct mdt_lock_handle *lh,
2553                               __u64 ibits, bool cos_incompat)
2554 {
2555         struct mdt_lock_handle tmp = *lh;
2556         int rc;
2557
2558         LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
2559         rc = mdt_object_lock_internal(info, o, &tmp, ibits, true, cos_incompat);
2560         if (rc == 0)
2561                 *lh = tmp;
2562
2563         return rc == 0;
2564 }
2565
2566 /**
2567  * Save a lock within request object.
2568  *
2569  * Keep the lock referenced until whether client ACK or transaction
2570  * commit happens or release the lock immediately depending on input
2571  * parameters. If COS is ON, a write lock is converted to COS lock
2572  * before saving.
2573  *
2574  * \param info thead info object
2575  * \param h lock handle
2576  * \param mode lock mode
2577  * \param decref force immediate lock releasing
2578  */
2579 static void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2580                           enum ldlm_mode mode, int decref)
2581 {
2582         ENTRY;
2583
2584         if (lustre_handle_is_used(h)) {
2585                 if (decref || !info->mti_has_trans ||
2586                     !(mode & (LCK_PW | LCK_EX))) {
2587                         mdt_fid_unlock(h, mode);
2588                 } else {
2589                         struct mdt_device *mdt = info->mti_mdt;
2590                         struct ldlm_lock *lock = ldlm_handle2lock(h);
2591                         struct ptlrpc_request *req = mdt_info_req(info);
2592                         int cos;
2593
2594                         cos = (mdt_cos_is_enabled(mdt) ||
2595                                mdt_slc_is_enabled(mdt));
2596
2597                         LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n",
2598                                  h->cookie);
2599
2600                         /* there is no request if mdt_object_unlock() is called
2601                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
2602                         if (likely(req != NULL)) {
2603                                 CDEBUG(D_HA, "request = %p reply state = %p"
2604                                        " transno = "LPD64"\n", req,
2605                                        req->rq_reply_state, req->rq_transno);
2606                                 if (cos) {
2607                                         ldlm_lock_downgrade(lock, LCK_COS);
2608                                         mode = LCK_COS;
2609                                 }
2610                                 ptlrpc_save_lock(req, h, mode, cos);
2611                         } else {
2612                                 mdt_fid_unlock(h, mode);
2613                         }
2614                         if (mdt_is_lock_sync(lock)) {
2615                                 CDEBUG(D_HA, "found sync-lock,"
2616                                        " async commit started\n");
2617                                 mdt_device_commit_async(info->mti_env,
2618                                                         mdt);
2619                         }
2620                         LDLM_LOCK_PUT(lock);
2621                 }
2622                 h->cookie = 0ull;
2623         }
2624
2625         EXIT;
2626 }
2627
2628 /**
2629  * Save cross-MDT lock in uncommitted_slc_locks
2630  *
2631  * Keep the lock referenced until transaction commit happens or release the lock
2632  * immediately depending on input parameters.
2633  *
2634  * \param info thead info object
2635  * \param h lock handle
2636  * \param mode lock mode
2637  * \param decref force immediate lock releasing
2638  */
2639 static void mdt_save_remote_lock(struct mdt_thread_info *info,
2640                                  struct lustre_handle *h, enum ldlm_mode mode,
2641                                  int decref)
2642 {
2643         ENTRY;
2644
2645         if (lustre_handle_is_used(h)) {
2646                 if (decref || !info->mti_has_trans ||
2647                     !(mode & (LCK_PW | LCK_EX))) {
2648                         ldlm_lock_decref_and_cancel(h, mode);
2649                 } else {
2650                         struct ldlm_lock *lock = ldlm_handle2lock(h);
2651                         struct ptlrpc_request *req = mdt_info_req(info);
2652
2653                         LASSERT(req != NULL);
2654                         tgt_save_slc_lock(lock, req->rq_transno);
2655                         ldlm_lock_decref(h, mode);
2656                 }
2657                 h->cookie = 0ull;
2658         }
2659
2660         EXIT;
2661 }
2662
2663 /**
2664  * Unlock mdt object.
2665  *
2666  * Immeditely release the regular lock and the PDO lock or save the
2667  * lock in request and keep them referenced until client ACK or
2668  * transaction commit.
2669  *
2670  * \param info thread info object
2671  * \param o mdt object
2672  * \param lh mdt lock handle referencing regular and PDO locks
2673  * \param decref force immediate lock releasing
2674  */
2675 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2676                        struct mdt_lock_handle *lh, int decref)
2677 {
2678         ENTRY;
2679
2680         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2681         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2682         mdt_save_remote_lock(info, &lh->mlh_rreg_lh, lh->mlh_rreg_mode, decref);
2683
2684         EXIT;
2685 }
2686
2687 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2688                                         const struct lu_fid *f,
2689                                         struct mdt_lock_handle *lh,
2690                                         __u64 ibits)
2691 {
2692         struct mdt_object *o;
2693
2694         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2695         if (!IS_ERR(o)) {
2696                 int rc;
2697
2698                 rc = mdt_object_lock(info, o, lh, ibits);
2699                 if (rc != 0) {
2700                         mdt_object_put(info->mti_env, o);
2701                         o = ERR_PTR(rc);
2702                 }
2703         }
2704         return o;
2705 }
2706
2707 void mdt_object_unlock_put(struct mdt_thread_info * info,
2708                            struct mdt_object * o,
2709                            struct mdt_lock_handle *lh,
2710                            int decref)
2711 {
2712         mdt_object_unlock(info, o, lh, decref);
2713         mdt_object_put(info->mti_env, o);
2714 }
2715
2716 /*
2717  * Generic code handling requests that have struct mdt_body passed in:
2718  *
2719  *  - extract mdt_body from request and save it in @info, if present;
2720  *
2721  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
2722  *  @info;
2723  *
2724  *  - if HABEO_CORPUS flag is set for this request type check whether object
2725  *  actually exists on storage (lu_object_exists()).
2726  *
2727  */
2728 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2729 {
2730         const struct mdt_body    *body;
2731         struct mdt_object        *obj;
2732         const struct lu_env      *env;
2733         struct req_capsule       *pill;
2734         int                       rc;
2735         ENTRY;
2736
2737         env = info->mti_env;
2738         pill = info->mti_pill;
2739
2740         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2741         if (body == NULL)
2742                 RETURN(-EFAULT);
2743
2744         if (!(body->mbo_valid & OBD_MD_FLID))
2745                 RETURN(0);
2746
2747         if (!fid_is_sane(&body->mbo_fid1)) {
2748                 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
2749                 RETURN(-EINVAL);
2750         }
2751
2752         obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
2753         if (!IS_ERR(obj)) {
2754                 if ((flags & HABEO_CORPUS) && !mdt_object_exists(obj)) {
2755                         mdt_object_put(env, obj);
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 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
2793 {
2794         lh->mlh_type = MDT_NUL_LOCK;
2795         lh->mlh_reg_lh.cookie = 0ull;
2796         lh->mlh_reg_mode = LCK_MINMODE;
2797         lh->mlh_pdo_lh.cookie = 0ull;
2798         lh->mlh_pdo_mode = LCK_MINMODE;
2799         lh->mlh_rreg_lh.cookie = 0ull;
2800         lh->mlh_rreg_mode = LCK_MINMODE;
2801 }
2802
2803 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
2804 {
2805         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2806         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2807 }
2808
2809 /*
2810  * Initialize fields of struct mdt_thread_info. Other fields are left in
2811  * uninitialized state, because it's too expensive to zero out whole
2812  * mdt_thread_info (> 1K) on each request arrival.
2813  */
2814 void mdt_thread_info_init(struct ptlrpc_request *req,
2815                           struct mdt_thread_info *info)
2816 {
2817         int i;
2818
2819         info->mti_pill = &req->rq_pill;
2820
2821         /* lock handle */
2822         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2823                 mdt_lock_handle_init(&info->mti_lh[i]);
2824
2825         /* mdt device: it can be NULL while CONNECT */
2826         if (req->rq_export) {
2827                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
2828                 info->mti_exp = req->rq_export;
2829         } else
2830                 info->mti_mdt = NULL;
2831         info->mti_env = req->rq_svc_thread->t_env;
2832         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
2833
2834         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
2835         info->mti_big_buf = LU_BUF_NULL;
2836         info->mti_body = NULL;
2837         info->mti_object = NULL;
2838         info->mti_dlm_req = NULL;
2839         info->mti_has_trans = 0;
2840         info->mti_cross_ref = 0;
2841         info->mti_opdata = 0;
2842         info->mti_big_lmm_used = 0;
2843
2844         info->mti_spec.no_create = 0;
2845         info->mti_spec.sp_rm_entry = 0;
2846         info->mti_spec.sp_permitted = 0;
2847         info->mti_spec.sp_migrate_close = 0;
2848
2849         info->mti_spec.u.sp_ea.eadata = NULL;
2850         info->mti_spec.u.sp_ea.eadatalen = 0;
2851 }
2852
2853 void mdt_thread_info_fini(struct mdt_thread_info *info)
2854 {
2855         int i;
2856
2857         if (info->mti_object != NULL) {
2858                 mdt_object_put(info->mti_env, info->mti_object);
2859                 info->mti_object = NULL;
2860         }
2861
2862         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
2863                 mdt_lock_handle_fini(&info->mti_lh[i]);
2864         info->mti_env = NULL;
2865         info->mti_pill = NULL;
2866         info->mti_exp = NULL;
2867
2868         if (unlikely(info->mti_big_buf.lb_buf != NULL))
2869                 lu_buf_free(&info->mti_big_buf);
2870 }
2871
2872 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
2873 {
2874         struct mdt_thread_info  *mti;
2875
2876         mti = mdt_th_info(tsi->tsi_env);
2877         LASSERT(mti != NULL);
2878
2879         mdt_thread_info_init(tgt_ses_req(tsi), mti);
2880         if (tsi->tsi_corpus != NULL) {
2881                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
2882                 lu_object_get(tsi->tsi_corpus);
2883         }
2884         mti->mti_body = tsi->tsi_mdt_body;
2885         mti->mti_dlm_req = tsi->tsi_dlm_req;
2886
2887         return mti;
2888 }
2889
2890 static int mdt_tgt_connect(struct tgt_session_info *tsi)
2891 {
2892         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2893         int                      rc;
2894
2895         ENTRY;
2896
2897         if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
2898             cfs_fail_val ==
2899             tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id) {
2900                 set_current_state(TASK_UNINTERRUPTIBLE);
2901                 schedule_timeout(msecs_to_jiffies(3 * MSEC_PER_SEC));
2902         }
2903
2904         rc = tgt_connect(tsi);
2905         if (rc != 0)
2906                 RETURN(rc);
2907
2908         rc = mdt_init_idmap(tsi);
2909         if (rc != 0)
2910                 GOTO(err, rc);
2911         RETURN(0);
2912 err:
2913         obd_disconnect(class_export_get(req->rq_export));
2914         return rc;
2915 }
2916
2917 enum mdt_it_code {
2918         MDT_IT_OPEN,
2919         MDT_IT_OCREAT,
2920         MDT_IT_CREATE,
2921         MDT_IT_GETATTR,
2922         MDT_IT_READDIR,
2923         MDT_IT_LOOKUP,
2924         MDT_IT_UNLINK,
2925         MDT_IT_TRUNC,
2926         MDT_IT_GETXATTR,
2927         MDT_IT_LAYOUT,
2928         MDT_IT_QUOTA,
2929         MDT_IT_NR
2930 };
2931
2932 static int mdt_intent_getattr(enum mdt_it_code opcode,
2933                               struct mdt_thread_info *info,
2934                               struct ldlm_lock **,
2935                               __u64);
2936
2937 static int mdt_intent_getxattr(enum mdt_it_code opcode,
2938                                 struct mdt_thread_info *info,
2939                                 struct ldlm_lock **lockp,
2940                                 __u64 flags);
2941
2942 static int mdt_intent_layout(enum mdt_it_code opcode,
2943                              struct mdt_thread_info *info,
2944                              struct ldlm_lock **,
2945                              __u64);
2946 static int mdt_intent_reint(enum mdt_it_code opcode,
2947                             struct mdt_thread_info *info,
2948                             struct ldlm_lock **,
2949                             __u64);
2950
2951 static struct mdt_it_flavor {
2952         const struct req_format *it_fmt;
2953         __u32                    it_flags;
2954         int                    (*it_act)(enum mdt_it_code ,
2955                                          struct mdt_thread_info *,
2956                                          struct ldlm_lock **,
2957                                          __u64);
2958         long                     it_reint;
2959 } mdt_it_flavor[] = {
2960         [MDT_IT_OPEN]     = {
2961                 .it_fmt   = &RQF_LDLM_INTENT,
2962                 /*.it_flags = HABEO_REFERO,*/
2963                 .it_flags = 0,
2964                 .it_act   = mdt_intent_reint,
2965                 .it_reint = REINT_OPEN
2966         },
2967         [MDT_IT_OCREAT]   = {
2968                 .it_fmt   = &RQF_LDLM_INTENT,
2969                 /*
2970                  * OCREAT is not a MUTABOR request as if the file
2971                  * already exists.
2972                  * We do the extra check of OBD_CONNECT_RDONLY in
2973                  * mdt_reint_open() when we really need to create
2974                  * the object.
