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.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2010, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/mdt/mdt_handler.c
  33  *
  34  * Lustre Metadata Target (mdt) request handler
  35  *
  36  * Author: Peter Braam <braam@clusterfs.com>
  37  * Author: Andreas Dilger <adilger@clusterfs.com>
  38  * Author: Phil Schwan <phil@clusterfs.com>
  39  * Author: Mike Shaver <shaver@clusterfs.com>
  40  * Author: Nikita Danilov <nikita@clusterfs.com>
  41  * Author: Huang Hua <huanghua@clusterfs.com>
  42  * Author: Yury Umanets <umka@clusterfs.com>
  43  */
  44
  45 #define DEBUG_SUBSYSTEM S_MDS
  46
  47 #include <linux/module.h>
  48 #include <linux/pagemap.h>
  49
  50 #include <dt_object.h>
  51 #include <lustre_acl.h>
  52 #include <lustre_export.h>
  53 #include <uapi/linux/lustre/lustre_ioctl.h>
  54 #include <lustre_lfsck.h>
  55 #include <lustre_log.h>
  56 #include <lustre_nodemap.h>
  57 #include <lustre_mds.h>
  58 #include <uapi/linux/lustre/lustre_param.h>
  59 #include <lustre_quota.h>
  60 #include <lustre_swab.h>
  61 #include <lustre_lmv.h>
  62 #include <obd.h>
  63 #include <obd_support.h>
  64 #include <lustre_barrier.h>
  65 #include <obd_cksum.h>
  66 #include <llog_swab.h>
  67
  68 #include "mdt_internal.h"
  69
  70 static unsigned int max_mod_rpcs_per_client = 8;
  71 module_param(max_mod_rpcs_per_client, uint, 0644);
  72 MODULE_PARM_DESC(max_mod_rpcs_per_client, "maximum number of modify RPCs in flight allowed per client");
  73
  74 mdl_mode_t mdt_mdl_lock_modes[] = {
  75         [LCK_MINMODE] = MDL_MINMODE,
  76         [LCK_EX]      = MDL_EX,
  77         [LCK_PW]      = MDL_PW,
  78         [LCK_PR]      = MDL_PR,
  79         [LCK_CW]      = MDL_CW,
  80         [LCK_CR]      = MDL_CR,
  81         [LCK_NL]      = MDL_NL,
  82         [LCK_GROUP]   = MDL_GROUP
  83 };
  84
  85 enum ldlm_mode mdt_dlm_lock_modes[] = {
  86         [MDL_MINMODE]   = LCK_MINMODE,
  87         [MDL_EX]        = LCK_EX,
  88         [MDL_PW]        = LCK_PW,
  89         [MDL_PR]        = LCK_PR,
  90         [MDL_CW]        = LCK_CW,
  91         [MDL_CR]        = LCK_CR,
  92         [MDL_NL]        = LCK_NL,
  93         [MDL_GROUP]     = LCK_GROUP
  94 };
  95
  96 static struct mdt_device *mdt_dev(struct lu_device *d);
  97
  98 static const struct lu_object_operations mdt_obj_ops;
  99
 100 /* Slab for MDT object allocation */
 101 static struct kmem_cache *mdt_object_kmem;
 102
 103 /* For HSM restore handles */
 104 struct kmem_cache *mdt_hsm_cdt_kmem;
 105
 106 /* For HSM request handles */
 107 struct kmem_cache *mdt_hsm_car_kmem;
 108
 109 static struct lu_kmem_descr mdt_caches[] = {
 110         {
 111                 .ckd_cache = &mdt_object_kmem,
 112                 .ckd_name  = "mdt_obj",
 113                 .ckd_size  = sizeof(struct mdt_object)
 114         },
 115         {
 116                 .ckd_cache      = &mdt_hsm_cdt_kmem,
 117                 .ckd_name       = "mdt_cdt_restore_handle",
 118                 .ckd_size       = sizeof(struct cdt_restore_handle)
 119         },
 120         {
 121                 .ckd_cache      = &mdt_hsm_car_kmem,
 122                 .ckd_name       = "mdt_cdt_agent_req",
 123                 .ckd_size       = sizeof(struct cdt_agent_req)
 124         },
 125         {
 126                 .ckd_cache = NULL
 127         }
 128 };
 129
 130 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag)
 131 {
 132         if (!rep)
 133                 return 0;
 134         return rep->lock_policy_res1 & op_flag;
 135 }
 136
 137 void mdt_clear_disposition(struct mdt_thread_info *info,
 138                            struct ldlm_reply *rep, __u64 op_flag)
 139 {
 140         if (info) {
 141                 info->mti_opdata &= ~op_flag;
 142                 tgt_opdata_clear(info->mti_env, op_flag);
 143         }
 144         if (rep)
 145                 rep->lock_policy_res1 &= ~op_flag;
 146 }
 147
 148 void mdt_set_disposition(struct mdt_thread_info *info,
 149                          struct ldlm_reply *rep, __u64 op_flag)
 150 {
 151         if (info) {
 152                 info->mti_opdata |= op_flag;
 153                 tgt_opdata_set(info->mti_env, op_flag);
 154         }
 155         if (rep)
 156                 rep->lock_policy_res1 |= op_flag;
 157 }
 158
 159 void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm)
 160 {
 161         lh->mlh_pdo_hash = 0;
 162         lh->mlh_reg_mode = lm;
 163         lh->mlh_rreg_mode = lm;
 164         lh->mlh_type = MDT_REG_LOCK;
 165 }
 166
 167 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode,
 168                        const struct lu_name *lname)
 169 {
 170         lh->mlh_reg_mode = lock_mode;
 171         lh->mlh_pdo_mode = LCK_MINMODE;
 172         lh->mlh_rreg_mode = lock_mode;
 173         lh->mlh_type = MDT_PDO_LOCK;
 174
 175         if (lu_name_is_valid(lname)) {
 176                 lh->mlh_pdo_hash = ll_full_name_hash(NULL, lname->ln_name,
 177                                                      lname->ln_namelen);
 178                 /* XXX Workaround for LU-2856
 179                  *
 180                  * Zero is a valid return value of full_name_hash, but
 181                  * several users of mlh_pdo_hash assume a non-zero
 182                  * hash value. We therefore map zero onto an
 183                  * arbitrary, but consistent value (1) to avoid
 184                  * problems further down the road. */
 185                 if (unlikely(lh->mlh_pdo_hash == 0))
 186                         lh->mlh_pdo_hash = 1;
 187         } else {
 188                 lh->mlh_pdo_hash = 0;
 189         }
 190 }
 191
 192 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 193                               struct mdt_lock_handle *lh)
 194 {
 195         mdl_mode_t mode;
 196         ENTRY;
 197
 198         /*
 199          * Any dir access needs couple of locks:
 200          *
 201          * 1) on part of dir we gonna take lookup/modify;
 202          *
 203          * 2) on whole dir to protect it from concurrent splitting and/or to
 204          * flush client's cache for readdir().
 205          *
 206          * so, for a given mode and object this routine decides what lock mode
 207          * to use for lock #2:
 208          *
 209          * 1) if caller's gonna lookup in dir then we need to protect dir from
 210          * being splitted only - LCK_CR
 211          *
 212          * 2) if caller's gonna modify dir then we need to protect dir from
 213          * being splitted and to flush cache - LCK_CW
 214          *
 215          * 3) if caller's gonna modify dir and that dir seems ready for
 216          * splitting then we need to protect it from any type of access
 217          * (lookup/modify/split) - LCK_EX --bzzz
 218          */
 219
 220         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 221         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 222
 223         /*
 224          * Ask underlaying level its opinion about preferable PDO lock mode
 225          * having access type passed as regular lock mode:
 226          *
 227          * - MDL_MINMODE means that lower layer does not want to specify lock
 228          * mode;
 229          *
 230          * - MDL_NL means that no PDO lock should be taken. This is used in some
 231          * cases. Say, for non-splittable directories no need to use PDO locks
 232          * at all.
 233          */
 234         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 235                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 236
 237         if (mode != MDL_MINMODE) {
 238                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 239         } else {
 240                 /*
 241                  * Lower layer does not want to specify locking mode. We do it
 242                  * our selves. No special protection is needed, just flush
 243                  * client's cache on modification and allow concurrent
 244                  * mondification.
 245                  */
 246                 switch (lh->mlh_reg_mode) {
 247                 case LCK_EX:
 248                         lh->mlh_pdo_mode = LCK_EX;
 249                         break;
 250                 case LCK_PR:
 251                         lh->mlh_pdo_mode = LCK_CR;
 252                         break;
 253                 case LCK_PW:
 254                         lh->mlh_pdo_mode = LCK_CW;
 255                         break;
 256                 default:
 257                         CERROR("Not expected lock type (0x%x)\n",
 258                                (int)lh->mlh_reg_mode);
 259                         LBUG();
 260                 }
 261         }
 262
 263         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 264         EXIT;
 265 }
 266
 267 static int mdt_lookup_fileset(struct mdt_thread_info *info, const char *fileset,
 268                               struct lu_fid *fid)
 269 {
 270         struct mdt_device *mdt = info->mti_mdt;
 271         struct lu_name *lname = &info->mti_name;
 272         char *filename = info->mti_filename;
 273         struct mdt_object *parent;
 274         u32 mode;
 275         int rc = 0;
 276
 277         LASSERT(!info->mti_cross_ref);
 278
 279         /*
 280          * We may want to allow this to mount a completely separate
 281          * fileset from the MDT in the future, but keeping it to
 282          * ROOT/ only for now avoid potential security issues.
 283          */
 284         *fid = mdt->mdt_md_root_fid;
 285
 286         while (rc == 0 && fileset != NULL && *fileset != '\0') {
 287                 const char *s1 = fileset;
 288                 const char *s2;
 289
 290                 while (*++s1 == '/')
 291                         ;
 292                 s2 = s1;
 293                 while (*s2 != '/' && *s2 != '\0')
 294                         s2++;
 295
 296                 if (s2 == s1)
 297                         break;
 298
 299                 fileset = s2;
 300
 301                 lname->ln_namelen = s2 - s1;
 302                 if (lname->ln_namelen > NAME_MAX) {
 303                         rc = -EINVAL;
 304                         break;
 305                 }
 306
 307                 /* reject .. as a path component */
 308                 if (lname->ln_namelen == 2 &&
 309                     strncmp(s1, "..", 2) == 0) {
 310                         rc = -EINVAL;
 311                         break;
 312                 }
 313
 314                 strncpy(filename, s1, lname->ln_namelen);
 315                 filename[lname->ln_namelen] = '\0';
 316                 lname->ln_name = filename;
 317
 318                 parent = mdt_object_find(info->mti_env, mdt, fid);
 319                 if (IS_ERR(parent)) {
 320                         rc = PTR_ERR(parent);
 321                         break;
 322                 }
 323                 /* Only got the fid of this obj by name */
 324                 fid_zero(fid);
 325                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
 326                                 fid, &info->mti_spec);
 327                 mdt_object_put(info->mti_env, parent);
 328         }
 329         if (!rc) {
 330                 parent = mdt_object_find(info->mti_env, mdt, fid);
 331                 if (IS_ERR(parent))
 332                         rc = PTR_ERR(parent);
 333                 else {
 334                         mode = lu_object_attr(&parent->mot_obj);
 335                         mdt_object_put(info->mti_env, parent);
 336                         if (!S_ISDIR(mode))
 337                                 rc = -ENOTDIR;
 338                 }
 339         }
 340
 341         return rc;
 342 }
 343
 344 static int mdt_get_root(struct tgt_session_info *tsi)
 345 {
 346         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
 347         struct mdt_device       *mdt = info->mti_mdt;
 348         struct mdt_body         *repbody;
 349         char                    *fileset = NULL, *buffer = NULL;
 350         int                      rc;
 351         struct obd_export       *exp = info->mti_exp;
 352         char                    *nodemap_fileset;
 353
 354         ENTRY;
 355
 356         rc = mdt_check_ucred(info);
 357         if (rc)
 358                 GOTO(out, rc = err_serious(rc));
 359
 360         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GET_ROOT_PACK))
 361                 GOTO(out, rc = err_serious(-ENOMEM));
 362
 363         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 364         if (req_capsule_get_size(info->mti_pill, &RMF_NAME, RCL_CLIENT) > 0) {
 365                 fileset = req_capsule_client_get(info->mti_pill, &RMF_NAME);
 366                 if (fileset == NULL)
 367                         GOTO(out, rc = err_serious(-EFAULT));
 368         }
 369
 370         nodemap_fileset = nodemap_get_fileset(exp->exp_target_data.ted_nodemap);
 371         if (nodemap_fileset && nodemap_fileset[0]) {
 372                 CDEBUG(D_INFO, "nodemap fileset is %s\n", nodemap_fileset);
 373                 if (fileset) {
 374                         /* consider fileset from client as a sub-fileset
 375                          * of the nodemap one */
 376                         OBD_ALLOC(buffer, PATH_MAX + 1);
 377                         if (buffer == NULL)
 378                                 GOTO(out, rc = err_serious(-ENOMEM));
 379                         if (snprintf(buffer, PATH_MAX + 1, "%s/%s",
 380                                      nodemap_fileset, fileset) >= PATH_MAX + 1)
 381                                 GOTO(out, rc = err_serious(-EINVAL));
 382                         fileset = buffer;
 383                 } else {
 384                         /* enforce fileset as specified in the nodemap */
 385                         fileset = nodemap_fileset;
 386                 }
 387         }
 388
 389         if (fileset) {
 390                 CDEBUG(D_INFO, "Getting fileset %s\n", fileset);
 391                 rc = mdt_lookup_fileset(info, fileset, &repbody->mbo_fid1);
 392                 if (rc < 0)
 393                         GOTO(out, rc = err_serious(rc));
 394         } else {
 395                 repbody->mbo_fid1 = mdt->mdt_md_root_fid;
 396         }
 397         repbody->mbo_valid |= OBD_MD_FLID;
 398
 399         EXIT;
 400 out:
 401         mdt_thread_info_fini(info);
 402         if (buffer)
 403                 OBD_FREE(buffer, PATH_MAX+1);
 404         return rc;
 405 }
 406
 407 static int mdt_statfs(struct tgt_session_info *tsi)
 408 {
 409         struct ptlrpc_request *req = tgt_ses_req(tsi);
 410         struct mdt_thread_info *info = tsi2mdt_info(tsi);
 411         struct mdt_device *mdt = info->mti_mdt;
 412         struct tg_grants_data *tgd = &mdt->mdt_lut.lut_tgd;
 413         struct md_device *next = mdt->mdt_child;
 414         struct ptlrpc_service_part *svcpt;
 415         struct obd_statfs *osfs;
 416         struct mdt_body *reqbody = NULL;
 417         struct mdt_statfs_cache *msf;
 418         ktime_t kstart = ktime_get();
 419         int rc;
 420
 421         ENTRY;
 422
 423         svcpt = req->rq_rqbd->rqbd_svcpt;
 424
 425         /* This will trigger a watchdog timeout */
 426         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 427                          (MDT_SERVICE_WATCHDOG_FACTOR *
 428                           at_get(&svcpt->scp_at_estimate)) + 1);
 429
 430         rc = mdt_check_ucred(info);
 431         if (rc)
 432                 GOTO(out, rc = err_serious(rc));
 433
 434         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
 435                 GOTO(out, rc = err_serious(-ENOMEM));
 436
 437         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 438         if (!osfs)
 439                 GOTO(out, rc = -EPROTO);
 440
 441         if (mdt_is_sum_statfs_client(req->rq_export) &&
 442                 lustre_packed_msg_size(req->rq_reqmsg) ==
 443                 req_capsule_fmt_size(req->rq_reqmsg->lm_magic,
 444                                      &RQF_MDS_STATFS_NEW, RCL_CLIENT)) {
 445                 req_capsule_extend(info->mti_pill, &RQF_MDS_STATFS_NEW);
 446                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
 447         }
 448
 449         if (reqbody && reqbody->mbo_valid & OBD_MD_FLAGSTATFS)
 450                 msf = &mdt->mdt_sum_osfs;
 451         else
 452                 msf = &mdt->mdt_osfs;
 453
 454         if (msf->msf_age + OBD_STATFS_CACHE_SECONDS <= ktime_get_seconds()) {
 455                         /** statfs data is too old, get up-to-date one */
 456                         if (reqbody && reqbody->mbo_valid & OBD_MD_FLAGSTATFS)
 457                                 rc = next->md_ops->mdo_statfs(info->mti_env,
 458                                                               next, osfs);
 459                         else
 460                                 rc = dt_statfs(info->mti_env, mdt->mdt_bottom,
 461                                                osfs);
 462                         if (rc)
 463                                 GOTO(out, rc);
 464                         spin_lock(&mdt->mdt_lock);
 465                         msf->msf_osfs = *osfs;
 466                         msf->msf_age = ktime_get_seconds();
 467                         spin_unlock(&mdt->mdt_lock);
 468         } else {
 469                         /** use cached statfs data */
 470                         spin_lock(&mdt->mdt_lock);
 471                         *osfs = msf->msf_osfs;
 472                         spin_unlock(&mdt->mdt_lock);
 473         }
 474
 475         /* at least try to account for cached pages.  its still racy and
 476          * might be under-reporting if clients haven't announced their
 477          * caches with brw recently */
 478         CDEBUG(D_SUPER | D_CACHE, "blocks cached %llu granted %llu"
 479                " pending %llu free %llu avail %llu\n",
 480                tgd->tgd_tot_dirty, tgd->tgd_tot_granted,
 481                tgd->tgd_tot_pending,
 482                osfs->os_bfree << tgd->tgd_blockbits,
 483                osfs->os_bavail << tgd->tgd_blockbits);
 484
 485         osfs->os_bavail -= min_t(u64, osfs->os_bavail,
 486                                  ((tgd->tgd_tot_dirty + tgd->tgd_tot_pending +
 487                                    osfs->os_bsize - 1) >> tgd->tgd_blockbits));
 488
 489         tgt_grant_sanity_check(mdt->mdt_lu_dev.ld_obd, __func__);
 490         CDEBUG(D_CACHE, "%llu blocks: %llu free, %llu avail; "
 491                "%llu objects: %llu free; state %x\n",
 492                osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
 493                osfs->os_files, osfs->os_ffree, osfs->os_state);
 494
 495         if (!exp_grant_param_supp(tsi->tsi_exp) &&
 496             tgd->tgd_blockbits > COMPAT_BSIZE_SHIFT) {
 497                 /* clients which don't support OBD_CONNECT_GRANT_PARAM
 498                  * should not see a block size > page size, otherwise
 499                  * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
 500                  * block size which is the biggest block size known to work
 501                  * with all client's page size. */
 502                 osfs->os_blocks <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
 503                 osfs->os_bfree  <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
 504                 osfs->os_bavail <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
 505                 osfs->os_bsize = 1 << COMPAT_BSIZE_SHIFT;
 506         }
 507         if (rc == 0)
 508                 mdt_counter_incr(req, LPROC_MDT_STATFS,
 509                                  ktime_us_delta(ktime_get(), kstart));
 510 out:
 511         mdt_thread_info_fini(info);
 512         RETURN(rc);
 513 }
 514
 515 __u32 mdt_lmm_dom_entry_check(struct lov_mds_md *lmm, int *is_dom_only)
 516 {
 517         struct lov_comp_md_v1 *comp_v1;
 518         struct lov_mds_md *v1;
 519         __u32 off;
 520         __u32 dom_stripesize = 0;
 521         int i;
 522         bool has_ost_stripes = false;
 523
 524         ENTRY;
 525
 526         if (is_dom_only)
 527                 *is_dom_only = 0;
 528
 529         if (le32_to_cpu(lmm->lmm_magic) != LOV_MAGIC_COMP_V1)
 530                 RETURN(0);
 531
 532         comp_v1 = (struct lov_comp_md_v1 *)lmm;
 533         off = le32_to_cpu(comp_v1->lcm_entries[0].lcme_offset);
 534         v1 = (struct lov_mds_md *)((char *)comp_v1 + off);
 535
 536         /* Fast check for DoM entry with no mirroring, should be the first */
 537         if (le16_to_cpu(comp_v1->lcm_mirror_count) == 0 &&
 538             lov_pattern(le32_to_cpu(v1->lmm_pattern)) != LOV_PATTERN_MDT)
 539                 RETURN(0);
 540
 541         /* check all entries otherwise */
 542         for (i = 0; i < le16_to_cpu(comp_v1->lcm_entry_count); i++) {
 543                 struct lov_comp_md_entry_v1 *lcme;
 544
 545                 lcme = &comp_v1->lcm_entries[i];
 546                 if (!(le32_to_cpu(lcme->lcme_flags) & LCME_FL_INIT))
 547                         continue;
 548
 549                 off = le32_to_cpu(lcme->lcme_offset);
 550                 v1 = (struct lov_mds_md *)((char *)comp_v1 + off);
 551
 552                 if (lov_pattern(le32_to_cpu(v1->lmm_pattern)) ==
 553                     LOV_PATTERN_MDT)
 554                         dom_stripesize = le32_to_cpu(v1->lmm_stripe_size);
 555                 else
 556                         has_ost_stripes = true;
 557
 558                 if (dom_stripesize && has_ost_stripes)
 559                         RETURN(dom_stripesize);
 560         }
 561         /* DoM-only case exits here */
 562         if (is_dom_only && dom_stripesize)
 563                 *is_dom_only = 1;
 564         RETURN(dom_stripesize);
 565 }
 566
 567 /**
 568  * Pack size attributes into the reply.
 569  */
 570 int mdt_pack_size2body(struct mdt_thread_info *info,
 571                         const struct lu_fid *fid, struct lustre_handle *lh)
 572 {
 573         struct mdt_body *b;
 574         struct md_attr *ma = &info->mti_attr;
 575         __u32 dom_stripe;
 576         bool dom_lock = false;
 577
 578         ENTRY;
 579
 580         LASSERT(ma->ma_attr.la_valid & LA_MODE);
 581
 582         if (!S_ISREG(ma->ma_attr.la_mode) ||
 583             !(ma->ma_valid & MA_LOV && ma->ma_lmm != NULL))
 584                 RETURN(-ENODATA);
 585
 586         dom_stripe = mdt_lmm_dom_stripesize(ma->ma_lmm);
 587         /* no DoM stripe, no size in reply */
 588         if (!dom_stripe)
 589                 RETURN(-ENOENT);
 590
 591         if (lustre_handle_is_used(lh)) {
 592                 struct ldlm_lock *lock;
 593
 594                 lock = ldlm_handle2lock(lh);
 595                 if (lock != NULL) {
 596                         dom_lock = ldlm_has_dom(lock);
 597                         LDLM_LOCK_PUT(lock);
 598                 }
 599         }
 600
 601         /* no DoM lock, no size in reply */
 602         if (!dom_lock)
 603                 RETURN(0);
 604
 605         /* Either DoM lock exists or LMM has only DoM stripe then
 606          * return size on body. */
 607         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 608
 609         mdt_dom_object_size(info->mti_env, info->mti_mdt, fid, b, dom_lock);
 610         RETURN(0);
 611 }
 612
 613 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
 614 /*
 615  * Pack ACL data into the reply. UIDs/GIDs are mapped and filtered by nodemap.
 616  *
 617  * \param       info    thread info object
 618  * \param       repbody reply to pack ACLs into
 619  * \param       o       mdt object of file to examine
 620  * \param       nodemap nodemap of client to reply to
 621  * \retval      0       success
 622  * \retval      -errno  error getting or parsing ACL from disk
 623  */
 624 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
 625                       struct mdt_object *o, struct lu_nodemap *nodemap)
 626 {
 627         const struct lu_env     *env = info->mti_env;
 628         struct md_object        *next = mdt_object_child(o);
 629         struct lu_buf           *buf = &info->mti_buf;
 630         struct mdt_device       *mdt = info->mti_mdt;
 631         struct req_capsule *pill = info->mti_pill;
 632         int rc;
 633
 634         ENTRY;
 635
 636         buf->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
 637         buf->lb_len = req_capsule_get_size(pill, &RMF_ACL, RCL_SERVER);
 638         if (buf->lb_len == 0)
 639                 RETURN(0);
 640
 641 again:
 642         rc = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_ACCESS);
 643         if (rc < 0) {
 644                 if (rc == -ENODATA) {
 645                         repbody->mbo_aclsize = 0;
 646                         repbody->mbo_valid |= OBD_MD_FLACL;
 647                         rc = 0;
 648                 } else if (rc == -EOPNOTSUPP) {
 649                         rc = 0;
 650                 } else {
 651                         if (rc == -ERANGE &&
 652                             exp_connect_large_acl(info->mti_exp) &&
 653                             buf->lb_buf != info->mti_big_acl) {
 654                                 if (info->mti_big_acl == NULL) {
 655                                         info->mti_big_aclsize =
 656                                                         min_t(unsigned int,
 657                                                               mdt->mdt_max_ea_size,
 658                                                               XATTR_SIZE_MAX);
 659                                         OBD_ALLOC_LARGE(info->mti_big_acl,
 660                                                         info->mti_big_aclsize);
 661                                         if (info->mti_big_acl == NULL) {
 662                                                 info->mti_big_aclsize = 0;
 663                                                 CERROR("%s: unable to grow "
 664                                                        DFID" ACL buffer\n",
 665                                                        mdt_obd_name(mdt),
 666                                                        PFID(mdt_object_fid(o)));
 667                                                 RETURN(-ENOMEM);
 668                                         }
 669                                 }
 670
 671                                 CDEBUG(D_INODE, "%s: grow the "DFID
 672                                        " ACL buffer to size %d\n",
 673                                        mdt_obd_name(mdt),
 674                                        PFID(mdt_object_fid(o)),
 675                                        info->mti_big_aclsize);
 676
 677                                 buf->lb_buf = info->mti_big_acl;
 678                                 buf->lb_len = info->mti_big_aclsize;
 679
 680                                 goto again;
 681                         }
 682
 683                         CERROR("%s: unable to read "DFID" ACL: rc = %d\n",
 684                                mdt_obd_name(mdt), PFID(mdt_object_fid(o)), rc);
 685                 }
 686         } else {
 687                 int client;
 688                 int server;
 689                 int acl_buflen;
 690                 int lmm_buflen = 0;
 691                 int lmmsize = 0;
 692
 693                 acl_buflen = req_capsule_get_size(pill, &RMF_ACL, RCL_SERVER);
 694                 if (acl_buflen >= rc)
 695                         goto map;
 696
 697                 /* If LOV/LMA EA is small, we can reuse part of their buffer */
 698                 client = ptlrpc_req_get_repsize(pill->rc_req);
 699                 server = lustre_packed_msg_size(pill->rc_req->rq_repmsg);
 700                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
 701                         lmm_buflen = req_capsule_get_size(pill, &RMF_MDT_MD,
 702                                                           RCL_SERVER);
 703                         lmmsize = repbody->mbo_eadatasize;
 704                 }
 705
 706                 if (client < server - acl_buflen - lmm_buflen + rc + lmmsize) {
 707                         CDEBUG(D_INODE, "%s: client prepared buffer size %d "
 708                                "is not big enough with the ACL size %d (%d)\n",
 709                                mdt_obd_name(mdt), client, rc,
 710                                server - acl_buflen - lmm_buflen + rc + lmmsize);
 711                         repbody->mbo_aclsize = 0;
 712                         repbody->mbo_valid &= ~OBD_MD_FLACL;
 713                         RETURN(-ERANGE);
 714                 }
 715
 716 map:
 717                 if (buf->lb_buf == info->mti_big_acl)
 718                         info->mti_big_acl_used = 1;
 719
 720                 rc = nodemap_map_acl(nodemap, buf->lb_buf,
 721                                      rc, NODEMAP_FS_TO_CLIENT);
 722                 /* if all ACLs mapped out, rc is still >= 0 */
 723                 if (rc < 0) {
 724                         CERROR("%s: nodemap_map_acl unable to parse "DFID
 725                                " ACL: rc = %d\n", mdt_obd_name(mdt),
 726                                PFID(mdt_object_fid(o)), rc);
 727                         repbody->mbo_aclsize = 0;
 728                         repbody->mbo_valid &= ~OBD_MD_FLACL;
 729                 } else {
 730                         repbody->mbo_aclsize = rc;
 731                         repbody->mbo_valid |= OBD_MD_FLACL;
 732                         rc = 0;
 733                 }
 734         }
 735
 736         RETURN(rc);
 737 }
 738 #endif
 739
 740 /* XXX Look into layout in MDT layer. */
 741 static inline bool mdt_hsm_is_released(struct lov_mds_md *lmm)
 742 {
 743         struct lov_comp_md_v1   *comp_v1;
 744         struct lov_mds_md       *v1;
 745         int                      i;
 746
 747         if (lmm->lmm_magic == LOV_MAGIC_COMP_V1) {
 748                 comp_v1 = (struct lov_comp_md_v1 *)lmm;
 749
 750                 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
 751                         v1 = (struct lov_mds_md *)((char *)comp_v1 +
 752                                 comp_v1->lcm_entries[i].lcme_offset);
 753                         /* We don't support partial release for now */
 754                         if (!(v1->lmm_pattern & LOV_PATTERN_F_RELEASED))
 755                                 return false;
 756                 }
 757                 return true;
 758         } else {
 759                 return (lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) ?