2975                  */
2976                 .it_flags = 0,
2977                 .it_act   = mdt_intent_reint,
2978                 .it_reint = REINT_OPEN
2979         },
2980         [MDT_IT_CREATE]   = {
2981                 .it_fmt   = &RQF_LDLM_INTENT,
2982                 .it_flags = MUTABOR,
2983                 .it_act   = mdt_intent_reint,
2984                 .it_reint = REINT_CREATE
2985         },
2986         [MDT_IT_GETATTR]  = {
2987                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2988                 .it_flags = HABEO_REFERO,
2989                 .it_act   = mdt_intent_getattr
2990         },
2991         [MDT_IT_READDIR]  = {
2992                 .it_fmt   = NULL,
2993                 .it_flags = 0,
2994                 .it_act   = NULL
2995         },
2996         [MDT_IT_LOOKUP]   = {
2997                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
2998                 .it_flags = HABEO_REFERO,
2999                 .it_act   = mdt_intent_getattr
3000         },
3001         [MDT_IT_UNLINK]   = {
3002                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
3003                 .it_flags = MUTABOR,
3004                 .it_act   = NULL,
3005                 .it_reint = REINT_UNLINK
3006         },
3007         [MDT_IT_TRUNC]    = {
3008                 .it_fmt   = NULL,
3009                 .it_flags = MUTABOR,
3010                 .it_act   = NULL
3011         },
3012         [MDT_IT_GETXATTR] = {
3013                 .it_fmt   = &RQF_LDLM_INTENT_GETXATTR,
3014                 .it_flags = HABEO_CORPUS,
3015                 .it_act   = mdt_intent_getxattr
3016         },
3017         [MDT_IT_LAYOUT] = {
3018                 .it_fmt   = &RQF_LDLM_INTENT_LAYOUT,
3019                 .it_flags = 0,
3020                 .it_act   = mdt_intent_layout
3021         }
3022 };
3023
3024 static int
3025 mdt_intent_lock_replace(struct mdt_thread_info *info,
3026                         struct ldlm_lock **lockp,
3027                         struct mdt_lock_handle *lh,
3028                         __u64 flags)
3029 {
3030         struct ptlrpc_request  *req = mdt_info_req(info);
3031         struct ldlm_lock       *lock = *lockp;
3032         struct ldlm_lock       *new_lock;
3033
3034         /* If possible resent found a lock, @lh is set to its handle */
3035         new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3036
3037         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3038                 lh->mlh_reg_lh.cookie = 0;
3039                 RETURN(0);
3040         }
3041
3042         LASSERTF(new_lock != NULL,
3043                  "lockh "LPX64"\n", lh->mlh_reg_lh.cookie);
3044
3045         /*
3046          * If we've already given this lock to a client once, then we should
3047          * have no readers or writers.  Otherwise, we should have one reader
3048          * _or_ writer ref (which will be zeroed below) before returning the
3049          * lock to a client.
3050          */
3051         if (new_lock->l_export == req->rq_export) {
3052                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3053         } else {
3054                 LASSERT(new_lock->l_export == NULL);
3055                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3056         }
3057
3058         *lockp = new_lock;
3059
3060         if (new_lock->l_export == req->rq_export) {
3061                 /*
3062                  * Already gave this to the client, which means that we
3063                  * reconstructed a reply.
3064                  */
3065                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3066                         MSG_RESENT);
3067
3068                 LDLM_LOCK_RELEASE(new_lock);
3069                 lh->mlh_reg_lh.cookie = 0;
3070                 RETURN(ELDLM_LOCK_REPLACED);
3071         }
3072
3073         /*
3074          * Fixup the lock to be given to the client.
3075          */
3076         lock_res_and_lock(new_lock);
3077         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3078          * possible blocking AST. */
3079         while (new_lock->l_readers > 0) {
3080                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3081                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3082                 new_lock->l_readers--;
3083         }
3084         while (new_lock->l_writers > 0) {
3085                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3086                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3087                 new_lock->l_writers--;
3088         }
3089
3090         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3091         new_lock->l_blocking_ast = lock->l_blocking_ast;
3092         new_lock->l_completion_ast = lock->l_completion_ast;
3093         new_lock->l_remote_handle = lock->l_remote_handle;
3094         new_lock->l_flags &= ~LDLM_FL_LOCAL;
3095
3096         unlock_res_and_lock(new_lock);
3097
3098         cfs_hash_add(new_lock->l_export->exp_lock_hash,
3099                      &new_lock->l_remote_handle,
3100                      &new_lock->l_exp_hash);
3101
3102         LDLM_LOCK_RELEASE(new_lock);
3103         lh->mlh_reg_lh.cookie = 0;
3104
3105         RETURN(ELDLM_LOCK_REPLACED);
3106 }
3107
3108 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3109                                     struct ldlm_lock *new_lock,
3110                                     struct mdt_lock_handle *lh,
3111                                     __u64 flags)
3112 {
3113         struct ptlrpc_request  *req = mdt_info_req(info);
3114         struct ldlm_request    *dlmreq;
3115
3116         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3117                 return;
3118
3119         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3120
3121         /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
3122          * lock was found by ldlm_handle_enqueue(); if so @lh must be
3123          * initialized. */
3124         if (flags & LDLM_FL_RESENT) {
3125                 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
3126                 lh->mlh_reg_mode = new_lock->l_granted_mode;
3127
3128                 LDLM_DEBUG(new_lock, "Restoring lock cookie");
3129                 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie "LPX64,
3130                           lh->mlh_reg_lh.cookie);
3131                 return;
3132         }
3133
3134         /*
3135          * If the xid matches, then we know this is a resent request, and allow
3136          * it. (It's probably an OPEN, for which we don't send a lock.
3137          */
3138         if (req_can_reconstruct(req, NULL))
3139                 return;
3140
3141         /*
3142          * This remote handle isn't enqueued, so we never received or processed
3143          * this request.  Clear MSG_RESENT, because it can be handled like any
3144          * normal request now.
3145          */
3146         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3147
3148         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64,
3149                   dlmreq->lock_handle[0].cookie);
3150 }
3151
3152 static int mdt_intent_getxattr(enum mdt_it_code opcode,
3153                                 struct mdt_thread_info *info,
3154                                 struct ldlm_lock **lockp,
3155                                 __u64 flags)
3156 {
3157         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3158         struct ldlm_reply      *ldlm_rep = NULL;
3159         int rc, grc;
3160
3161         /*
3162          * Initialize lhc->mlh_reg_lh either from a previously granted lock
3163          * (for the resend case) or a new lock. Below we will use it to
3164          * replace the original lock.
3165          */
3166         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3167         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3168                 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3169                 rc = mdt_object_lock(info, info->mti_object, lhc,
3170                                      MDS_INODELOCK_XATTR);
3171                 if (rc)
3172                         return rc;
3173         }
3174
3175         grc = mdt_getxattr(info);
3176
3177         rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3178
3179         if (mdt_info_req(info)->rq_repmsg != NULL)
3180                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3181         if (ldlm_rep == NULL)
3182                 RETURN(err_serious(-EFAULT));
3183
3184         ldlm_rep->lock_policy_res2 = grc;
3185
3186         return rc;
3187 }
3188
3189 static int mdt_intent_getattr(enum mdt_it_code opcode,
3190                               struct mdt_thread_info *info,
3191                               struct ldlm_lock **lockp,
3192                               __u64 flags)
3193 {
3194         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3195         __u64                   child_bits;
3196         struct ldlm_reply      *ldlm_rep;
3197         struct mdt_body        *reqbody;
3198         struct mdt_body        *repbody;
3199         int                     rc, rc2;
3200         ENTRY;
3201
3202         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3203         LASSERT(reqbody);
3204
3205         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3206         LASSERT(repbody);
3207
3208         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
3209         repbody->mbo_eadatasize = 0;
3210         repbody->mbo_aclsize = 0;
3211
3212         switch (opcode) {
3213         case MDT_IT_LOOKUP:
3214                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3215                 break;
3216         case MDT_IT_GETATTR:
3217                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3218                              MDS_INODELOCK_PERM;
3219                 break;
3220         default:
3221                 CERROR("Unsupported intent (%d)\n", opcode);
3222                 GOTO(out_shrink, rc = -EINVAL);
3223         }
3224
3225         rc = mdt_init_ucred_intent_getattr(info, reqbody);
3226         if (rc)
3227                 GOTO(out_shrink, rc);
3228
3229         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3230         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3231
3232         /* Get lock from request for possible resent case. */
3233         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3234
3235         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3236         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3237
3238         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3239                 ldlm_rep->lock_policy_res2 = 0;
3240         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3241             ldlm_rep->lock_policy_res2) {
3242                 lhc->mlh_reg_lh.cookie = 0ull;
3243                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3244         }
3245
3246         rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3247         EXIT;
3248 out_ucred:
3249         mdt_exit_ucred(info);
3250 out_shrink:
3251         mdt_client_compatibility(info);
3252         rc2 = mdt_fix_reply(info);
3253         if (rc == 0)
3254                 rc = rc2;
3255         return rc;
3256 }
3257
3258 static int mdt_intent_layout(enum mdt_it_code opcode,
3259                              struct mdt_thread_info *info,
3260                              struct ldlm_lock **lockp,
3261                              __u64 flags)
3262 {
3263         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_LAYOUT];
3264         struct layout_intent *layout;
3265         struct lu_fid *fid;
3266         struct mdt_object *obj = NULL;
3267         int layout_size = 0;
3268         int rc = 0;
3269         ENTRY;
3270
3271         if (opcode != MDT_IT_LAYOUT) {
3272                 CERROR("%s: Unknown intent (%d)\n", mdt_obd_name(info->mti_mdt),
3273                         opcode);
3274                 RETURN(-EINVAL);
3275         }
3276
3277         layout = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3278         if (layout == NULL)
3279                 RETURN(-EPROTO);
3280
3281         if (layout->li_opc != LAYOUT_INTENT_ACCESS) {
3282                 CERROR("%s: Unsupported layout intent opc %d\n",
3283                        mdt_obd_name(info->mti_mdt), layout->li_opc);
3284                 RETURN(-EINVAL);
3285         }
3286
3287         fid = &info->mti_tmp_fid2;
3288         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3289
3290         /* Get lock from request for possible resent case. */
3291         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3292
3293         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3294         if (IS_ERR(obj))
3295                 GOTO(out, rc = PTR_ERR(obj));
3296
3297         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3298                 layout_size = mdt_attr_get_eabuf_size(info, obj);
3299                 if (layout_size < 0)
3300                         GOTO(out_obj, rc = layout_size);
3301
3302                 if (layout_size > info->mti_mdt->mdt_max_mdsize)
3303                         info->mti_mdt->mdt_max_mdsize = layout_size;
3304         }
3305
3306         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3307         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
3308                              layout_size);
3309         rc = req_capsule_server_pack(info->mti_pill);
3310         GOTO(out_obj, rc);
3311
3312 out_obj:
3313         mdt_object_put(info->mti_env, obj);
3314
3315         if (rc == 0 && lustre_handle_is_used(&lhc->mlh_reg_lh))
3316                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3317
3318 out:
3319         lhc->mlh_reg_lh.cookie = 0;
3320
3321         return rc;
3322 }
3323
3324 static int mdt_intent_reint(enum mdt_it_code opcode,
3325                             struct mdt_thread_info *info,
3326                             struct ldlm_lock **lockp,
3327                             __u64 flags)
3328 {
3329         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3330         struct ldlm_reply      *rep = NULL;
3331         long                    opc;
3332         int                     rc;
3333
3334         static const struct req_format *intent_fmts[REINT_MAX] = {
3335                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3336                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3337         };
3338
3339         ENTRY;
3340
3341         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
3342         if (opc < 0)
3343                 RETURN(opc);
3344
3345         if (mdt_it_flavor[opcode].it_reint != opc) {
3346                 CERROR("Reint code %ld doesn't match intent: %d\n",
3347                        opc, opcode);
3348                 RETURN(err_serious(-EPROTO));
3349         }
3350
3351         /* Get lock from request for possible resent case. */
3352         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3353
3354         rc = mdt_reint_internal(info, lhc, opc);
3355
3356         /* Check whether the reply has been packed successfully. */
3357         if (mdt_info_req(info)->rq_repmsg != NULL)
3358                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3359         if (rep == NULL)
3360                 RETURN(err_serious(-EFAULT));
3361
3362         /* MDC expects this in any case */
3363         if (rc != 0)
3364                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3365
3366         /* the open lock or the lock for cross-ref object should be
3367          * returned to the client */
3368         if (lustre_handle_is_used(&lhc->mlh_reg_lh) &&
3369             (rc == 0 || rc == -MDT_EREMOTE_OPEN)) {
3370                 rep->lock_policy_res2 = 0;
3371                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags);
3372                 RETURN(rc);
3373         }
3374
3375         rep->lock_policy_res2 = clear_serious(rc);
3376
3377         if (rep->lock_policy_res2 == -ENOENT &&
3378             mdt_get_disposition(rep, DISP_LOOKUP_NEG) &&
3379             !mdt_get_disposition(rep, DISP_OPEN_CREATE))
3380                 rep->lock_policy_res2 = 0;
3381
3382         lhc->mlh_reg_lh.cookie = 0ull;
3383         if (rc == -ENOTCONN || rc == -ENODEV ||
3384             rc == -EOVERFLOW) { /**< if VBR failure then return error */
3385                 /*
3386                  * If it is the disconnect error (ENODEV & ENOCONN), the error
3387                  * will be returned by rq_status, and client at ptlrpc layer
3388                  * will detect this, then disconnect, reconnect the import
3389                  * immediately, instead of impacting the following the rpc.
3390                  */
3391                 RETURN(rc);
3392         }
3393         /*
3394          * For other cases, the error will be returned by intent, and client
3395          * will retrieve the result from intent.
3396          */
3397         RETURN(ELDLM_LOCK_ABORTED);
3398 }
3399
3400 static int mdt_intent_code(enum ldlm_intent_flags itcode)
3401 {
3402         int rc;
3403
3404         switch (itcode) {
3405         case IT_OPEN:
3406                 rc = MDT_IT_OPEN;
3407                 break;
3408         case IT_OPEN|IT_CREAT:
3409                 rc = MDT_IT_OCREAT;
3410                 break;
3411         case IT_CREAT:
3412                 rc = MDT_IT_CREATE;
3413                 break;
3414         case IT_READDIR:
3415                 rc = MDT_IT_READDIR;
3416                 break;
3417         case IT_GETATTR:
3418                 rc = MDT_IT_GETATTR;
3419                 break;
3420         case IT_LOOKUP:
3421                 rc = MDT_IT_LOOKUP;
3422                 break;
3423         case IT_UNLINK:
3424                 rc = MDT_IT_UNLINK;
3425                 break;
3426         case IT_TRUNC:
3427                 rc = MDT_IT_TRUNC;
3428                 break;
3429         case IT_GETXATTR:
3430                 rc = MDT_IT_GETXATTR;
3431                 break;
3432         case IT_LAYOUT:
3433                 rc = MDT_IT_LAYOUT;
3434                 break;
3435         case IT_QUOTA_DQACQ:
3436         case IT_QUOTA_CONN:
3437                 rc = MDT_IT_QUOTA;
3438                 break;
3439         default:
3440                 CERROR("Unknown intent opcode: 0x%08x\n", itcode);
3441                 rc = -EINVAL;
3442                 break;
3443         }
3444         return rc;
3445 }
3446
3447 static int mdt_intent_opc(enum ldlm_intent_flags itopc,
3448                           struct mdt_thread_info *info,
3449                           struct ldlm_lock **lockp, __u64 flags)
3450 {
3451         struct req_capsule      *pill = info->mti_pill;
3452         struct ptlrpc_request   *req = mdt_info_req(info);
3453         struct mdt_it_flavor    *flv;
3454         int opc;
3455         int rc;
3456         ENTRY;
3457
3458         opc = mdt_intent_code(itopc);
3459         if (opc < 0)
3460                 RETURN(-EINVAL);
3461
3462         if (opc == MDT_IT_QUOTA) {
3463                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
3464
3465                 if (qmt == NULL)
3466                         RETURN(-EOPNOTSUPP);
3467
3468                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
3469                 /* pass the request to quota master */
3470                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
3471                                                  mdt_info_req(info), lockp,
3472                                                  flags);
3473                 RETURN(rc);
3474         }
3475
3476         flv = &mdt_it_flavor[opc];
3477         if (flv->it_fmt != NULL)
3478                 req_capsule_extend(pill, flv->it_fmt);
3479
3480         rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3481         if (rc < 0)
3482                 RETURN(rc);
3483
3484         if (flv->it_flags & MUTABOR &&
3485             exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
3486                 RETURN(-EROFS);
3487
3488         if (flv->it_act != NULL) {
3489                 struct ldlm_reply *rep;
3490
3491                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_INTENT_DELAY, 10);
3492
3493                 /* execute policy */
3494                 rc = flv->it_act(opc, info, lockp, flags);
3495
3496                 /* Check whether the reply has been packed successfully. */
3497                 if (req->rq_repmsg != NULL) {
3498                         rep = req_capsule_server_get(info->mti_pill,
3499                                                      &RMF_DLM_REP);
3500                         rep->lock_policy_res2 =
3501                                 ptlrpc_status_hton(rep->lock_policy_res2);
3502                 }
3503         }
3504
3505         RETURN(rc);
3506 }
3507
3508 static int mdt_intent_policy(struct ldlm_namespace *ns,
3509                              struct ldlm_lock **lockp, void *req_cookie,
3510                              enum ldlm_mode mode, __u64 flags, void *data)
3511 {
3512         struct tgt_session_info *tsi;
3513         struct mdt_thread_info  *info;
3514         struct ptlrpc_request   *req  =  req_cookie;
3515         struct ldlm_intent      *it;
3516         struct req_capsule      *pill;
3517         int rc;
3518
3519         ENTRY;
3520
3521         LASSERT(req != NULL);
3522
3523         tsi = tgt_ses_info(req->rq_svc_thread->t_env);
3524
3525         info = tsi2mdt_info(tsi);
3526         LASSERT(info != NULL);
3527         pill = info->mti_pill;
3528         LASSERT(pill->rc_req == req);
3529
3530         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3531                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
3532                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3533                 if (it != NULL) {
3534                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
3535                         if (rc == 0)
3536                                 rc = ELDLM_OK;
3537
3538                         /* Lock without inodebits makes no sense and will oops
3539                          * later in ldlm. Let's check it now to see if we have
3540                          * ibits corrupted somewhere in mdt_intent_opc().
3541                          * The case for client miss to set ibits has been
3542                          * processed by others. */
3543                         LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
3544                                         lr_type == LDLM_IBITS,
3545                                      info->mti_dlm_req->lock_desc.\
3546                                         l_policy_data.l_inodebits.bits != 0));
3547                 } else
3548                         rc = err_serious(-EFAULT);
3549         } else {
3550                 /* No intent was provided */
3551                 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3552                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
3553                 rc = req_capsule_server_pack(pill);
3554                 if (rc)
3555                         rc = err_serious(rc);
3556         }
3557         mdt_thread_info_fini(info);
3558         RETURN(rc);
3559 }
3560
3561 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
3562 {
3563         struct seq_server_site  *ss = mdt_seq_site(mdt);
3564
3565         if (ss->ss_node_id == 0)
3566                 return;
3567
3568         if (ss->ss_client_seq != NULL) {
3569                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3570                 ss->ss_client_seq->lcs_exp = NULL;
3571         }
3572
3573         if (ss->ss_server_fld != NULL) {
3574                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
3575                 ss->ss_server_fld->lsf_control_exp = NULL;
3576         }
3577 }
3578
3579 static void mdt_seq_fini_cli(struct mdt_device *mdt)
3580 {
3581         struct seq_server_site *ss = mdt_seq_site(mdt);
3582
3583         if (ss == NULL)
3584                 return;
3585
3586         if (ss->ss_server_seq != NULL)
3587                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
3588 }
3589
3590 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
3591 {
3592         mdt_seq_fini_cli(mdt);
3593         mdt_deregister_seq_exp(mdt);
3594
3595         return seq_site_fini(env, mdt_seq_site(mdt));
3596 }
3597
3598 /**
3599  * It will retrieve its FLDB entries from MDT0, and it only happens
3600  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
3601  * and it needs to refresh FLDB from the MDT0.
3602  **/
3603 static int mdt_register_lwp_callback(void *data)
3604 {
3605         struct lu_env           env;
3606         struct mdt_device       *mdt = data;
3607         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
3608         int                     rc;
3609         ENTRY;
3610
3611         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
3612
3613         rc = lu_env_init(&env, LCT_MD_THREAD);
3614         if (rc < 0) {
3615                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
3616                 RETURN(rc);
3617         }
3618
3619         /* Allocate new sequence now to avoid creating local transaction
3620          * in the normal transaction process */
3621         rc = seq_server_check_and_alloc_super(&env,
3622                                               mdt_seq_site(mdt)->ss_server_seq);
3623         if (rc < 0)
3624                 GOTO(out, rc);
3625
3626         if (fld->lsf_new) {
3627                 rc = fld_update_from_controller(&env, fld);
3628                 if (rc != 0) {
3629                         CERROR("%s: cannot update controller: rc = %d\n",
3630                                mdt_obd_name(mdt), rc);
3631                         GOTO(out, rc);
3632                 }
3633         }
3634 out:
3635         lu_env_fini(&env);
3636         RETURN(rc);
3637 }
3638
3639 static int mdt_register_seq_exp(struct mdt_device *mdt)
3640 {
3641         struct seq_server_site  *ss = mdt_seq_site(mdt);
3642         char                    *lwp_name = NULL;
3643         int                     rc;
3644
3645         if (ss->ss_node_id == 0)
3646                 return 0;
3647
3648         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
3649         if (lwp_name == NULL)
3650                 GOTO(out_free, rc = -ENOMEM);
3651
3652         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
3653         if (rc != 0)
3654                 GOTO(out_free, rc);
3655
3656         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
3657                                       NULL, NULL);
3658         if (rc != 0)
3659                 GOTO(out_free, rc);
3660
3661         rc = lustre_register_lwp_item(lwp_name,
3662                                       &ss->ss_server_fld->lsf_control_exp,
3663                                       mdt_register_lwp_callback, mdt);
3664         if (rc != 0) {
3665                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3666                 ss->ss_client_seq->lcs_exp = NULL;
3667                 GOTO(out_free, rc);
3668         }
3669 out_free:
3670         if (lwp_name != NULL)
3671                 OBD_FREE(lwp_name, MAX_OBD_NAME);
3672
3673         return rc;
3674 }
3675
3676 /*
3677  * Init client sequence manager which is used by local MDS to talk to sequence
3678  * controller on remote node.
3679  */
3680 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
3681 {
3682         struct seq_server_site  *ss = mdt_seq_site(mdt);
3683         int                     rc;
3684         char                    *prefix;
3685         ENTRY;
3686
3687         /* check if this is adding the first MDC and controller is not yet
3688          * initialized. */
3689         OBD_ALLOC_PTR(ss->ss_client_seq);
3690         if (ss->ss_client_seq == NULL)
3691                 RETURN(-ENOMEM);
3692
3693         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3694         if (prefix == NULL) {
3695                 OBD_FREE_PTR(ss->ss_client_seq);
3696                 ss->ss_client_seq = NULL;
3697                 RETURN(-ENOMEM);
3698         }
3699
3700         /* Note: seq_client_fini will be called in seq_site_fini */
3701         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
3702         rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
3703                              prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
3704                                                             NULL);
3705         OBD_FREE(prefix, MAX_OBD_NAME + 5);
3706         if (rc != 0) {
3707                 OBD_FREE_PTR(ss->ss_client_seq);
3708                 ss->ss_client_seq = NULL;
3709                 RETURN(rc);
3710         }
3711
3712         rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
3713
3714         RETURN(rc);
3715 }
3716
3717 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
3718 {
3719         struct seq_server_site  *ss;
3720         int                     rc;
3721         ENTRY;
3722
3723         ss = mdt_seq_site(mdt);
3724         /* init sequence controller server(MDT0) */
3725         if (ss->ss_node_id == 0) {
3726                 OBD_ALLOC_PTR(ss->ss_control_seq);
3727                 if (ss->ss_control_seq == NULL)
3728                         RETURN(-ENOMEM);
3729
3730                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
3731                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
3732                                      ss);
3733                 if (rc)
3734                         GOTO(out_seq_fini, rc);
3735         }
3736
3737         /* Init normal sequence server */
3738         OBD_ALLOC_PTR(ss->ss_server_seq);
3739         if (ss->ss_server_seq == NULL)
3740                 GOTO(out_seq_fini, rc = -ENOMEM);
3741
3742         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
3743                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
3744         if (rc)
3745                 GOTO(out_seq_fini, rc);
3746
3747         /* init seq client for seq server to talk to seq controller(MDT0) */
3748         rc = mdt_seq_init_cli(env, mdt);
3749         if (rc != 0)
3750                 GOTO(out_seq_fini, rc);
3751
3752         if (ss->ss_node_id != 0)
3753                 /* register controller export through lwp */
3754                 rc = mdt_register_seq_exp(mdt);
3755
3756         EXIT;
3757 out_seq_fini:
3758         if (rc)
3759                 mdt_seq_fini(env, mdt);
3760
3761         return rc;
3762 }
3763
3764 /*
3765  * FLD wrappers
3766  */
3767 static int mdt_fld_fini(const struct lu_env *env,
3768                         struct mdt_device *m)
3769 {
3770         struct seq_server_site *ss = mdt_seq_site(m);
3771         ENTRY;
3772
3773         if (ss && ss->ss_server_fld) {
3774                 fld_server_fini(env, ss->ss_server_fld);
3775                 OBD_FREE_PTR(ss->ss_server_fld);
3776                 ss->ss_server_fld = NULL;
3777         }
3778
3779         RETURN(0);
3780 }
3781
3782 static int mdt_fld_init(const struct lu_env *env,
3783                         const char *uuid,
3784                         struct mdt_device *m)
3785 {
3786         struct seq_server_site *ss;
3787         int rc;
3788         ENTRY;
3789
3790         ss = mdt_seq_site(m);
3791
3792         OBD_ALLOC_PTR(ss->ss_server_fld);
3793         if (ss->ss_server_fld == NULL)
3794                 RETURN(rc = -ENOMEM);
3795
3796         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
3797                              LU_SEQ_RANGE_MDT);
3798         if (rc) {
3799                 OBD_FREE_PTR(ss->ss_server_fld);
3800                 ss->ss_server_fld = NULL;
3801                 RETURN(rc);
3802         }
3803
3804         RETURN(0);
3805 }
3806
3807 static void mdt_stack_pre_fini(const struct lu_env *env,
3808                            struct mdt_device *m, struct lu_device *top)
3809 {
3810         struct lustre_cfg_bufs  *bufs;
3811         struct lustre_cfg       *lcfg;
3812         struct mdt_thread_info  *info;
3813         ENTRY;
3814
3815         LASSERT(top);
3816
3817         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3818         LASSERT(info != NULL);
3819
3820         bufs = &info->mti_u.bufs;
3821
3822         LASSERT(m->mdt_child_exp);
3823         LASSERT(m->mdt_child_exp->exp_obd);
3824
3825         /* process cleanup, pass mdt obd name to get obd umount flags */
3826         /* XXX: this is needed because all layers are referenced by
3827          * objects (some of them are pinned by osd, for example *
3828          * the proper solution should be a model where object used
3829          * by osd only doesn't have mdt/mdd slices -bzzz */
3830         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
3831         lustre_cfg_bufs_set_string(bufs, 1, NULL);
3832         lcfg = lustre_cfg_new(LCFG_PRE_CLEANUP, bufs);
3833         if (lcfg == NULL)
3834                 RETURN_EXIT;
3835
3836         top->ld_ops->ldo_process_config(env, top, lcfg);
3837         lustre_cfg_free(lcfg);
3838         EXIT;
3839 }
3840
3841 static void mdt_stack_fini(const struct lu_env *env,
3842                            struct mdt_device *m, struct lu_device *top)
3843 {
3844         struct obd_device       *obd = mdt2obd_dev(m);
3845         struct lustre_cfg_bufs  *bufs;
3846         struct lustre_cfg       *lcfg;
3847         struct mdt_thread_info  *info;
3848         char                     flags[3] = "";
3849         ENTRY;
3850
3851         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
3852         LASSERT(info != NULL);
3853
3854         lu_dev_del_linkage(top->ld_site, top);
3855
3856         lu_site_purge(env, top->ld_site, -1);
3857
3858         bufs = &info->mti_u.bufs;
3859         /* process cleanup, pass mdt obd name to get obd umount flags */
3860         /* another purpose is to let all layers to release their objects */
3861         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
3862         if (obd->obd_force)
3863                 strcat(flags, "F");
3864         if (obd->obd_fail)
3865                 strcat(flags, "A");
3866         lustre_cfg_bufs_set_string(bufs, 1, flags);
3867         lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
3868         if (lcfg == NULL)
3869                 RETURN_EXIT;
3870
3871         LASSERT(top);
3872         top->ld_ops->ldo_process_config(env, top, lcfg);
3873         lustre_cfg_free(lcfg);
3874
3875         lu_site_purge(env, top->ld_site, -1);
3876
3877         m->mdt_child = NULL;
3878         m->mdt_bottom = NULL;
3879
3880         obd_disconnect(m->mdt_child_exp);
3881         m->mdt_child_exp = NULL;
3882
3883         obd_disconnect(m->mdt_bottom_exp);
3884         m->mdt_child_exp = NULL;
3885 }
3886
3887 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
3888                                const char *next, struct obd_export **exp)
3889 {
3890         struct obd_connect_data *data = NULL;
3891         struct obd_device       *obd;
3892         int                      rc;
3893         ENTRY;
3894
3895         OBD_ALLOC_PTR(data);
3896         if (data == NULL)
3897                 GOTO(out, rc = -ENOMEM);
3898
3899         obd = class_name2obd(next);
3900         if (obd == NULL) {
3901                 CERROR("%s: can't locate next device: %s\n",
3902                        mdt_obd_name(m), next);
3903                 GOTO(out, rc = -ENOTCONN);
3904         }
3905
3906         data->ocd_connect_flags = OBD_CONNECT_VERSION;
3907         data->ocd_version = LUSTRE_VERSION_CODE;
3908
3909         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
3910         if (rc) {
3911                 CERROR("%s: cannot connect to next dev %s (%d)\n",
3912                        mdt_obd_name(m), next, rc);
3913                 GOTO(out, rc);
3914         }
3915
3916 out:
3917         if (data)
3918                 OBD_FREE_PTR(data);
3919         RETURN(rc);
3920 }
3921
3922 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
3923                           struct lustre_cfg *cfg)
3924 {
3925         char                   *dev = lustre_cfg_string(cfg, 0);
3926         int                     rc, name_size, uuid_size;
3927         char                   *name, *uuid, *p;
3928         struct lustre_cfg_bufs *bufs;
3929         struct lustre_cfg      *lcfg;
3930         struct obd_device      *obd;
3931         struct lustre_profile  *lprof;
3932         struct lu_site         *site;
3933         ENTRY;
3934
3935         /* in 1.8 we had the only device in the stack - MDS.