 760                         true : false;
 761         }
 762 }
 763
 764 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 765                         const struct lu_attr *attr, const struct lu_fid *fid)
 766 {
 767         struct md_attr *ma = &info->mti_attr;
 768         struct obd_export *exp = info->mti_exp;
 769         struct lu_nodemap *nodemap = NULL;
 770
 771         LASSERT(ma->ma_valid & MA_INODE);
 772
 773         if (attr->la_valid & LA_ATIME) {
 774                 b->mbo_atime = attr->la_atime;
 775                 b->mbo_valid |= OBD_MD_FLATIME;
 776         }
 777         if (attr->la_valid & LA_MTIME) {
 778                 b->mbo_mtime = attr->la_mtime;
 779                 b->mbo_valid |= OBD_MD_FLMTIME;
 780         }
 781         if (attr->la_valid & LA_CTIME) {
 782                 b->mbo_ctime = attr->la_ctime;
 783                 b->mbo_valid |= OBD_MD_FLCTIME;
 784         }
 785         if (attr->la_valid & LA_BTIME) {
 786                 b->mbo_btime = attr->la_btime;
 787                 b->mbo_valid |= OBD_MD_FLBTIME;
 788         }
 789         if (attr->la_valid & LA_FLAGS) {
 790                 b->mbo_flags = attr->la_flags;
 791                 b->mbo_valid |= OBD_MD_FLFLAGS;
 792         }
 793         if (attr->la_valid & LA_NLINK) {
 794                 b->mbo_nlink = attr->la_nlink;
 795                 b->mbo_valid |= OBD_MD_FLNLINK;
 796         }
 797         if (attr->la_valid & (LA_UID|LA_GID)) {
 798                 nodemap = nodemap_get_from_exp(exp);
 799                 if (IS_ERR(nodemap))
 800                         goto out;
 801         }
 802         if (attr->la_valid & LA_UID) {
 803                 b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 804                                             NODEMAP_FS_TO_CLIENT,
 805                                             attr->la_uid);
 806                 b->mbo_valid |= OBD_MD_FLUID;
 807         }
 808         if (attr->la_valid & LA_GID) {
 809                 b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 810                                             NODEMAP_FS_TO_CLIENT,
 811                                             attr->la_gid);
 812                 b->mbo_valid |= OBD_MD_FLGID;
 813         }
 814
 815         if (attr->la_valid & LA_PROJID) {
 816                 /* TODO, nodemap for project id */
 817                 b->mbo_projid = attr->la_projid;
 818                 b->mbo_valid |= OBD_MD_FLPROJID;
 819         }
 820
 821         b->mbo_mode = attr->la_mode;
 822         if (attr->la_valid & LA_MODE)
 823                 b->mbo_valid |= OBD_MD_FLMODE;
 824         if (attr->la_valid & LA_TYPE)
 825                 b->mbo_valid |= OBD_MD_FLTYPE;
 826
 827         if (fid != NULL) {
 828                 b->mbo_fid1 = *fid;
 829                 b->mbo_valid |= OBD_MD_FLID;
 830                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid=%#llx\n",
 831                        PFID(fid), b->mbo_nlink, b->mbo_mode, b->mbo_valid);
 832         }
 833
 834         if (!(attr->la_valid & LA_TYPE))
 835                 return;
 836
 837         b->mbo_rdev   = attr->la_rdev;
 838         b->mbo_size   = attr->la_size;
 839         b->mbo_blocks = attr->la_blocks;
 840
 841         if (!S_ISREG(attr->la_mode)) {
 842                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 843         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
 844                 /* means no objects are allocated on osts. */
 845                 LASSERT(!(ma->ma_valid & MA_LOV));
 846                 /* just ignore blocks occupied by extend attributes on MDS */
 847                 b->mbo_blocks = 0;
 848                 /* if no object is allocated on osts, the size on mds is valid.
 849                  * b=22272 */
 850                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 851         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL) {
 852                 if (mdt_hsm_is_released(ma->ma_lmm)) {
 853                         /* A released file stores its size on MDS. */
 854                         /* But return 1 block for released file, unless tools
 855                          * like tar will consider it fully sparse. (LU-3864)
 856                          */
 857                         if (unlikely(b->mbo_size == 0))
 858                                 b->mbo_blocks = 0;
 859                         else
 860                                 b->mbo_blocks = 1;
 861                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 862                 } else if (info->mti_som_valid) { /* som is valid */
 863                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 864                 } else if (ma->ma_valid & MA_SOM) { /* lsom is valid */
 865                         b->mbo_valid |= OBD_MD_FLLAZYSIZE | OBD_MD_FLLAZYBLOCKS;
 866                         b->mbo_size = ma->ma_som.ms_size;
 867                         b->mbo_blocks = ma->ma_som.ms_blocks;
 868                 }
 869         }
 870
 871         if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE ||
 872                             b->mbo_valid & OBD_MD_FLLAZYSIZE))
 873                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 874                        PFID(fid), (unsigned long long)b->mbo_size);
 875
 876 out:
 877         if (!IS_ERR_OR_NULL(nodemap))
 878                 nodemap_putref(nodemap);
 879 }
 880
 881 static inline int mdt_body_has_lov(const struct lu_attr *la,
 882                                    const struct mdt_body *body)
 883 {
 884         return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
 885                (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
 886 }
 887
 888 void mdt_client_compatibility(struct mdt_thread_info *info)
 889 {
 890         struct mdt_body       *body;
 891         struct ptlrpc_request *req = mdt_info_req(info);
 892         struct obd_export     *exp = req->rq_export;
 893         struct md_attr        *ma = &info->mti_attr;
 894         struct lu_attr        *la = &ma->ma_attr;
 895         ENTRY;
 896
 897         if (exp_connect_layout(exp))
 898                 /* the client can deal with 16-bit lmm_stripe_count */
 899                 RETURN_EXIT;
 900
 901         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 902
 903         if (!mdt_body_has_lov(la, body))
 904                 RETURN_EXIT;
 905
 906         /* now we have a reply with a lov for a client not compatible with the
 907          * layout lock so we have to clean the layout generation number */
 908         if (S_ISREG(la->la_mode))
 909                 ma->ma_lmm->lmm_layout_gen = 0;
 910         EXIT;
 911 }
 912
 913 static int mdt_attr_get_eabuf_size(struct mdt_thread_info *info,
 914                                    struct mdt_object *o)
 915 {
 916         const struct lu_env *env = info->mti_env;
 917         int rc, rc2;
 918
 919         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 920                           XATTR_NAME_LOV);
 921
 922         if (rc == -ENODATA)
 923                 rc = 0;
 924
 925         if (rc < 0)
 926                 goto out;
 927
 928         /* Is it a directory? Let's check for the LMV as well */
 929         if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
 930                 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 931                                    XATTR_NAME_LMV);
 932
 933                 if (rc2 == -ENODATA)
 934                         rc2 = mo_xattr_get(env, mdt_object_child(o),
 935                                            &LU_BUF_NULL,
 936                                            XATTR_NAME_DEFAULT_LMV);
 937
 938                 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
 939                         rc = rc2;
 940         }
 941
 942 out:
 943         return rc;
 944 }
 945
 946 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
 947                       const char *name)
 948 {
 949         const struct lu_env *env = info->mti_env;
 950         int rc;
 951         ENTRY;
 952
 953         LASSERT(info->mti_big_lmm_used == 0);
 954         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
 955         if (rc < 0)
 956                 RETURN(rc);
 957
 958         /* big_lmm may need to be grown */
 959         if (info->mti_big_lmmsize < rc) {
 960                 int size = size_roundup_power2(rc);
 961
 962                 if (info->mti_big_lmmsize > 0) {
 963                         /* free old buffer */
 964                         LASSERT(info->mti_big_lmm);
 965                         OBD_FREE_LARGE(info->mti_big_lmm,
 966                                        info->mti_big_lmmsize);
 967                         info->mti_big_lmm = NULL;
 968                         info->mti_big_lmmsize = 0;
 969                 }
 970
 971                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
 972                 if (info->mti_big_lmm == NULL)
 973                         RETURN(-ENOMEM);
 974                 info->mti_big_lmmsize = size;
 975         }
 976         LASSERT(info->mti_big_lmmsize >= rc);
 977
 978         info->mti_buf.lb_buf = info->mti_big_lmm;
 979         info->mti_buf.lb_len = info->mti_big_lmmsize;
 980         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
 981
 982         RETURN(rc);
 983 }
 984
 985 int __mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
 986                      struct md_attr *ma, const char *name)
 987 {
 988         struct md_object *next = mdt_object_child(o);
 989         struct lu_buf    *buf = &info->mti_buf;
 990         int rc;
 991
 992         if (strcmp(name, XATTR_NAME_LOV) == 0) {
 993                 buf->lb_buf = ma->ma_lmm;
 994                 buf->lb_len = ma->ma_lmm_size;
 995                 LASSERT(!(ma->ma_valid & MA_LOV));
 996         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 997                 buf->lb_buf = ma->ma_lmv;
 998                 buf->lb_len = ma->ma_lmv_size;
 999                 LASSERT(!(ma->ma_valid & MA_LMV));
1000         } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
1001                 buf->lb_buf = ma->ma_default_lmv;
1002                 buf->lb_len = ma->ma_default_lmv_size;
1003                 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
1004         } else {
1005                 return -EINVAL;
1006         }
1007
1008         LASSERT(buf->lb_buf);
1009
1010         rc = mo_xattr_get(info->mti_env, next, buf, name);
1011         if (rc > 0) {
1012
1013 got:
1014                 if (strcmp(name, XATTR_NAME_LOV) == 0) {
1015                         if (info->mti_big_lmm_used)
1016                                 ma->ma_lmm = info->mti_big_lmm;
1017
1018                         /* NOT return LOV EA with hole to old client. */
1019                         if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
1020                                      LOV_PATTERN_F_HOLE) &&
1021                             !(exp_connect_flags(info->mti_exp) &
1022                               OBD_CONNECT_LFSCK)) {
1023                                 return -EIO;
1024                         } else {
1025                                 ma->ma_lmm_size = rc;
1026                                 ma->ma_valid |= MA_LOV;
1027                         }
1028                 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
1029                         if (info->mti_big_lmm_used)
1030                                 ma->ma_lmv = info->mti_big_lmm;
1031
1032                         ma->ma_lmv_size = rc;
1033                         ma->ma_valid |= MA_LMV;
1034                 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
1035                         ma->ma_default_lmv_size = rc;
1036                         ma->ma_valid |= MA_LMV_DEF;
1037                 }
1038
1039                 /* Update mdt_max_mdsize so all clients will be aware that */
1040                 if (info->mti_mdt->mdt_max_mdsize < rc)
1041                         info->mti_mdt->mdt_max_mdsize = rc;
1042
1043                 rc = 0;
1044         } else if (rc == -ENODATA) {
1045                 /* no LOV EA */
1046                 rc = 0;
1047         } else if (rc == -ERANGE) {
1048                 /* Default LMV has fixed size, so it must be able to fit
1049                  * in the original buffer */
1050                 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
1051                         return rc;
1052                 rc = mdt_big_xattr_get(info, o, name);
1053                 if (rc > 0) {
1054                         info->mti_big_lmm_used = 1;
1055                         goto got;
1056                 }
1057         }
1058
1059         return rc;
1060 }
1061
1062 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
1063                    struct md_attr *ma, const char *name)
1064 {
1065         int rc;
1066
1067         if (!info->mti_big_lmm) {
1068                 OBD_ALLOC(info->mti_big_lmm, PAGE_SIZE);
1069                 if (!info->mti_big_lmm)
1070                         return -ENOMEM;
1071                 info->mti_big_lmmsize = PAGE_SIZE;
1072         }
1073
1074         if (strcmp(name, XATTR_NAME_LOV) == 0) {
1075                 ma->ma_lmm = info->mti_big_lmm;
1076                 ma->ma_lmm_size = info->mti_big_lmmsize;
1077                 ma->ma_valid &= ~MA_LOV;
1078         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
1079                 ma->ma_lmv = info->mti_big_lmm;
1080                 ma->ma_lmv_size = info->mti_big_lmmsize;
1081                 ma->ma_valid &= ~MA_LMV;
1082         } else {
1083                 LBUG();
1084         }
1085
1086         LASSERT(!info->mti_big_lmm_used);
1087         rc = __mdt_stripe_get(info, o, ma, name);
1088         /* since big_lmm is always used here, clear 'used' flag to avoid
1089          * assertion in mdt_big_xattr_get().
1090          */
1091         info->mti_big_lmm_used = 0;
1092
1093         return rc;
1094 }
1095
1096 int mdt_attr_get_pfid(struct mdt_thread_info *info, struct mdt_object *o,
1097                       struct lu_fid *pfid)
1098 {
1099         struct lu_buf           *buf = &info->mti_buf;
1100         struct link_ea_header   *leh;
1101         struct link_ea_entry    *lee;
1102         int                      rc;
1103         ENTRY;
1104
1105         buf->lb_buf = info->mti_big_lmm;
1106         buf->lb_len = info->mti_big_lmmsize;
1107         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
1108                           buf, XATTR_NAME_LINK);
1109         /* ignore errors, MA_PFID won't be set and it is
1110          * up to the caller to treat this as an error */
1111         if (rc == -ERANGE || buf->lb_len == 0) {
1112                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
1113                 buf->lb_buf = info->mti_big_lmm;
1114                 buf->lb_len = info->mti_big_lmmsize;
1115         }
1116
1117         if (rc < 0)
1118                 RETURN(rc);
1119         if (rc < sizeof(*leh)) {
1120                 CERROR("short LinkEA on "DFID": rc = %d\n",
1121                        PFID(mdt_object_fid(o)), rc);
1122                 RETURN(-ENODATA);
1123         }
1124
1125         leh = (struct link_ea_header *) buf->lb_buf;
1126         lee = (struct link_ea_entry *)(leh + 1);
1127         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
1128                 leh->leh_magic = LINK_EA_MAGIC;
1129                 leh->leh_reccount = __swab32(leh->leh_reccount);
1130                 leh->leh_len = __swab64(leh->leh_len);
1131         }
1132         if (leh->leh_magic != LINK_EA_MAGIC)
1133                 RETURN(-EINVAL);
1134         if (leh->leh_reccount == 0)
1135                 RETURN(-ENODATA);
1136
1137         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
1138         fid_be_to_cpu(pfid, pfid);
1139
1140         RETURN(0);
1141 }
1142
1143 int mdt_attr_get_pfid_name(struct mdt_thread_info *info, struct mdt_object *o,
1144                            struct lu_fid *pfid, struct lu_name *lname)
1145 {
1146         struct lu_buf *buf = &info->mti_buf;
1147         struct link_ea_header *leh;
1148         struct link_ea_entry *lee;
1149         int reclen;
1150         int rc;
1151
1152         buf->lb_buf = info->mti_xattr_buf;
1153         buf->lb_len = sizeof(info->mti_xattr_buf);
1154         rc = mo_xattr_get(info->mti_env, mdt_object_child(o), buf,
1155                           XATTR_NAME_LINK);
1156         if (rc == -ERANGE) {
1157                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
1158                 buf->lb_buf = info->mti_big_lmm;
1159                 buf->lb_len = info->mti_big_lmmsize;
1160         }
1161         if (rc < 0)
1162                 return rc;
1163
1164         if (rc < sizeof(*leh)) {
1165                 CERROR("short LinkEA on "DFID": rc = %d\n",
1166                        PFID(mdt_object_fid(o)), rc);
1167                 return -ENODATA;
1168         }
1169
1170         leh = (struct link_ea_header *)buf->lb_buf;
1171         lee = (struct link_ea_entry *)(leh + 1);
1172         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
1173                 leh->leh_magic = LINK_EA_MAGIC;
1174                 leh->leh_reccount = __swab32(leh->leh_reccount);
1175                 leh->leh_len = __swab64(leh->leh_len);
1176         }
1177         if (leh->leh_magic != LINK_EA_MAGIC)
1178                 return -EINVAL;
1179
1180         if (leh->leh_reccount == 0)
1181                 return -ENODATA;
1182
1183         linkea_entry_unpack(lee, &reclen, lname, pfid);
1184
1185         return 0;
1186 }
1187
1188 int mdt_attr_get_complex(struct mdt_thread_info *info,
1189                          struct mdt_object *o, struct md_attr *ma)
1190 {
1191         const struct lu_env *env = info->mti_env;
1192         struct md_object    *next = mdt_object_child(o);
1193         struct lu_buf       *buf = &info->mti_buf;
1194         int                  need = ma->ma_need;
1195         int                  rc = 0, rc2;
1196         u32                  mode;
1197         ENTRY;
1198
1199         ma->ma_valid = 0;
1200
1201         if (mdt_object_exists(o) == 0)
1202                 GOTO(out, rc = -ENOENT);
1203         mode = lu_object_attr(&next->mo_lu);
1204
1205         if (need & MA_INODE) {
1206                 ma->ma_need = MA_INODE;
1207                 rc = mo_attr_get(env, next, ma);
1208                 if (rc)
1209                         GOTO(out, rc);
1210
1211                 if (S_ISREG(mode))
1212                         (void) mdt_get_som(info, o, ma);
1213                 ma->ma_valid |= MA_INODE;
1214         }
1215
1216         if (need & MA_PFID) {
1217                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
1218                 if (rc == 0)
1219                         ma->ma_valid |= MA_PFID;
1220                 /* ignore this error, parent fid is not mandatory */
1221                 rc = 0;
1222         }
1223
1224         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
1225                 rc = __mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
1226                 if (rc)
1227                         GOTO(out, rc);
1228         }
1229
1230         if (need & MA_LMV && S_ISDIR(mode)) {
1231                 rc = __mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
1232                 if (rc != 0)
1233                         GOTO(out, rc);
1234         }
1235
1236         if (need & MA_LMV_DEF && S_ISDIR(mode)) {
1237                 rc = __mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
1238                 if (rc != 0)
1239                         GOTO(out, rc);
1240         }
1241
1242         /*
1243          * In the handle of MA_INODE, we may already get the SOM attr.
1244          */
1245         if (need & MA_SOM && S_ISREG(mode) && !(ma->ma_valid & MA_SOM)) {
1246                 rc = mdt_get_som(info, o, ma);
1247                 if (rc != 0)
1248                         GOTO(out, rc);
1249         }
1250
1251         if (need & MA_HSM && S_ISREG(mode)) {
1252                 buf->lb_buf = info->mti_xattr_buf;
1253                 buf->lb_len = sizeof(info->mti_xattr_buf);
1254                 BUILD_BUG_ON(sizeof(struct hsm_attrs) >
1255                              sizeof(info->mti_xattr_buf));
1256                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
1257                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
1258                 if (rc2 == 0)
1259                         ma->ma_valid |= MA_HSM;
1260                 else if (rc2 < 0 && rc2 != -ENODATA)
1261                         GOTO(out, rc = rc2);
1262         }
1263
1264 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1265         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
1266                 buf->lb_buf = ma->ma_acl;
1267                 buf->lb_len = ma->ma_acl_size;
1268                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
1269                 if (rc2 > 0) {
1270                         ma->ma_acl_size = rc2;
1271                         ma->ma_valid |= MA_ACL_DEF;
1272                 } else if (rc2 == -ENODATA) {
1273                         /* no ACLs */
1274                         ma->ma_acl_size = 0;
1275                 } else
1276                         GOTO(out, rc = rc2);
1277         }
1278 #endif
1279 out:
1280         ma->ma_need = need;
1281         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = %#llx ma_lmm=%p\n",
1282                rc, ma->ma_valid, ma->ma_lmm);
1283         RETURN(rc);
1284 }
1285
1286 static int mdt_getattr_internal(struct mdt_thread_info *info,
1287                                 struct mdt_object *o, int ma_need)
1288 {
1289         struct mdt_device *mdt = info->mti_mdt;
1290         struct md_object *next = mdt_object_child(o);
1291         const struct mdt_body *reqbody = info->mti_body;
1292         struct ptlrpc_request *req = mdt_info_req(info);
1293         struct md_attr *ma = &info->mti_attr;
1294         struct lu_attr *la = &ma->ma_attr;
1295         struct req_capsule *pill = info->mti_pill;
1296         const struct lu_env *env = info->mti_env;
1297         struct mdt_body *repbody;
1298         struct lu_buf *buffer = &info->mti_buf;
1299         struct obd_export *exp = info->mti_exp;
1300         ktime_t kstart = ktime_get();
1301         int rc;
1302
1303         ENTRY;
1304
1305         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
1306                 RETURN(err_serious(-ENOMEM));
1307
1308         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1309
1310         ma->ma_valid = 0;
1311
1312         if (mdt_object_remote(o)) {
1313                 /* This object is located on remote node.*/
1314                 /* Return -ENOTSUPP for old client */
1315                 if (!mdt_is_dne_client(req->rq_export))
1316                         GOTO(out, rc = -ENOTSUPP);
1317
1318                 repbody->mbo_fid1 = *mdt_object_fid(o);
1319                 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
1320                 GOTO(out, rc = 0);
1321         }
1322
1323         if (reqbody->mbo_eadatasize > 0) {
1324                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
1325                 if (buffer->lb_buf == NULL)
1326                         GOTO(out, rc = -EPROTO);
1327                 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
1328                                                       RCL_SERVER);
1329         } else {
1330                 buffer->lb_buf = NULL;
1331                 buffer->lb_len = 0;
1332                 ma_need &= ~(MA_LOV | MA_LMV);
1333                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
1334                        mdt_obd_name(info->mti_mdt),
1335                        req->rq_export->exp_client_uuid.uuid);
1336         }
1337
1338         /* from 2.12.58 intent_getattr pack default LMV in reply */
1339         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1340             ((reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA)) ==
1341                     (OBD_MD_MEA | OBD_MD_DEFAULT_MEA)) &&
1342             req_capsule_has_field(&req->rq_pill, &RMF_DEFAULT_MDT_MD,
1343                                   RCL_SERVER)) {
1344                 ma->ma_lmv = buffer->lb_buf;
1345                 ma->ma_lmv_size = buffer->lb_len;
1346                 ma->ma_default_lmv = req_capsule_server_get(pill,
1347                                                 &RMF_DEFAULT_MDT_MD);
1348                 ma->ma_default_lmv_size = req_capsule_get_size(pill,
1349                                                 &RMF_DEFAULT_MDT_MD,
1350                                                 RCL_SERVER);
1351                 ma->ma_need = MA_INODE;
1352                 if (ma->ma_lmv_size > 0)
1353                         ma->ma_need |= MA_LMV;
1354                 if (ma->ma_default_lmv_size > 0)
1355                         ma->ma_need |= MA_LMV_DEF;
1356         } else if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1357                    (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
1358                 /* If it is dir and client require MEA, then we got MEA */
1359                 /* Assumption: MDT_MD size is enough for lmv size. */
1360                 ma->ma_lmv = buffer->lb_buf;
1361                 ma->ma_lmv_size = buffer->lb_len;
1362                 ma->ma_need = MA_INODE;
1363                 if (ma->ma_lmv_size > 0) {
1364                         if (reqbody->mbo_valid & OBD_MD_MEA) {
1365                                 ma->ma_need |= MA_LMV;
1366                         } else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA) {
1367                                 ma->ma_need |= MA_LMV_DEF;
1368                                 ma->ma_default_lmv = buffer->lb_buf;
1369                                 ma->ma_lmv = NULL;
1370                                 ma->ma_default_lmv_size = buffer->lb_len;
1371                                 ma->ma_lmv_size = 0;
1372                         }
1373                 }
1374         } else {
1375                 ma->ma_lmm = buffer->lb_buf;
1376                 ma->ma_lmm_size = buffer->lb_len;
1377                 ma->ma_need = MA_INODE | MA_HSM;
1378                 if (ma->ma_lmm_size > 0) {
1379                         ma->ma_need |= MA_LOV;
1380                         /* Older clients may crash if they getattr overstriped
1381                          * files
1382                          */
1383                         if (!exp_connect_overstriping(exp) &&
1384                             mdt_lmm_is_overstriping(ma->ma_lmm))
1385                                 RETURN(-EOPNOTSUPP);
1386                 }
1387         }
1388
1389         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1390             reqbody->mbo_valid & OBD_MD_FLDIREA  &&
1391             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
1392                 /* get default stripe info for this dir. */
1393                 ma->ma_need |= MA_LOV_DEF;
1394         }
1395         ma->ma_need |= ma_need;
1396
1397         rc = mdt_attr_get_complex(info, o, ma);
1398         if (unlikely(rc)) {
1399                 CDEBUG(rc == -ENOENT ? D_OTHER : D_ERROR,
1400                        "%s: getattr error for "DFID": rc = %d\n",
1401                        mdt_obd_name(info->mti_mdt),
1402                        PFID(mdt_object_fid(o)), rc);
1403                 RETURN(rc);
1404         }
1405
1406         /* if file is released, check if a restore is running */
1407         if (ma->ma_valid & MA_HSM) {
1408                 repbody->mbo_valid |= OBD_MD_TSTATE;
1409                 if ((ma->ma_hsm.mh_flags & HS_RELEASED) &&
1410                     mdt_hsm_restore_is_running(info, mdt_object_fid(o)))
1411                         repbody->mbo_t_state = MS_RESTORE;
1412         }
1413
1414         if (unlikely(!(ma->ma_valid & MA_INODE)))
1415                 RETURN(-EFAULT);
1416
1417         mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
1418
1419         if (mdt_body_has_lov(la, reqbody)) {
1420                 u32 stripe_count = 1;
1421
1422                 if (ma->ma_valid & MA_LOV) {
1423                         LASSERT(ma->ma_lmm_size);
1424                         repbody->mbo_eadatasize = ma->ma_lmm_size;
1425                         if (S_ISDIR(la->la_mode))
1426                                 repbody->mbo_valid |= OBD_MD_FLDIREA;
1427                         else
1428                                 repbody->mbo_valid |= OBD_MD_FLEASIZE;
1429                         mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
1430                 }
1431                 if (ma->ma_valid & MA_LMV) {
1432                         struct lmv_mds_md_v1 *lmv = &ma->ma_lmv->lmv_md_v1;
1433                         u32 magic = le32_to_cpu(lmv->lmv_magic);
1434
1435                         /* Return -ENOTSUPP for old client */
1436                         if (!mdt_is_striped_client(req->rq_export))
1437                                 RETURN(-ENOTSUPP);
1438
1439                         LASSERT(S_ISDIR(la->la_mode));
1440                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1441                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1442                         repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
1443
1444                         stripe_count = le32_to_cpu(lmv->lmv_stripe_count);
1445                         if (magic == LMV_MAGIC_STRIPE && lmv_is_restriping(lmv))
1446                                 mdt_restripe_migrate_add(info, o);
1447                         else if (magic == LMV_MAGIC_V1 &&
1448                                  lmv_is_restriping(lmv))
1449                                 mdt_restripe_update_add(info, o);
1450                 }
1451                 if (ma->ma_valid & MA_LMV_DEF) {
1452                         /* Return -ENOTSUPP for old client */
1453                         if (!mdt_is_striped_client(req->rq_export))
1454                                 RETURN(-ENOTSUPP);
1455                         LASSERT(S_ISDIR(la->la_mode));
1456                         /*
1457                          * when ll_dir_getstripe() gets default LMV, it
1458                          * checks mbo_eadatasize.