3936          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
3937          * in 2.3 OSD is instantiated by obd_mount.c, so we need
3938          * to generate names and setup MDT, MDD. MDT will be using
3939          * generated name to connect to MDD. for MDD the next device
3940          * will be LOD with name taken from so called "profile" which
3941          * is generated by mount_option line
3942          *
3943          * 1.8 MGS generates config. commands like this:
3944          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
3945          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
3946          * 2.0 MGS generates config. commands like this:
3947          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
3948          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
3949          *                    3:lustre-MDT0000-mdtlov  4:f
3950          *
3951          * we generate MDD name from MDT one, just replacing T with D
3952          *
3953          * after all the preparations, the logical equivalent will be
3954          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
3955          *                    3:lustre-MDT0000-mdtlov  4:f
3956          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
3957          *                    3:lustre-MDD0000  4:f
3958          *
3959          *  notice we build the stack from down to top: MDD first, then MDT */
3960
3961         name_size = MAX_OBD_NAME;
3962         uuid_size = MAX_OBD_NAME;
3963
3964         OBD_ALLOC(name, name_size);
3965         OBD_ALLOC(uuid, uuid_size);
3966         if (name == NULL || uuid == NULL)
3967                 GOTO(cleanup_mem, rc = -ENOMEM);
3968
3969         OBD_ALLOC_PTR(bufs);
3970         if (!bufs)
3971                 GOTO(cleanup_mem, rc = -ENOMEM);
3972
3973         strcpy(name, dev);
3974         p = strstr(name, "-MDT");
3975         if (p == NULL)
3976                 GOTO(free_bufs, rc = -ENOMEM);
3977         p[3] = 'D';
3978
3979         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
3980
3981         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
3982         if (lprof == NULL || lprof->lp_dt == NULL) {
3983                 CERROR("can't find the profile: %s\n",
3984                        lustre_cfg_string(cfg, 0));
3985                 GOTO(free_bufs, rc = -EINVAL);
3986         }
3987
3988         lustre_cfg_bufs_reset(bufs, name);
3989         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
3990         lustre_cfg_bufs_set_string(bufs, 2, uuid);
3991         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
3992
3993         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
3994         if (lcfg == NULL)
3995                 GOTO(put_profile, rc = -ENOMEM);
3996
3997         rc = class_attach(lcfg);
3998         if (rc)
3999                 GOTO(lcfg_cleanup, rc);
4000
4001         obd = class_name2obd(name);
4002         if (!obd) {
4003                 CERROR("Can not find obd %s (%s in config)\n",
4004                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
4005                 GOTO(lcfg_cleanup, rc = -EINVAL);
4006         }
4007
4008         lustre_cfg_free(lcfg);
4009
4010         lustre_cfg_bufs_reset(bufs, name);
4011         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4012         lustre_cfg_bufs_set_string(bufs, 2, dev);
4013         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4014
4015         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4016         if (lcfg == NULL)
4017                 GOTO(class_detach, rc = -ENOMEM);
4018
4019         rc = class_setup(obd, lcfg);
4020         if (rc)
4021                 GOTO(class_detach, rc);
4022
4023         /* connect to MDD we just setup */
4024         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
4025         if (rc)
4026                 GOTO(class_detach, rc);
4027
4028         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
4029         LASSERT(site);
4030         LASSERT(mdt_lu_site(mdt) == NULL);
4031         mdt->mdt_lu_dev.ld_site = site;
4032         site->ls_top_dev = &mdt->mdt_lu_dev;
4033         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
4034
4035         /* now connect to bottom OSD */
4036         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
4037         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
4038         if (rc)
4039                 GOTO(class_detach, rc);
4040         mdt->mdt_bottom =
4041                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
4042
4043         rc = lu_env_refill((struct lu_env *)env);
4044         if (rc != 0)
4045                 CERROR("Failure to refill session: '%d'\n", rc);
4046
4047         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
4048
4049         EXIT;
4050 class_detach:
4051         if (rc)
4052                 class_detach(obd, lcfg);
4053 lcfg_cleanup:
4054         lustre_cfg_free(lcfg);
4055 put_profile:
4056         class_put_profile(lprof);
4057 free_bufs:
4058         OBD_FREE_PTR(bufs);
4059 cleanup_mem:
4060         if (name)
4061                 OBD_FREE(name, name_size);
4062         if (uuid)
4063                 OBD_FREE(uuid, uuid_size);
4064         RETURN(rc);
4065 }
4066
4067 /* setup quota master target on MDT0 */
4068 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
4069                           struct lustre_cfg *cfg)
4070 {
4071         struct obd_device       *obd;
4072         char                    *dev = lustre_cfg_string(cfg, 0);
4073         char                    *qmtname, *uuid, *p;
4074         struct lustre_cfg_bufs  *bufs;
4075         struct lustre_cfg       *lcfg;
4076         struct lustre_profile   *lprof;
4077         struct obd_connect_data *data;
4078         int                      rc;
4079         ENTRY;
4080
4081         LASSERT(mdt->mdt_qmt_exp == NULL);
4082         LASSERT(mdt->mdt_qmt_dev == NULL);
4083
4084         /* quota master is on MDT0 only for now */
4085         if (mdt->mdt_seq_site.ss_node_id != 0)
4086                 RETURN(0);
4087
4088         /* MGS generates config commands which look as follows:
4089          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4090          *                    3:lustre-MDT0000-mdtlov  4:f
4091          *
4092          * We generate the QMT name from the MDT one, just replacing MD with QM
4093          * after all the preparations, the logical equivalent will be:
4094          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
4095          *                    3:lustre-MDT0000-osd  4:f */
4096         OBD_ALLOC(qmtname, MAX_OBD_NAME);
4097         OBD_ALLOC(uuid, UUID_MAX);
4098         OBD_ALLOC_PTR(bufs);
4099         OBD_ALLOC_PTR(data);
4100         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
4101                 GOTO(cleanup_mem, rc = -ENOMEM);
4102
4103         strcpy(qmtname, dev);
4104         p = strstr(qmtname, "-MDT");
4105         if (p == NULL)
4106                 GOTO(cleanup_mem, rc = -ENOMEM);
4107         /* replace MD with QM */
4108         p[1] = 'Q';
4109         p[2] = 'M';
4110
4111         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
4112
4113         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4114         if (lprof == NULL || lprof->lp_dt == NULL) {
4115                 CERROR("can't find profile for %s\n",
4116                        lustre_cfg_string(cfg, 0));
4117                 GOTO(cleanup_mem, rc = -EINVAL);
4118         }
4119
4120         lustre_cfg_bufs_reset(bufs, qmtname);
4121         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
4122         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4123         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4124
4125         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4126         if (lcfg == NULL)
4127                 GOTO(put_profile, rc = -ENOMEM);
4128
4129         rc = class_attach(lcfg);
4130         if (rc)
4131                 GOTO(lcfg_cleanup, rc);
4132
4133         obd = class_name2obd(qmtname);
4134         if (!obd) {
4135                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
4136                        lustre_cfg_string(cfg, 0));
4137                 GOTO(lcfg_cleanup, rc = -EINVAL);
4138         }
4139
4140         lustre_cfg_free(lcfg);
4141
4142         lustre_cfg_bufs_reset(bufs, qmtname);
4143         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4144         lustre_cfg_bufs_set_string(bufs, 2, dev);
4145
4146         /* for quota, the next device should be the OSD device */
4147         lustre_cfg_bufs_set_string(bufs, 3,
4148                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
4149
4150         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4151         if (lcfg == NULL)
4152                 GOTO(class_detach, rc = -ENOMEM);
4153
4154         rc = class_setup(obd, lcfg);
4155         if (rc)
4156                 GOTO(class_detach, rc);
4157
4158         mdt->mdt_qmt_dev = obd->obd_lu_dev;
4159
4160         /* configure local quota objects */
4161         rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
4162                                                    &mdt->mdt_lu_dev,
4163                                                    mdt->mdt_qmt_dev);
4164         if (rc)
4165                 GOTO(class_cleanup, rc);
4166
4167         /* connect to quota master target */
4168         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4169         data->ocd_version = LUSTRE_VERSION_CODE;
4170         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
4171                          data, NULL);
4172         if (rc) {
4173                 CERROR("cannot connect to quota master device %s (%d)\n",
4174                        qmtname, rc);
4175                 GOTO(class_cleanup, rc);
4176         }
4177
4178         EXIT;
4179 class_cleanup:
4180         if (rc) {
4181                 class_manual_cleanup(obd);
4182                 mdt->mdt_qmt_dev = NULL;
4183         }
4184 class_detach:
4185         if (rc)
4186                 class_detach(obd, lcfg);
4187 lcfg_cleanup:
4188         lustre_cfg_free(lcfg);
4189 put_profile:
4190         class_put_profile(lprof);
4191 cleanup_mem:
4192         if (bufs)
4193                 OBD_FREE_PTR(bufs);
4194         if (qmtname)
4195                 OBD_FREE(qmtname, MAX_OBD_NAME);
4196         if (uuid)
4197                 OBD_FREE(uuid, UUID_MAX);
4198         if (data)
4199                 OBD_FREE_PTR(data);
4200         return rc;
4201 }
4202
4203 /* Shutdown quota master target associated with mdt */
4204 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
4205 {
4206         ENTRY;
4207
4208         if (mdt->mdt_qmt_exp == NULL)
4209                 RETURN_EXIT;
4210         LASSERT(mdt->mdt_qmt_dev != NULL);
4211
4212         /* the qmt automatically shuts down when the mdt disconnects */
4213         obd_disconnect(mdt->mdt_qmt_exp);
4214         mdt->mdt_qmt_exp = NULL;
4215         mdt->mdt_qmt_dev = NULL;
4216         EXIT;
4217 }
4218
4219 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
4220  * for now. This will be removed along with converting rest of MDT code
4221  * to use tgt_session_info */
4222 static int mdt_tgt_getxattr(struct tgt_session_info *tsi)
4223 {
4224         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
4225         int                      rc;
4226
4227         rc = mdt_getxattr(info);
4228
4229         mdt_thread_info_fini(info);
4230         return rc;
4231 }
4232
4233 static struct tgt_handler mdt_tgt_handlers[] = {
4234 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4235                 0,                      MDS_CONNECT,    mdt_tgt_connect,
4236                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
4237 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4238                 0,                      MDS_DISCONNECT, tgt_disconnect,
4239                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
4240 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4241                 HABEO_REFERO,           MDS_SET_INFO,   mdt_set_info,
4242                 &RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION),
4243 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
4244 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_GETSTATUS,  mdt_getstatus),
4245 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETATTR,    mdt_getattr),
4246 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_GETATTR_NAME,
4247                                                         mdt_getattr_name),
4248 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETXATTR,   mdt_tgt_getxattr),
4249 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_STATFS,     mdt_statfs),
4250 TGT_MDT_HDL(0           | MUTABOR,      MDS_REINT,      mdt_reint),
4251 TGT_MDT_HDL(HABEO_CORPUS,               MDS_CLOSE,      mdt_close),
4252 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_READPAGE,   mdt_readpage),
4253 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_SYNC,       mdt_sync),
4254 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
4255 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_PROGRESS,
4256                                                         mdt_hsm_progress),
4257 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_REGISTER,
4258                                                         mdt_hsm_ct_register),
4259 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_UNREGISTER,
4260                                                         mdt_hsm_ct_unregister),
4261 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_STATE_GET,
4262                                                         mdt_hsm_state_get),
4263 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_STATE_SET,
4264                                                         mdt_hsm_state_set),
4265 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_ACTION, mdt_hsm_action),
4266 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_REQUEST,
4267                                                         mdt_hsm_request),
4268 TGT_MDT_HDL(HABEO_CLAVIS | HABEO_CORPUS | HABEO_REFERO | MUTABOR,
4269             MDS_SWAP_LAYOUTS,
4270             mdt_swap_layouts),
4271 };
4272
4273 static struct tgt_handler mdt_sec_ctx_ops[] = {
4274 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
4275 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
4276 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
4277 };
4278
4279 static struct tgt_handler mdt_quota_ops[] = {
4280 TGT_QUOTA_HDL(HABEO_REFERO,             QUOTA_DQACQ,      mdt_quota_dqacq),
4281 };
4282
4283 static struct tgt_opc_slice mdt_common_slice[] = {
4284         {
4285                 .tos_opc_start  = MDS_FIRST_OPC,
4286                 .tos_opc_end    = MDS_LAST_OPC,
4287                 .tos_hs         = mdt_tgt_handlers
4288         },
4289         {
4290                 .tos_opc_start  = OBD_FIRST_OPC,
4291                 .tos_opc_end    = OBD_LAST_OPC,
4292                 .tos_hs         = tgt_obd_handlers
4293         },
4294         {
4295                 .tos_opc_start  = LDLM_FIRST_OPC,
4296                 .tos_opc_end    = LDLM_LAST_OPC,
4297                 .tos_hs         = tgt_dlm_handlers
4298         },
4299         {
4300                 .tos_opc_start  = SEC_FIRST_OPC,
4301                 .tos_opc_end    = SEC_LAST_OPC,
4302                 .tos_hs         = mdt_sec_ctx_ops
4303         },
4304         {
4305                 .tos_opc_start  = OUT_UPDATE_FIRST_OPC,
4306                 .tos_opc_end    = OUT_UPDATE_LAST_OPC,
4307                 .tos_hs         = tgt_out_handlers
4308         },
4309         {
4310                 .tos_opc_start  = FLD_FIRST_OPC,
4311                 .tos_opc_end    = FLD_LAST_OPC,
4312                 .tos_hs         = fld_handlers
4313         },
4314         {
4315                 .tos_opc_start  = SEQ_FIRST_OPC,
4316                 .tos_opc_end    = SEQ_LAST_OPC,
4317                 .tos_hs         = seq_handlers
4318         },
4319         {
4320                 .tos_opc_start  = QUOTA_DQACQ,
4321                 .tos_opc_end    = QUOTA_LAST_OPC,
4322                 .tos_hs         = mdt_quota_ops
4323         },
4324         {
4325                 .tos_opc_start  = LLOG_FIRST_OPC,
4326                 .tos_opc_end    = LLOG_LAST_OPC,
4327                 .tos_hs         = tgt_llog_handlers
4328         },
4329         {
4330                 .tos_opc_start  = LFSCK_FIRST_OPC,
4331                 .tos_opc_end    = LFSCK_LAST_OPC,
4332                 .tos_hs         = tgt_lfsck_handlers
4333         },
4334
4335         {
4336                 .tos_hs         = NULL
4337         }
4338 };
4339
4340 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4341 {
4342         struct md_device        *next = m->mdt_child;
4343         struct lu_device        *d    = &m->mdt_lu_dev;
4344         struct obd_device       *obd  = mdt2obd_dev(m);
4345         struct lfsck_stop        stop;
4346         ENTRY;
4347
4348         stop.ls_status = LS_PAUSED;
4349         stop.ls_flags = 0;
4350         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
4351
4352         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
4353         target_recovery_fini(obd);
4354         ping_evictor_stop();
4355
4356         if (m->mdt_opts.mo_coordinator)
4357                 mdt_hsm_cdt_stop(m);
4358
4359         mdt_hsm_cdt_fini(m);
4360
4361         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
4362         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4363
4364         if (m->mdt_namespace != NULL)
4365                 ldlm_namespace_free_prior(m->mdt_namespace, NULL,
4366                                           d->ld_obd->obd_force);
4367
4368         obd_exports_barrier(obd);
4369         obd_zombie_barrier();
4370
4371         mdt_procfs_fini(m);
4372
4373         tgt_fini(env, &m->mdt_lut);
4374         mdt_fs_cleanup(env, m);
4375         upcall_cache_cleanup(m->mdt_identity_cache);
4376         m->mdt_identity_cache = NULL;
4377
4378         if (m->mdt_namespace != NULL) {
4379                 ldlm_namespace_free_post(m->mdt_namespace);
4380                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4381         }
4382
4383         mdt_quota_fini(env, m);
4384
4385         cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
4386
4387         mdt_seq_fini(env, m);
4388         mdt_fld_fini(env, m);
4389
4390         /*
4391          * Finish the stack
4392          */
4393         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4394
4395         LASSERT(atomic_read(&d->ld_ref) == 0);
4396
4397         server_put_mount(mdt_obd_name(m), true);
4398
4399         EXIT;
4400 }
4401
4402 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4403
4404 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4405                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
4406 {
4407         struct mdt_thread_info    *info;
4408         struct obd_device         *obd;
4409         const char                *dev = lustre_cfg_string(cfg, 0);
4410         const char                *num = lustre_cfg_string(cfg, 2);
4411         struct lustre_mount_info  *lmi = NULL;
4412         struct lustre_sb_info     *lsi;
4413         struct lu_site            *s;
4414         struct seq_server_site    *ss_site;
4415         const char                *identity_upcall = "NONE";
4416         struct md_device          *next;
4417         int                        rc;
4418         long                       node_id;
4419         mntopt_t                   mntopts;
4420         ENTRY;
4421
4422         lu_device_init(&m->mdt_lu_dev, ldt);
4423         /*
4424          * Environment (env) might be missing mdt_thread_key values at that
4425          * point, if device is allocated when mdt_thread_key is in QUIESCENT
4426          * mode.