1459                          */
1460                         if (!(ma->ma_valid & MA_LMV))
1461                                 repbody->mbo_eadatasize =
1462                                         ma->ma_default_lmv_size;
1463                         repbody->mbo_valid |= (OBD_MD_FLDIREA |
1464                                                OBD_MD_DEFAULT_MEA);
1465                 }
1466                 CDEBUG(D_VFSTRACE,
1467                        "dirent count %llu stripe count %u MDT count %d\n",
1468                        ma->ma_attr.la_dirent_count, stripe_count,
1469                        atomic_read(&mdt->mdt_mds_mds_conns) + 1);
1470                 if (ma->ma_attr.la_dirent_count != LU_DIRENT_COUNT_UNSET &&
1471                     ma->ma_attr.la_dirent_count >
1472                         mdt->mdt_restriper.mdr_dir_split_count &&
1473                     !fid_is_root(mdt_object_fid(o)) &&
1474                     mdt->mdt_enable_dir_auto_split &&
1475                     !o->mot_restriping &&
1476                     stripe_count < atomic_read(&mdt->mdt_mds_mds_conns) + 1)
1477                         mdt_auto_split_add(info, o);
1478         } else if (S_ISLNK(la->la_mode) &&
1479                    reqbody->mbo_valid & OBD_MD_LINKNAME) {
1480                 buffer->lb_buf = ma->ma_lmm;
1481                 /* eadatasize from client includes NULL-terminator, so
1482                  * there is no need to read it */
1483                 buffer->lb_len = reqbody->mbo_eadatasize - 1;
1484                 rc = mo_readlink(env, next, buffer);
1485                 if (unlikely(rc <= 0)) {
1486                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
1487                                mdt_obd_name(info->mti_mdt),
1488                                PFID(mdt_object_fid(o)), rc);
1489                         rc = -EFAULT;
1490                 } else {
1491                         int print_limit = min_t(int, PAGE_SIZE - 128, rc);
1492
1493                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
1494                                 rc -= 2;
1495                         repbody->mbo_valid |= OBD_MD_LINKNAME;
1496                         /* we need to report back size with NULL-terminator
1497                          * because client expects that */
1498                         repbody->mbo_eadatasize = rc + 1;
1499                         if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
1500                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
1501                                        "on "DFID ", expected %d\n",
1502                                        mdt_obd_name(info->mti_mdt),
1503                                        rc, PFID(mdt_object_fid(o)),
1504                                        reqbody->mbo_eadatasize - 1);
1505                         /* NULL terminate */
1506                         ((char *)ma->ma_lmm)[rc] = 0;
1507
1508                         /* If the total CDEBUG() size is larger than a page, it
1509                          * will print a warning to the console, avoid this by
1510                          * printing just the last part of the symlink. */
1511                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
1512                                print_limit < rc ? "..." : "", print_limit,
1513                                (char *)ma->ma_lmm + rc - print_limit, rc);
1514                         rc = 0;
1515                 }
1516         }
1517
1518         if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
1519                 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1520                 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1521                 CDEBUG(D_INODE, "changing the max MD size to %u\n",
1522                        repbody->mbo_max_mdsize);
1523         }
1524
1525 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1526         if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1527                  (reqbody->mbo_valid & OBD_MD_FLACL)) {
1528                 struct lu_nodemap *nodemap = nodemap_get_from_exp(exp);
1529                 if (IS_ERR(nodemap))
1530                         RETURN(PTR_ERR(nodemap));
1531
1532                 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
1533                 nodemap_putref(nodemap);
1534         }
1535 #endif
1536
1537 out:
1538         if (rc == 0)
1539                 mdt_counter_incr(req, LPROC_MDT_GETATTR,
1540                                  ktime_us_delta(ktime_get(), kstart));
1541
1542         RETURN(rc);
1543 }
1544
1545 static int mdt_getattr(struct tgt_session_info *tsi)
1546 {
1547         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1548         struct mdt_object       *obj = info->mti_object;
1549         struct req_capsule      *pill = info->mti_pill;
1550         struct mdt_body         *reqbody;
1551         struct mdt_body         *repbody;
1552         int rc, rc2;
1553         ENTRY;
1554
1555         if (unlikely(info->mti_object == NULL))
1556                 RETURN(-EPROTO);
1557
1558         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1559         LASSERT(reqbody);
1560         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1561
1562         /* Special case for Data-on-MDT files to get data version */
1563         if (unlikely(reqbody->mbo_valid & OBD_MD_FLDATAVERSION)) {
1564                 rc = mdt_data_version_get(tsi);
1565                 GOTO(out, rc);
1566         }
1567
1568         /* Unlike intent case where we need to pre-fill out buffers early on
1569          * in intent policy for ldlm reasons, here we can have a much better
1570          * guess at EA size by just reading it from disk.
1571          * Exceptions are readdir and (missing) directory striping */
1572         /* Readlink */
1573         if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1574                 /* No easy way to know how long is the symlink, but it cannot
1575                  * be more than PATH_MAX, so we allocate +1 */
1576                 rc = PATH_MAX + 1;
1577         /* A special case for fs ROOT: getattr there might fetch
1578          * default EA for entire fs, not just for this dir!
1579          */
1580         } else if (lu_fid_eq(mdt_object_fid(obj),
1581                              &info->mti_mdt->mdt_md_root_fid) &&
1582                    (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1583                    (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1584                                                                  MDS_GETATTR)) {
1585                 /* Should the default strping be bigger, mdt_fix_reply
1586                  * will reallocate */
1587                 rc = DEF_REP_MD_SIZE;
1588         } else {
1589                 /* Read the actual EA size from disk */
1590                 rc = mdt_attr_get_eabuf_size(info, obj);
1591         }
1592
1593         if (rc < 0)
1594                 GOTO(out, rc = err_serious(rc));
1595
1596         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1597
1598         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
1599          * by default. If the target object has more ACL entries, then
1600          * enlarge the buffer when necessary. */
1601         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
1602                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
1603
1604         rc = req_capsule_server_pack(pill);
1605         if (unlikely(rc != 0))
1606                 GOTO(out, rc = err_serious(rc));
1607
1608         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1609         LASSERT(repbody != NULL);
1610         repbody->mbo_eadatasize = 0;
1611         repbody->mbo_aclsize = 0;
1612
1613         rc = mdt_check_ucred(info);
1614         if (unlikely(rc))
1615                 GOTO(out_shrink, rc);
1616
1617         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1618
1619         rc = mdt_getattr_internal(info, obj, 0);
1620         EXIT;
1621 out_shrink:
1622         mdt_client_compatibility(info);
1623         rc2 = mdt_fix_reply(info);
1624         if (rc == 0)
1625                 rc = rc2;
1626 out:
1627         mdt_thread_info_fini(info);
1628         return rc;
1629 }
1630
1631 /**
1632  * Handler of layout intent RPC requiring the layout modification
1633  *
1634  * \param[in]  info     thread environment
1635  * \param[in]  obj      object
1636  * \param[out] lhc      object ldlm lock handle
1637  * \param[in]  layout   layout change descriptor
1638  *
1639  * \retval 0    on success
1640  * \retval < 0  error code
1641  */
1642 int mdt_layout_change(struct mdt_thread_info *info, struct mdt_object *obj,
1643                       struct mdt_lock_handle *lhc,
1644                       struct md_layout_change *layout)
1645 {
1646         int rc;
1647
1648         ENTRY;
1649
1650         if (!mdt_object_exists(obj))
1651                 RETURN(-ENOENT);
1652
1653         if (!S_ISREG(lu_object_attr(&obj->mot_obj)))
1654                 RETURN(-EINVAL);
1655
1656         rc = mo_permission(info->mti_env, NULL, mdt_object_child(obj), NULL,
1657                            MAY_WRITE);
1658         if (rc)
1659                 RETURN(rc);
1660
1661         rc = mdt_check_resent_lock(info, obj, lhc);
1662         if (rc < 0)
1663                 RETURN(rc);
1664
1665         if (rc > 0) {
1666                 /* not resent */
1667                 __u64 lockpart = MDS_INODELOCK_LAYOUT;
1668
1669                 /* take layout lock to prepare layout change */
1670                 if (layout->mlc_opc == MD_LAYOUT_WRITE)
1671                         lockpart |= MDS_INODELOCK_UPDATE;
1672
1673                 mdt_lock_handle_init(lhc);
1674                 mdt_lock_reg_init(lhc, LCK_EX);
1675                 rc = mdt_reint_object_lock(info, obj, lhc, lockpart, false);
1676                 if (rc)
1677                         RETURN(rc);
1678         }
1679
1680         mutex_lock(&obj->mot_som_mutex);
1681         rc = mo_layout_change(info->mti_env, mdt_object_child(obj), layout);
1682         mutex_unlock(&obj->mot_som_mutex);
1683
1684         if (rc)
1685                 mdt_object_unlock(info, obj, lhc, 1);
1686
1687         RETURN(rc);
1688 }
1689
1690 /**
1691  * Exchange MOF_LOV_CREATED flags between two objects after a
1692  * layout swap. No assumption is made on whether o1 or o2 have
1693  * created objects or not.
1694  *
1695  * \param[in,out] o1    First swap layout object
1696  * \param[in,out] o2    Second swap layout object
1697  */
1698 static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2)
1699 {
1700         unsigned int o1_lov_created = o1->mot_lov_created;
1701
1702         mutex_lock(&o1->mot_lov_mutex);
1703         mutex_lock(&o2->mot_lov_mutex);
1704
1705         o1->mot_lov_created = o2->mot_lov_created;
1706         o2->mot_lov_created = o1_lov_created;
1707
1708         mutex_unlock(&o2->mot_lov_mutex);
1709         mutex_unlock(&o1->mot_lov_mutex);
1710 }
1711
1712 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1713 {
1714         struct mdt_thread_info  *info;
1715         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1716         struct obd_export       *exp = req->rq_export;
1717         struct mdt_object       *o1, *o2, *o;
1718         struct mdt_lock_handle  *lh1, *lh2;
1719         struct mdc_swap_layouts *msl;
1720         int                      rc;
1721         ENTRY;
1722
1723         /* client does not support layout lock, so layout swaping
1724          * is disabled.
1725          * FIXME: there is a problem for old clients which don't support
1726          * layout lock yet. If those clients have already opened the file
1727          * they won't be notified at all so that old layout may still be
1728          * used to do IO. This can be fixed after file release is landed by
1729          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1730         if (!exp_connect_layout(exp))
1731                 RETURN(-EOPNOTSUPP);
1732
1733         info = tsi2mdt_info(tsi);
1734         if (unlikely(info->mti_object == NULL))
1735                 RETURN(-EPROTO);
1736
1737         if (info->mti_dlm_req != NULL)
1738                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1739
1740         o1 = info->mti_object;
1741         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1742                                 &info->mti_body->mbo_fid2);
1743         if (IS_ERR(o))
1744                 GOTO(out, rc = PTR_ERR(o));
1745
1746         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1747                 GOTO(put, rc = -ENOENT);
1748
1749         rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1750         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1751                 GOTO(put, rc);
1752
1753         if (rc < 0)
1754                 swap(o1, o2);
1755
1756         /* permission check. Make sure the calling process having permission
1757          * to write both files. */
1758         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1759                            MAY_WRITE);
1760         if (rc < 0)
1761                 GOTO(put, rc);
1762
1763         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1764                            MAY_WRITE);
1765         if (rc < 0)
1766                 GOTO(put, rc);
1767
1768         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1769         if (msl == NULL)
1770                 GOTO(put, rc = -EPROTO);
1771
1772         lh1 = &info->mti_lh[MDT_LH_NEW];
1773         mdt_lock_reg_init(lh1, LCK_EX);
1774         lh2 = &info->mti_lh[MDT_LH_OLD];
1775         mdt_lock_reg_init(lh2, LCK_EX);
1776
1777         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1778                              MDS_INODELOCK_XATTR);
1779         if (rc < 0)
1780                 GOTO(put, rc);
1781
1782         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1783                              MDS_INODELOCK_XATTR);
1784         if (rc < 0)
1785                 GOTO(unlock1, rc);
1786
1787         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1788                              mdt_object_child(o2), msl->msl_flags);
1789         if (rc < 0)
1790                 GOTO(unlock2, rc);
1791
1792         mdt_swap_lov_flag(o1, o2);
1793
1794 unlock2:
1795         mdt_object_unlock(info, o2, lh2, rc);
1796 unlock1:
1797         mdt_object_unlock(info, o1, lh1, rc);
1798 put:
1799         mdt_object_put(info->mti_env, o);
1800 out:
1801         mdt_thread_info_fini(info);
1802         RETURN(rc);
1803 }
1804
1805 static int mdt_raw_lookup(struct mdt_thread_info *info,
1806                           struct mdt_object *parent,
1807                           const struct lu_name *lname,
1808                           struct ldlm_reply *ldlm_rep)
1809 {
1810         struct lu_fid   *child_fid = &info->mti_tmp_fid1;
1811         int              rc;
1812         ENTRY;
1813
1814         LASSERT(!info->mti_cross_ref);
1815
1816         /* Only got the fid of this obj by name */
1817         fid_zero(child_fid);
1818         rc = mdo_lookup(info->mti_env, mdt_object_child(info->mti_object),
1819                         lname, child_fid, &info->mti_spec);
1820         if (rc == 0) {
1821                 struct mdt_body *repbody;
1822
1823                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1824                 repbody->mbo_fid1 = *child_fid;
1825                 repbody->mbo_valid = OBD_MD_FLID;
1826                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1827         } else if (rc == -ENOENT) {
1828                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1829         }
1830
1831         RETURN(rc);
1832 }
1833
1834 /*
1835  * UPDATE lock should be taken against parent, and be released before exit;
1836  * child_bits lock should be taken against child, and be returned back:
1837  *            (1)normal request should release the child lock;
1838  *            (2)intent request will grant the lock to client.
1839  */
1840 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1841                                  struct mdt_lock_handle *lhc,
1842                                  __u64 child_bits,
1843                                  struct ldlm_reply *ldlm_rep)
1844 {
1845         struct ptlrpc_request *req = mdt_info_req(info);
1846         struct mdt_body *reqbody = NULL;
1847         struct mdt_object *parent = info->mti_object;
1848         struct mdt_object *child = NULL;
1849         struct lu_fid *child_fid = &info->mti_tmp_fid1;
1850         struct lu_name *lname = NULL;
1851         struct mdt_lock_handle *lhp = NULL;
1852         struct ldlm_lock *lock;
1853         struct req_capsule *pill = info->mti_pill;
1854         __u64 try_bits = 0;
1855         bool is_resent;
1856         int ma_need = 0;
1857         int rc;
1858
1859         ENTRY;
1860
1861         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1862         LASSERT(ergo(is_resent,
1863                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1864
1865         if (parent == NULL)
1866                 RETURN(-ENOENT);
1867
1868         if (info->mti_cross_ref) {
1869                 /* Only getattr on the child. Parent is on another node. */
1870                 mdt_set_disposition(info, ldlm_rep,
1871                                     DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1872                 child = parent;
1873                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1874                        "ldlm_rep = %p\n",
1875                        PFID(mdt_object_fid(child)), ldlm_rep);
1876
1877                 rc = mdt_check_resent_lock(info, child, lhc);
1878                 if (rc < 0) {
1879                         RETURN(rc);
1880                 } else if (rc > 0) {
1881                         mdt_lock_handle_init(lhc);
1882                         mdt_lock_reg_init(lhc, LCK_PR);
1883
1884                         /*
1885                          * Object's name entry is on another MDS, it will
1886                          * request PERM lock only because LOOKUP lock is owned
1887                          * by the MDS where name entry resides.
1888                          *
1889                          * TODO: it should try layout lock too. - Jinshan
1890                          */
1891                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
1892                                         MDS_INODELOCK_LAYOUT);
1893                         child_bits |= MDS_INODELOCK_PERM;
1894
1895                         rc = mdt_object_lock(info, child, lhc, child_bits);
1896                         if (rc < 0)
1897                                 RETURN(rc);
1898                 }
1899
1900                 /* Finally, we can get attr for child. */
1901                 if (!mdt_object_exists(child)) {
1902                         LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1903                                         &child->mot_obj,
1904                                         "remote object doesn't exist.");
1905                         mdt_object_unlock(info, child, lhc, 1);
1906                         RETURN(-ENOENT);
1907                 }
1908
1909                 rc = mdt_getattr_internal(info, child, 0);
1910                 if (unlikely(rc != 0)) {
1911                         mdt_object_unlock(info, child, lhc, 1);
1912                         RETURN(rc);
1913                 }
1914
1915                 rc = mdt_pack_secctx_in_reply(info, child);
1916                 if (unlikely(rc)) {
1917                         mdt_object_unlock(info, child, lhc, 1);
1918                         RETURN(rc);
1919                 }
1920
1921                 rc = mdt_pack_encctx_in_reply(info, child);
1922                 if (unlikely(rc))
1923                         mdt_object_unlock(info, child, lhc, 1);
1924                 RETURN(rc);
1925         }
1926
1927         lname = &info->mti_name;
1928         mdt_name_unpack(pill, &RMF_NAME, lname, MNF_FIX_ANON);
1929
1930         if (lu_name_is_valid(lname)) {
1931                 if (mdt_object_remote(parent)) {
1932                         CERROR("%s: parent "DFID" is on remote target\n",
1933                                mdt_obd_name(info->mti_mdt),
1934                                PFID(mdt_object_fid(parent)));
1935                         RETURN(-EPROTO);
1936                 }
1937
1938                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1939                        "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1940                        PNAME(lname), ldlm_rep);
1941         } else {
1942                 reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1943                 if (unlikely(reqbody == NULL))
1944                         RETURN(err_serious(-EPROTO));
1945
1946                 *child_fid = reqbody->mbo_fid2;
1947                 if (unlikely(!fid_is_sane(child_fid)))
1948                         RETURN(err_serious(-EINVAL));
1949
1950                 if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1951                         mdt_object_get(info->mti_env, parent);
1952                         child = parent;
1953                 } else {
1954                         child = mdt_object_find(info->mti_env, info->mti_mdt,
1955                                                 child_fid);
1956                         if (IS_ERR(child))
1957                                 RETURN(PTR_ERR(child));
1958                 }
1959
1960                 if (mdt_object_remote(child)) {
1961                         CERROR("%s: child "DFID" is on remote target\n",
1962                                mdt_obd_name(info->mti_mdt),
1963                                PFID(mdt_object_fid(child)));
1964                         GOTO(out_child, rc = -EPROTO);
1965                 }
1966
1967                 /* don't fetch LOOKUP lock if it's remote object */
1968                 rc = mdt_is_remote_object(info, parent, child);
1969                 if (rc < 0)
1970                         GOTO(out_child, rc);
1971                 if (rc)
1972                         child_bits &= ~MDS_INODELOCK_LOOKUP;
1973
1974                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1975                        "ldlm_rep = %p\n",
1976                        PFID(mdt_object_fid(parent)),
1977                        PFID(&reqbody->mbo_fid2), ldlm_rep);
1978         }
1979
1980         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1981
1982         if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1983                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1984                                 &parent->mot_obj,
1985                                 "Parent doesn't exist!");
1986                 GOTO(out_child, rc = -ESTALE);
1987         }
1988
1989         if (lu_name_is_valid(lname)) {
1990                 /* Always allow to lookup ".." */
1991                 if (unlikely(lname->ln_namelen == 2 &&
1992                              lname->ln_name[0] == '.' &&
1993                              lname->ln_name[1] == '.'))
1994                         info->mti_spec.sp_permitted = 1;
1995
1996                 if (info->mti_body->mbo_valid == OBD_MD_FLID) {
1997                         rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1998
1999                         RETURN(rc);
2000                 }
2001
2002                 /* step 1: lock parent only if parent is a directory */
2003                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
2004                         lhp = &info->mti_lh[MDT_LH_PARENT];
2005                         mdt_lock_pdo_init(lhp, LCK_PR, lname);
2006                         rc = mdt_object_lock(info, parent, lhp,
2007                                              MDS_INODELOCK_UPDATE);
2008                         if (unlikely(rc != 0))
2009                                 RETURN(rc);
2010                 }
2011
2012                 /* step 2: lookup child's fid by name */
2013                 fid_zero(child_fid);
2014                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
2015                                 child_fid, &info->mti_spec);
2016                 if (rc == -ENOENT)
2017                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
2018
2019                 if (rc != 0)
2020                         GOTO(unlock_parent, rc);
2021
2022                 child = mdt_object_find(info->mti_env, info->mti_mdt,
2023                                         child_fid);
2024                 if (unlikely(IS_ERR(child)))
2025                         GOTO(unlock_parent, rc = PTR_ERR(child));
2026         }
2027
2028         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
2029
2030         /* step 3: lock child regardless if it is local or remote. */
2031         LASSERT(child);
2032
2033         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
2034         if (!mdt_object_exists(child)) {
2035                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
2036                                 &child->mot_obj,
2037                                 "Object doesn't exist!");
2038                 GOTO(out_child, rc = -ENOENT);
2039         }
2040
2041         rc = mdt_check_resent_lock(info, child, lhc);
2042         if (rc < 0) {
2043                 GOTO(out_child, rc);
2044         } else if (rc > 0) {
2045                 mdt_lock_handle_init(lhc);
2046                 mdt_lock_reg_init(lhc, LCK_PR);
2047
2048                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
2049                     !mdt_object_remote(child)) {
2050                         struct md_attr *ma = &info->mti_attr;
2051
2052                         ma->ma_valid = 0;
2053                         ma->ma_need = MA_INODE;
2054                         rc = mdt_attr_get_complex(info, child, ma);
2055                         if (unlikely(rc != 0))
2056                                 GOTO(out_child, rc);
2057
2058                         /* If the file has not been changed for some time, we
2059                          * return not only a LOOKUP lock, but also an UPDATE
2060                          * lock and this might save us RPC on later STAT. For
2061                          * directories, it also let negative dentry cache start
2062                          * working for this dir. */
2063                         if (ma->ma_valid & MA_INODE &&
2064                             ma->ma_attr.la_valid & LA_CTIME &&
2065                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
2066                             ma->ma_attr.la_ctime < ktime_get_real_seconds())
2067                                 child_bits |= MDS_INODELOCK_UPDATE;
2068                 }
2069
2070                 /* layout lock must be granted in a best-effort way
2071                  * for IT operations */
2072                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
2073                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
2074                     !mdt_object_remote(child) && ldlm_rep != NULL) {
2075                         if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
2076                             exp_connect_layout(info->mti_exp)) {
2077                                 /* try to grant layout lock for regular file. */
2078                                 try_bits = MDS_INODELOCK_LAYOUT;
2079                         }
2080                         /* Acquire DOM lock in advance for data-on-mdt file */
2081                         if (child != parent)
2082                                 try_bits |= MDS_INODELOCK_DOM;
2083                 }
2084
2085                 if (try_bits != 0) {
2086                         /* try layout lock, it may fail to be granted due to
2087                          * contention at LOOKUP or UPDATE */
2088                         rc = mdt_object_lock_try(info, child, lhc, &child_bits,
2089                                                  try_bits, false);
2090                         if (child_bits & MDS_INODELOCK_LAYOUT)
2091                                 ma_need |= MA_LOV;
2092                 } else {
2093                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
2094                          * client will enqueue the lock to the remote MDT */
2095                         if (mdt_object_remote(child))
2096                                 child_bits &= ~MDS_INODELOCK_UPDATE;
2097                         rc = mdt_object_lock(info, child, lhc, child_bits);
2098                 }
2099                 if (unlikely(rc != 0))
2100                         GOTO(out_child, rc);
2101         }
2102
2103         /* finally, we can get attr for child. */
2104         rc = mdt_getattr_internal(info, child, ma_need);
2105         if (unlikely(rc != 0)) {
2106                 mdt_object_unlock(info, child, lhc, 1);
2107                 GOTO(out_child, rc);
2108         }
2109
2110         rc = mdt_pack_secctx_in_reply(info, child);
2111         if (unlikely(rc)) {
2112                 mdt_object_unlock(info, child, lhc, 1);
2113                 GOTO(out_child, rc);
2114         }
2115
2116         rc = mdt_pack_encctx_in_reply(info, child);
2117         if (unlikely(rc)) {
2118                 mdt_object_unlock(info, child, lhc, 1);
2119                 GOTO(out_child, rc);
2120         }
2121
2122         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2123         if (lock) {
2124                 /* Debugging code. */
2125                 LDLM_DEBUG(lock, "Returning lock to client");
2126                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
2127                                          &lock->l_resource->lr_name),
2128                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
2129                          PLDLMRES(lock->l_resource),
2130                          PFID(mdt_object_fid(child)));
2131
2132                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
2133                     !mdt_object_remote(child) && child != parent) {
2134                         mdt_object_put(info->mti_env, child);
2135                         rc = mdt_pack_size2body(info, child_fid,
2136                                                 &lhc->mlh_reg_lh);
2137                         if (rc != 0 && child_bits & MDS_INODELOCK_DOM) {
2138                                 /* DOM lock was taken in advance but this is
2139                                  * not DoM file. Drop the lock.
2140                                  */
2141                                 lock_res_and_lock(lock);
2142                                 ldlm_inodebits_drop(lock, MDS_INODELOCK_DOM);
2143                                 unlock_res_and_lock(lock);
2144                         }
2145                         LDLM_LOCK_PUT(lock);
2146                         GOTO(unlock_parent, rc = 0);
2147                 }
2148                 LDLM_LOCK_PUT(lock);
2149         }
2150
2151         EXIT;
2152 out_child:
2153         if (child)
2154                 mdt_object_put(info->mti_env, child);
2155 unlock_parent:
2156         if (lhp)
2157                 mdt_object_unlock(info, parent, lhp, 1);
2158         return rc;
2159 }
2160
2161 /* normal handler: should release the child lock */
2162 static int mdt_getattr_name(struct tgt_session_info *tsi)
2163 {
2164         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
2165         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
2166         struct mdt_body        *reqbody;
2167         struct mdt_body        *repbody;
2168         int rc, rc2;
2169         ENTRY;
2170
2171         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
2172         LASSERT(reqbody != NULL);
2173         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
2174         LASSERT(repbody != NULL);
2175
2176         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
2177         repbody->mbo_eadatasize = 0;
2178         repbody->mbo_aclsize = 0;
2179
2180         rc = mdt_init_ucred_intent_getattr(info, reqbody);
2181         if (unlikely(rc))
2182                 GOTO(out_shrink, rc);
2183
2184         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
2185         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
2186                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
2187                 lhc->mlh_reg_lh.cookie = 0;
2188         }
2189         mdt_exit_ucred(info);
2190         EXIT;
2191 out_shrink:
2192         mdt_client_compatibility(info);
2193         rc2 = mdt_fix_reply(info);
2194         if (rc == 0)
2195                 rc = rc2;
2196         mdt_thread_info_fini(info);
2197         return rc;
2198 }
2199
2200 static int mdt_rmfid_unlink(struct mdt_thread_info *info,
2201                             const struct lu_fid *pfid,
2202                             const struct lu_name *name,
2203                             struct mdt_object *obj, s64 ctime)
2204 {
2205         struct lu_fid *child_fid = &info->mti_tmp_fid1;
2206         struct ldlm_enqueue_info *einfo = &info->mti_einfo[0];
2207         struct mdt_device *mdt = info->mti_mdt;
2208         struct md_attr *ma = &info->mti_attr;
2209         struct mdt_lock_handle *parent_lh;
2210         struct mdt_lock_handle *child_lh;
2211         struct mdt_object *pobj;
2212         bool cos_incompat = false;
2213         int rc;
2214         ENTRY;
2215
2216         pobj = mdt_object_find(info->mti_env, mdt, pfid);
2217         if (IS_ERR(pobj))
2218                 GOTO(out, rc = PTR_ERR(pobj));
2219
2220         parent_lh = &info->mti_lh[MDT_LH_PARENT];
2221         mdt_lock_pdo_init(parent_lh, LCK_PW, name);
2222         rc = mdt_object_lock(info, pobj, parent_lh, MDS_INODELOCK_UPDATE);
2223         if (rc != 0)
2224                 GOTO(put_parent, rc);
2225
2226         if (mdt_object_remote(pobj))
2227                 cos_incompat = true;
2228
2229         rc = mdo_lookup(info->mti_env, mdt_object_child(pobj),
2230                         name, child_fid, &info->mti_spec);
2231         if (rc != 0)
2232                 GOTO(unlock_parent, rc);
2233
2234         if (!lu_fid_eq(child_fid, mdt_object_fid(obj)))
2235                 GOTO(unlock_parent, rc = -EREMCHG);
2236
2237         child_lh = &info->mti_lh[MDT_LH_CHILD];
2238         mdt_lock_reg_init(child_lh, LCK_EX);
2239         rc = mdt_reint_striped_lock(info, obj, child_lh,
2240                                     MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE,
2241                                     einfo, cos_incompat);
2242         if (rc != 0)
2243                 GOTO(unlock_parent, rc);
2244
2245         if (atomic_read(&obj->mot_open_count)) {
2246                 CDEBUG(D_OTHER, "object "DFID" open, skip\n",
2247                        PFID(mdt_object_fid(obj)));
2248                 GOTO(unlock_child, rc = -EBUSY);
2249         }
2250
2251         ma->ma_need = 0;
2252         ma->ma_valid = MA_INODE;
2253         ma->ma_attr.la_valid = LA_CTIME;
2254         ma->ma_attr.la_ctime = ctime;
2255
2256         mutex_lock(&obj->mot_lov_mutex);
2257
2258         rc = mdo_unlink(info->mti_env, mdt_object_child(pobj),
2259                         mdt_object_child(obj), name, ma, 0);
2260
2261         mutex_unlock(&obj->mot_lov_mutex);
2262
2263 unlock_child:
2264         mdt_reint_striped_unlock(info, obj, child_lh, einfo, 1);
2265 unlock_parent:
2266         mdt_object_unlock(info, pobj, parent_lh, 1);
2267 put_parent:
2268         mdt_object_put(info->mti_env, pobj);
2269 out:
2270         RETURN(rc);
2271 }
2272
2273 static int mdt_rmfid_check_permission(struct mdt_thread_info *info,
2274                                         struct mdt_object *obj)
2275 {
2276         struct lu_ucred *uc = lu_ucred(info->mti_env);
2277         struct md_attr *ma = &info->mti_attr;
2278         struct lu_attr *la = &ma->ma_attr;
2279         int rc = 0;
2280         ENTRY;
2281
2282         ma->ma_need = MA_INODE;
2283         rc = mo_attr_get(info->mti_env, mdt_object_child(obj), ma);
2284         if (rc)
2285                 GOTO(out, rc);
2286
2287         if (la->la_flags & LUSTRE_IMMUTABLE_FL)
2288                         rc = -EACCES;
2289
2290         if (md_capable(uc, CFS_CAP_DAC_OVERRIDE))
2291                 RETURN(0);
2292         if (uc->uc_fsuid == la->la_uid) {
2293                 if ((la->la_mode & S_IWUSR) == 0)
2294                         rc = -EACCES;
2295         } else if (uc->uc_fsgid == la->la_gid) {
2296                 if ((la->la_mode & S_IWGRP) == 0)
2297                         rc = -EACCES;
2298         } else if ((la->la_mode & S_IWOTH) == 0) {
2299                         rc = -EACCES;
2300         }
2301
2302 out:
2303         RETURN(rc);
2304 }
2305
2306 static int mdt_rmfid_one(struct mdt_thread_info *info, struct lu_fid *fid,
2307                          s64 ctime)
2308 {
2309         struct mdt_device *mdt = info->mti_mdt;
2310         struct mdt_object *obj = NULL;
2311         struct linkea_data ldata = { NULL };
2312         struct lu_buf *buf = &info->mti_big_buf;
2313         struct lu_name *name = &info->mti_name;
2314         struct lu_fid *pfid = &info->mti_tmp_fid1;
2315         struct link_ea_header *leh;
2316         struct link_ea_entry *lee;
2317         int reclen, count, rc = 0;
2318         ENTRY;
2319
2320         if (!fid_is_sane(fid))
2321                 GOTO(out, rc = -EINVAL);
2322
2323         if (!fid_is_namespace_visible(fid))
2324                 GOTO(out, rc = -EINVAL);
2325
2326         obj = mdt_object_find(info->mti_env, mdt, fid);
2327         if (IS_ERR(obj))
2328                 GOTO(out, rc = PTR_ERR(obj));
2329
2330         if (mdt_object_remote(obj))
2331                 GOTO(out, rc = -EREMOTE);
2332         if (!mdt_object_exists(obj) || lu_object_is_dying(&obj->mot_header))
2333                 GOTO(out, rc = -ENOENT);
2334
2335         rc = mdt_rmfid_check_permission(info, obj);
2336         if (rc)
2337                 GOTO(out, rc);
2338
2339         /* take LinkEA */
2340         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
2341         if (!buf->lb_buf)
2342                 GOTO(out, rc = -ENOMEM);
2343
2344         ldata.ld_buf = buf;
2345         rc = mdt_links_read(info, obj, &ldata);
2346         if (rc)
2347                 GOTO(out, rc);
2348
2349         leh = buf->lb_buf;
2350         lee = (struct link_ea_entry *)(leh + 1);
2351         for (count = 0; count < leh->leh_reccount; count++) {
2352                 /* remove every hardlink */
2353                 linkea_entry_unpack(lee, &reclen, name, pfid);
2354                 lee = (struct link_ea_entry *) ((char *)lee + reclen);
2355                 rc = mdt_rmfid_unlink(info, pfid, name, obj, ctime);
2356                 if (rc)
2357                         break;
2358         }
2359
2360 out:
2361         if (obj && !IS_ERR(obj))
2362                 mdt_object_put(info->mti_env, obj);
2363         if (info->mti_big_buf.lb_buf)
2364                 lu_buf_free(&info->mti_big_buf);
2365
2366         RETURN(rc);
2367 }
2368
2369 static int mdt_rmfid(struct tgt_session_info *tsi)
2370 {
2371         struct mdt_thread_info *mti = tsi2mdt_info(tsi);
2372         struct mdt_body *reqbody;
2373         struct lu_fid *fids, *rfids;
2374         int bufsize, rc;
2375         __u32 *rcs;
2376         int i, nr;
2377         ENTRY;
2378
2379         reqbody = req_capsule_client_get(tsi->tsi_pill, &RMF_MDT_BODY);
2380         if (reqbody == NULL)
2381                 RETURN(-EPROTO);
2382         bufsize = req_capsule_get_size(tsi->tsi_pill, &RMF_FID_ARRAY,
2383                                        RCL_CLIENT);
2384         nr = bufsize / sizeof(struct lu_fid);
2385         if (nr * sizeof(struct lu_fid) != bufsize)
2386                 RETURN(-EINVAL);
2387         req_capsule_set_size(tsi->tsi_pill, &RMF_RCS,
2388                              RCL_SERVER, nr * sizeof(__u32));
2389         req_capsule_set_size(tsi->tsi_pill, &RMF_FID_ARRAY,
2390                              RCL_SERVER, nr * sizeof(struct lu_fid));
2391         rc = req_capsule_server_pack(tsi->tsi_pill);
2392         if (rc)
2393                 GOTO(out, rc = err_serious(rc));
2394         fids = req_capsule_client_get(tsi->tsi_pill, &RMF_FID_ARRAY);
2395         if (fids == NULL)
2396                 RETURN(-EPROTO);
2397         rcs = req_capsule_server_get(tsi->tsi_pill, &RMF_RCS);
2398         LASSERT(rcs);
2399         rfids = req_capsule_server_get(tsi->tsi_pill, &RMF_FID_ARRAY);
2400         LASSERT(rfids);
2401
2402         mdt_init_ucred(mti, reqbody);
2403         for (i = 0; i < nr; i++) {
2404                 rfids[i] = fids[i];
2405                 rcs[i] = mdt_rmfid_one(mti, fids + i, reqbody->mbo_ctime);
2406         }
2407         mdt_exit_ucred(mti);
2408
2409 out:
2410         RETURN(rc);
2411 }
2412
2413 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2414                          void *karg, void __user *uarg);
2415
2416 static int mdt_set_info(struct tgt_session_info *tsi)
2417 {
2418         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2419         char                    *key;
2420         void                    *val;
2421         int                      keylen, vallen, rc = 0;
2422
2423         ENTRY;
2424
2425         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
2426         if (key == NULL) {
2427                 DEBUG_REQ(D_HA, req, "no set_info key");
2428                 RETURN(err_serious(-EFAULT));
2429         }
2430
2431         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
2432                                       RCL_CLIENT);
2433
2434         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
2435         if (val == NULL) {
2436                 DEBUG_REQ(D_HA, req, "no set_info val");
2437                 RETURN(err_serious(-EFAULT));
2438         }
2439
2440         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
2441                                       RCL_CLIENT);
2442
2443         /* Swab any part of val you need to here */
2444         if (KEY_IS(KEY_READ_ONLY)) {
2445                 spin_lock(&req->rq_export->exp_lock);
2446                 if (*(__u32 *)val)
2447                         *exp_connect_flags_ptr(req->rq_export) |=
2448                                 OBD_CONNECT_RDONLY;
2449                 else
2450                         *exp_connect_flags_ptr(req->rq_export) &=
2451                                 ~OBD_CONNECT_RDONLY;
2452                 spin_unlock(&req->rq_export->exp_lock);
2453         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2454                 struct changelog_setinfo *cs = val;
2455
2456                 if (vallen != sizeof(*cs)) {
2457                         CERROR("%s: bad changelog_clear setinfo size %d\n",
2458                                tgt_name(tsi->tsi_tgt), vallen);
2459                         RETURN(-EINVAL);
2460                 }
2461                 if (ptlrpc_req_need_swab(req)) {
2462                         __swab64s(&cs->cs_recno);
2463                         __swab32s(&cs->cs_id);
2464                 }
2465
2466                 if (!mdt_is_rootadmin(tsi2mdt_info(tsi)))
2467                         RETURN(-EACCES);
2468                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
2469                                    vallen, val, NULL);
2470         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2471                 if (vallen > 0)
2472                         obd_export_evict_by_nid(req->rq_export->exp_obd, val);
2473         } else {
2474                 RETURN(-EINVAL);
2475         }
2476         RETURN(rc);
2477 }
2478
2479 static int mdt_readpage(struct tgt_session_info *tsi)
2480 {
2481         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
2482         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
2483         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
2484         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
2485         struct mdt_body         *repbody;
2486         int                      rc;
2487         int                      i;
2488
2489         ENTRY;
2490
2491         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
2492                 RETURN(err_serious(-ENOMEM));
2493
2494         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
2495         if (repbody == NULL || reqbody == NULL)
2496                 RETURN(err_serious(-EFAULT));
2497
2498         /*
2499          * prepare @rdpg before calling lower layers and transfer itself. Here
2500          * reqbody->size contains offset of where to start to read and
2501          * reqbody->nlink contains number bytes to read.
2502          */
2503         rdpg->rp_hash = reqbody->mbo_size;
2504         if (rdpg->rp_hash != reqbody->mbo_size) {
2505                 CERROR("Invalid hash: %#llx != %#llx\n",
2506                        rdpg->rp_hash, reqbody->mbo_size);
2507                 RETURN(-EFAULT);
2508         }
2509
2510         rdpg->rp_attrs = reqbody->mbo_mode;
2511         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
2512                 rdpg->rp_attrs |= LUDA_64BITHASH;
2513         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
2514                                 exp_max_brw_size(tsi->tsi_exp));
2515         rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >>
2516                           PAGE_SHIFT;
2517         OBD_ALLOC_PTR_ARRAY(rdpg->rp_pages, rdpg->rp_npages);
2518         if (rdpg->rp_pages == NULL)
2519                 RETURN(-ENOMEM);
2520
2521         for (i = 0; i < rdpg->rp_npages; ++i) {
2522                 rdpg->rp_pages[i] = alloc_page(GFP_NOFS);
2523                 if (rdpg->rp_pages[i] == NULL)
2524                         GOTO(free_rdpg, rc = -ENOMEM);
2525         }
2526
2527         /* call lower layers to fill allocated pages with directory data */
2528         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
2529         if (rc < 0)
2530                 GOTO(free_rdpg, rc);
2531
2532         /* send pages to client */
2533         rc = tgt_sendpage(tsi, rdpg, rc);
2534
2535         EXIT;
2536 free_rdpg:
2537
2538         for (i = 0; i < rdpg->rp_npages; i++)
2539                 if (rdpg->rp_pages[i] != NULL)
2540                         __free_page(rdpg->rp_pages[i]);
2541         OBD_FREE_PTR_ARRAY(rdpg->rp_pages, rdpg->rp_npages);
2542
2543         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
2544                 RETURN(0);
2545
2546         return rc;
2547 }
2548
2549 static int mdt_fix_attr_ucred(struct mdt_thread_info *info, __u32 op)
2550 {
2551         struct lu_ucred *uc = mdt_ucred_check(info);
2552         struct lu_attr *attr = &info->mti_attr.ma_attr;
2553
2554         if (uc == NULL)
2555                 return -EINVAL;
2556
2557         if (op != REINT_SETATTR) {
2558                 if ((attr->la_valid & LA_UID) && (attr->la_uid != -1))
2559                         attr->la_uid = uc->uc_fsuid;
2560                 /* for S_ISGID, inherit gid from his parent, such work will be
2561                  * done in cmm/mdd layer, here set all cases as uc->uc_fsgid. */
2562                 if ((attr->la_valid & LA_GID) && (attr->la_gid != -1))
2563                         attr->la_gid = uc->uc_fsgid;
2564         }
2565
2566         return 0;
2567 }
2568
2569 static inline bool mdt_is_readonly_open(struct mdt_thread_info *info, __u32 op)
2570 {
2571         return op == REINT_OPEN &&
2572              !(info->mti_spec.sp_cr_flags & (MDS_FMODE_WRITE | MDS_OPEN_CREAT));
2573 }
2574
2575 static void mdt_preset_secctx_size(struct mdt_thread_info *info)
2576 {
2577         struct req_capsule *pill = info->mti_pill;
2578
2579         if (req_capsule_has_field(pill, &RMF_FILE_SECCTX,
2580                                   RCL_SERVER) &&
2581             req_capsule_has_field(pill, &RMF_FILE_SECCTX_NAME,
2582                                   RCL_CLIENT)) {
2583                 if (req_capsule_get_size(pill, &RMF_FILE_SECCTX_NAME,
2584                                          RCL_CLIENT) != 0)
2585                         /* pre-set size in server part with max size */
2586                         req_capsule_set_size(pill, &RMF_FILE_SECCTX,
2587                                              RCL_SERVER,
2588                                              OBD_MAX_DEFAULT_EA_SIZE);
2589                 else
2590                         req_capsule_set_size(pill, &RMF_FILE_SECCTX,
2591                                              RCL_SERVER, 0);
2592         }
2593 }
2594
2595 static void mdt_preset_encctx_size(struct mdt_thread_info *info)
2596 {
2597         struct req_capsule *pill = info->mti_pill;
2598
2599         if (req_capsule_has_field(pill, &RMF_FILE_ENCCTX,
2600                                   RCL_SERVER))
2601                 /* pre-set size in server part with max size */
2602                 req_capsule_set_size(pill, &RMF_FILE_ENCCTX,
2603                                      RCL_SERVER,
2604                                      info->mti_mdt->mdt_max_mdsize);
2605 }
2606
2607 static int mdt_reint_internal(struct mdt_thread_info *info,
2608                               struct mdt_lock_handle *lhc,
2609                               __u32 op)
2610 {
2611         struct req_capsule      *pill = info->mti_pill;
2612         struct mdt_body         *repbody;
2613         int                      rc = 0, rc2;
2614
2615         ENTRY;
2616
2617         rc = mdt_reint_unpack(info, op);
2618         if (rc != 0) {
2619                 CERROR("Can't unpack reint, rc %d\n", rc);
2620                 RETURN(err_serious(rc));
2621         }
2622
2623
2624         /* check if the file system is set to readonly. O_RDONLY open
2625          * is still allowed even the file system is set to readonly mode */
2626         if (mdt_rdonly(info->mti_exp) && !mdt_is_readonly_open(info, op))
2627                 RETURN(err_serious(-EROFS));
2628
2629         /* for replay (no_create) lmm is not needed, client has it already */
2630         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2631                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2632                                      DEF_REP_MD_SIZE);
2633
2634         /* llog cookies are always 0, the field is kept for compatibility */
2635         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2636                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
2637
2638         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
2639          * by default. If the target object has more ACL entries, then
2640          * enlarge the buffer when necessary. */
2641         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
2642                 req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
2643                                      LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
2644
2645         mdt_preset_secctx_size(info);
2646         mdt_preset_encctx_size(info);
2647
2648         rc = req_capsule_server_pack(pill);
2649         if (rc != 0) {
2650                 CERROR("Can't pack response, rc %d\n", rc);
2651                 RETURN(err_serious(rc));
2652         }
2653
2654         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
2655                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
2656                 LASSERT(repbody);
2657                 repbody->mbo_eadatasize = 0;
2658                 repbody->mbo_aclsize = 0;
2659         }
2660
2661         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
2662
2663         /* for replay no cookkie / lmm need, because client have this already */
2664         if (info->mti_spec.no_create)
2665                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2666                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
2667
2668         rc = mdt_init_ucred_reint(info);
2669         if (rc)
2670                 GOTO(out_shrink, rc);
2671
2672         rc = mdt_fix_attr_ucred(info, op);
2673         if (rc != 0)
2674                 GOTO(out_ucred, rc = err_serious(rc));
2675
2676         rc = mdt_check_resent(info, mdt_reconstruct, lhc);
2677         if (rc < 0) {
2678                 GOTO(out_ucred, rc);
2679         } else if (rc == 1) {
2680                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt");
2681                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
2682                 GOTO(out_ucred, rc);
2683         }
2684         rc = mdt_reint_rec(info, lhc);
2685         EXIT;
2686 out_ucred:
2687         mdt_exit_ucred(info);
2688 out_shrink:
2689         mdt_client_compatibility(info);
2690
2691         rc2 = mdt_fix_reply(info);
2692         if (rc == 0)
2693                 rc = rc2;
2694
2695         /*
2696          * Data-on-MDT optimization - read data along with OPEN and return it
2697          * in reply when possible.
2698          */
2699         if (rc == 0 && op == REINT_OPEN && !req_is_replay(pill->rc_req))
2700                 rc = mdt_dom_read_on_open(info, info->mti_mdt,
2701                                           &lhc->mlh_reg_lh);
2702
2703         return rc;
2704 }
2705
2706 static long mdt_reint_opcode(struct ptlrpc_request *req,
2707                              const struct req_format **fmt)
2708 {
2709         struct mdt_device       *mdt;
2710         struct mdt_rec_reint    *rec;
2711         long                     opc;
2712
2713         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
2714         if (rec != NULL) {
2715                 opc = rec->rr_opcode;
2716                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
2717                 if (opc < REINT_MAX && fmt[opc] != NULL)
2718                         req_capsule_extend(&req->rq_pill, fmt[opc]);
2719                 else {
2720                         mdt = mdt_exp2dev(req->rq_export);
2721                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
2722                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
2723                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
2724                         opc = err_serious(-EFAULT);
2725                 }
2726         } else {
2727                 opc = err_serious(-EFAULT);
2728         }
2729         return opc;
2730 }
2731
2732 static int mdt_reint(struct tgt_session_info *tsi)
2733 {
2734         long opc;
2735         int  rc;
2736         static const struct req_format *reint_fmts[REINT_MAX] = {
2737                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
2738                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
2739                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
2740                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
2741                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
2742                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
2743                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
2744                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
2745                 [REINT_MIGRATE]  = &RQF_MDS_REINT_MIGRATE,
2746                 [REINT_RESYNC]   = &RQF_MDS_REINT_RESYNC,
2747         };
2748
2749         ENTRY;
2750
2751         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
2752         if (opc >= 0) {
2753                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2754                 /*
2755                  * No lock possible here from client to pass it to reint code
2756                  * path.
2757                  */
2758                 rc = mdt_reint_internal(info, NULL, opc);
2759                 mdt_thread_info_fini(info);
2760         } else {
2761                 rc = opc;
2762         }
2763
2764         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
2765         RETURN(rc);
2766 }
2767
2768 /* this should sync the whole device */
2769 int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
2770 {
2771         struct dt_device *dt = mdt->mdt_bottom;
2772         int rc;
2773         ENTRY;
2774
2775         rc = dt->dd_ops->dt_sync(env, dt);
2776         RETURN(rc);
2777 }
2778
2779 /* this should sync this object */
2780 static int mdt_object_sync(const struct lu_env *env, struct obd_export *exp,
2781                            struct mdt_object *mo)
2782 {
2783         int rc = 0;
2784
2785         ENTRY;
2786
2787         if (!mdt_object_exists(mo)) {
2788                 CWARN("%s: non existing object "DFID": rc = %d\n",
2789                       exp->exp_obd->obd_name, PFID(mdt_object_fid(mo)),
2790                       -ESTALE);
2791                 RETURN(-ESTALE);
2792         }
2793
2794         if (S_ISREG(lu_object_attr(&mo->mot_obj))) {
2795                 struct lu_target *tgt = tgt_ses_info(env)->tsi_tgt;
2796                 dt_obj_version_t version;
2797
2798                 version = dt_version_get(env, mdt_obj2dt(mo));
2799                 if (version > tgt->lut_obd->obd_last_committed)
2800                         rc = mo_object_sync(env, mdt_object_child(mo));
2801         } else {
2802                 rc = mo_object_sync(env, mdt_object_child(mo));
2803         }
2804
2805         RETURN(rc);
2806 }
2807
2808 static int mdt_sync(struct tgt_session_info *tsi)
2809 {
2810         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2811         struct req_capsule      *pill = tsi->tsi_pill;
2812         struct mdt_body         *body;
2813         ktime_t                  kstart = ktime_get();
2814         int                      rc;
2815
2816         ENTRY;
2817
2818         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
2819                 RETURN(err_serious(-ENOMEM));
2820
2821         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
2822                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
2823         } else {
2824                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2825
2826                 if (unlikely(info->mti_object == NULL))
2827                         RETURN(-EPROTO);
2828
2829                 /* sync an object */
2830                 rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp,
2831                                      info->mti_object);
2832                 if (rc == 0) {
2833                         const struct lu_fid *fid;
2834                         struct lu_attr *la = &info->mti_attr.ma_attr;
2835
2836                         info->mti_attr.ma_need = MA_INODE;
2837                         info->mti_attr.ma_valid = 0;
2838                         rc = mdt_attr_get_complex(info, info->mti_object,
2839                                                   &info->mti_attr);
2840                         if (rc == 0) {
2841                                 body = req_capsule_server_get(pill,
2842                                                               &RMF_MDT_BODY);
2843                                 fid = mdt_object_fid(info->mti_object);
2844                                 mdt_pack_attr2body(info, body, la, fid);
2845                         }
2846                 }
2847                 mdt_thread_info_fini(info);
2848         }
2849         if (rc == 0)
2850                 mdt_counter_incr(req, LPROC_MDT_SYNC,
2851                                  ktime_us_delta(ktime_get(), kstart));
2852
2853         RETURN(rc);
2854 }
2855
2856 static int mdt_data_sync(struct tgt_session_info *tsi)
2857 {
2858         struct mdt_thread_info *info;
2859         struct mdt_device *mdt = mdt_exp2dev(tsi->tsi_exp);
2860         struct ost_body *body = tsi->tsi_ost_body;
2861         struct ost_body *repbody;
2862         struct mdt_object *mo = NULL;
2863         struct md_attr *ma;
2864         int rc = 0;
2865
2866         ENTRY;
2867
2868         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2869
2870         /* if no fid is specified then do nothing,
2871          * device sync is done via MDS_SYNC */
2872         if (fid_is_zero(&tsi->tsi_fid))
2873                 RETURN(0);
2874
2875         mo = mdt_object_find(tsi->tsi_env, mdt, &tsi->tsi_fid);
2876         if (IS_ERR(mo))
2877                 RETURN(PTR_ERR(mo));
2878
2879         rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp, mo);
2880         if (rc)
2881                 GOTO(put, rc);
2882
2883         repbody->oa.o_oi = body->oa.o_oi;
2884         repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
2885
2886         info = tsi2mdt_info(tsi);
2887         ma = &info->mti_attr;
2888         ma->ma_need = MA_INODE;
2889         ma->ma_valid = 0;
2890         rc = mdt_attr_get_complex(info, mo, ma);
2891         if (rc == 0)
2892                 obdo_from_la(&repbody->oa, &ma->ma_attr, VALID_FLAGS);
2893         else
2894                 rc = 0;
2895         mdt_thread_info_fini(info);
2896
2897         EXIT;
2898 put:
2899         if (mo != NULL)
2900                 mdt_object_put(tsi->tsi_env, mo);
2901         return rc;
2902 }
2903
2904 /*
2905  * Handle quota control requests to consult current usage/limit, but also
2906  * to configure quota enforcement
2907  */
2908 static int mdt_quotactl(struct tgt_session_info *tsi)
2909 {
2910         struct obd_export *exp  = tsi->tsi_exp;
2911         struct req_capsule *pill = tsi->tsi_pill;
2912         struct obd_quotactl *oqctl, *repoqc;
2913         int id, rc;
2914         struct mdt_device *mdt = mdt_exp2dev(exp);
2915         struct lu_device *qmt = mdt->mdt_qmt_dev;
2916         struct lu_nodemap *nodemap;
2917         ENTRY;
2918
2919         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
2920         if (!oqctl)
2921                 RETURN(err_serious(-EPROTO));
2922
2923         rc = req_capsule_server_pack(pill);
2924         if (rc)
2925                 RETURN(err_serious(rc));
2926
2927         nodemap = nodemap_get_from_exp(exp);
2928         if (IS_ERR(nodemap))
2929                 RETURN(PTR_ERR(nodemap));
2930
2931         switch (oqctl->qc_cmd) {
2932                 /* master quotactl */
2933         case Q_SETINFO:
2934         case Q_SETQUOTA:
2935         case LUSTRE_Q_SETDEFAULT:
2936         case LUSTRE_Q_SETQUOTAPOOL:
2937         case LUSTRE_Q_SETINFOPOOL:
2938                 if (!nodemap_can_setquota(nodemap))
2939                         GOTO(out_nodemap, rc = -EPERM);
2940                 /* fallthrough */
2941         case Q_GETINFO:
2942         case Q_GETQUOTA:
2943         case LUSTRE_Q_GETDEFAULT:
2944         case LUSTRE_Q_GETQUOTAPOOL:
2945         case LUSTRE_Q_GETINFOPOOL:
2946                 if (qmt == NULL)
2947                         GOTO(out_nodemap, rc = -EOPNOTSUPP);
2948                 /* slave quotactl */
2949                 /* fallthrough */
2950         case Q_GETOINFO:
2951         case Q_GETOQUOTA:
2952                 break;
2953         default:
2954                 rc = -EFAULT;
2955                 CERROR("%s: unsupported quotactl command %d: rc = %d\n",
2956                        mdt_obd_name(mdt), oqctl->qc_cmd, rc);
2957                 GOTO(out_nodemap, rc);
2958         }
2959
2960         id = oqctl->qc_id;
2961         switch (oqctl->qc_type) {
2962         case USRQUOTA:
2963                 id = nodemap_map_id(nodemap, NODEMAP_UID,
2964                                     NODEMAP_CLIENT_TO_FS, id);
2965                 break;
2966         case GRPQUOTA:
2967                 id = nodemap_map_id(nodemap, NODEMAP_GID,
2968                                     NODEMAP_CLIENT_TO_FS, id);
2969                 break;
2970         case PRJQUOTA:
2971                 /* todo: check/map project id */
2972                 id = oqctl->qc_id;
2973                 break;
2974         default:
2975                 GOTO(out_nodemap, rc = -EOPNOTSUPP);
2976         }
2977         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2978         if (repoqc == NULL)
2979                 GOTO(out_nodemap, rc = err_serious(-EFAULT));
2980
2981         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA)
2982                 barrier_exit(tsi->tsi_tgt->lut_bottom);
2983
2984         if (oqctl->qc_id != id)
2985                 swap(oqctl->qc_id, id);
2986
2987         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA) {
2988                 if (unlikely(!barrier_entry(tsi->tsi_tgt->lut_bottom)))
2989                         RETURN(-EINPROGRESS);
2990         }
2991
2992         switch (oqctl->qc_cmd) {
2993
2994         case Q_GETINFO:
2995         case Q_SETINFO:
2996         case Q_SETQUOTA:
2997         case Q_GETQUOTA:
2998         case LUSTRE_Q_SETDEFAULT:
2999         case LUSTRE_Q_GETDEFAULT:
3000         case LUSTRE_Q_SETQUOTAPOOL:
3001         case LUSTRE_Q_GETQUOTAPOOL:
3002         case LUSTRE_Q_SETINFOPOOL:
3003         case LUSTRE_Q_GETINFOPOOL:
3004                 /* forward quotactl request to QMT */
3005                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
3006                 break;
3007
3008         case Q_GETOINFO:
3009         case Q_GETOQUOTA:
3010                 /* slave quotactl */
3011                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
3012                                    oqctl);
3013                 break;
3014
3015         default:
3016                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
3017                 GOTO(out_nodemap, rc = -EFAULT);
3018         }
3019
3020         if (oqctl->qc_id != id)
3021                 swap(oqctl->qc_id, id);
3022
3023         QCTL_COPY(repoqc, oqctl);
3024         EXIT;
3025
3026 out_nodemap:
3027         nodemap_putref(nodemap);
3028
3029         return rc;
3030 }
3031
3032 /** clone llog ctxt from child (mdd)
3033  * This allows remote llog (replicator) access.