4427          *
4428          * Usually device allocation path doesn't use module key values, but
4429          * mdt has to do a lot of work here, so allocate key value.
4430          */
4431         rc = lu_env_refill((struct lu_env *)env);
4432         if (rc != 0)
4433                 RETURN(rc);
4434
4435         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4436         LASSERT(info != NULL);
4437
4438         obd = class_name2obd(dev);
4439         LASSERT(obd != NULL);
4440
4441         m->mdt_max_mdsize = MAX_MD_SIZE; /* 4 stripes */
4442         m->mdt_opts.mo_evict_tgt_nids = 1;
4443         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4444
4445         /* default is coordinator off, it is started through conf_param
4446          * or /proc */
4447         m->mdt_opts.mo_coordinator = 0;
4448
4449         lmi = server_get_mount(dev);
4450         if (lmi == NULL) {
4451                 CERROR("Cannot get mount info for %s!\n", dev);
4452                 RETURN(-EFAULT);
4453         } else {
4454                 lsi = s2lsi(lmi->lmi_sb);
4455                 /* CMD is supported only in IAM mode */
4456                 LASSERT(num);
4457                 node_id = simple_strtol(num, NULL, 10);
4458                 obd->u.obt.obt_magic = OBT_MAGIC;
4459                 if (lsi->lsi_lmd != NULL &&
4460                     lsi->lsi_lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
4461                         m->mdt_skip_lfsck = 1;
4462         }
4463
4464         m->mdt_squash.rsi_uid = 0;
4465         m->mdt_squash.rsi_gid = 0;
4466         INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
4467         init_rwsem(&m->mdt_squash.rsi_sem);
4468         spin_lock_init(&m->mdt_osfs_lock);
4469         m->mdt_osfs_age = cfs_time_shift_64(-1000);
4470         m->mdt_enable_remote_dir = 0;
4471         m->mdt_enable_remote_dir_gid = 0;
4472
4473         atomic_set(&m->mdt_mds_mds_conns, 0);
4474         atomic_set(&m->mdt_async_commit_count, 0);
4475
4476         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
4477         m->mdt_lu_dev.ld_obd = obd;
4478         /* Set this lu_device to obd for error handling purposes. */
4479         obd->obd_lu_dev = &m->mdt_lu_dev;
4480
4481         /* init the stack */
4482         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
4483         if (rc) {
4484                 CERROR("%s: Can't init device stack, rc %d\n",
4485                        mdt_obd_name(m), rc);
4486                 GOTO(err_lmi, rc);
4487         }
4488
4489         s = mdt_lu_site(m);
4490         ss_site = mdt_seq_site(m);
4491         s->ld_seq_site = ss_site;
4492         ss_site->ss_lu = s;
4493
4494         /* set server index */
4495         ss_site->ss_node_id = node_id;
4496
4497         /* failover is the default
4498          * FIXME: we do not failout mds0/mgs, which may cause some problems.
4499          * assumed whose ss_node_id == 0 XXX
4500          * */
4501         obd->obd_replayable = 1;
4502         /* No connection accepted until configurations will finish */
4503         obd->obd_no_conn = 1;
4504
4505         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4506                 char *str = lustre_cfg_string(cfg, 4);
4507                 if (strchr(str, 'n')) {
4508                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
4509                         obd->obd_replayable = 0;
4510                 }
4511         }
4512
4513         rc = mdt_fld_init(env, mdt_obd_name(m), m);
4514         if (rc)
4515                 GOTO(err_fini_stack, rc);
4516
4517         rc = mdt_seq_init(env, m);
4518         if (rc)
4519                 GOTO(err_fini_fld, rc);
4520
4521         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
4522                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
4523         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4524                                               LDLM_NAMESPACE_SERVER,
4525                                               LDLM_NAMESPACE_GREEDY,
4526                                               LDLM_NS_TYPE_MDT);
4527         if (m->mdt_namespace == NULL)
4528                 GOTO(err_fini_seq, rc = -ENOMEM);
4529
4530         m->mdt_namespace->ns_lvbp = m;
4531         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
4532
4533         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4534         /* set obd_namespace for compatibility with old code */
4535         obd->obd_namespace = m->mdt_namespace;
4536
4537         rc = mdt_hsm_cdt_init(m);
4538         if (rc != 0) {
4539                 CERROR("%s: error initializing coordinator, rc %d\n",
4540                        mdt_obd_name(m), rc);
4541                 GOTO(err_free_ns, rc);
4542         }
4543
4544         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
4545                       OBD_FAIL_MDS_ALL_REQUEST_NET,
4546                       OBD_FAIL_MDS_ALL_REPLY_NET);
4547         if (rc)
4548                 GOTO(err_free_hsm, rc);
4549
4550         rc = mdt_fs_setup(env, m, obd, lsi);
4551         if (rc)
4552                 GOTO(err_tgt, rc);
4553
4554         tgt_adapt_sptlrpc_conf(&m->mdt_lut, 1);
4555
4556         next = m->mdt_child;
4557         rc = next->md_ops->mdo_iocontrol(env, next, OBD_IOC_GET_MNTOPT, 0,
4558                                          &mntopts);
4559         if (rc)
4560                 GOTO(err_fs_cleanup, rc);
4561
4562         if (mntopts & MNTOPT_USERXATTR)
4563                 m->mdt_opts.mo_user_xattr = 1;
4564         else
4565                 m->mdt_opts.mo_user_xattr = 0;
4566
4567         rc = next->md_ops->mdo_maxeasize_get(env, next, &m->mdt_max_ea_size);
4568         if (rc)
4569                 GOTO(err_fs_cleanup, rc);
4570
4571         if (mntopts & MNTOPT_ACL)
4572                 m->mdt_opts.mo_acl = 1;
4573         else
4574                 m->mdt_opts.mo_acl = 0;
4575
4576         /* XXX: to support suppgid for ACL, we enable identity_upcall
4577          * by default, otherwise, maybe got unexpected -EACCESS. */
4578         if (m->mdt_opts.mo_acl)
4579                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4580
4581         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
4582                                                 identity_upcall,
4583                                                 &mdt_identity_upcall_cache_ops);
4584         if (IS_ERR(m->mdt_identity_cache)) {
4585                 rc = PTR_ERR(m->mdt_identity_cache);
4586                 m->mdt_identity_cache = NULL;
4587                 GOTO(err_fs_cleanup, rc);
4588         }
4589
4590         rc = mdt_procfs_init(m, dev);
4591         if (rc) {
4592                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4593                 GOTO(err_recovery, rc);
4594         }
4595
4596         rc = mdt_quota_init(env, m, cfg);
4597         if (rc)
4598                 GOTO(err_procfs, rc);
4599
4600         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
4601         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
4602                            "mdt_ldlm_client", m->mdt_ldlm_client);
4603
4604         ping_evictor_start();
4605
4606         /* recovery will be started upon mdt_prepare()
4607          * when the whole stack is complete and ready
4608          * to serve the requests */
4609
4610         /* Reduce the initial timeout on an MDS because it doesn't need such
4611          * a long timeout as an OST does. Adaptive timeouts will adjust this
4612          * value appropriately. */
4613         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4614                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4615
4616         RETURN(0);
4617 err_procfs:
4618         mdt_procfs_fini(m);
4619 err_recovery:
4620         target_recovery_fini(obd);
4621         upcall_cache_cleanup(m->mdt_identity_cache);
4622         m->mdt_identity_cache = NULL;
4623 err_fs_cleanup:
4624         mdt_fs_cleanup(env, m);
4625 err_tgt:
4626         tgt_fini(env, &m->mdt_lut);
4627 err_free_hsm:
4628         mdt_hsm_cdt_fini(m);
4629 err_free_ns:
4630         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4631         obd->obd_namespace = m->mdt_namespace = NULL;
4632 err_fini_seq:
4633         mdt_seq_fini(env, m);
4634 err_fini_fld:
4635         mdt_fld_fini(env, m);
4636 err_fini_stack:
4637         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4638 err_lmi:
4639         if (lmi)
4640                 server_put_mount(dev, true);
4641         return(rc);
4642 }
4643
4644 /* For interoperability, the left element is old parameter, the right one
4645  * is the new version of the parameter, if some parameter is deprecated,
4646  * the new version should be set as NULL. */
4647 static struct cfg_interop_param mdt_interop_param[] = {
4648         { "mdt.group_upcall",   NULL },
4649         { "mdt.quota_type",     NULL },
4650         { "mdd.quota_type",     NULL },
4651         { "mdt.rootsquash",     "mdt.root_squash" },
4652         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
4653         { NULL }
4654 };
4655
4656 /* used by MGS to process specific configurations */
4657 static int mdt_process_config(const struct lu_env *env,
4658                               struct lu_device *d, struct lustre_cfg *cfg)
4659 {
4660         struct mdt_device *m = mdt_dev(d);
4661         struct md_device *md_next = m->mdt_child;
4662         struct lu_device *next = md2lu_dev(md_next);
4663         int rc;
4664         ENTRY;
4665
4666         switch (cfg->lcfg_command) {
4667         case LCFG_PARAM: {
4668                 struct obd_device          *obd = d->ld_obd;
4669
4670                 /* For interoperability */
4671                 struct cfg_interop_param   *ptr = NULL;
4672                 struct lustre_cfg          *old_cfg = NULL;
4673                 char                       *param = NULL;
4674
4675                 param = lustre_cfg_string(cfg, 1);
4676                 if (param == NULL) {
4677                         CERROR("param is empty\n");
4678                         rc = -EINVAL;
4679                         break;
4680                 }
4681
4682                 ptr = class_find_old_param(param, mdt_interop_param);
4683                 if (ptr != NULL) {
4684                         if (ptr->new_param == NULL) {
4685                                 rc = 0;