3034  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
3035  * context was originally set up, or we can handle them directly.
3036  * I choose the latter, but that means I need any llog
3037  * contexts set up by child to be accessable by the mdt.  So we clone the
3038  * context into our context list here.
3039  */
3040 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
3041                                int idx)
3042 {
3043         struct md_device  *next = mdt->mdt_child;
3044         struct llog_ctxt *ctxt;
3045         int rc;
3046
3047         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
3048                 return 0;
3049
3050         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
3051         if (rc || ctxt == NULL) {
3052                 return 0;
3053         }
3054
3055         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
3056         if (rc)
3057                 CERROR("Can't set mdt ctxt %d\n", rc);
3058
3059         return rc;
3060 }
3061
3062 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
3063                                  struct mdt_device *mdt, int idx)
3064 {
3065         struct llog_ctxt *ctxt;
3066
3067         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
3068         if (ctxt == NULL)
3069                 return 0;
3070         /* Put once for the get we just did, and once for the clone */
3071         llog_ctxt_put(ctxt);
3072         llog_ctxt_put(ctxt);
3073         return 0;
3074 }
3075
3076 /*
3077  * sec context handlers
3078  */
3079 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
3080 {
3081         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
3082
3083         return 0;
3084 }
3085
3086 /*
3087  * quota request handlers
3088  */
3089 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
3090 {
3091         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
3092         struct lu_device        *qmt = mdt->mdt_qmt_dev;
3093         int                      rc;
3094         ENTRY;
3095
3096         if (qmt == NULL)
3097                 RETURN(err_serious(-EOPNOTSUPP));
3098
3099         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
3100         RETURN(rc);
3101 }
3102
3103 struct mdt_object *mdt_object_new(const struct lu_env *env,
3104                                   struct mdt_device *d,
3105                                   const struct lu_fid *f)
3106 {
3107         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
3108         struct lu_object *o;
3109         struct mdt_object *m;
3110         ENTRY;
3111
3112         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
3113         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
3114         if (unlikely(IS_ERR(o)))
3115                 m = (struct mdt_object *)o;
3116         else
3117                 m = mdt_obj(o);
3118         RETURN(m);
3119 }
3120
3121 struct mdt_object *mdt_object_find(const struct lu_env *env,
3122                                    struct mdt_device *d,
3123                                    const struct lu_fid *f)
3124 {
3125         struct lu_object *o;
3126         struct mdt_object *m;
3127         ENTRY;
3128
3129         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
3130         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
3131         if (unlikely(IS_ERR(o)))
3132                 m = (struct mdt_object *)o;
3133         else
3134                 m = mdt_obj(o);
3135
3136         RETURN(m);
3137 }
3138
3139 /**
3140  * Asyncronous commit for mdt device.
3141  *
3142  * Pass asynchonous commit call down the MDS stack.
3143  *
3144  * \param env environment
3145  * \param mdt the mdt device
3146  */
3147 static void mdt_device_commit_async(const struct lu_env *env,
3148                                     struct mdt_device *mdt)
3149 {
3150         struct dt_device *dt = mdt->mdt_bottom;
3151         int rc;
3152         ENTRY;
3153
3154         rc = dt->dd_ops->dt_commit_async(env, dt);
3155         if (unlikely(rc != 0))
3156                 CWARN("%s: async commit start failed: rc = %d\n",
3157                       mdt_obd_name(mdt), rc);
3158         atomic_inc(&mdt->mdt_async_commit_count);
3159         EXIT;
3160 }
3161
3162 /**
3163  * Mark the lock as "synchonous".
3164  *
3165  * Mark the lock to deffer transaction commit to the unlock time.
3166  *
3167  * \param lock the lock to mark as "synchonous"
3168  *
3169  * \see mdt_is_lock_sync
3170  * \see mdt_save_lock
3171  */
3172 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
3173 {
3174         lock->l_ast_data = (void*)1;
3175 }
3176
3177 /**
3178  * Check whehter the lock "synchonous" or not.
3179  *
3180  * \param lock the lock to check
3181  * \retval 1 the lock is "synchonous"
3182  * \retval 0 the lock isn't "synchronous"
3183  *
3184  * \see mdt_set_lock_sync
3185  * \see mdt_save_lock
3186  */
3187 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
3188 {
3189         return lock->l_ast_data != NULL;
3190 }
3191
3192 /**
3193  * Blocking AST for mdt locks.
3194  *
3195  * Starts transaction commit if in case of COS lock conflict or
3196  * deffers such a commit to the mdt_save_lock.
3197  *
3198  * \param lock the lock which blocks a request or cancelling lock
3199  * \param desc unused
3200  * \param data unused
3201  * \param flag indicates whether this cancelling or blocking callback
3202  * \retval 0
3203  * \see ldlm_blocking_ast_nocheck
3204  */
3205 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
3206                      void *data, int flag)
3207 {
3208         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
3209         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
3210         struct ldlm_cb_set_arg *arg = data;
3211         bool commit_async = false;
3212         int rc;
3213         ENTRY;
3214
3215         if (flag == LDLM_CB_CANCELING)
3216                 RETURN(0);
3217
3218         lock_res_and_lock(lock);
3219         if (lock->l_blocking_ast != mdt_blocking_ast) {
3220                 unlock_res_and_lock(lock);
3221                 RETURN(0);
3222         }
3223
3224         /* A blocking ast may be sent from ldlm_lock_decref_internal
3225          * when the last reference to a local lock was released and
3226          * during blocking event from ldlm_work_bl_ast_lock().
3227          * The 'data' parameter is l_ast_data in the first case and
3228          * callback arguments in the second one. Distinguish them by that.
3229          */
3230         if (!data || data == lock->l_ast_data || !arg->bl_desc)
3231                 goto skip_cos_checks;
3232
3233         if (lock->l_req_mode & (LCK_PW | LCK_EX)) {
3234                 if (mdt_cos_is_enabled(mdt)) {
3235                         if (!arg->bl_desc->bl_same_client)
3236                                 mdt_set_lock_sync(lock);
3237                 } else if (mdt_slc_is_enabled(mdt) &&
3238                            arg->bl_desc->bl_cos_incompat) {
3239                         mdt_set_lock_sync(lock);
3240                         /*
3241                          * we may do extra commit here, but there is a small
3242                          * window to miss a commit: lock was unlocked (saved),
3243                          * then a conflict lock queued and we come here, but
3244                          * REP-ACK not received, so lock was not converted to
3245                          * COS mode yet.
3246                          * Fortunately this window is quite small, so the
3247                          * extra commit should be rare (not to say distributed
3248                          * operation is rare too).
3249                          */
3250                         commit_async = true;
3251                 }
3252         } else if (lock->l_req_mode == LCK_COS) {
3253                 commit_async = true;
3254         }
3255
3256 skip_cos_checks:
3257         rc = ldlm_blocking_ast_nocheck(lock);
3258
3259         if (commit_async) {
3260                 struct lu_env env;
3261
3262                 rc = lu_env_init(&env, LCT_LOCAL);
3263                 if (unlikely(rc != 0))
3264                         CWARN("%s: lu_env initialization failed, cannot "
3265                               "start asynchronous commit: rc = %d\n",
3266                               obd->obd_name, rc);
3267                 else
3268                         mdt_device_commit_async(&env, mdt);
3269                 lu_env_fini(&env);
3270         }
3271         RETURN(rc);
3272 }
3273
3274 /*
3275  * Blocking AST for cross-MDT lock
3276  *
3277  * Discard lock from uncommitted_slc_locks and cancel it.
3278  *
3279  * \param lock  the lock which blocks a request or cancelling lock
3280  * \param desc  unused
3281  * \param data  unused
3282  * \param flag  indicates whether this cancelling or blocking callback
3283  * \retval      0 on success
3284  * \retval      negative number on error
3285  */
3286 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
3287                             void *data, int flag)
3288 {
3289         int rc = 0;
3290         ENTRY;
3291
3292         switch (flag) {
3293         case LDLM_CB_BLOCKING: {
3294                 struct lustre_handle lockh;
3295
3296                 ldlm_lock2handle(lock, &lockh);
3297                 rc = ldlm_cli_cancel(&lockh,
3298                         ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
3299                 if (rc < 0) {
3300                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
3301                         RETURN(rc);
3302                 }
3303                 break;
3304         }
3305         case LDLM_CB_CANCELING: {
3306                 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
3307                 struct mdt_device *mdt =
3308                                 mdt_dev(obd->obd_lu_dev->ld_site->ls_top_dev);
3309
3310                 LDLM_DEBUG(lock, "Revoke remote lock\n");
3311
3312                 /* discard slc lock here so that it can be cleaned anytime,
3313                  * especially for cleanup_resource() */
3314                 tgt_discard_slc_lock(&mdt->mdt_lut, lock);
3315
3316                 /* once we cache lock, l_ast_data is set to mdt_object */
3317                 if (lock->l_ast_data != NULL) {
3318                         struct mdt_object *mo = lock->l_ast_data;
3319                         struct lu_env env;
3320
3321                         rc = lu_env_init(&env, LCT_MD_THREAD);
3322                         if (unlikely(rc != 0)) {
3323                                 CWARN("%s: lu_env initialization failed, object %p "DFID" is leaked!: rc = %d\n",
3324                                       obd->obd_name, mo,
3325                                       PFID(mdt_object_fid(mo)), rc);
3326                                 RETURN(rc);
3327                         }
3328
3329                         if (lock->l_policy_data.l_inodebits.bits &
3330                             (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)) {
3331                                 rc = mo_invalidate(&env, mdt_object_child(mo));
3332                                 mo->mot_cache_attr = 0;
3333                         }
3334                         mdt_object_put(&env, mo);
3335                         lu_env_fini(&env);
3336                 }
3337                 break;
3338         }
3339         default:
3340                 LBUG();
3341         }
3342
3343         RETURN(rc);
3344 }
3345
3346 int mdt_check_resent_lock(struct mdt_thread_info *info,
3347                           struct mdt_object *mo,
3348                           struct mdt_lock_handle *lhc)
3349 {
3350         /* the lock might already be gotten in ldlm_handle_enqueue() */
3351         if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
3352                 struct ptlrpc_request *req = mdt_info_req(info);
3353                 struct ldlm_lock      *lock;
3354
3355                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
3356                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
3357                 if (lock == NULL) {
3358                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
3359                          * a resend case, however, it could be already destroyed
3360                          * due to client eviction or a raced cancel RPC. */
3361                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx",
3362                                           lhc->mlh_reg_lh.cookie);
3363                         RETURN(-ESTALE);
3364                 }
3365
3366                 if (!fid_res_name_eq(mdt_object_fid(mo),
3367                                      &lock->l_resource->lr_name)) {
3368                         CWARN("%s: Although resent, but still not "
3369                               "get child lock:"DFID"\n",
3370                               info->mti_exp->exp_obd->obd_name,
3371                               PFID(mdt_object_fid(mo)));
3372                         LDLM_LOCK_PUT(lock);
3373                         RETURN(-EPROTO);
3374                 }
3375                 LDLM_LOCK_PUT(lock);
3376                 return 0;
3377         }
3378         return 1;
3379 }
3380
3381 static void mdt_remote_object_lock_created_cb(struct ldlm_lock *lock)
3382 {
3383         mdt_object_get(NULL, lock->l_ast_data);
3384 }
3385
3386 int mdt_remote_object_lock_try(struct mdt_thread_info *mti,
3387                                struct mdt_object *o, const struct lu_fid *fid,
3388                                struct lustre_handle *lh, enum ldlm_mode mode,
3389                                __u64 *ibits, __u64 trybits, bool cache)
3390 {
3391         struct ldlm_enqueue_info *einfo = &mti->mti_remote_einfo;
3392         union ldlm_policy_data *policy = &mti->mti_policy;
3393         struct ldlm_res_id *res_id = &mti->mti_res_id;
3394         int rc = 0;
3395         ENTRY;
3396
3397         LASSERT(mdt_object_remote(o));
3398
3399         fid_build_reg_res_name(fid, res_id);
3400
3401         memset(einfo, 0, sizeof(*einfo));
3402         einfo->ei_type = LDLM_IBITS;
3403         einfo->ei_mode = mode;
3404         einfo->ei_cb_bl = mdt_remote_blocking_ast;
3405         einfo->ei_cb_cp = ldlm_completion_ast;
3406         einfo->ei_enq_slave = 0;
3407         einfo->ei_res_id = res_id;
3408
3409         if (cache) {
3410                 /*
3411                  * if we cache lock, couple lock with mdt_object, so that object
3412                  * can be easily found in lock ASTs.
3413                  */
3414                 einfo->ei_cbdata = o;
3415                 einfo->ei_cb_created = mdt_remote_object_lock_created_cb;
3416         }
3417
3418         memset(policy, 0, sizeof(*policy));
3419         policy->l_inodebits.bits = *ibits;
3420         policy->l_inodebits.try_bits = trybits;
3421
3422         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
3423                             policy);
3424
3425         /* Return successfully acquired bits to a caller */
3426         if (rc == 0) {
3427                 struct ldlm_lock *lock = ldlm_handle2lock(lh);
3428
3429                 LASSERT(lock);
3430                 *ibits = lock->l_policy_data.l_inodebits.bits;
3431                 LDLM_LOCK_PUT(lock);
3432         }
3433         RETURN(rc);
3434 }
3435
3436 int mdt_remote_object_lock(struct mdt_thread_info *mti, struct mdt_object *o,
3437                            const struct lu_fid *fid, struct lustre_handle *lh,
3438                            enum ldlm_mode mode, __u64 ibits, bool cache)
3439 {
3440         return mdt_remote_object_lock_try(mti, o, fid, lh, mode, &ibits, 0,
3441                                           cache);
3442 }
3443
3444 int mdt_object_local_lock(struct mdt_thread_info *info, struct mdt_object *o,
3445                           struct mdt_lock_handle *lh, __u64 *ibits,
3446                           __u64 trybits, bool cos_incompat)
3447 {
3448         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
3449         union ldlm_policy_data *policy = &info->mti_policy;
3450         struct ldlm_res_id *res_id = &info->mti_res_id;
3451         __u64 dlmflags = 0, *cookie = NULL;
3452         int rc;
3453         ENTRY;
3454
3455         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
3456         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
3457         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
3458         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
3459
3460         if (cos_incompat) {
3461                 LASSERT(lh->mlh_reg_mode == LCK_PW ||
3462                         lh->mlh_reg_mode == LCK_EX);
3463                 dlmflags |= LDLM_FL_COS_INCOMPAT;
3464         } else if (mdt_cos_is_enabled(info->mti_mdt)) {
3465                 dlmflags |= LDLM_FL_COS_ENABLED;
3466         }
3467
3468         /* Only enqueue LOOKUP lock for remote object */
3469         LASSERT(ergo(mdt_object_remote(o), *ibits == MDS_INODELOCK_LOOKUP));
3470
3471         /* Lease lock are granted with LDLM_FL_CANCEL_ON_BLOCK */
3472         if (lh->mlh_type == MDT_REG_LOCK && lh->mlh_reg_mode == LCK_EX &&
3473             *ibits == MDS_INODELOCK_OPEN)
3474                 dlmflags |= LDLM_FL_CANCEL_ON_BLOCK;
3475
3476         if (lh->mlh_type == MDT_PDO_LOCK) {
3477                 /* check for exists after object is locked */
3478                 if (mdt_object_exists(o) == 0) {
3479                         /* Non-existent object shouldn't have PDO lock */
3480                         RETURN(-ESTALE);
3481                 } else {
3482                         /* Non-dir object shouldn't have PDO lock */
3483                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
3484                                 RETURN(-ENOTDIR);
3485                 }
3486         }
3487
3488         fid_build_reg_res_name(mdt_object_fid(o), res_id);
3489         dlmflags |= LDLM_FL_ATOMIC_CB;
3490
3491         if (info->mti_exp)
3492                 cookie = &info->mti_exp->exp_handle.h_cookie;
3493
3494         /*
3495          * Take PDO lock on whole directory and build correct @res_id for lock
3496          * on part of directory.
3497          */
3498         if (lh->mlh_pdo_hash != 0) {
3499                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
3500                 mdt_lock_pdo_mode(info, o, lh);
3501                 if (lh->mlh_pdo_mode != LCK_NL) {
3502                         /*
3503                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
3504                          * is never going to be sent to client and we do not
3505                          * want it slowed down due to possible cancels.
3506                          */
3507                         policy->l_inodebits.bits =
3508                                 *ibits & MDS_INODELOCK_UPDATE;
3509                         policy->l_inodebits.try_bits =
3510                                 trybits & MDS_INODELOCK_UPDATE;
3511                         /* at least one of them should be set */
3512                         LASSERT(policy->l_inodebits.bits |
3513                                 policy->l_inodebits.try_bits);
3514                         rc = mdt_fid_lock(info->mti_env, ns, &lh->mlh_pdo_lh,
3515                                           lh->mlh_pdo_mode, policy, res_id,
3516                                           dlmflags, cookie);
3517                         if (unlikely(rc != 0))
3518                                 GOTO(out_unlock, rc);
3519                 }
3520
3521                 /*
3522                  * Finish res_id initializing by name hash marking part of
3523                  * directory which is taking modification.
3524                  */
3525                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
3526         }
3527
3528         policy->l_inodebits.bits = *ibits;
3529         policy->l_inodebits.try_bits = trybits;
3530
3531         /*
3532          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
3533          * going to be sent to client. If it is - mdt_intent_policy() path will
3534          * fix it up and turn FL_LOCAL flag off.
3535          */
3536         rc = mdt_fid_lock(info->mti_env, ns, &lh->mlh_reg_lh, lh->mlh_reg_mode,
3537                           policy, res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
3538                           cookie);
3539 out_unlock:
3540         if (rc != 0)
3541                 mdt_object_unlock(info, o, lh, 1);
3542         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
3543                    lh->mlh_pdo_hash != 0 &&
3544                    (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
3545                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
3546
3547         /* Return successfully acquired bits to a caller */
3548         if (rc == 0) {
3549                 struct ldlm_lock *lock = ldlm_handle2lock(&lh->mlh_reg_lh);
3550
3551                 LASSERT(lock);
3552                 *ibits = lock->l_policy_data.l_inodebits.bits;
3553                 LDLM_LOCK_PUT(lock);
3554         }
3555         RETURN(rc);
3556 }
3557
3558 static int
3559 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
3560                          struct mdt_lock_handle *lh, __u64 *ibits,
3561                          __u64 trybits, bool cos_incompat)
3562 {
3563         struct mdt_lock_handle *local_lh = NULL;
3564         int rc;
3565         ENTRY;
3566
3567         if (!mdt_object_remote(o)) {
3568                 rc = mdt_object_local_lock(info, o, lh, ibits, trybits,
3569                                            cos_incompat);
3570                 RETURN(rc);
3571         }
3572
3573         /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
3574         *ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
3575                     MDS_INODELOCK_XATTR);
3576
3577         /* Only enqueue LOOKUP lock for remote object */
3578         if (*ibits & MDS_INODELOCK_LOOKUP) {
3579                 __u64 local = MDS_INODELOCK_LOOKUP;
3580
3581                 rc = mdt_object_local_lock(info, o, lh, &local, 0,
3582                                            cos_incompat);
3583                 if (rc != ELDLM_OK)
3584                         RETURN(rc);
3585
3586                 local_lh = lh;
3587         }
3588
3589         if ((*ibits | trybits) & MDS_INODELOCK_UPDATE) {
3590                 /* Sigh, PDO needs to enqueue 2 locks right now, but
3591                  * enqueue RPC can only request 1 lock, to avoid extra
3592                  * RPC, so it will instead enqueue EX lock for remote
3593                  * object anyway XXX*/
3594                 if (lh->mlh_type == MDT_PDO_LOCK &&
3595                     lh->mlh_pdo_hash != 0) {
3596                         CDEBUG(D_INFO,
3597                                "%s: "DFID" convert PDO lock to EX lock.\n",
3598                                mdt_obd_name(info->mti_mdt),
3599                                PFID(mdt_object_fid(o)));
3600                         lh->mlh_pdo_hash = 0;
3601                         lh->mlh_rreg_mode = LCK_EX;
3602                         lh->mlh_type = MDT_REG_LOCK;
3603                 }
3604
3605                 rc = mdt_remote_object_lock_try(info, o, mdt_object_fid(o),
3606                                                 &lh->mlh_rreg_lh,
3607                                                 lh->mlh_rreg_mode,
3608                                                 ibits, trybits, false);
3609                 if (rc != ELDLM_OK) {
3610                         if (local_lh != NULL)
3611                                 mdt_object_unlock(info, o, local_lh, rc);
3612                         RETURN(rc);
3613                 }
3614         }
3615
3616         /* other components like LFSCK can use lockless access
3617          * and populate cache, so we better invalidate it */
3618         mo_invalidate(info->mti_env, mdt_object_child(o));
3619
3620         RETURN(0);
3621 }
3622
3623 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
3624                     struct mdt_lock_handle *lh, __u64 ibits)
3625 {
3626         return mdt_object_lock_internal(info, o, lh, &ibits, 0, false);
3627 }
3628
3629 int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
3630                           struct mdt_lock_handle *lh, __u64 ibits,
3631                           bool cos_incompat)
3632 {
3633         LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
3634         return mdt_object_lock_internal(info, o, lh, &ibits, 0,
3635                                         cos_incompat);
3636 }
3637
3638 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
3639                         struct mdt_lock_handle *lh, __u64 *ibits,
3640                         __u64 trybits, bool cos_incompat)
3641 {
3642         bool trylock_only = *ibits == 0;
3643         int rc;
3644
3645         LASSERT(!(*ibits & trybits));
3646         rc = mdt_object_lock_internal(info, o, lh, ibits, trybits,
3647                                       cos_incompat);
3648         if (rc && trylock_only) { /* clear error for try ibits lock only */
3649                 LASSERT(*ibits == 0);
3650                 rc = 0;
3651         }
3652         return rc;
3653 }
3654
3655 /**
3656  * Save a lock within request object.
3657  *
3658  * Keep the lock referenced until whether client ACK or transaction
3659  * commit happens or release the lock immediately depending on input
3660  * parameters. If COS is ON, a write lock is converted to COS lock
3661  * before saving.
3662  *
3663  * \param info thead info object
3664  * \param h lock handle
3665  * \param mode lock mode
3666  * \param decref force immediate lock releasing
3667  */
3668 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
3669                    enum ldlm_mode mode, int decref)
3670 {
3671         ENTRY;
3672
3673         if (lustre_handle_is_used(h)) {
3674                 if (decref || !info->mti_has_trans ||
3675                     !(mode & (LCK_PW | LCK_EX))) {
3676                         mdt_fid_unlock(h, mode);
3677                 } else {
3678                         struct mdt_device *mdt = info->mti_mdt;
3679                         struct ldlm_lock *lock = ldlm_handle2lock(h);
3680                         struct ptlrpc_request *req = mdt_info_req(info);
3681                         bool cos = mdt_cos_is_enabled(mdt);
3682                         bool convert_lock = !cos && mdt_slc_is_enabled(mdt);
3683
3684                         LASSERTF(lock != NULL, "no lock for cookie %#llx\n",
3685                                  h->cookie);
3686
3687                         /* there is no request if mdt_object_unlock() is called
3688                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
3689                         if (likely(req != NULL)) {
3690                                 LDLM_DEBUG(lock, "save lock request %p reply "
3691                                         "state %p transno %lld\n", req,
3692                                         req->rq_reply_state, req->rq_transno);
3693                                 if (cos) {
3694                                         ldlm_lock_mode_downgrade(lock, LCK_COS);
3695                                         mode = LCK_COS;
3696                                 }
3697                                 if (req->rq_export->exp_disconnected)
3698                                         mdt_fid_unlock(h, mode);
3699                                 else
3700                                         ptlrpc_save_lock(req, h, mode, cos,
3701                                                          convert_lock);
3702                         } else {
3703                                 mdt_fid_unlock(h, mode);
3704                         }
3705                         if (mdt_is_lock_sync(lock)) {
3706                                 CDEBUG(D_HA, "found sync-lock,"
3707                                        " async commit started\n");
3708                                 mdt_device_commit_async(info->mti_env,
3709                                                         mdt);
3710                         }
3711                         LDLM_LOCK_PUT(lock);
3712                 }
3713                 h->cookie = 0ull;
3714         }
3715
3716         EXIT;
3717 }
3718
3719 /**
3720  * Save cross-MDT lock in uncommitted_slc_locks
3721  *
3722  * Keep the lock referenced until transaction commit happens or release the lock
3723  * immediately depending on input parameters.
3724  *
3725  * \param info thead info object
3726  * \param h lock handle
3727  * \param mode lock mode
3728  * \param decref force immediate lock releasing
3729  */
3730 static void mdt_save_remote_lock(struct mdt_thread_info *info,
3731                                  struct mdt_object *o, struct lustre_handle *h,
3732                                  enum ldlm_mode mode, int decref)
3733 {
3734         ENTRY;
3735
3736         if (lustre_handle_is_used(h)) {
3737                 struct ldlm_lock *lock = ldlm_handle2lock(h);
3738                 struct ptlrpc_request *req = mdt_info_req(info);
3739
3740                 if (o != NULL &&
3741                     (lock->l_policy_data.l_inodebits.bits &
3742                      (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)))
3743                         mo_invalidate(info->mti_env, mdt_object_child(o));
3744
3745                 if (decref || !info->mti_has_trans || !req ||
3746                     !(mode & (LCK_PW | LCK_EX))) {
3747                         ldlm_lock_decref_and_cancel(h, mode);
3748                         LDLM_LOCK_PUT(lock);
3749                 } else {
3750                         tgt_save_slc_lock(&info->mti_mdt->mdt_lut, lock,
3751                                           req->rq_transno);
3752                         ldlm_lock_decref(h, mode);
3753                 }
3754                 h->cookie = 0ull;
3755         }
3756
3757         EXIT;
3758 }
3759
3760 /**
3761  * Unlock mdt object.
3762  *
3763  * Immeditely release the regular lock and the PDO lock or save the
3764  * lock in request and keep them referenced until client ACK or
3765  * transaction commit.
3766  *
3767  * \param info thread info object
3768  * \param o mdt object
3769  * \param lh mdt lock handle referencing regular and PDO locks
3770  * \param decref force immediate lock releasing
3771  *
3772  * XXX o is not used and may be NULL, see hsm_cdt_request_completed().
3773  */
3774 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
3775                        struct mdt_lock_handle *lh, int decref)
3776 {
3777         ENTRY;
3778
3779         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
3780         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
3781         mdt_save_remote_lock(info, o, &lh->mlh_rreg_lh, lh->mlh_rreg_mode,
3782                              decref);
3783
3784         EXIT;
3785 }
3786
3787 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
3788                                         const struct lu_fid *f,
3789                                         struct mdt_lock_handle *lh,
3790                                         __u64 ibits)
3791 {
3792         struct mdt_object *o;
3793
3794         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
3795         if (!IS_ERR(o)) {
3796                 int rc;
3797
3798                 rc = mdt_object_lock(info, o, lh, ibits);
3799                 if (rc != 0) {
3800                         mdt_object_put(info->mti_env, o);
3801                         o = ERR_PTR(rc);
3802                 }
3803         }
3804         return o;
3805 }
3806
3807 void mdt_object_unlock_put(struct mdt_thread_info * info,
3808                            struct mdt_object * o,
3809                            struct mdt_lock_handle *lh,
3810                            int decref)
3811 {
3812         mdt_object_unlock(info, o, lh, decref);
3813         mdt_object_put(info->mti_env, o);
3814 }
3815
3816 /*
3817  * Generic code handling requests that have struct mdt_body passed in:
3818  *
3819  *  - extract mdt_body from request and save it in @info, if present;
3820  *
3821  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
3822  *  @info;
3823  *
3824  *  - if HAS_BODY flag is set for this request type check whether object
3825  *  actually exists on storage (lu_object_exists()).
3826  *
3827  */
3828 static int mdt_body_unpack(struct mdt_thread_info *info,
3829                            enum tgt_handler_flags flags)
3830 {
3831         const struct mdt_body    *body;
3832         struct mdt_object        *obj;
3833         const struct lu_env      *env;
3834         struct req_capsule       *pill;
3835         int                       rc;
3836         ENTRY;
3837
3838         env = info->mti_env;
3839         pill = info->mti_pill;
3840
3841         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
3842         if (body == NULL)
3843                 RETURN(-EFAULT);
3844
3845         if (!(body->mbo_valid & OBD_MD_FLID))
3846                 RETURN(0);
3847
3848         if (!fid_is_sane(&body->mbo_fid1)) {
3849                 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
3850                 RETURN(-EINVAL);
3851         }
3852
3853         obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
3854         if (!IS_ERR(obj)) {
3855                 if ((flags & HAS_BODY) && !mdt_object_exists(obj)) {
3856                         mdt_object_put(env, obj);
3857                         rc = -ENOENT;
3858                 } else {
3859                         info->mti_object = obj;
3860                         rc = 0;
3861                 }
3862         } else
3863                 rc = PTR_ERR(obj);
3864
3865         RETURN(rc);
3866 }
3867
3868 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info,
3869                                    enum tgt_handler_flags flags)
3870 {
3871         struct req_capsule *pill = info->mti_pill;
3872         int rc;
3873
3874         ENTRY;
3875
3876         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
3877                 rc = mdt_body_unpack(info, flags);
3878         else
3879                 rc = 0;
3880
3881         if (rc == 0 && (flags & HAS_REPLY)) {
3882                 /* Pack reply. */
3883                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
3884                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
3885                                              DEF_REP_MD_SIZE);
3886                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
3887                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
3888                                              RCL_SERVER, 0);
3889
3890                 /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
3891                  * by default. If the target object has more ACL entries, then
3892                  * enlarge the buffer when necessary. */
3893                 if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
3894                         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
3895                                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
3896
3897                 mdt_preset_secctx_size(info);
3898                 mdt_preset_encctx_size(info);
3899
3900                 rc = req_capsule_server_pack(pill);
3901                 if (rc)
3902                         CWARN("%s: cannot pack response: rc = %d\n",
3903                                       mdt_obd_name(info->mti_mdt), rc);
3904         }
3905         RETURN(rc);
3906 }
3907
3908 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
3909 {
3910         lh->mlh_type = MDT_NUL_LOCK;
3911         lh->mlh_reg_lh.cookie = 0ull;
3912         lh->mlh_reg_mode = LCK_MINMODE;
3913         lh->mlh_pdo_lh.cookie = 0ull;
3914         lh->mlh_pdo_mode = LCK_MINMODE;
3915         lh->mlh_rreg_lh.cookie = 0ull;
3916         lh->mlh_rreg_mode = LCK_MINMODE;
3917 }
3918
3919 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
3920 {
3921         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
3922         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
3923 }
3924
3925 /*
3926  * Initialize fields of struct mdt_thread_info. Other fields are left in
3927  * uninitialized state, because it's too expensive to zero out whole
3928  * mdt_thread_info (> 1K) on each request arrival.
3929  */
3930 void mdt_thread_info_init(struct ptlrpc_request *req,
3931                           struct mdt_thread_info *info)
3932 {
3933         int i;
3934
3935         info->mti_pill = &req->rq_pill;
3936
3937         /* lock handle */
3938         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3939                 mdt_lock_handle_init(&info->mti_lh[i]);
3940
3941         /* mdt device: it can be NULL while CONNECT */
3942         if (req->rq_export) {
3943                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
3944                 info->mti_exp = req->rq_export;
3945         } else
3946                 info->mti_mdt = NULL;
3947         info->mti_env = req->rq_svc_thread->t_env;
3948         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
3949
3950         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
3951         info->mti_big_buf = LU_BUF_NULL;
3952         info->mti_body = NULL;
3953         info->mti_object = NULL;
3954         info->mti_dlm_req = NULL;
3955         info->mti_has_trans = 0;
3956         info->mti_cross_ref = 0;
3957         info->mti_opdata = 0;
3958         info->mti_big_lmm_used = 0;
3959         info->mti_big_acl_used = 0;
3960         info->mti_som_valid = 0;
3961
3962         info->mti_spec.no_create = 0;
3963         info->mti_spec.sp_rm_entry = 0;
3964         info->mti_spec.sp_permitted = 0;
3965
3966         info->mti_spec.u.sp_ea.eadata = NULL;
3967         info->mti_spec.u.sp_ea.eadatalen = 0;
3968 }
3969
3970 void mdt_thread_info_fini(struct mdt_thread_info *info)
3971 {
3972         int i;
3973
3974         if (info->mti_object != NULL) {
3975                 mdt_object_put(info->mti_env, info->mti_object);
3976                 info->mti_object = NULL;
3977         }
3978
3979         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3980                 mdt_lock_handle_fini(&info->mti_lh[i]);
3981         info->mti_env = NULL;
3982         info->mti_pill = NULL;
3983         info->mti_exp = NULL;
3984         info->mti_mdt = NULL;
3985
3986         if (unlikely(info->mti_big_buf.lb_buf != NULL))
3987                 lu_buf_free(&info->mti_big_buf);
3988 }
3989
3990 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
3991 {
3992         struct mdt_thread_info  *mti;
3993
3994         mti = mdt_th_info(tsi->tsi_env);
3995         LASSERT(mti != NULL);
3996
3997         mdt_thread_info_init(tgt_ses_req(tsi), mti);
3998         if (tsi->tsi_corpus != NULL) {
3999                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
4000                 lu_object_get(tsi->tsi_corpus);
4001         }
4002         mti->mti_body = tsi->tsi_mdt_body;
4003         mti->mti_dlm_req = tsi->tsi_dlm_req;
4004
4005         return mti;
4006 }
4007
4008 static int mdt_tgt_connect(struct tgt_session_info *tsi)
4009 {
4010         if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
4011             cfs_fail_val ==
4012             tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id)
4013                 schedule_timeout_uninterruptible(cfs_time_seconds(3));
4014
4015         return tgt_connect(tsi);
4016 }
4017
4018 static int mdt_intent_glimpse(enum ldlm_intent_flags it_opc,
4019                               struct mdt_thread_info *info,
4020                               struct ldlm_lock **lockp, __u64 flags)
4021 {
4022         return mdt_glimpse_enqueue(info, info->mti_mdt->mdt_namespace,
4023                                    lockp, flags);
4024 }
4025 static int mdt_intent_brw(enum ldlm_intent_flags it_opc,
4026                           struct mdt_thread_info *info,
4027                           struct ldlm_lock **lockp, __u64 flags)
4028 {
4029         return mdt_brw_enqueue(info, info->mti_mdt->mdt_namespace,
4030                                lockp, flags);
4031 }
4032
4033 int mdt_intent_lock_replace(struct mdt_thread_info *info,
4034                             struct ldlm_lock **lockp,
4035                             struct mdt_lock_handle *lh,
4036                             __u64 flags, int result)
4037 {
4038         struct ptlrpc_request  *req = mdt_info_req(info);
4039         struct ldlm_lock       *lock = *lockp;
4040         struct ldlm_lock       *new_lock;
4041
4042         /* If possible resent found a lock, @lh is set to its handle */
4043         new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
4044
4045         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
4046                 lh->mlh_reg_lh.cookie = 0;
4047                 RETURN(0);
4048         }
4049
4050         if (new_lock == NULL && (flags & LDLM_FL_RESENT)) {
4051                 /* Lock is pinned by ldlm_handle_enqueue0() as it is
4052                  * a resend case, however, it could be already destroyed
4053                  * due to client eviction or a raced cancel RPC. */
4054                 LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx\n",
4055                                   lh->mlh_reg_lh.cookie);
4056                 lh->mlh_reg_lh.cookie = 0;
4057                 RETURN(-ESTALE);
4058         }
4059
4060         LASSERTF(new_lock != NULL,
4061                  "lockh %#llx flags %#llx : rc = %d\n",
4062                  lh->mlh_reg_lh.cookie, flags, result);
4063
4064         /*
4065          * If we've already given this lock to a client once, then we should
4066          * have no readers or writers.  Otherwise, we should have one reader
4067          * _or_ writer ref (which will be zeroed below) before returning the
4068          * lock to a client.
4069          */
4070         if (new_lock->l_export == req->rq_export) {
4071                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
4072         } else {
4073                 LASSERT(new_lock->l_export == NULL);
4074                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
4075         }
4076
4077         *lockp = new_lock;
4078
4079         if (new_lock->l_export == req->rq_export) {
4080                 /*
4081                  * Already gave this to the client, which means that we
4082                  * reconstructed a reply.
4083                  */
4084                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
4085                         MSG_RESENT);
4086
4087                 LDLM_LOCK_RELEASE(new_lock);
4088                 lh->mlh_reg_lh.cookie = 0;
4089                 RETURN(ELDLM_LOCK_REPLACED);
4090         }
4091
4092         /*
4093          * Fixup the lock to be given to the client.
4094          */
4095         lock_res_and_lock(new_lock);
4096         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
4097          * possible blocking AST. */
4098         while (new_lock->l_readers > 0) {
4099                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
4100                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
4101                 new_lock->l_readers--;
4102         }
4103         while (new_lock->l_writers > 0) {
4104                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
4105                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
4106                 new_lock->l_writers--;
4107         }
4108
4109         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
4110         new_lock->l_blocking_ast = lock->l_blocking_ast;
4111         new_lock->l_completion_ast = lock->l_completion_ast;
4112         if (ldlm_has_dom(new_lock))
4113                 new_lock->l_glimpse_ast = ldlm_server_glimpse_ast;
4114         new_lock->l_remote_handle = lock->l_remote_handle;
4115         new_lock->l_flags &= ~LDLM_FL_LOCAL;
4116
4117         unlock_res_and_lock(new_lock);
4118
4119         cfs_hash_add(new_lock->l_export->exp_lock_hash,
4120                      &new_lock->l_remote_handle,
4121                      &new_lock->l_exp_hash);
4122
4123         LDLM_LOCK_RELEASE(new_lock);
4124         lh->mlh_reg_lh.cookie = 0;
4125
4126         RETURN(ELDLM_LOCK_REPLACED);
4127 }
4128
4129 void mdt_intent_fixup_resent(struct mdt_thread_info *info,
4130                              struct ldlm_lock *new_lock,
4131                              struct mdt_lock_handle *lh, __u64 flags)
4132 {
4133         struct ptlrpc_request  *req = mdt_info_req(info);
4134         struct ldlm_request    *dlmreq;
4135
4136         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
4137                 return;
4138
4139         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
4140
4141         /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
4142          * lock was found by ldlm_handle_enqueue(); if so @lh must be
4143          * initialized. */
4144         if (flags & LDLM_FL_RESENT) {
4145                 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
4146                 lh->mlh_reg_mode = new_lock->l_granted_mode;
4147
4148                 LDLM_DEBUG(new_lock, "Restoring lock cookie");
4149                 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie %#llx",
4150                           lh->mlh_reg_lh.cookie);
4151                 return;
4152         }
4153
4154         /*
4155          * If the xid matches, then we know this is a resent request, and allow
4156          * it. (It's probably an OPEN, for which we don't send a lock.
4157          */
4158         if (req_can_reconstruct(req, NULL) != 0)
4159                 return;
4160
4161         /*
4162          * This remote handle isn't enqueued, so we never received or processed
4163          * this request.  Clear MSG_RESENT, because it can be handled like any
4164          * normal request now.
4165          */
4166         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
4167
4168         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle %#llx",
4169                   dlmreq->lock_handle[0].cookie);
4170 }
4171
4172 static int mdt_intent_getxattr(enum ldlm_intent_flags it_opc,
4173                                struct mdt_thread_info *info,
4174                                struct ldlm_lock **lockp,
4175                                __u64 flags)
4176 {
4177         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4178         struct ldlm_reply      *ldlm_rep = NULL;
4179         int rc;
4180         ENTRY;
4181
4182         /*
4183          * Initialize lhc->mlh_reg_lh either from a previously granted lock
4184          * (for the resend case) or a new lock. Below we will use it to
4185          * replace the original lock.
4186          */
4187         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4188         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
4189                 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
4190                 rc = mdt_object_lock(info, info->mti_object, lhc,
4191                                      MDS_INODELOCK_XATTR);
4192                 if (rc)
4193                         return rc;
4194         }
4195
4196         rc = mdt_getxattr(info);
4197
4198         if (mdt_info_req(info)->rq_repmsg != NULL)
4199                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4200
4201         if (ldlm_rep == NULL ||
4202             OBD_FAIL_CHECK(OBD_FAIL_MDS_XATTR_REP)) {
4203                 mdt_object_unlock(info,  info->mti_object, lhc, 1);
4204                 if (is_serious(rc))
4205                         RETURN(rc);
4206                 else
4207                         RETURN(err_serious(-EFAULT));
4208         }
4209
4210         ldlm_rep->lock_policy_res2 = clear_serious(rc);
4211
4212         /* This is left for interop instead of adding a new interop flag.
4213          * LU-7433 */
4214 #if LUSTRE_VERSION_CODE > OBD_OCD_VERSION(3, 0, 0, 0)
4215         if (ldlm_rep->lock_policy_res2) {
4216                 mdt_object_unlock(info, info->mti_object, lhc, 1);
4217                 RETURN(ELDLM_LOCK_ABORTED);
4218         }
4219 #endif
4220
4221         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4222         RETURN(rc);
4223 }
4224
4225 static int mdt_intent_getattr(enum ldlm_intent_flags it_opc,
4226                               struct mdt_thread_info *info,
4227                               struct ldlm_lock **lockp,
4228                               __u64 flags)
4229 {
4230         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4231         __u64                   child_bits;
4232         struct ldlm_reply      *ldlm_rep;
4233         struct mdt_body        *reqbody;
4234         struct mdt_body        *repbody;
4235         int                     rc, rc2;
4236         ENTRY;
4237
4238         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
4239         LASSERT(reqbody);
4240
4241         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
4242         LASSERT(repbody);
4243
4244         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
4245         repbody->mbo_eadatasize = 0;
4246         repbody->mbo_aclsize = 0;
4247
4248         switch (it_opc) {
4249         case IT_LOOKUP:
4250                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
4251                 break;
4252         case IT_GETATTR:
4253                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
4254                              MDS_INODELOCK_PERM;
4255                 break;
4256         default:
4257                 CERROR("%s: unsupported intent %#x\n",
4258                        mdt_obd_name(info->mti_mdt), (unsigned int)it_opc);
4259                 GOTO(out_shrink, rc = -EINVAL);
4260         }
4261
4262         rc = mdt_init_ucred_intent_getattr(info, reqbody);
4263         if (rc)
4264                 GOTO(out_shrink, rc);
4265
4266         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4267         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
4268
4269         /* Get lock from request for possible resent case. */
4270         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4271
4272         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
4273         ldlm_rep->lock_policy_res2 = clear_serious(rc);
4274
4275         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
4276                 ldlm_rep->lock_policy_res2 = 0;
4277         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
4278             ldlm_rep->lock_policy_res2) {
4279                 lhc->mlh_reg_lh.cookie = 0ull;
4280                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
4281         }
4282
4283         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4284         EXIT;
4285 out_ucred:
4286         mdt_exit_ucred(info);
4287 out_shrink:
4288         mdt_client_compatibility(info);
4289         rc2 = mdt_fix_reply(info);
4290         if (rc == 0)
4291                 rc = rc2;
4292         return rc;
4293 }
4294
4295 static int mdt_intent_layout(enum ldlm_intent_flags it_opc,
4296                              struct mdt_thread_info *info,
4297                              struct ldlm_lock **lockp,
4298                              __u64 flags)
4299 {
4300         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4301         struct md_layout_change layout = { .mlc_opc = MD_LAYOUT_NOP };
4302         struct layout_intent *intent;
4303         struct ldlm_reply *ldlm_rep;
4304         struct lu_fid *fid = &info->mti_tmp_fid2;
4305         struct mdt_object *obj = NULL;
4306         int layout_size = 0;
4307         struct lu_buf *buf = &layout.mlc_buf;
4308         int rc = 0;
4309
4310         ENTRY;
4311
4312         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
4313
4314         intent = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
4315         if (intent == NULL)
4316                 RETURN(-EPROTO);
4317
4318         CDEBUG(D_INFO, DFID "got layout change request from client: "
4319                "opc:%u flags:%#x extent "DEXT"\n",
4320                PFID(fid), intent->li_opc, intent->li_flags,
4321                PEXT(&intent->li_extent));
4322
4323         switch (intent->li_opc) {
4324         case LAYOUT_INTENT_TRUNC:
4325         case LAYOUT_INTENT_WRITE:
4326                 layout.mlc_opc = MD_LAYOUT_WRITE;
4327                 layout.mlc_intent = intent;
4328                 break;
4329         case LAYOUT_INTENT_ACCESS:
4330                 break;
4331         case LAYOUT_INTENT_READ:
4332         case LAYOUT_INTENT_GLIMPSE:
4333         case LAYOUT_INTENT_RELEASE:
4334         case LAYOUT_INTENT_RESTORE:
4335                 CERROR("%s: Unsupported layout intent opc %d\n",
4336                        mdt_obd_name(info->mti_mdt), intent->li_opc);
4337                 RETURN(-ENOTSUPP);
4338         default:
4339                 CERROR("%s: Unknown layout intent opc %d\n",
4340                        mdt_obd_name(info->mti_mdt), intent->li_opc);
4341                 RETURN(-EINVAL);
4342         }
4343
4344         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
4345         if (IS_ERR(obj))
4346                 RETURN(PTR_ERR(obj));
4347
4348         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
4349                 /* if layout is going to be changed don't use the current EA
4350                  * size but the maximum one. That buffer will be shrinked
4351                  * to the actual size in req_capsule_shrink() before reply.
4352                  */
4353                 if (layout.mlc_opc == MD_LAYOUT_WRITE) {
4354                         layout_size = info->mti_mdt->mdt_max_mdsize;
4355                 } else {
4356                         layout_size = mdt_attr_get_eabuf_size(info, obj);
4357                         if (layout_size < 0)
4358                                 GOTO(out, rc = layout_size);
4359
4360                         if (layout_size > info->mti_mdt->mdt_max_mdsize)
4361                                 info->mti_mdt->mdt_max_mdsize = layout_size;
4362                 }
4363                 CDEBUG(D_INFO, "%s: layout_size %d\n",
4364                        mdt_obd_name(info->mti_mdt), layout_size);
4365         }
4366
4367         /*
4368          * set reply buffer size, so that ldlm_handle_enqueue0()->
4369          * ldlm_lvbo_fill() will fill the reply buffer with lovea.
4370          */
4371         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
4372                              layout_size);
4373         rc = req_capsule_server_pack(info->mti_pill);
4374         if (rc)
4375                 GOTO(out, rc);
4376
4377         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4378         if (!ldlm_rep)
4379                 GOTO(out, rc = -EPROTO);
4380
4381         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
4382
4383         /* take lock in ldlm_lock_enqueue() for LAYOUT_INTENT_ACCESS */
4384         if (layout.mlc_opc == MD_LAYOUT_NOP)
4385                 GOTO(out, rc = 0);
4386
4387         rc = mdt_check_resent(info, mdt_reconstruct_generic, lhc);
4388         if (rc < 0)
4389                 GOTO(out, rc);
4390         if (rc == 1) {
4391                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
4392                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
4393                 GOTO(out, rc);
4394         }
4395
4396         buf->lb_buf = NULL;
4397         buf->lb_len = 0;
4398         if (unlikely(req_is_replay(mdt_info_req(info)))) {
4399                 buf->lb_buf = req_capsule_client_get(info->mti_pill,
4400                                                      &RMF_EADATA);
4401                 buf->lb_len = req_capsule_get_size(info->mti_pill,
4402                                                      &RMF_EADATA, RCL_CLIENT);
4403                 /*
4404                  * If it's a replay of layout write intent RPC, the client has
4405                  * saved the extended lovea when it get reply then.
4406                  */
4407                 if (buf->lb_len > 0)
4408                         mdt_fix_lov_magic(info, buf->lb_buf);
4409         }
4410
4411         /* Get lock from request for possible resent case. */
4412         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4413         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
4414
4415         /*
4416          * Instantiate some layout components, if @buf contains lovea, then it's
4417          * a replay of the layout intent write RPC.
4418          */
4419         rc = mdt_layout_change(info, obj, lhc, &layout);
4420         ldlm_rep->lock_policy_res2 = clear_serious(rc);
4421
4422         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
4423                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4424                 if (rc == ELDLM_LOCK_REPLACED &&
4425                     (*lockp)->l_granted_mode == LCK_EX)
4426                         ldlm_lock_mode_downgrade(*lockp, LCK_CR);
4427         }
4428
4429         EXIT;
4430 out:
4431         mdt_object_put(info->mti_env, obj);
4432         return rc;
4433 }
4434
4435 static int mdt_intent_open(enum ldlm_intent_flags it_opc,
4436                            struct mdt_thread_info *info,
4437                            struct ldlm_lock **lockp,
4438                            __u64 flags)
4439 {
4440         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4441         struct ldlm_reply      *rep = NULL;
4442         long                    opc;
4443         int                     rc;
4444
4445         static const struct req_format *intent_fmts[REINT_MAX] = {
4446                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
4447                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
4448         };
4449
4450         ENTRY;
4451
4452         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
4453         if (opc < 0)
4454                 RETURN(opc);
4455
4456         /* Get lock from request for possible resent case. */
4457         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4458
4459         rc = mdt_reint_internal(info, lhc, opc);
4460
4461         /* Check whether the reply has been packed successfully. */
4462         if (mdt_info_req(info)->rq_repmsg != NULL)
4463                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4464         if (rep == NULL) {
4465                 if (is_serious(rc))
4466                         RETURN(rc);
4467                 else
4468                         RETURN(err_serious(-EFAULT));
4469         }
4470
4471         /* MDC expects this in any case */
4472         if (rc != 0)
4473                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
4474
4475         /* the open lock or the lock for cross-ref object should be
4476          * returned to the client */
4477         if (lustre_handle_is_used(&lhc->mlh_reg_lh) &&
4478             (rc == 0 || rc == -MDT_EREMOTE_OPEN)) {
4479                 rep->lock_policy_res2 = 0;
4480                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4481                 RETURN(rc);
4482         }
4483
4484         rep->lock_policy_res2 = clear_serious(rc);
4485
4486         if (rep->lock_policy_res2 == -ENOENT &&
4487             mdt_get_disposition(rep, DISP_LOOKUP_NEG) &&
4488             !mdt_get_disposition(rep, DISP_OPEN_CREATE))
4489                 rep->lock_policy_res2 = 0;
4490
4491         lhc->mlh_reg_lh.cookie = 0ull;
4492         if (rc == -ENOTCONN || rc == -ENODEV ||
4493             rc == -EOVERFLOW) { /**< if VBR failure then return error */
4494                 /*
4495                  * If it is the disconnect error (ENODEV & ENOCONN), the error
4496                  * will be returned by rq_status, and client at ptlrpc layer
4497                  * will detect this, then disconnect, reconnect the import
4498                  * immediately, instead of impacting the following the rpc.
4499                  */
4500                 RETURN(rc);
4501         }
4502         /*
4503          * For other cases, the error will be returned by intent, and client
4504          * will retrieve the result from intent.
4505          */
4506         RETURN(ELDLM_LOCK_ABORTED);
4507 }
4508
4509 static int mdt_intent_opc(enum ldlm_intent_flags it_opc,
4510                           struct mdt_thread_info *info,
4511                           struct ldlm_lock **lockp,
4512                           u64 flags /* LDLM_FL_* */)
4513 {
4514         struct req_capsule *pill = info->mti_pill;
4515         struct ptlrpc_request *req = mdt_info_req(info);
4516         const struct req_format *it_format;
4517         int (*it_handler)(enum ldlm_intent_flags,
4518                           struct mdt_thread_info *,
4519                           struct ldlm_lock **,
4520                           u64);
4521         enum tgt_handler_flags it_handler_flags = 0;
4522         struct ldlm_reply *rep;
4523         bool check_mdt_object = false;
4524         int rc;
4525         ENTRY;
4526
4527         switch (it_opc) {
4528         case IT_OPEN:
4529         case IT_OPEN|IT_CREAT:
4530                 /*
4531                  * OCREAT is not a IS_MUTABLE request since the file may
4532                  * already exist. We do the extra check of
4533                  * OBD_CONNECT_RDONLY in mdt_reint_open() when we
4534                  * really need to create the object.
4535                  */
4536                 it_format = &RQF_LDLM_INTENT;
4537                 it_handler = &mdt_intent_open;
4538                 break;
4539         case IT_GETATTR:
4540                 check_mdt_object = true;
4541                 /* fallthrough */
4542         case IT_LOOKUP:
4543                 it_format = &RQF_LDLM_INTENT_GETATTR;
4544                 it_handler = &mdt_intent_getattr;
4545                 it_handler_flags = HAS_REPLY;
4546                 break;
4547         case IT_GETXATTR:
4548                 check_mdt_object = true;
4549                 it_format = &RQF_LDLM_INTENT_GETXATTR;
4550                 it_handler = &mdt_intent_getxattr;
4551                 it_handler_flags = HAS_BODY;
4552                 break;
4553         case IT_LAYOUT:
4554                 it_format = &RQF_LDLM_INTENT_LAYOUT;
4555                 it_handler = &mdt_intent_layout;
4556                 break;
4557         case IT_GLIMPSE:
4558                 it_format = &RQF_LDLM_INTENT;
4559                 it_handler = &mdt_intent_glimpse;
4560                 break;
4561         case IT_BRW:
4562                 it_format = &RQF_LDLM_INTENT;
4563                 it_handler = &mdt_intent_brw;
4564                 break;
4565         case IT_QUOTA_DQACQ:
4566         case IT_QUOTA_CONN: {
4567                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
4568
4569                 if (qmt == NULL)
4570                         RETURN(-EOPNOTSUPP);
4571
4572                 if (mdt_rdonly(req->rq_export))
4573                         RETURN(-EROFS);
4574
4575                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
4576                 /* pass the request to quota master */
4577                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
4578                                                  mdt_info_req(info), lockp,
4579                                                  flags);
4580                 RETURN(rc);
4581         }
4582         default:
4583                 CERROR("%s: unknown intent code %#x\n",
4584                        mdt_obd_name(info->mti_mdt), it_opc);
4585                 RETURN(-EPROTO);
4586         }
4587
4588         req_capsule_extend(pill, it_format);
4589
4590         rc = mdt_unpack_req_pack_rep(info, it_handler_flags);
4591         if (rc < 0)
4592                 RETURN(rc);
4593
4594         if (unlikely(info->mti_object == NULL && check_mdt_object))
4595                 RETURN(-EPROTO);
4596
4597         if (it_handler_flags & IS_MUTABLE && mdt_rdonly(req->rq_export))
4598                 RETURN(-EROFS);
4599
4600         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_INTENT_DELAY, 10);
4601
4602         /* execute policy */
4603         rc = (*it_handler)(it_opc, info, lockp, flags);
4604
4605         /* Check whether the reply has been packed successfully. */
4606         if (req->rq_repmsg != NULL) {
4607                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4608                 rep->lock_policy_res2 =
4609                         ptlrpc_status_hton(rep->lock_policy_res2);
4610         }
4611
4612         RETURN(rc);
4613 }
4614
4615 static void mdt_ptlrpc_stats_update(struct ptlrpc_request *req,
4616                                     enum ldlm_intent_flags it_opc)
4617 {
4618         struct lprocfs_stats *srv_stats = ptlrpc_req2svc(req)->srv_stats;
4619
4620         /* update stats when IT code is known */
4621         if (srv_stats != NULL)
4622                 lprocfs_counter_incr(srv_stats,
4623                                 PTLRPC_LAST_CNTR + (it_opc == IT_GLIMPSE ?
4624                                 LDLM_GLIMPSE_ENQUEUE : LDLM_IBITS_ENQUEUE));
4625 }
4626
4627 static int mdt_intent_policy(const struct lu_env *env,
4628                              struct ldlm_namespace *ns,
4629                              struct ldlm_lock **lockp,
4630                              void *req_cookie,
4631                              enum ldlm_mode mode,
4632                              __u64 flags, void *data)
4633 {
4634         struct tgt_session_info *tsi;
4635         struct mdt_thread_info  *info;
4636         struct ptlrpc_request   *req  =  req_cookie;
4637         struct ldlm_intent      *it;
4638         struct req_capsule      *pill;
4639         const struct ldlm_lock_desc *ldesc;
4640         int rc;
4641
4642         ENTRY;
4643
4644         LASSERT(req != NULL);
4645
4646         tsi = tgt_ses_info(env);
4647
4648         info = tsi2mdt_info(tsi);
4649         LASSERT(info != NULL);
4650         pill = info->mti_pill;
4651         LASSERT(pill->rc_req == req);
4652         ldesc = &info->mti_dlm_req->lock_desc;
4653
4654         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
4655                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
4656                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
4657                 if (it != NULL) {
4658                         mdt_ptlrpc_stats_update(req, it->opc);
4659                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
4660                         if (rc == 0)
4661                                 rc = ELDLM_OK;
4662
4663                         /* Lock without inodebits makes no sense and will oops
4664                          * later in ldlm. Let's check it now to see if we have
4665                          * ibits corrupted somewhere in mdt_intent_opc().