4686                                 CWARN("For interoperability, skip this %s."
4687                                       " It is obsolete.\n", ptr->old_param);
4688                                 break;
4689                         }
4690
4691                         CWARN("Found old param %s, changed it to %s.\n",
4692                               ptr->old_param, ptr->new_param);
4693
4694                         old_cfg = cfg;
4695                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
4696                         if (IS_ERR(cfg)) {
4697                                 rc = PTR_ERR(cfg);
4698                                 break;
4699                         }
4700                 }
4701
4702                 rc = class_process_proc_param(PARAM_MDT, obd->obd_vars,
4703                                               cfg, obd);
4704                 if (rc > 0 || rc == -ENOSYS) {
4705                         /* is it an HSM var ? */
4706                         rc = class_process_proc_param(PARAM_HSM,
4707                                                       hsm_cdt_get_proc_vars(),
4708                                                       cfg, obd);
4709                         if (rc > 0 || rc == -ENOSYS)
4710                                 /* we don't understand; pass it on */
4711                                 rc = next->ld_ops->ldo_process_config(env, next,
4712                                                                       cfg);
4713                 }
4714
4715                 if (old_cfg != NULL)
4716                         lustre_cfg_free(cfg);
4717
4718                 break;
4719         }
4720         default:
4721                 /* others are passed further */
4722                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
4723                 break;
4724         }
4725         RETURN(rc);
4726 }
4727
4728 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
4729                                           const struct lu_object_header *hdr,
4730                                           struct lu_device *d)
4731 {
4732         struct mdt_object *mo;
4733
4734         ENTRY;
4735
4736         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
4737         if (mo != NULL) {
4738                 struct lu_object *o;
4739                 struct lu_object_header *h;
4740
4741                 o = &mo->mot_obj;
4742                 h = &mo->mot_header;
4743                 lu_object_header_init(h);
4744                 lu_object_init(o, h, d);
4745                 lu_object_add_top(h, o);
4746                 o->lo_ops = &mdt_obj_ops;
4747                 spin_lock_init(&mo->mot_write_lock);
4748                 mutex_init(&mo->mot_lov_mutex);
4749                 init_rwsem(&mo->mot_open_sem);
4750                 RETURN(o);
4751         }
4752         RETURN(NULL);
4753 }
4754
4755 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
4756                            const struct lu_object_conf *unused)
4757 {
4758         struct mdt_device *d = mdt_dev(o->lo_dev);
4759         struct lu_device  *under;
4760         struct lu_object  *below;
4761         int                rc = 0;
4762         ENTRY;
4763
4764         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
4765                PFID(lu_object_fid(o)));
4766
4767         under = &d->mdt_child->md_lu_dev;
4768         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
4769         if (below != NULL) {
4770                 lu_object_add(o, below);
4771         } else
4772                 rc = -ENOMEM;
4773
4774         RETURN(rc);
4775 }
4776
4777 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
4778 {
4779         struct mdt_object *mo = mdt_obj(o);
4780         struct lu_object_header *h;
4781         ENTRY;
4782
4783         h = o->lo_header;
4784         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
4785                PFID(lu_object_fid(o)));
4786
4787         LASSERT(atomic_read(&mo->mot_open_count) == 0);
4788         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
4789
4790         lu_object_fini(o);
4791         lu_object_header_fini(h);
4792         OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
4793
4794         EXIT;
4795 }
4796
4797 static int mdt_object_print(const struct lu_env *env, void *cookie,
4798                             lu_printer_t p, const struct lu_object *o)
4799 {
4800         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
4801
4802         return (*p)(env, cookie,
4803                     LUSTRE_MDT_NAME"-object@%p(flags=%d, writecount=%d)",
4804                     mdto, mdto->mot_flags, mdto->mot_write_count);
4805 }
4806
4807 static int mdt_prepare(const struct lu_env *env,
4808                 struct lu_device *pdev,
4809                 struct lu_device *cdev)
4810 {
4811         struct mdt_device *mdt = mdt_dev(cdev);
4812         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
4813         struct obd_device *obd = cdev->ld_obd;
4814         int rc;
4815
4816         ENTRY;
4817
4818         LASSERT(obd);
4819
4820         rc = next->ld_ops->ldo_prepare(env, cdev, next);
4821         if (rc)
4822                 RETURN(rc);
4823
4824         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
4825         if (rc)
4826                 RETURN(rc);
4827
4828         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
4829         if (rc)
4830                 RETURN(rc);
4831
4832         rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
4833         /* The LFSCK instance is registered just now, so it must be there when
4834          * register the namespace to such instance. */
4835         LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
4836
4837         if (mdt->mdt_seq_site.ss_node_id == 0) {
4838                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
4839                                                          &mdt->mdt_md_root_fid);
4840                 if (rc)
4841                         RETURN(rc);
4842         }
4843
4844         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
4845
4846         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
4847         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
4848         LASSERT(obd->obd_no_conn);
4849         spin_lock(&obd->obd_dev_lock);
4850         obd->obd_no_conn = 0;
4851         spin_unlock(&obd->obd_dev_lock);
4852
4853         if (obd->obd_recovering == 0)
4854                 mdt_postrecov(env, mdt);
4855
4856         RETURN(rc);
4857 }
4858
4859 const struct lu_device_operations mdt_lu_ops = {
4860         .ldo_object_alloc   = mdt_object_alloc,
4861         .ldo_process_config = mdt_process_config,
4862         .ldo_prepare        = mdt_prepare,
4863 };
4864
4865 static const struct lu_object_operations mdt_obj_ops = {
4866         .loo_object_init    = mdt_object_init,
4867         .loo_object_free    = mdt_object_free,
4868         .loo_object_print   = mdt_object_print
4869 };
4870
4871 static int mdt_obd_set_info_async(const struct lu_env *env,
4872                                   struct obd_export *exp,
4873                                   __u32 keylen, void *key,
4874                                   __u32 vallen, void *val,
4875                                   struct ptlrpc_request_set *set)
4876 {
4877         int rc;
4878
4879         ENTRY;
4880
4881         if (KEY_IS(KEY_SPTLRPC_CONF)) {
4882                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp), 0);
4883                 RETURN(rc);
4884         }
4885
4886         RETURN(0);
4887 }
4888
4889 /**
4890  * Match client and server connection feature flags.
4891  *
4892  * Compute the compatibility flags for a connection request based on
4893  * features mutually supported by client and server.
4894  *
4895  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
4896  * must not be updated here, otherwise a partially initialized value may
4897  * be exposed. After the connection request is successfully processed,
4898  * the top-level MDT connect request handler atomically updates the export
4899  * connect flags from the obd_connect_data::ocd_connect_flags field of the
4900  * reply. \see mdt_connect().
4901  *
4902  * \param exp   the obd_export associated with this client/target pair
4903  * \param mdt   the target device for the connection
4904  * \param data  stores data for this connect request
4905  *
4906  * \retval 0       success
4907  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
4908  * \retval -EBADE  client and server feature requirements are incompatible
4909  */
4910 static int mdt_connect_internal(struct obd_export *exp,
4911                                 struct mdt_device *mdt,
4912                                 struct obd_connect_data *data)
4913 {
4914         LASSERT(data != NULL);
4915
4916         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
4917         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
4918
4919         if (!(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
4920             !(data->ocd_connect_flags & OBD_CONNECT_IBITS)) {
4921                 CWARN("%s: client %s does not support ibits lock, either "
4922                       "very old or an invalid client: flags "LPX64"\n",
4923                       mdt_obd_name(mdt), exp->exp_client_uuid.uuid,
4924                       data->ocd_connect_flags);
4925                 return -EBADE;
4926         }
4927
4928         if (!mdt->mdt_opts.mo_acl)
4929                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
4930
4931         if (!mdt->mdt_opts.mo_user_xattr)
4932                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
4933
4934         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
4935                 data->ocd_brw_size = min(data->ocd_brw_size,
4936                                          (__u32)MD_MAX_BRW_SIZE);
4937                 if (data->ocd_brw_size == 0) {
4938                         CERROR("%s: cli %s/%p ocd_connect_flags: "LPX64
4939                                " ocd_version: %x ocd_grant: %d "
4940                                "ocd_index: %u ocd_brw_size is "
4941                                "unexpectedly zero, network data "
4942                                "corruption? Refusing connection of this"
4943                                " client\n",
4944                                mdt_obd_name(mdt),
4945                                exp->exp_client_uuid.uuid,
4946                                exp, data->ocd_connect_flags, data->ocd_version,
4947                                data->ocd_grant, data->ocd_index);
4948                         return -EPROTO;
4949                 }
4950         }
4951
4952         /* NB: Disregard the rule against updating
4953          * exp_connect_data.ocd_connect_flags in this case, since
4954          * tgt_client_new() needs to know if this is a lightweight
4955          * connection, and it is safe to expose this flag before
4956          * connection processing completes. */
4957         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
4958                 spin_lock(&exp->exp_lock);
4959                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
4960                 spin_unlock(&exp->exp_lock);
4961         }
4962
4963         data->ocd_version = LUSTRE_VERSION_CODE;
4964
4965         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
4966                 CWARN("%s: MDS requires FID support, but client not\n",
4967                       mdt_obd_name(mdt));
4968                 return -EBADE;
4969         }
4970
4971         if (OCD_HAS_FLAG(data, PINGLESS)) {
4972                 if (ptlrpc_pinger_suppress_pings()) {
4973                         spin_lock(&exp->exp_obd->obd_dev_lock);
4974                         list_del_init(&exp->exp_obd_chain_timed);
4975                         spin_unlock(&exp->exp_obd->obd_dev_lock);
4976                 } else {
4977                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
4978                 }
4979         }
4980
4981         data->ocd_max_easize = mdt->mdt_max_ea_size;
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 client supports
4986          * multiple modify RPCs, and it is safe to expose this flag before
4987          * connection processing completes. */
4988         if (data->ocd_connect_flags & OBD_CONNECT_MULTIMODRPCS) {
4989                 data->ocd_maxmodrpcs = max_mod_rpcs_per_client;
4990                 spin_lock(&exp->exp_lock);
4991                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_MULTIMODRPCS;
4992                 spin_unlock(&exp->exp_lock);
4993         }
4994
4995         return 0;
4996 }
4997
4998 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
4999                                    struct mdt_thread_info *info,
5000                                    struct mdt_file_data *mfd)
5001 {
5002         struct lu_context ses;
5003         int rc;
5004         ENTRY;
5005
5006         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
5007         if (rc)
5008                 RETURN(rc);
5009
5010         env->le_ses = &ses;
5011         lu_context_enter(&ses);
5012
5013         mdt_ucred(info)->uc_valid = UCRED_OLD;
5014         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
5015
5016         lu_context_exit(&ses);
5017         lu_context_fini(&ses);
5018         env->le_ses = NULL;
5019
5020         RETURN(rc);
5021 }
5022
5023 static int mdt_export_cleanup(struct obd_export *exp)
5024 {
5025         struct list_head         closing_list;
5026         struct mdt_export_data  *med = &exp->exp_mdt_data;
5027         struct obd_device       *obd = exp->exp_obd;
5028         struct mdt_device       *mdt;
5029         struct mdt_thread_info  *info;
5030         struct lu_env            env;
5031         struct mdt_file_data    *mfd, *n;
5032         int rc = 0;
5033         ENTRY;
5034
5035         INIT_LIST_HEAD(&closing_list);
5036         spin_lock(&med->med_open_lock);
5037         while (!list_empty(&med->med_open_head)) {
5038                 struct list_head *tmp = med->med_open_head.next;
5039                 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
5040
5041                 /* Remove mfd handle so it can't be found again.
5042                  * We are consuming the mfd_list reference here. */
5043                 class_handle_unhash(&mfd->mfd_handle);
5044                 list_move_tail(&mfd->mfd_list, &closing_list);
5045         }
5046         spin_unlock(&med->med_open_lock);
5047         mdt = mdt_dev(obd->obd_lu_dev);
5048         LASSERT(mdt != NULL);
5049
5050         rc = lu_env_init(&env, LCT_MD_THREAD);
5051         if (rc)
5052                 RETURN(rc);
5053
5054         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5055         LASSERT(info != NULL);
5056         memset(info, 0, sizeof *info);
5057         info->mti_env = &env;
5058         info->mti_mdt = mdt;
5059         info->mti_exp = exp;
5060
5061         if (!list_empty(&closing_list)) {
5062                 struct md_attr *ma = &info->mti_attr;
5063
5064                 /* Close any open files (which may also cause orphan
5065                  * unlinking). */
5066                 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5067                         list_del_init(&mfd->mfd_list);
5068                         ma->ma_need = ma->ma_valid = 0;
5069
5070                         /* This file is being closed due to an eviction, it
5071                          * could have been modified and now dirty regarding to
5072                          * HSM archive, check this!
5073                          * The logic here is to mark a file dirty if there's a
5074                          * chance it was dirtied before the client was evicted,
5075                          * so that we don't have to wait for a release attempt
5076                          * before finding out the file was actually dirty and
5077                          * fail the release. Aggressively marking it dirty here
5078                          * will cause the policy engine to attempt to
5079                          * re-archive it; when rearchiving, we can compare the
5080                          * current version to the HSM data_version and make the
5081                          * archive request into a noop if it's not actually
5082                          * dirty.