4666                          * The case for client miss to set ibits has been
4667                          * processed by others. */
4668                         LASSERT(ergo(ldesc->l_resource.lr_type == LDLM_IBITS,
4669                                 ldesc->l_policy_data.l_inodebits.bits != 0));
4670                 } else {
4671                         rc = err_serious(-EFAULT);
4672                 }
4673         } else {
4674                 /* No intent was provided */
4675                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
4676                 rc = req_capsule_server_pack(pill);
4677                 if (rc)
4678                         rc = err_serious(rc);
4679         }
4680         mdt_thread_info_fini(info);
4681         RETURN(rc);
4682 }
4683
4684 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
4685 {
4686         struct seq_server_site  *ss = mdt_seq_site(mdt);
4687
4688         if (ss->ss_node_id == 0)
4689                 return;
4690
4691         if (ss->ss_client_seq != NULL) {
4692                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
4693                 ss->ss_client_seq->lcs_exp = NULL;
4694         }
4695
4696         if (ss->ss_server_fld != NULL) {
4697                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
4698                 ss->ss_server_fld->lsf_control_exp = NULL;
4699         }
4700 }
4701
4702 static void mdt_seq_fini_cli(struct mdt_device *mdt)
4703 {
4704         struct seq_server_site *ss = mdt_seq_site(mdt);
4705
4706         if (ss == NULL)
4707                 return;
4708
4709         if (ss->ss_server_seq != NULL)
4710                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
4711 }
4712
4713 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
4714 {
4715         mdt_seq_fini_cli(mdt);
4716         mdt_deregister_seq_exp(mdt);
4717
4718         return seq_site_fini(env, mdt_seq_site(mdt));
4719 }
4720
4721 /**
4722  * It will retrieve its FLDB entries from MDT0, and it only happens
4723  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
4724  * and it needs to refresh FLDB from the MDT0.
4725  **/
4726 static int mdt_register_lwp_callback(void *data)
4727 {
4728         struct lu_env           env;
4729         struct mdt_device       *mdt = data;
4730         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
4731         int                     rc;
4732         ENTRY;
4733
4734         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
4735
4736         rc = lu_env_init(&env, LCT_MD_THREAD);
4737         if (rc < 0) {
4738                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
4739                 RETURN(rc);
4740         }
4741
4742         /* Allocate new sequence now to avoid creating local transaction
4743          * in the normal transaction process */
4744         rc = seq_server_check_and_alloc_super(&env,
4745                                               mdt_seq_site(mdt)->ss_server_seq);
4746         if (rc < 0)
4747                 GOTO(out, rc);
4748
4749         if (fld->lsf_new) {
4750                 rc = fld_update_from_controller(&env, fld);
4751                 if (rc != 0) {
4752                         CERROR("%s: cannot update controller: rc = %d\n",
4753                                mdt_obd_name(mdt), rc);
4754                         GOTO(out, rc);
4755                 }
4756         }
4757 out:
4758         lu_env_fini(&env);
4759         RETURN(rc);
4760 }
4761
4762 static int mdt_register_seq_exp(struct mdt_device *mdt)
4763 {
4764         struct seq_server_site  *ss = mdt_seq_site(mdt);
4765         char                    *lwp_name = NULL;
4766         int                     rc;
4767
4768         if (ss->ss_node_id == 0)
4769                 return 0;
4770
4771         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
4772         if (lwp_name == NULL)
4773                 GOTO(out_free, rc = -ENOMEM);
4774
4775         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
4776         if (rc != 0)
4777                 GOTO(out_free, rc);
4778
4779         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
4780                                       NULL, NULL);
4781         if (rc != 0)
4782                 GOTO(out_free, rc);
4783
4784         rc = lustre_register_lwp_item(lwp_name,
4785                                       &ss->ss_server_fld->lsf_control_exp,
4786                                       mdt_register_lwp_callback, mdt);
4787         if (rc != 0) {
4788                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
4789                 ss->ss_client_seq->lcs_exp = NULL;
4790                 GOTO(out_free, rc);
4791         }
4792 out_free:
4793         if (lwp_name != NULL)
4794                 OBD_FREE(lwp_name, MAX_OBD_NAME);
4795
4796         return rc;
4797 }
4798
4799 /*
4800  * Init client sequence manager which is used by local MDS to talk to sequence
4801  * controller on remote node.
4802  */
4803 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
4804 {
4805         struct seq_server_site *ss = mdt_seq_site(mdt);
4806         char *prefix;
4807         ENTRY;
4808
4809         /* check if this is adding the first MDC and controller is not yet
4810          * initialized. */
4811         OBD_ALLOC_PTR(ss->ss_client_seq);
4812         if (ss->ss_client_seq == NULL)
4813                 RETURN(-ENOMEM);
4814
4815         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
4816         if (prefix == NULL) {
4817                 OBD_FREE_PTR(ss->ss_client_seq);
4818                 ss->ss_client_seq = NULL;
4819                 RETURN(-ENOMEM);
4820         }
4821
4822         /* Note: seq_client_fini will be called in seq_site_fini */
4823         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
4824         seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
4825                         prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
4826                                                             NULL);
4827         OBD_FREE(prefix, MAX_OBD_NAME + 5);
4828
4829         RETURN(seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq));
4830 }
4831
4832 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
4833 {
4834         struct seq_server_site  *ss;
4835         int                     rc;
4836         ENTRY;
4837
4838         ss = mdt_seq_site(mdt);
4839         /* init sequence controller server(MDT0) */
4840         if (ss->ss_node_id == 0) {
4841                 OBD_ALLOC_PTR(ss->ss_control_seq);
4842                 if (ss->ss_control_seq == NULL)
4843                         RETURN(-ENOMEM);
4844
4845                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
4846                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
4847                                      ss);
4848                 if (rc)
4849                         GOTO(out_seq_fini, rc);
4850         }
4851
4852         /* Init normal sequence server */
4853         OBD_ALLOC_PTR(ss->ss_server_seq);
4854         if (ss->ss_server_seq == NULL)
4855                 GOTO(out_seq_fini, rc = -ENOMEM);
4856
4857         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
4858                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
4859         if (rc)
4860                 GOTO(out_seq_fini, rc);
4861
4862         /* init seq client for seq server to talk to seq controller(MDT0) */
4863         rc = mdt_seq_init_cli(env, mdt);
4864         if (rc != 0)
4865                 GOTO(out_seq_fini, rc);
4866
4867         if (ss->ss_node_id != 0)
4868                 /* register controller export through lwp */
4869                 rc = mdt_register_seq_exp(mdt);
4870
4871         EXIT;
4872 out_seq_fini:
4873         if (rc)
4874                 mdt_seq_fini(env, mdt);
4875
4876         return rc;
4877 }
4878
4879 /*
4880  * FLD wrappers
4881  */
4882 static int mdt_fld_fini(const struct lu_env *env,
4883                         struct mdt_device *m)
4884 {
4885         struct seq_server_site *ss = mdt_seq_site(m);
4886         ENTRY;
4887
4888         if (ss && ss->ss_server_fld) {
4889                 fld_server_fini(env, ss->ss_server_fld);
4890                 OBD_FREE_PTR(ss->ss_server_fld);
4891                 ss->ss_server_fld = NULL;
4892         }
4893
4894         RETURN(0);
4895 }
4896
4897 static int mdt_fld_init(const struct lu_env *env,
4898                         const char *uuid,
4899                         struct mdt_device *m)
4900 {
4901         struct seq_server_site *ss;
4902         int rc;
4903         ENTRY;
4904
4905         ss = mdt_seq_site(m);
4906
4907         OBD_ALLOC_PTR(ss->ss_server_fld);
4908         if (ss->ss_server_fld == NULL)
4909                 RETURN(rc = -ENOMEM);
4910
4911         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
4912                              LU_SEQ_RANGE_MDT);
4913         if (rc) {
4914                 OBD_FREE_PTR(ss->ss_server_fld);
4915                 ss->ss_server_fld = NULL;
4916                 RETURN(rc);
4917         }
4918
4919         RETURN(0);
4920 }
4921
4922 static void mdt_stack_pre_fini(const struct lu_env *env,
4923                            struct mdt_device *m, struct lu_device *top)
4924 {
4925         struct lustre_cfg_bufs  *bufs;
4926         struct lustre_cfg       *lcfg;
4927         struct mdt_thread_info  *info;
4928         ENTRY;
4929
4930         LASSERT(top);
4931
4932         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4933         LASSERT(info != NULL);
4934
4935         bufs = &info->mti_u.bufs;
4936
4937         LASSERT(m->mdt_child_exp);
4938         LASSERT(m->mdt_child_exp->exp_obd);
4939
4940         /* process cleanup, pass mdt obd name to get obd umount flags */
4941         /* XXX: this is needed because all layers are referenced by
4942          * objects (some of them are pinned by osd, for example *
4943          * the proper solution should be a model where object used
4944          * by osd only doesn't have mdt/mdd slices -bzzz */
4945         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4946         lustre_cfg_bufs_set_string(bufs, 1, NULL);
4947         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4948         if (!lcfg)
4949                 RETURN_EXIT;
4950         lustre_cfg_init(lcfg, LCFG_PRE_CLEANUP, bufs);
4951
4952         top->ld_ops->ldo_process_config(env, top, lcfg);
4953         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4954         EXIT;
4955 }
4956
4957 static void mdt_stack_fini(const struct lu_env *env,
4958                            struct mdt_device *m, struct lu_device *top)
4959 {
4960         struct obd_device       *obd = mdt2obd_dev(m);
4961         struct lustre_cfg_bufs  *bufs;
4962         struct lustre_cfg       *lcfg;
4963         struct mdt_thread_info  *info;
4964         char                     flags[3] = "";
4965         ENTRY;
4966
4967         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4968         LASSERT(info != NULL);
4969
4970         lu_dev_del_linkage(top->ld_site, top);
4971
4972         lu_site_purge(env, top->ld_site, -1);
4973
4974         bufs = &info->mti_u.bufs;
4975         /* process cleanup, pass mdt obd name to get obd umount flags */
4976         /* another purpose is to let all layers to release their objects */
4977         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4978         if (obd->obd_force)
4979                 strcat(flags, "F");
4980         if (obd->obd_fail)
4981                 strcat(flags, "A");
4982         lustre_cfg_bufs_set_string(bufs, 1, flags);
4983         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4984         if (!lcfg)
4985                 RETURN_EXIT;
4986         lustre_cfg_init(lcfg, LCFG_CLEANUP, bufs);
4987
4988         LASSERT(top);
4989         top->ld_ops->ldo_process_config(env, top, lcfg);
4990         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4991
4992         lu_site_purge(env, top->ld_site, -1);
4993
4994         m->mdt_child = NULL;
4995         m->mdt_bottom = NULL;
4996
4997         obd_disconnect(m->mdt_child_exp);
4998         m->mdt_child_exp = NULL;
4999
5000         obd_disconnect(m->mdt_bottom_exp);
5001         m->mdt_child_exp = NULL;
5002 }
5003
5004 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
5005                                const char *next, struct obd_export **exp)
5006 {
5007         struct obd_connect_data *data = NULL;
5008         struct obd_device       *obd;
5009         int                      rc;
5010         ENTRY;
5011
5012         OBD_ALLOC_PTR(data);
5013         if (data == NULL)
5014                 GOTO(out, rc = -ENOMEM);
5015
5016         obd = class_name2obd(next);
5017         if (obd == NULL) {
5018                 CERROR("%s: can't locate next device: %s\n",
5019                        mdt_obd_name(m), next);
5020                 GOTO(out, rc = -ENOTCONN);
5021         }
5022
5023         data->ocd_connect_flags = OBD_CONNECT_VERSION;
5024         data->ocd_version = LUSTRE_VERSION_CODE;
5025
5026         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
5027         if (rc) {
5028                 CERROR("%s: cannot connect to next dev %s (%d)\n",
5029                        mdt_obd_name(m), next, rc);
5030                 GOTO(out, rc);
5031         }
5032
5033 out:
5034         if (data)
5035                 OBD_FREE_PTR(data);
5036         RETURN(rc);
5037 }
5038
5039 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
5040                           struct lustre_cfg *cfg)
5041 {
5042         char                   *dev = lustre_cfg_string(cfg, 0);
5043         int                     rc, name_size, uuid_size;
5044         char                   *name, *uuid, *p;
5045         struct lustre_cfg_bufs *bufs;
5046         struct lustre_cfg      *lcfg;
5047         struct obd_device      *obd;
5048         struct lustre_profile  *lprof;
5049         struct lu_site         *site;
5050         ENTRY;
5051
5052         /* in 1.8 we had the only device in the stack - MDS.
5053          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
5054          * in 2.3 OSD is instantiated by obd_mount.c, so we need
5055          * to generate names and setup MDT, MDD. MDT will be using
5056          * generated name to connect to MDD. for MDD the next device
5057          * will be LOD with name taken from so called "profile" which
5058          * is generated by mount_option line
5059          *
5060          * 1.8 MGS generates config. commands like this:
5061          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
5062          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
5063          * 2.0 MGS generates config. commands like this:
5064          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
5065          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
5066          *                    3:lustre-MDT0000-mdtlov  4:f
5067          *
5068          * we generate MDD name from MDT one, just replacing T with D
5069          *
5070          * after all the preparations, the logical equivalent will be
5071          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
5072          *                    3:lustre-MDT0000-mdtlov  4:f
5073          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
5074          *                    3:lustre-MDD0000  4:f
5075          *
5076          *  notice we build the stack from down to top: MDD first, then MDT */
5077
5078         name_size = MAX_OBD_NAME;
5079         uuid_size = MAX_OBD_NAME;
5080
5081         OBD_ALLOC(name, name_size);
5082         OBD_ALLOC(uuid, uuid_size);
5083         if (name == NULL || uuid == NULL)
5084                 GOTO(cleanup_mem, rc = -ENOMEM);
5085
5086         OBD_ALLOC_PTR(bufs);
5087         if (!bufs)
5088                 GOTO(cleanup_mem, rc = -ENOMEM);
5089
5090         strcpy(name, dev);
5091         p = strstr(name, "-MDT");
5092         if (p == NULL)
5093                 GOTO(free_bufs, rc = -ENOMEM);
5094         p[3] = 'D';
5095
5096         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
5097
5098         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
5099         if (lprof == NULL || lprof->lp_dt == NULL) {
5100                 CERROR("can't find the profile: %s\n",
5101                        lustre_cfg_string(cfg, 0));
5102                 GOTO(free_bufs, rc = -EINVAL);
5103         }
5104
5105         lustre_cfg_bufs_reset(bufs, name);
5106         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
5107         lustre_cfg_bufs_set_string(bufs, 2, uuid);
5108         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
5109
5110         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5111         if (!lcfg)
5112                 GOTO(put_profile, rc = -ENOMEM);
5113         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
5114
5115         rc = class_attach(lcfg);
5116         if (rc)
5117                 GOTO(lcfg_cleanup, rc);
5118
5119         obd = class_name2obd(name);
5120         if (!obd) {
5121                 CERROR("Can not find obd %s (%s in config)\n",
5122                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
5123                 GOTO(lcfg_cleanup, rc = -EINVAL);
5124         }
5125
5126         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5127
5128         lustre_cfg_bufs_reset(bufs, name);
5129         lustre_cfg_bufs_set_string(bufs, 1, uuid);
5130         lustre_cfg_bufs_set_string(bufs, 2, dev);
5131         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
5132
5133         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5134         if (!lcfg)
5135                 GOTO(class_detach, rc = -ENOMEM);
5136         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
5137
5138         rc = class_setup(obd, lcfg);
5139         if (rc)
5140                 GOTO(class_detach, rc);
5141
5142         /* connect to MDD we just setup */
5143         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
5144         if (rc)
5145                 GOTO(class_detach, rc);
5146
5147         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
5148         LASSERT(site);
5149         LASSERT(mdt_lu_site(mdt) == NULL);
5150         mdt->mdt_lu_dev.ld_site = site;
5151         site->ls_top_dev = &mdt->mdt_lu_dev;
5152         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
5153
5154         /* now connect to bottom OSD */
5155         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
5156         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
5157         if (rc)
5158                 GOTO(class_detach, rc);
5159         mdt->mdt_bottom =
5160                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
5161
5162         rc = lu_env_refill((struct lu_env *)env);
5163         if (rc != 0)
5164                 CERROR("Failure to refill session: '%d'\n", rc);
5165
5166         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
5167
5168         EXIT;
5169 class_detach:
5170         if (rc)
5171                 class_detach(obd, lcfg);
5172 lcfg_cleanup:
5173         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5174 put_profile:
5175         class_put_profile(lprof);
5176 free_bufs:
5177         OBD_FREE_PTR(bufs);
5178 cleanup_mem:
5179         if (name)
5180                 OBD_FREE(name, name_size);
5181         if (uuid)
5182                 OBD_FREE(uuid, uuid_size);
5183         RETURN(rc);
5184 }
5185
5186 /* setup quota master target on MDT0 */
5187 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
5188                           struct lustre_cfg *cfg)
5189 {
5190         struct obd_device       *obd;
5191         char                    *dev = lustre_cfg_string(cfg, 0);
5192         char                    *qmtname, *uuid, *p;
5193         struct lustre_cfg_bufs  *bufs;
5194         struct lustre_cfg       *lcfg;
5195         struct lustre_profile   *lprof;
5196         struct obd_connect_data *data;
5197         int                      rc;
5198         ENTRY;
5199
5200         LASSERT(mdt->mdt_qmt_exp == NULL);
5201         LASSERT(mdt->mdt_qmt_dev == NULL);
5202
5203         /* quota master is on MDT0 only for now */
5204         if (mdt->mdt_seq_site.ss_node_id != 0)
5205                 RETURN(0);
5206
5207         /* MGS generates config commands which look as follows:
5208          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
5209          *                    3:lustre-MDT0000-mdtlov  4:f
5210          *
5211          * We generate the QMT name from the MDT one, just replacing MD with QM
5212          * after all the preparations, the logical equivalent will be:
5213          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
5214          *                    3:lustre-MDT0000-osd  4:f */
5215         OBD_ALLOC(qmtname, MAX_OBD_NAME);
5216         OBD_ALLOC(uuid, UUID_MAX);
5217         OBD_ALLOC_PTR(bufs);
5218         OBD_ALLOC_PTR(data);
5219         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
5220                 GOTO(cleanup_mem, rc = -ENOMEM);
5221
5222         strcpy(qmtname, dev);
5223         p = strstr(qmtname, "-MDT");
5224         if (p == NULL)
5225                 GOTO(cleanup_mem, rc = -ENOMEM);
5226         /* replace MD with QM */
5227         p[1] = 'Q';
5228         p[2] = 'M';
5229
5230         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
5231
5232         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
5233         if (lprof == NULL || lprof->lp_dt == NULL) {
5234                 CERROR("can't find profile for %s\n",
5235                        lustre_cfg_string(cfg, 0));
5236                 GOTO(cleanup_mem, rc = -EINVAL);
5237         }
5238
5239         lustre_cfg_bufs_reset(bufs, qmtname);
5240         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
5241         lustre_cfg_bufs_set_string(bufs, 2, uuid);
5242         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
5243
5244         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5245         if (!lcfg)
5246                 GOTO(put_profile, rc = -ENOMEM);
5247         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
5248
5249         rc = class_attach(lcfg);
5250         if (rc)
5251                 GOTO(lcfg_cleanup, rc);
5252
5253         obd = class_name2obd(qmtname);
5254         if (!obd) {
5255                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
5256                        lustre_cfg_string(cfg, 0));
5257                 GOTO(lcfg_cleanup, rc = -EINVAL);
5258         }
5259
5260         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5261
5262         lustre_cfg_bufs_reset(bufs, qmtname);
5263         lustre_cfg_bufs_set_string(bufs, 1, uuid);
5264         lustre_cfg_bufs_set_string(bufs, 2, dev);
5265
5266         /* for quota, the next device should be the OSD device */
5267         lustre_cfg_bufs_set_string(bufs, 3,
5268                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
5269
5270         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5271         if (!lcfg)
5272                 GOTO(class_detach, rc = -ENOMEM);
5273         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
5274
5275         rc = class_setup(obd, lcfg);
5276         if (rc)
5277                 GOTO(class_detach, rc);
5278
5279         mdt->mdt_qmt_dev = obd->obd_lu_dev;
5280
5281         /* configure local quota objects */
5282         if (OBD_FAIL_CHECK(OBD_FAIL_QUOTA_INIT))
5283                 rc = -EBADF;
5284         else
5285                 rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
5286                                                            &mdt->mdt_lu_dev,
5287                                                            mdt->mdt_qmt_dev);
5288         if (rc)
5289                 GOTO(class_cleanup, rc);
5290
5291         /* connect to quota master target */
5292         data->ocd_connect_flags = OBD_CONNECT_VERSION;
5293         data->ocd_version = LUSTRE_VERSION_CODE;
5294         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
5295                          data, NULL);
5296         if (rc) {
5297                 CERROR("cannot connect to quota master device %s (%d)\n",
5298                        qmtname, rc);
5299                 GOTO(class_cleanup, rc);
5300         }
5301
5302         EXIT;
5303 class_cleanup:
5304         if (rc) {
5305                 class_manual_cleanup(obd);
5306                 mdt->mdt_qmt_dev = NULL;
5307                 GOTO(lcfg_cleanup, rc);
5308         }
5309 class_detach:
5310         if (rc)
5311                 class_detach(obd, lcfg);
5312 lcfg_cleanup:
5313         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5314 put_profile:
5315         class_put_profile(lprof);
5316 cleanup_mem:
5317         if (bufs)
5318                 OBD_FREE_PTR(bufs);
5319         if (qmtname)
5320                 OBD_FREE(qmtname, MAX_OBD_NAME);
5321         if (uuid)
5322                 OBD_FREE(uuid, UUID_MAX);
5323         if (data)
5324                 OBD_FREE_PTR(data);
5325         return rc;
5326 }
5327
5328 /* Shutdown quota master target associated with mdt */
5329 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
5330 {
5331         ENTRY;
5332
5333         if (mdt->mdt_qmt_exp == NULL)
5334                 RETURN_EXIT;
5335         LASSERT(mdt->mdt_qmt_dev != NULL);
5336
5337         /* the qmt automatically shuts down when the mdt disconnects */
5338         obd_disconnect(mdt->mdt_qmt_exp);
5339         mdt->mdt_qmt_exp = NULL;
5340         mdt->mdt_qmt_dev = NULL;
5341         EXIT;
5342 }
5343
5344 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
5345  * for now. This will be removed along with converting rest of MDT code
5346  * to use tgt_session_info */
5347 static int mdt_tgt_getxattr(struct tgt_session_info *tsi)
5348 {
5349         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
5350         int                      rc;
5351
5352         if (unlikely(info->mti_object == NULL))
5353                 return -EPROTO;
5354
5355         rc = mdt_getxattr(info);
5356
5357         mdt_thread_info_fini(info);
5358         return rc;
5359 }
5360
5361 static int mdt_llog_open(struct tgt_session_info *tsi)
5362 {
5363         ENTRY;
5364
5365         if (!mdt_is_rootadmin(tsi2mdt_info(tsi)))
5366                 RETURN(err_serious(-EACCES));
5367
5368         RETURN(tgt_llog_open(tsi));
5369 }
5370
5371 #define OBD_FAIL_OST_READ_NET   OBD_FAIL_OST_BRW_NET
5372 #define OBD_FAIL_OST_WRITE_NET  OBD_FAIL_OST_BRW_NET
5373 #define OST_BRW_READ    OST_READ
5374 #define OST_BRW_WRITE   OST_WRITE
5375
5376 static struct tgt_handler mdt_tgt_handlers[] = {
5377 TGT_RPC_HANDLER(MDS_FIRST_OPC,
5378                 0,                      MDS_CONNECT,    mdt_tgt_connect,
5379                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
5380 TGT_RPC_HANDLER(MDS_FIRST_OPC,
5381                 0,                      MDS_DISCONNECT, tgt_disconnect,
5382                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
5383 TGT_RPC_HANDLER(MDS_FIRST_OPC,
5384                 HAS_REPLY,              MDS_SET_INFO,   mdt_set_info,
5385                 &RQF_MDT_SET_INFO, LUSTRE_MDS_VERSION),
5386 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
5387 TGT_MDT_HDL(HAS_REPLY,          MDS_GET_ROOT,   mdt_get_root),
5388 TGT_MDT_HDL(HAS_BODY,           MDS_GETATTR,    mdt_getattr),
5389 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_GETATTR_NAME,
5390                                                         mdt_getattr_name),
5391 TGT_MDT_HDL(HAS_BODY,           MDS_GETXATTR,   mdt_tgt_getxattr),
5392 TGT_MDT_HDL(HAS_REPLY,          MDS_STATFS,     mdt_statfs),
5393 TGT_MDT_HDL(IS_MUTABLE,         MDS_REINT,      mdt_reint),
5394 TGT_MDT_HDL(HAS_BODY,           MDS_CLOSE,      mdt_close),
5395 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_READPAGE,   mdt_readpage),
5396 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_SYNC,       mdt_sync),
5397 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
5398 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_PROGRESS,
5399                                                         mdt_hsm_progress),
5400 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_CT_REGISTER,
5401                                                         mdt_hsm_ct_register),
5402 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_CT_UNREGISTER,
5403                                                         mdt_hsm_ct_unregister),
5404 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_STATE_GET,
5405                                                         mdt_hsm_state_get),
5406 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_STATE_SET,
5407                                                         mdt_hsm_state_set),
5408 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_ACTION,       mdt_hsm_action),
5409 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_REQUEST,
5410                                                         mdt_hsm_request),
5411 TGT_MDT_HDL(HAS_KEY | HAS_BODY | HAS_REPLY | IS_MUTABLE,
5412             MDS_SWAP_LAYOUTS,
5413             mdt_swap_layouts),
5414 TGT_MDT_HDL(IS_MUTABLE,         MDS_RMFID,      mdt_rmfid),
5415 };
5416
5417 static struct tgt_handler mdt_io_ops[] = {
5418 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY, OST_BRW_READ, tgt_brw_read,
5419                                                         mdt_hp_brw),
5420 TGT_OST_HDL_HP(HAS_BODY | IS_MUTABLE,    OST_BRW_WRITE, tgt_brw_write,
5421                                                         mdt_hp_brw),
5422 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY | IS_MUTABLE,
5423                                          OST_PUNCH,     mdt_punch_hdl,
5424                                                         mdt_hp_punch),
5425 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SYNC,     mdt_data_sync),
5426 };
5427
5428 static struct tgt_handler mdt_sec_ctx_ops[] = {
5429 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
5430 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
5431 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
5432 };
5433
5434 static struct tgt_handler mdt_quota_ops[] = {
5435 TGT_QUOTA_HDL(HAS_REPLY,                QUOTA_DQACQ,      mdt_quota_dqacq),
5436 };
5437
5438 static struct tgt_handler mdt_llog_handlers[] = {
5439         TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_CREATE,      mdt_llog_open),
5440         TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_NEXT_BLOCK,  tgt_llog_next_block),
5441         TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
5442         TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_PREV_BLOCK,  tgt_llog_prev_block),
5443 };
5444
5445 static struct tgt_opc_slice mdt_common_slice[] = {
5446         {
5447                 .tos_opc_start  = MDS_FIRST_OPC,
5448                 .tos_opc_end    = MDS_LAST_OPC,
5449                 .tos_hs         = mdt_tgt_handlers
5450         },
5451         {
5452                 .tos_opc_start  = OBD_FIRST_OPC,
5453                 .tos_opc_end    = OBD_LAST_OPC,
5454                 .tos_hs         = tgt_obd_handlers
5455         },
5456         {
5457                 .tos_opc_start  = LDLM_FIRST_OPC,
5458                 .tos_opc_end    = LDLM_LAST_OPC,
5459                 .tos_hs         = tgt_dlm_handlers
5460         },
5461         {
5462                 .tos_opc_start  = SEC_FIRST_OPC,
5463                 .tos_opc_end    = SEC_LAST_OPC,
5464                 .tos_hs         = mdt_sec_ctx_ops
5465         },
5466         {
5467                 .tos_opc_start  = OUT_UPDATE_FIRST_OPC,
5468                 .tos_opc_end    = OUT_UPDATE_LAST_OPC,
5469                 .tos_hs         = tgt_out_handlers
5470         },
5471         {
5472                 .tos_opc_start  = FLD_FIRST_OPC,
5473                 .tos_opc_end    = FLD_LAST_OPC,
5474                 .tos_hs         = fld_handlers
5475         },
5476         {
5477                 .tos_opc_start  = SEQ_FIRST_OPC,
5478                 .tos_opc_end    = SEQ_LAST_OPC,
5479                 .tos_hs         = seq_handlers
5480         },
5481         {
5482                 .tos_opc_start  = QUOTA_DQACQ,
5483                 .tos_opc_end    = QUOTA_LAST_OPC,
5484                 .tos_hs         = mdt_quota_ops
5485         },
5486         {
5487                 .tos_opc_start  = LLOG_FIRST_OPC,
5488                 .tos_opc_end    = LLOG_LAST_OPC,
5489                 .tos_hs         = mdt_llog_handlers
5490         },
5491         {
5492                 .tos_opc_start  = LFSCK_FIRST_OPC,
5493                 .tos_opc_end    = LFSCK_LAST_OPC,
5494                 .tos_hs         = tgt_lfsck_handlers
5495         },
5496         {
5497                 .tos_opc_start  = OST_FIRST_OPC,
5498                 .tos_opc_end    = OST_LAST_OPC,
5499                 .tos_hs         = mdt_io_ops
5500         },
5501         {
5502                 .tos_hs         = NULL
5503         }
5504 };
5505
5506 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
5507 {
5508         struct md_device *next = m->mdt_child;
5509         struct lu_device *d = &m->mdt_lu_dev;
5510         struct obd_device *obd = mdt2obd_dev(m);
5511         struct lfsck_stop stop;
5512
5513         ENTRY;
5514         stop.ls_status = LS_PAUSED;
5515         stop.ls_flags = 0;
5516         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
5517
5518         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
5519
5520         mdt_restriper_stop(m);
5521         ping_evictor_stop();
5522
5523         /* Remove the HSM /proc entry so the coordinator cannot be
5524          * restarted by a user while it's shutting down.