5083                          */
5084                         if (mfd->mfd_mode & FMODE_WRITE)
5085                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
5086
5087                         /* Don't unlink orphan on failover umount, LU-184 */
5088                         if (exp->exp_flags & OBD_OPT_FAILOVER) {
5089                                 ma->ma_valid = MA_FLAGS;
5090                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5091                         }
5092                         mdt_mfd_close(info, mfd);
5093                 }
5094         }
5095         info->mti_mdt = NULL;
5096         /* cleanup client slot early */
5097         /* Do not erase record for recoverable client. */
5098         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5099                 tgt_client_del(&env, exp);
5100         lu_env_fini(&env);
5101
5102         RETURN(rc);
5103 }
5104
5105 static inline void mdt_enable_slc(struct mdt_device *mdt)
5106 {
5107         if (mdt->mdt_lut.lut_sync_lock_cancel == NEVER_SYNC_ON_CANCEL)
5108                 mdt->mdt_lut.lut_sync_lock_cancel = BLOCKING_SYNC_ON_CANCEL;
5109 }
5110
5111 static inline void mdt_disable_slc(struct mdt_device *mdt)
5112 {
5113         if (mdt->mdt_lut.lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL)
5114                 mdt->mdt_lut.lut_sync_lock_cancel = NEVER_SYNC_ON_CANCEL;
5115 }
5116
5117 static int mdt_obd_disconnect(struct obd_export *exp)
5118 {
5119         int rc;
5120         ENTRY;
5121
5122         LASSERT(exp);
5123         class_export_get(exp);
5124
5125         if ((exp_connect_flags(exp) & OBD_CONNECT_MDS_MDS) &&
5126             !(exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT)) {
5127                 struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
5128
5129                 if (atomic_dec_and_test(&mdt->mdt_mds_mds_conns))
5130                         mdt_disable_slc(mdt);
5131         }
5132
5133         rc = server_disconnect_export(exp);
5134         if (rc != 0)
5135                 CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
5136
5137         rc = mdt_export_cleanup(exp);
5138         nodemap_del_member(exp);
5139         class_export_put(exp);
5140         RETURN(rc);
5141 }
5142
5143 /* mds_connect copy */
5144 static int mdt_obd_connect(const struct lu_env *env,
5145                            struct obd_export **exp, struct obd_device *obd,
5146                            struct obd_uuid *cluuid,
5147                            struct obd_connect_data *data,
5148                            void *localdata)
5149 {
5150         struct obd_export       *lexp;
5151         struct lustre_handle    conn = { 0 };
5152         struct mdt_device       *mdt;
5153         int                      rc;
5154         lnet_nid_t              *client_nid = localdata;
5155         ENTRY;
5156
5157         LASSERT(env != NULL);
5158         if (!exp || !obd || !cluuid)
5159                 RETURN(-EINVAL);
5160
5161         mdt = mdt_dev(obd->obd_lu_dev);
5162
5163         if ((data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5164             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
5165                 atomic_inc(&mdt->mdt_mds_mds_conns);
5166                 mdt_enable_slc(mdt);
5167         }
5168
5169         /*
5170          * first, check whether the stack is ready to handle requests
5171          * XXX: probably not very appropriate method is used now
5172          *      at some point we should find a better one
5173          */
5174         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) && data != NULL &&
5175             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) &&
5176             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
5177                 rc = obd_get_info(env, mdt->mdt_child_exp,
5178                                   sizeof(KEY_OSP_CONNECTED),
5179                                   KEY_OSP_CONNECTED, NULL, NULL);
5180                 if (rc)
5181                         RETURN(-EAGAIN);
5182                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
5183         }
5184
5185         rc = class_connect(&conn, obd, cluuid);
5186         if (rc)
5187                 RETURN(rc);
5188
5189         lexp = class_conn2export(&conn);
5190         LASSERT(lexp != NULL);
5191
5192         rc = nodemap_add_member(*client_nid, lexp);
5193         if (rc != 0 && rc != -EEXIST)
5194                 GOTO(out, rc);
5195
5196         rc = mdt_connect_internal(lexp, mdt, data);
5197         if (rc == 0) {
5198                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5199
5200                 LASSERT(lcd);
5201                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5202                 rc = tgt_client_new(env, lexp);
5203                 if (rc == 0)
5204                         mdt_export_stats_init(obd, lexp, localdata);
5205         }
5206 out:
5207         if (rc != 0) {
5208                 class_disconnect(lexp);
5209                 nodemap_del_member(lexp);
5210                 *exp = NULL;
5211         } else {
5212                 *exp = lexp;
5213         }
5214
5215         RETURN(rc);
5216 }
5217
5218 static int mdt_obd_reconnect(const struct lu_env *env,
5219                              struct obd_export *exp, struct obd_device *obd,
5220                              struct obd_uuid *cluuid,
5221                              struct obd_connect_data *data,
5222                              void *localdata)
5223 {
5224         lnet_nid_t             *client_nid = localdata;
5225         int                     rc;
5226         ENTRY;
5227
5228         if (exp == NULL || obd == NULL || cluuid == NULL)
5229                 RETURN(-EINVAL);
5230
5231         rc = nodemap_add_member(*client_nid, exp);
5232         if (rc != 0 && rc != -EEXIST)
5233                 RETURN(rc);
5234
5235         rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
5236         if (rc == 0)
5237                 mdt_export_stats_init(obd, exp, localdata);
5238         else
5239                 nodemap_del_member(exp);
5240
5241         RETURN(rc);
5242 }
5243
5244 /* FIXME: Can we avoid using these two interfaces? */
5245 static int mdt_init_export(struct obd_export *exp)
5246 {
5247         struct mdt_export_data *med = &exp->exp_mdt_data;
5248         int                     rc;
5249         ENTRY;
5250
5251         INIT_LIST_HEAD(&med->med_open_head);
5252         spin_lock_init(&med->med_open_lock);
5253         mutex_init(&med->med_idmap_mutex);
5254         med->med_idmap = NULL;
5255         spin_lock(&exp->exp_lock);
5256         exp->exp_connecting = 1;
5257         spin_unlock(&exp->exp_lock);
5258
5259         /* self-export doesn't need client data and ldlm initialization */
5260         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5261                                      &exp->exp_client_uuid)))
5262                 RETURN(0);
5263
5264         rc = tgt_client_alloc(exp);
5265         if (rc)
5266                 GOTO(err, rc);
5267
5268         rc = ldlm_init_export(exp);
5269         if (rc)
5270                 GOTO(err_free, rc);
5271
5272         RETURN(rc);
5273
5274 err_free:
5275         tgt_client_free(exp);
5276 err:
5277         CERROR("%s: Failed to initialize export: rc = %d\n",
5278                exp->exp_obd->obd_name, rc);
5279         return rc;
5280 }
5281
5282 static int mdt_destroy_export(struct obd_export *exp)
5283 {
5284         ENTRY;
5285
5286         if (exp_connect_rmtclient(exp))
5287                 mdt_cleanup_idmap(&exp->exp_mdt_data);
5288
5289         target_destroy_export(exp);
5290         /* destroy can be called from failed obd_setup, so
5291          * checking uuid is safer than obd_self_export */
5292         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5293                                      &exp->exp_client_uuid)))
5294                 RETURN(0);
5295
5296         ldlm_destroy_export(exp);
5297         tgt_client_free(exp);
5298
5299         LASSERT(list_empty(&exp->exp_outstanding_replies));
5300         LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
5301
5302         RETURN(0);
5303 }
5304
5305 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
5306                    struct linkea_data *ldata)
5307 {
5308         int rc;
5309
5310         LASSERT(ldata->ld_buf->lb_buf != NULL);
5311
5312         if (!mdt_object_exists(mdt_obj))
5313                 return -ENODATA;
5314
5315         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5316                           ldata->ld_buf, XATTR_NAME_LINK);
5317         if (rc == -ERANGE) {
5318                 /* Buf was too small, figure out what we need. */
5319                 lu_buf_free(ldata->ld_buf);
5320                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5321                                   ldata->ld_buf, XATTR_NAME_LINK);
5322                 if (rc < 0)
5323                         return rc;
5324                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
5325                 if (ldata->ld_buf->lb_buf == NULL)
5326                         return -ENOMEM;
5327                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5328                                   ldata->ld_buf, XATTR_NAME_LINK);
5329         }
5330         if (rc < 0)
5331                 return rc;
5332
5333         return linkea_init(ldata);
5334 }
5335
5336 /**
5337  * Given an MDT object, try to look up the full path to the object.
5338  * Part of the MDT layer implementation of lfs fid2path.
5339  *
5340  * \param[in]     info  Per-thread common data shared by MDT level handlers.
5341  * \param[in]     obj   Object to do path lookup of
5342  * \param[in,out] fp    User-provided struct to store path information
5343  *
5344  * \retval 0 Lookup successful, path information stored in fp
5345  * \retval -EAGAIN Lookup failed, usually because object is being moved
5346  * \retval negative errno if there was a problem
5347  */
5348 static int mdt_path_current(struct mdt_thread_info *info,
5349                             struct mdt_object *obj,
5350                             struct getinfo_fid2path *fp)
5351 {
5352         struct mdt_device       *mdt = info->mti_mdt;
5353         struct mdt_object       *mdt_obj;
5354         struct link_ea_header   *leh;
5355         struct link_ea_entry    *lee;
5356         struct lu_name          *tmpname = &info->mti_name;
5357         struct lu_fid           *tmpfid = &info->mti_tmp_fid1;
5358         struct lu_buf           *buf = &info->mti_big_buf;
5359         char                    *ptr;
5360         int                     reclen;
5361         struct linkea_data      ldata = { NULL };
5362         int                     rc = 0;
5363         bool                    first = true;
5364         ENTRY;
5365
5366         /* temp buffer for path element, the buffer will be finally freed
5367          * in mdt_thread_info_fini */
5368         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
5369         if (buf->lb_buf == NULL)
5370                 RETURN(-ENOMEM);
5371
5372         ldata.ld_buf = buf;
5373         ptr = fp->gf_path + fp->gf_pathlen - 1;
5374         *ptr = 0;
5375         --ptr;
5376         *tmpfid = fp->gf_fid = *mdt_object_fid(obj);
5377
5378         /* root FID only exists on MDT0, and fid2path should also ends at MDT0,
5379          * so checking root_fid can only happen on MDT0. */
5380         while (!lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid)) {
5381                 struct lu_buf           lmv_buf;
5382
5383                 mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
5384                 if (IS_ERR(mdt_obj))
5385                         GOTO(out, rc = PTR_ERR(mdt_obj));
5386
5387                 if (!mdt_object_exists(mdt_obj)) {
5388                         mdt_object_put(info->mti_env, mdt_obj);
5389                         GOTO(out, rc = -ENOENT);
5390                 }
5391
5392                 if (mdt_object_remote(mdt_obj)) {
5393                         mdt_object_put(info->mti_env, mdt_obj);
5394                         GOTO(remote_out, rc = -EREMOTE);
5395                 }
5396
5397                 rc = mdt_links_read(info, mdt_obj, &ldata);
5398                 if (rc != 0) {
5399                         mdt_object_put(info->mti_env, mdt_obj);
5400                         GOTO(out, rc);
5401                 }
5402
5403                 leh = buf->lb_buf;
5404                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
5405                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
5406                 /* If set, use link #linkno for path lookup, otherwise use
5407                    link #0.  Only do this for the final path element. */
5408                 if (first && fp->gf_linkno < leh->leh_reccount) {
5409                         int count;
5410                         for (count = 0; count < fp->gf_linkno; count++) {
5411                                 lee = (struct link_ea_entry *)
5412                                      ((char *)lee + reclen);
5413                                 linkea_entry_unpack(lee, &reclen, tmpname,
5414                                                     tmpfid);
5415                         }
5416                         if (fp->gf_linkno < leh->leh_reccount - 1)
5417                                 /* indicate to user there are more links */
5418                                 fp->gf_linkno++;
5419                 }
5420
5421                 lmv_buf.lb_buf = info->mti_xattr_buf;
5422                 lmv_buf.lb_len = sizeof(info->mti_xattr_buf);
5423                 /* Check if it is slave stripes */
5424                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5425                                   &lmv_buf, XATTR_NAME_LMV);
5426                 mdt_object_put(info->mti_env, mdt_obj);
5427                 if (rc > 0) {
5428                         union lmv_mds_md *lmm = lmv_buf.lb_buf;
5429
5430                         /* For slave stripes, get its master */
5431                         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE) {
5432                                 fp->gf_fid = *tmpfid;
5433                                 continue;
5434                         }
5435                 } else if (rc < 0 && rc != -ENODATA) {
5436                         GOTO(out, rc);
5437                 }
5438
5439                 rc = 0;
5440
5441                 /* Pack the name in the end of the buffer */
5442                 ptr -= tmpname->ln_namelen;
5443                 if (ptr - 1 <= fp->gf_path)
5444                         GOTO(out, rc = -EOVERFLOW);
5445                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
5446                 *(--ptr) = '/';
5447
5448                 /* keep the last resolved fid to the client, so the
5449                  * client will build the left path on another MDT for
5450                  * remote object */
5451                 fp->gf_fid = *tmpfid;
5452
5453                 first = false;
5454         }
5455
5456 remote_out:
5457         ptr++; /* skip leading / */
5458         memmove(fp->gf_path, ptr, fp->gf_path + fp->gf_pathlen - ptr);
5459
5460 out:
5461         RETURN(rc);
5462 }
5463
5464 /**
5465  * Given an MDT object, use mdt_path_current to get the path.
5466  * Essentially a wrapper to retry mdt_path_current a set number of times
5467  * if -EAGAIN is returned (usually because an object is being moved).
5468  *
5469  * Part of the MDT layer implementation of lfs fid2path.
5470  *
5471  * \param[in]     info  Per-thread common data shared by mdt level handlers.
5472  * \param[in]     obj   Object to do path lookup of
5473  * \param[in,out] fp    User-provided struct for arguments and to store path
5474  *                      information
5475  *
5476  * \retval 0 Lookup successful, path information stored in fp
5477  * \retval negative errno if there was a problem
5478  */
5479 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
5480                     struct getinfo_fid2path *fp)
5481 {
5482         struct mdt_device       *mdt = info->mti_mdt;
5483         int                     tries = 3;
5484         int                     rc = -EAGAIN;
5485         ENTRY;
5486
5487         if (fp->gf_pathlen < 3)
5488                 RETURN(-EOVERFLOW);
5489
5490         if (lu_fid_eq(&mdt->mdt_md_root_fid, mdt_object_fid(obj))) {
5491                 fp->gf_path[0] = '\0';
5492                 RETURN(0);
5493         }
5494
5495         /* Retry multiple times in case file is being moved */
5496         while (tries-- && rc == -EAGAIN)
5497                 rc = mdt_path_current(info, obj, fp);
5498
5499         RETURN(rc);
5500 }
5501
5502 /**
5503  * Get the full path of the provided FID, as of changelog record recno.
5504  *
5505  * This checks sanity and looks up object for user provided FID
5506  * before calling the actual path lookup code.
5507  *
5508  * Part of the MDT layer implementation of lfs fid2path.
5509  *
5510  * \param[in]     info  Per-thread common data shared by mdt level handlers.
5511  * \param[in,out] fp    User-provided struct for arguments and to store path
5512  *                      information
5513  *
5514  * \retval 0 Lookup successful, path information and recno stored in fp
5515  * \retval -ENOENT, object does not exist
5516  * \retval negative errno if there was a problem
5517  */
5518 static int mdt_fid2path(struct mdt_thread_info *info,
5519                         struct getinfo_fid2path *fp)
5520 {
5521         struct mdt_device *mdt = info->mti_mdt;
5522         struct mdt_object *obj;
5523         int    rc;
5524         ENTRY;
5525
5526         CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
5527                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
5528
5529         if (!fid_is_sane(&fp->gf_fid))
5530                 RETURN(-EINVAL);
5531
5532         if (!fid_is_namespace_visible(&fp->gf_fid)) {
5533                 CWARN("%s: "DFID" is invalid, sequence should be "
5534                       ">= "LPX64"\n", mdt_obd_name(mdt),
5535                       PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
5536                 RETURN(-EINVAL);
5537         }
5538
5539         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
5540         if (IS_ERR(obj)) {
5541                 rc = PTR_ERR(obj);
5542                 CDEBUG(D_IOCTL, "cannot find "DFID": rc = %d\n",
5543                        PFID(&fp->gf_fid), rc);
5544                 RETURN(rc);
5545         }
5546
5547         if (mdt_object_remote(obj))
5548                 rc = -EREMOTE;
5549         else if (!mdt_object_exists(obj))
5550                 rc = -ENOENT;
5551         else
5552                 rc = 0;
5553
5554         if (rc < 0) {
5555                 mdt_object_put(info->mti_env, obj);
5556                 CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n",
5557                        PFID(&fp->gf_fid), rc);
5558                 RETURN(rc);
5559         }
5560
5561         rc = mdt_path(info, obj, fp);
5562
5563         CDEBUG(D_INFO, "fid "DFID", path %s recno "LPX64" linkno %u\n",
5564                PFID(&fp->gf_fid), fp->gf_path, fp->gf_recno, fp->gf_linkno);
5565
5566         mdt_object_put(info->mti_env, obj);
5567
5568         RETURN(rc);
5569 }
5570
5571 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key,
5572                             void *val, int vallen)
5573 {
5574         struct getinfo_fid2path *fpout, *fpin;
5575         int rc = 0;
5576
5577         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
5578         fpout = val;
5579
5580         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5581                 lustre_swab_fid2path(fpin);
5582
5583         memcpy(fpout, fpin, sizeof(*fpin));
5584         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
5585                 RETURN(-EINVAL);
5586
5587         rc = mdt_fid2path(info, fpout);
5588         RETURN(rc);
5589 }
5590
5591 int mdt_get_info(struct tgt_session_info *tsi)
5592 {
5593         char    *key;
5594         int      keylen;
5595         __u32   *vallen;
5596         void    *valout;
5597         int      rc;
5598
5599         ENTRY;
5600
5601         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
5602         if (key == NULL) {
5603                 CDEBUG(D_IOCTL, "No GETINFO key\n");
5604                 RETURN(err_serious(-EFAULT));
5605         }
5606         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
5607                                       RCL_CLIENT);
5608
5609         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
5610         if (vallen == NULL) {
5611                 CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n",
5612                                 tgt_name(tsi->tsi_tgt));
5613                 RETURN(err_serious(-EFAULT));
5614         }
5615
5616         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
5617                              *vallen);
5618         rc = req_capsule_server_pack(tsi->tsi_pill);
5619         if (rc)
5620                 RETURN(err_serious(rc));
5621
5622         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
5623         if (valout == NULL) {
5624                 CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n",
5625                                 tgt_name(tsi->tsi_tgt));
5626                 RETURN(err_serious(-EFAULT));
5627         }
5628
5629         if (KEY_IS(KEY_FID2PATH)) {
5630                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
5631
5632                 rc = mdt_rpc_fid2path(info, key, valout, *vallen);
5633                 mdt_thread_info_fini(info);
5634         } else {
5635                 rc = -EINVAL;
5636         }
5637         RETURN(rc);
5638 }
5639
5640 /* Pass the ioc down */
5641 static int mdt_ioc_child(struct lu_env *env, struct mdt_device *mdt,
5642                          unsigned int cmd, int len, void *data)
5643 {
5644         struct lu_context ioctl_session;
5645         struct md_device *next = mdt->mdt_child;
5646         int rc;
5647         ENTRY;
5648
5649         rc = lu_context_init(&ioctl_session, LCT_SERVER_SESSION);
5650         if (rc)
5651                 RETURN(rc);
5652         ioctl_session.lc_thread = (struct ptlrpc_thread *)current;
5653         lu_context_enter(&ioctl_session);
5654         env->le_ses = &ioctl_session;
5655
5656         LASSERT(next->md_ops->mdo_iocontrol);
5657         rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data);
5658
5659         lu_context_exit(&ioctl_session);
5660         lu_context_fini(&ioctl_session);
5661         RETURN(rc);
5662 }
5663
5664 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
5665 {
5666         struct obd_ioctl_data *data = karg;
5667         struct lu_fid *fid;
5668         __u64 version;
5669         struct mdt_object *obj;
5670         struct mdt_lock_handle  *lh;
5671         int rc;
5672         ENTRY;
5673
5674         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
5675             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
5676                 RETURN(-EINVAL);
5677
5678         fid = (struct lu_fid *)data->ioc_inlbuf1;
5679
5680         if (!fid_is_sane(fid))
5681                 RETURN(-EINVAL);
5682
5683         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
5684
5685         lh = &mti->mti_lh[MDT_LH_PARENT];
5686         mdt_lock_reg_init(lh, LCK_CR);
5687
5688         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
5689         if (IS_ERR(obj))
5690                 RETURN(PTR_ERR(obj));
5691
5692         if (mdt_object_remote(obj)) {
5693                 rc = -EREMOTE;
5694                 /**
5695                  * before calling version get the correct MDS should be
5696                  * fid, this is error to find remote object here
5697                  */
5698                 CERROR("nonlocal object "DFID"\n", PFID(fid));
5699         } else if (!mdt_object_exists(obj)) {
5700                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
5701                 rc = -ENOENT;
5702         } else {
5703                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
5704                *(__u64 *)data->ioc_inlbuf2 = version;
5705                 rc = 0;
5706         }
5707         mdt_object_unlock_put(mti, obj, lh, 1);
5708         RETURN(rc);
5709 }
5710
5711 /* ioctls on obd dev */
5712 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
5713                          void *karg, void *uarg)
5714 {
5715         struct lu_env      env;
5716         struct obd_device *obd = exp->exp_obd;
5717         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
5718         struct dt_device  *dt = mdt->mdt_bottom;
5719         int rc;
5720
5721         ENTRY;
5722         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
5723         rc = lu_env_init(&env, LCT_MD_THREAD);
5724         if (rc)
5725                 RETURN(rc);
5726
5727         switch (cmd) {
5728         case OBD_IOC_SYNC:
5729                 rc = mdt_device_sync(&env, mdt);
5730                 break;
5731         case OBD_IOC_SET_READONLY:
5732                 rc = dt->dd_ops->dt_ro(&env, dt);
5733                 break;
5734         case OBD_IOC_ABORT_RECOVERY:
5735                 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
5736                 obd->obd_abort_recovery = 1;
5737                 target_stop_recovery_thread(obd);
5738                 rc = 0;
5739                 break;
5740         case OBD_IOC_CHANGELOG_REG:
5741         case OBD_IOC_CHANGELOG_DEREG:
5742         case OBD_IOC_CHANGELOG_CLEAR:
5743                 rc = mdt_ioc_child(&env, mdt, cmd, len, karg);
5744                 break;
5745         case OBD_IOC_START_LFSCK: {
5746                 struct md_device *next = mdt->mdt_child;
5747                 struct obd_ioctl_data *data = karg;
5748                 struct lfsck_start_param lsp;
5749
5750                 if (unlikely(data == NULL)) {
5751                         rc = -EINVAL;
5752                         break;
5753                 }
5754
5755                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
5756                 lsp.lsp_index_valid = 0;
5757                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
5758                 break;
5759         }
5760         case OBD_IOC_STOP_LFSCK: {
5761                 struct md_device        *next = mdt->mdt_child;
5762                 struct obd_ioctl_data   *data = karg;
5763                 struct lfsck_stop        stop;
5764
5765                 stop.ls_status = LS_STOPPED;
5766                 /* Old lfsck utils may pass NULL @stop. */
5767                 if (data->ioc_inlbuf1 == NULL)
5768                         stop.ls_flags = 0;
5769                 else
5770                         stop.ls_flags =
5771                         ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
5772
5773                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
5774                 break;
5775         }
5776         case OBD_IOC_GET_OBJ_VERSION: {
5777                 struct mdt_thread_info *mti;
5778                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5779                 memset(mti, 0, sizeof *mti);
5780                 mti->mti_env = &env;
5781                 mti->mti_mdt = mdt;
5782                 mti->mti_exp = exp;
5783
5784                 rc = mdt_ioc_version_get(mti, karg);
5785                 break;
5786         }
5787         case OBD_IOC_CATLOGLIST: {
5788                 struct mdt_thread_info *mti;
5789
5790                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5791                 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
5792                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
5793                                        &mti->mti_tmp_fid1);
5794                 break;
5795          }
5796         default:
5797                 rc = -EOPNOTSUPP;
5798                 CERROR("%s: Not supported cmd = %d, rc = %d\n",
5799                         mdt_obd_name(mdt), cmd, rc);
5800         }
5801
5802         lu_env_fini(&env);
5803         RETURN(rc);
5804 }
5805
5806 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
5807 {
5808         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
5809         int rc;
5810         ENTRY;
5811
5812         if (!mdt->mdt_skip_lfsck) {
5813                 struct lfsck_start_param lsp;
5814
5815                 lsp.lsp_start = NULL;
5816                 lsp.lsp_index_valid = 0;
5817                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
5818                                                            OBD_IOC_START_LFSCK,
5819                                                            0, &lsp);
5820                 if (rc != 0 && rc != -EALREADY)
5821                         CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
5822                               mdt_obd_name(mdt), rc);
5823         }
5824
5825         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
5826         RETURN(rc);
5827 }
5828
5829 static int mdt_obd_postrecov(struct obd_device *obd)
5830 {
5831         struct lu_env env;
5832         int rc;
5833
5834         rc = lu_env_init(&env, LCT_MD_THREAD);
5835         if (rc)
5836                 RETURN(rc);
5837         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
5838         lu_env_fini(&env);
5839         return rc;
5840 }
5841
5842 static struct obd_ops mdt_obd_device_ops = {
5843         .o_owner          = THIS_MODULE,
5844         .o_set_info_async = mdt_obd_set_info_async,
5845         .o_connect        = mdt_obd_connect,
5846         .o_reconnect      = mdt_obd_reconnect,
5847         .o_disconnect     = mdt_obd_disconnect,
5848         .o_init_export    = mdt_init_export,
5849         .o_destroy_export = mdt_destroy_export,
5850         .o_iocontrol      = mdt_iocontrol,
5851         .o_postrecov      = mdt_obd_postrecov,
5852 };
5853
5854 static struct lu_device* mdt_device_fini(const struct lu_env *env,
5855                                          struct lu_device *d)
5856 {
5857         struct mdt_device *m = mdt_dev(d);
5858         ENTRY;
5859
5860         mdt_fini(env, m);
5861         RETURN(NULL);
5862 }
5863
5864 static struct lu_device *mdt_device_free(const struct lu_env *env,
5865                                          struct lu_device *d)
5866 {
5867         struct mdt_device *m = mdt_dev(d);
5868         ENTRY;
5869
5870         lu_device_fini(&m->mdt_lu_dev);
5871         OBD_FREE_PTR(m);
5872
5873         RETURN(NULL);
5874 }
5875
5876 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
5877                                           struct lu_device_type *t,
5878                                           struct lustre_cfg *cfg)
5879 {
5880         struct lu_device  *l;
5881         struct mdt_device *m;
5882
5883         OBD_ALLOC_PTR(m);
5884         if (m != NULL) {
5885                 int rc;
5886
5887                 l = &m->mdt_lu_dev;
5888                 rc = mdt_init0(env, m, t, cfg);
5889                 if (rc != 0) {
5890                         mdt_device_free(env, l);
5891                         l = ERR_PTR(rc);
5892                         return l;
5893                 }
5894         } else
5895                 l = ERR_PTR(-ENOMEM);
5896         return l;
5897 }
5898
5899 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
5900 LU_KEY_INIT(mdt, struct mdt_thread_info);
5901
5902 static void mdt_key_fini(const struct lu_context *ctx,
5903                          struct lu_context_key *key, void* data)
5904 {
5905         struct mdt_thread_info *info = data;
5906
5907         if (info->mti_big_lmm) {
5908                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
5909                 info->mti_big_lmm = NULL;
5910                 info->mti_big_lmmsize = 0;
5911         }
5912         OBD_FREE_PTR(info);
5913 }
5914
5915 /* context key: mdt_thread_key */
5916 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
5917
5918 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
5919 {
5920         return lu_ucred(info->mti_env);
5921 }
5922
5923 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
5924 {
5925         return lu_ucred_check(info->mti_env);
5926 }
5927
5928 /**
5929  * Enable/disable COS (Commit On Sharing).
5930  *
5931  * Set/Clear the COS flag in mdt options.
5932  *
5933  * \param mdt mdt device
5934  * \param val 0 disables COS, other values enable COS
5935  */
5936 void mdt_enable_cos(struct mdt_device *mdt, int val)
5937 {
5938         struct lu_env env;
5939         int rc;
5940
5941         mdt->mdt_opts.mo_cos = !!val;
5942         rc = lu_env_init(&env, LCT_LOCAL);
5943         if (unlikely(rc != 0)) {
5944                 CWARN("%s: lu_env initialization failed, cannot "
5945                       "sync: rc = %d\n", mdt_obd_name(mdt), rc);
5946                 return;
5947         }
5948         mdt_device_sync(&env, mdt);
5949         lu_env_fini(&env);
5950 }
5951
5952 /**
5953  * Check COS (Commit On Sharing) status.
5954  *
5955  * Return COS flag status.
5956  *
5957  * \param mdt mdt device
5958  */
5959 int mdt_cos_is_enabled(struct mdt_device *mdt)
5960 {
5961         return mdt->mdt_opts.mo_cos != 0;
5962 }
5963
5964 static struct lu_device_type_operations mdt_device_type_ops = {
5965         .ldto_device_alloc = mdt_device_alloc,
5966         .ldto_device_free  = mdt_device_free,
5967         .ldto_device_fini  = mdt_device_fini
5968 };
5969
5970 static struct lu_device_type mdt_device_type = {
5971         .ldt_tags     = LU_DEVICE_MD,
5972         .ldt_name     = LUSTRE_MDT_NAME,
5973         .ldt_ops      = &mdt_device_type_ops,
5974         .ldt_ctx_tags = LCT_MD_THREAD
5975 };
5976
5977 static int __init mdt_init(void)
5978 {
5979         int rc;
5980
5981         CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") ==
5982                  FID_NOBRACE_LEN + 1);
5983         CLASSERT(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") ==
5984                  FID_LEN + 1);
5985         rc = lu_kmem_init(mdt_caches);
5986         if (rc)
5987                 return rc;
5988
5989         rc = mds_mod_init();
5990         if (rc)
5991                 GOTO(lu_fini, rc);
5992
5993         rc = class_register_type(&mdt_obd_device_ops, NULL, true, NULL,
5994                                  LUSTRE_MDT_NAME, &mdt_device_type);
5995         if (rc)
5996                 GOTO(mds_fini, rc);
5997 lu_fini:
5998         if (rc)
5999                 lu_kmem_fini(mdt_caches);
6000 mds_fini:
6001         if (rc)
6002                 mds_mod_exit();
6003         return rc;
6004 }
6005
6006 static void __exit mdt_exit(void)
6007 {
6008         class_unregister_type(LUSTRE_MDT_NAME);
6009         mds_mod_exit();
6010         lu_kmem_fini(mdt_caches);
6011 }
6012
6013 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
6014 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
6015 MODULE_VERSION(LUSTRE_VERSION_STRING);
6016 MODULE_LICENSE("GPL");
6017
6018 module_init(mdt_init);
6019 module_exit(mdt_exit);