5525          */
5526         mdt_hsm_cdt_stop(m);
5527
5528         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
5529         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
5530
5531         if (m->mdt_namespace != NULL)
5532                 ldlm_namespace_free_prior(m->mdt_namespace, NULL,
5533                                           d->ld_obd->obd_force);
5534
5535         obd_exports_barrier(obd);
5536         obd_zombie_barrier();
5537
5538         mdt_quota_fini(env, m);
5539
5540         cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
5541
5542         /* Calling the cleanup functions in the same order as in the mdt_init0
5543          * error path
5544          */
5545         mdt_tunables_fini(m);
5546
5547         target_recovery_fini(obd);
5548         upcall_cache_cleanup(m->mdt_identity_cache);
5549         m->mdt_identity_cache = NULL;
5550
5551         mdt_fs_cleanup(env, m);
5552
5553         tgt_fini(env, &m->mdt_lut);
5554
5555         mdt_hsm_cdt_fini(m);
5556
5557         if (m->mdt_los != NULL) {
5558                 local_oid_storage_fini(env, m->mdt_los);
5559                 m->mdt_los = NULL;
5560         }
5561
5562         if (m->mdt_namespace != NULL) {
5563                 ldlm_namespace_free_post(m->mdt_namespace);
5564                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
5565         }
5566
5567         if (m->mdt_md_root != NULL) {
5568                 mdt_object_put(env, m->mdt_md_root);
5569                 m->mdt_md_root = NULL;
5570         }
5571
5572         mdt_seq_fini(env, m);
5573
5574         mdt_fld_fini(env, m);
5575
5576         /*
5577          * Finish the stack
5578          */
5579         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
5580
5581         LASSERT(atomic_read(&d->ld_ref) == 0);
5582
5583         server_put_mount(mdt_obd_name(m), true);
5584
5585         EXIT;
5586 }
5587
5588 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
5589
5590 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
5591                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
5592 {
5593         const struct dt_device_param *dt_conf;
5594         struct mdt_thread_info *info;
5595         struct obd_device *obd;
5596         const char *dev = lustre_cfg_string(cfg, 0);
5597         const char *num = lustre_cfg_string(cfg, 2);
5598         struct tg_grants_data *tgd = &m->mdt_lut.lut_tgd;
5599         struct lustre_mount_info *lmi = NULL;
5600         struct lustre_sb_info *lsi;
5601         struct lu_site *s;
5602         struct seq_server_site *ss_site;
5603         const char *identity_upcall = "NONE";
5604         struct md_device *next;
5605         struct lu_fid fid;
5606         int rc;
5607         long node_id;
5608         mntopt_t mntopts;
5609         ENTRY;
5610
5611         lu_device_init(&m->mdt_lu_dev, ldt);
5612         /*
5613          * Environment (env) might be missing mdt_thread_key values at that
5614          * point, if device is allocated when mdt_thread_key is in QUIESCENT
5615          * mode.
5616          *
5617          * Usually device allocation path doesn't use module key values, but
5618          * mdt has to do a lot of work here, so allocate key value.
5619          */
5620         rc = lu_env_refill((struct lu_env *)env);
5621         if (rc != 0)
5622                 RETURN(rc);
5623
5624         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5625         LASSERT(info != NULL);
5626
5627         obd = class_name2obd(dev);
5628         LASSERT(obd != NULL);
5629
5630         m->mdt_max_mdsize = MAX_MD_SIZE_OLD;
5631         m->mdt_opts.mo_evict_tgt_nids = 1;
5632         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
5633
5634         lmi = server_get_mount(dev);
5635         if (lmi == NULL) {
5636                 CERROR("Cannot get mount info for %s!\n", dev);
5637                 RETURN(-EFAULT);
5638         } else {
5639                 lsi = s2lsi(lmi->lmi_sb);
5640                 LASSERT(lsi->lsi_lmd);
5641                 /* CMD is supported only in IAM mode */
5642                 LASSERT(num);
5643                 rc = kstrtol(num, 10, &node_id);
5644                 if (rc)
5645                         RETURN(rc);
5646
5647                 obd->u.obt.obt_magic = OBT_MAGIC;
5648                 if (lsi->lsi_lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
5649                         m->mdt_skip_lfsck = 1;
5650         }
5651
5652         /* DoM files get IO lock at open optionally by default */
5653         m->mdt_opts.mo_dom_lock = ALWAYS_DOM_LOCK_ON_OPEN;
5654         /* DoM files are read at open and data is packed in the reply */
5655         m->mdt_opts.mo_dom_read_open = 1;
5656
5657         m->mdt_squash.rsi_uid = 0;
5658         m->mdt_squash.rsi_gid = 0;
5659         INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
5660         spin_lock_init(&m->mdt_squash.rsi_lock);
5661         spin_lock_init(&m->mdt_lock);
5662         m->mdt_enable_remote_dir = 1;
5663         m->mdt_enable_striped_dir = 1;
5664         m->mdt_enable_dir_migration = 1;
5665         m->mdt_enable_dir_restripe = 0;
5666         m->mdt_enable_dir_auto_split = 0;
5667         m->mdt_enable_remote_dir_gid = 0;
5668         m->mdt_enable_chprojid_gid = 0;
5669         m->mdt_enable_remote_rename = 1;
5670         m->mdt_dir_restripe_nsonly = 1;
5671
5672         atomic_set(&m->mdt_mds_mds_conns, 0);
5673         atomic_set(&m->mdt_async_commit_count, 0);
5674
5675         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
5676         m->mdt_lu_dev.ld_obd = obd;
5677         /* Set this lu_device to obd for error handling purposes. */
5678         obd->obd_lu_dev = &m->mdt_lu_dev;
5679
5680         /* init the stack */
5681         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
5682         if (rc) {
5683                 CERROR("%s: Can't init device stack, rc %d\n",
5684                        mdt_obd_name(m), rc);
5685                 GOTO(err_lmi, rc);
5686         }
5687
5688         s = mdt_lu_site(m);
5689         ss_site = mdt_seq_site(m);
5690         s->ld_seq_site = ss_site;
5691         ss_site->ss_lu = s;
5692
5693         /* set server index */
5694         ss_site->ss_node_id = node_id;
5695
5696         /* failover is the default
5697          * FIXME: we do not failout mds0/mgs, which may cause some problems.
5698          * assumed whose ss_node_id == 0 XXX
5699          * */
5700         obd->obd_replayable = 1;
5701         /* No connection accepted until configurations will finish */
5702         obd->obd_no_conn = 1;
5703
5704         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
5705                 char *str = lustre_cfg_string(cfg, 4);
5706                 if (strchr(str, 'n')) {
5707                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
5708                         obd->obd_replayable = 0;
5709                 }
5710         }
5711
5712         rc = mdt_fld_init(env, mdt_obd_name(m), m);
5713         if (rc)
5714                 GOTO(err_fini_stack, rc);
5715
5716         rc = mdt_seq_init(env, m);
5717         if (rc)
5718                 GOTO(err_fini_fld, rc);
5719
5720         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
5721                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
5722         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
5723                                               LDLM_NAMESPACE_SERVER,
5724                                               LDLM_NAMESPACE_GREEDY,
5725                                               LDLM_NS_TYPE_MDT);
5726         if (m->mdt_namespace == NULL)
5727                 GOTO(err_fini_seq, rc = -ENOMEM);
5728
5729         m->mdt_namespace->ns_lvbp = m;
5730         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
5731
5732         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
5733         /* set obd_namespace for compatibility with old code */
5734         obd->obd_namespace = m->mdt_namespace;
5735
5736         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
5737                       OBD_FAIL_MDS_ALL_REQUEST_NET,
5738                       OBD_FAIL_MDS_ALL_REPLY_NET);
5739         if (rc)
5740                 GOTO(err_free_ns, rc);
5741
5742         /* Amount of available space excluded from granting and reserved
5743          * for metadata. It is a percentage of the total MDT size. */
5744         tgd->tgd_reserved_pcnt = 10;
5745
5746         if (ONE_MB_BRW_SIZE < (1U << tgd->tgd_blockbits))
5747                 m->mdt_brw_size = 1U << tgd->tgd_blockbits;
5748         else
5749                 m->mdt_brw_size = ONE_MB_BRW_SIZE;
5750
5751         rc = mdt_fs_setup(env, m, obd, lsi);
5752         if (rc)
5753                 GOTO(err_tgt, rc);
5754
5755         fid.f_seq = FID_SEQ_LOCAL_NAME;
5756         fid.f_oid = 1;
5757         fid.f_ver = 0;
5758         rc = local_oid_storage_init(env, m->mdt_bottom, &fid, &m->mdt_los);
5759         if (rc != 0)
5760                 GOTO(err_fs_cleanup, rc);
5761
5762         rc = mdt_hsm_cdt_init(m);
5763         if (rc != 0) {
5764                 CERROR("%s: error initializing coordinator, rc %d\n",
5765                        mdt_obd_name(m), rc);
5766                 GOTO(err_los_fini, rc);
5767         }
5768
5769         tgt_adapt_sptlrpc_conf(&m->mdt_lut);
5770
5771         next = m->mdt_child;
5772         dt_conf = next->md_ops->mdo_dtconf_get(env, next);
5773
5774         mntopts = dt_conf->ddp_mntopts;
5775
5776         if (mntopts & MNTOPT_USERXATTR)
5777                 m->mdt_opts.mo_user_xattr = 1;
5778         else
5779                 m->mdt_opts.mo_user_xattr = 0;
5780
5781         m->mdt_max_ea_size = dt_conf->ddp_max_ea_size;
5782
5783         if (mntopts & MNTOPT_ACL)
5784                 m->mdt_opts.mo_acl = 1;
5785         else
5786                 m->mdt_opts.mo_acl = 0;
5787
5788         /* XXX: to support suppgid for ACL, we enable identity_upcall
5789          * by default, otherwise, maybe got unexpected -EACCESS. */
5790         if (m->mdt_opts.mo_acl)
5791                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
5792
5793         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
5794                                                 identity_upcall,
5795                                                 &mdt_identity_upcall_cache_ops);
5796         if (IS_ERR(m->mdt_identity_cache)) {
5797                 rc = PTR_ERR(m->mdt_identity_cache);
5798                 m->mdt_identity_cache = NULL;
5799                 GOTO(err_free_hsm, rc);
5800         }
5801
5802         rc = mdt_tunables_init(m, dev);
5803         if (rc) {
5804                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
5805                 GOTO(err_recovery, rc);
5806         }
5807
5808         rc = mdt_quota_init(env, m, cfg);
5809         if (rc)
5810                 GOTO(err_procfs, rc);
5811
5812         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
5813         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
5814                            "mdt_ldlm_client", m->mdt_ldlm_client);
5815
5816         ping_evictor_start();
5817
5818         /* recovery will be started upon mdt_prepare()
5819          * when the whole stack is complete and ready
5820          * to serve the requests */
5821
5822         /* Reduce the initial timeout on an MDS because it doesn't need such
5823          * a long timeout as an OST does. Adaptive timeouts will adjust this
5824          * value appropriately. */
5825         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
5826                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
5827
5828         if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_LOCAL_RECOV))
5829                 m->mdt_lut.lut_local_recovery = 1;
5830
5831         rc = mdt_restriper_start(m);
5832         if (rc)
5833                 GOTO(err_ping_evictor, rc);
5834
5835         RETURN(0);
5836
5837 err_ping_evictor:
5838         ping_evictor_stop();
5839 err_procfs:
5840         mdt_tunables_fini(m);
5841 err_recovery:
5842         upcall_cache_cleanup(m->mdt_identity_cache);
5843         m->mdt_identity_cache = NULL;
5844 err_free_hsm:
5845         mdt_hsm_cdt_fini(m);
5846 err_los_fini:
5847         local_oid_storage_fini(env, m->mdt_los);
5848         m->mdt_los = NULL;
5849 err_fs_cleanup:
5850         mdt_fs_cleanup(env, m);
5851 err_tgt:
5852         /* keep recoverable clients */
5853         obd->obd_fail = 1;
5854         target_recovery_fini(obd);
5855         obd_exports_barrier(obd);
5856         obd_zombie_barrier();
5857         tgt_fini(env, &m->mdt_lut);
5858 err_free_ns:
5859         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
5860         obd->obd_namespace = m->mdt_namespace = NULL;
5861 err_fini_seq:
5862         mdt_seq_fini(env, m);
5863 err_fini_fld:
5864         mdt_fld_fini(env, m);
5865 err_fini_stack:
5866         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
5867 err_lmi:
5868         if (lmi)
5869                 server_put_mount(dev, true);
5870         return(rc);
5871 }
5872
5873 /* For interoperability, the left element is old parameter, the right one
5874  * is the new version of the parameter, if some parameter is deprecated,
5875  * the new version should be set as NULL. */
5876 static struct cfg_interop_param mdt_interop_param[] = {
5877         { "mdt.group_upcall",   NULL },
5878         { "mdt.quota_type",     NULL },
5879         { "mdd.quota_type",     NULL },
5880         { "mdt.som",            NULL },
5881         { "mdt.rootsquash",     "mdt.root_squash" },
5882         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
5883         { NULL }
5884 };
5885
5886 /* used by MGS to process specific configurations */
5887 static int mdt_process_config(const struct lu_env *env,
5888                               struct lu_device *d, struct lustre_cfg *cfg)
5889 {
5890         struct mdt_device *m = mdt_dev(d);
5891         struct md_device *md_next = m->mdt_child;
5892         struct lu_device *next = md2lu_dev(md_next);
5893         int rc;
5894         ENTRY;
5895
5896         switch (cfg->lcfg_command) {
5897         case LCFG_PARAM: {
5898                 struct obd_device *obd = d->ld_obd;
5899                 /* For interoperability */
5900                 struct cfg_interop_param *ptr = NULL;
5901                 struct lustre_cfg *old_cfg = NULL;
5902                 char *param = NULL;
5903                 ssize_t count;
5904
5905                 param = lustre_cfg_string(cfg, 1);
5906                 if (param == NULL) {
5907                         CERROR("param is empty\n");
5908                         rc = -EINVAL;
5909                         break;
5910                 }
5911
5912                 ptr = class_find_old_param(param, mdt_interop_param);
5913                 if (ptr != NULL) {
5914                         if (ptr->new_param == NULL) {
5915                                 rc = 0;
5916                                 CWARN("For interoperability, skip this %s."
5917                                       " It is obsolete.\n", ptr->old_param);
5918                                 break;
5919                         }
5920
5921                         CWARN("Found old param %s, changed it to %s.\n",
5922                               ptr->old_param, ptr->new_param);
5923
5924                         old_cfg = cfg;
5925                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
5926                         if (IS_ERR(cfg)) {
5927                                 rc = PTR_ERR(cfg);
5928                                 break;
5929                         }
5930                 }
5931
5932                 count = class_modify_config(cfg, PARAM_MDT,
5933                                             &obd->obd_kset.kobj);
5934                 if (count < 0) {
5935                         struct coordinator *cdt = &m->mdt_coordinator;
5936
5937                         /* is it an HSM var ? */
5938                         count = class_modify_config(cfg, PARAM_HSM,
5939                                                     &cdt->cdt_hsm_kobj);
5940                         if (count < 0)
5941                                 /* we don't understand; pass it on */
5942                                 rc = next->ld_ops->ldo_process_config(env, next,
5943                                                                       cfg);
5944                         else
5945                                 rc = count > 0 ? 0 : count;
5946                 } else {
5947                         rc = count > 0 ? 0 : count;
5948                 }
5949
5950                 if (old_cfg)
5951                         OBD_FREE(cfg, lustre_cfg_len(cfg->lcfg_bufcount,
5952                                                      cfg->lcfg_buflens));
5953                 break;
5954         }
5955         default:
5956                 /* others are passed further */
5957                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
5958                 break;
5959         }
5960         RETURN(rc);
5961 }
5962
5963 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
5964                                           const struct lu_object_header *hdr,
5965                                           struct lu_device *d)
5966 {
5967         struct mdt_object *mo;
5968
5969         ENTRY;
5970
5971         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
5972         if (mo != NULL) {
5973                 struct lu_object *o;
5974                 struct lu_object_header *h;
5975
5976                 o = &mo->mot_obj;
5977                 h = &mo->mot_header;
5978                 lu_object_header_init(h);
5979                 lu_object_init(o, h, d);
5980                 lu_object_add_top(h, o);
5981                 o->lo_ops = &mdt_obj_ops;
5982                 spin_lock_init(&mo->mot_write_lock);
5983                 mutex_init(&mo->mot_som_mutex);
5984                 mutex_init(&mo->mot_lov_mutex);
5985                 init_rwsem(&mo->mot_dom_sem);
5986                 init_rwsem(&mo->mot_open_sem);
5987                 atomic_set(&mo->mot_open_count, 0);
5988                 mo->mot_restripe_offset = 0;
5989                 INIT_LIST_HEAD(&mo->mot_restripe_linkage);
5990                 RETURN(o);
5991         }
5992         RETURN(NULL);
5993 }
5994
5995 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
5996                            const struct lu_object_conf *unused)
5997 {
5998         struct mdt_device *d = mdt_dev(o->lo_dev);
5999         struct lu_device  *under;
6000         struct lu_object  *below;
6001         int                rc = 0;
6002         ENTRY;
6003
6004         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
6005                PFID(lu_object_fid(o)));
6006
6007         under = &d->mdt_child->md_lu_dev;
6008         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
6009         if (below != NULL) {
6010                 lu_object_add(o, below);
6011         } else
6012                 rc = -ENOMEM;
6013
6014         RETURN(rc);
6015 }
6016
6017 static void mdt_object_free_rcu(struct rcu_head *head)
6018 {
6019         struct mdt_object *mo = container_of(head, struct mdt_object,
6020                                              mot_header.loh_rcu);
6021
6022         kmem_cache_free(mdt_object_kmem, mo);
6023 }
6024
6025 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
6026 {
6027         struct mdt_object *mo = mdt_obj(o);
6028         struct lu_object_header *h;
6029         ENTRY;
6030
6031         h = o->lo_header;
6032         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
6033                PFID(lu_object_fid(o)));
6034
6035         LASSERT(atomic_read(&mo->mot_open_count) == 0);
6036         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
6037
6038         lu_object_fini(o);
6039         lu_object_header_fini(h);
6040         OBD_FREE_PRE(mo, sizeof(*mo), "slab-freed");
6041         call_rcu(&mo->mot_header.loh_rcu, mdt_object_free_rcu);
6042
6043         EXIT;
6044 }
6045
6046 static int mdt_object_print(const struct lu_env *env, void *cookie,
6047                             lu_printer_t p, const struct lu_object *o)
6048 {
6049         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
6050
6051         return (*p)(env, cookie,
6052                     LUSTRE_MDT_NAME"-object@%p(%s %s, writecount=%d)",
6053                     mdto, mdto->mot_lov_created ? "lov_created" : "",
6054                     mdto->mot_cache_attr ? "cache_attr" : "",
6055                     mdto->mot_write_count);
6056 }
6057
6058 static int mdt_prepare(const struct lu_env *env,
6059                 struct lu_device *pdev,
6060                 struct lu_device *cdev)
6061 {
6062         struct mdt_device *mdt = mdt_dev(cdev);
6063         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
6064         struct obd_device *obd = cdev->ld_obd;
6065         int rc;
6066
6067         ENTRY;
6068
6069         LASSERT(obd);
6070
6071         rc = next->ld_ops->ldo_prepare(env, cdev, next);
6072         if (rc)
6073                 RETURN(rc);
6074
6075         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
6076         if (rc)
6077                 RETURN(rc);
6078
6079         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
6080         if (rc)
6081                 RETURN(rc);
6082
6083         rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
6084         /* The LFSCK instance is registered just now, so it must be there when
6085          * register the namespace to such instance. */
6086         LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
6087
6088         if (mdt->mdt_seq_site.ss_node_id == 0) {
6089                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
6090                                                          &mdt->mdt_md_root_fid);
6091                 if (rc)
6092                         RETURN(rc);
6093         }
6094
6095         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
6096
6097         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
6098         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
6099         LASSERT(obd->obd_no_conn);
6100         spin_lock(&obd->obd_dev_lock);
6101         obd->obd_no_conn = 0;
6102         spin_unlock(&obd->obd_dev_lock);
6103
6104         if (obd->obd_recovering == 0)
6105                 mdt_postrecov(env, mdt);
6106
6107         RETURN(rc);
6108 }
6109
6110 const struct lu_device_operations mdt_lu_ops = {
6111         .ldo_object_alloc   = mdt_object_alloc,
6112         .ldo_process_config = mdt_process_config,
6113         .ldo_prepare        = mdt_prepare,
6114 };
6115
6116 static const struct lu_object_operations mdt_obj_ops = {
6117         .loo_object_init    = mdt_object_init,
6118         .loo_object_free    = mdt_object_free,
6119         .loo_object_print   = mdt_object_print
6120 };
6121
6122 static int mdt_obd_set_info_async(const struct lu_env *env,
6123                                   struct obd_export *exp,
6124                                   __u32 keylen, void *key,
6125                                   __u32 vallen, void *val,
6126                                   struct ptlrpc_request_set *set)
6127 {
6128         int rc;
6129
6130         ENTRY;
6131
6132         if (KEY_IS(KEY_SPTLRPC_CONF)) {
6133                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp));
6134                 RETURN(rc);
6135         }
6136
6137         RETURN(0);
6138 }
6139
6140 static inline void mdt_enable_slc(struct mdt_device *mdt)
6141 {
6142         if (mdt->mdt_lut.lut_sync_lock_cancel == SYNC_LOCK_CANCEL_NEVER)
6143                 mdt->mdt_lut.lut_sync_lock_cancel = SYNC_LOCK_CANCEL_BLOCKING;
6144 }
6145
6146 static inline void mdt_disable_slc(struct mdt_device *mdt)
6147 {
6148         if (mdt->mdt_lut.lut_sync_lock_cancel == SYNC_LOCK_CANCEL_BLOCKING)
6149                 mdt->mdt_lut.lut_sync_lock_cancel = SYNC_LOCK_CANCEL_NEVER;
6150 }
6151
6152 /**
6153  * Match client and server connection feature flags.
6154  *
6155  * Compute the compatibility flags for a connection request based on
6156  * features mutually supported by client and server.
6157  *
6158  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
6159  * must not be updated here, otherwise a partially initialized value may
6160  * be exposed. After the connection request is successfully processed,
6161  * the top-level MDT connect request handler atomically updates the export
6162  * connect flags from the obd_connect_data::ocd_connect_flags field of the
6163  * reply. \see mdt_connect().
6164  *
6165  * Before 2.7.50 clients will send a struct obd_connect_data_v1 rather than a
6166  * full struct obd_connect_data. So care must be taken when accessing fields
6167  * that are not present in struct obd_connect_data_v1. See LU-16.
6168  *
6169  * \param exp   the obd_export associated with this client/target pair
6170  * \param mdt   the target device for the connection
6171  * \param data  stores data for this connect request
6172  *
6173  * \retval 0       success
6174  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
6175  * \retval -EBADE  client and server feature requirements are incompatible
6176  */
6177 static int mdt_connect_internal(const struct lu_env *env,
6178                                 struct obd_export *exp,
6179                                 struct mdt_device *mdt,
6180                                 struct obd_connect_data *data, bool reconnect)
6181 {
6182         const char *obd_name = mdt_obd_name(mdt);
6183         LASSERT(data != NULL);
6184
6185         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
6186
6187         if (mdt->mdt_bottom->dd_rdonly &&
6188             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
6189             !(data->ocd_connect_flags & OBD_CONNECT_RDONLY))
6190                 RETURN(-EACCES);
6191
6192         if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
6193                 data->ocd_connect_flags2 &= MDT_CONNECT_SUPPORTED2;
6194
6195         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
6196
6197         if (!mdt->mdt_opts.mo_acl)
6198                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
6199
6200         if (!mdt->mdt_opts.mo_user_xattr)
6201                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
6202
6203         if (OCD_HAS_FLAG(data, BRW_SIZE)) {
6204                 data->ocd_brw_size = min(data->ocd_brw_size,
6205                                          mdt->mdt_brw_size);
6206                 if (data->ocd_brw_size == 0) {
6207                         CERROR("%s: cli %s/%p ocd_connect_flags: %#llx "
6208                                "ocd_version: %x ocd_grant: %d ocd_index: %u "
6209                                "ocd_brw_size unexpectedly zero, network data "
6210                                "corruption? Refusing to connect this client\n",
6211                                obd_name, exp->exp_client_uuid.uuid,
6212                                exp, data->ocd_connect_flags, data->ocd_version,
6213                                data->ocd_grant, data->ocd_index);
6214                         return -EPROTO;
6215                 }
6216         }
6217
6218         if (OCD_HAS_FLAG(data, GRANT_PARAM)) {
6219                 struct dt_device_param *ddp = &mdt->mdt_lut.lut_dt_conf;
6220
6221                 /* client is reporting its page size, for future use */
6222                 exp->exp_target_data.ted_pagebits = data->ocd_grant_blkbits;
6223                 data->ocd_grant_blkbits  = mdt->mdt_lut.lut_tgd.tgd_blockbits;
6224                 /* ddp_inodespace may not be power-of-two value, eg. for ldiskfs
6225                  * it's LDISKFS_DIR_REC_LEN(20) = 28. */
6226                 data->ocd_grant_inobits = fls(ddp->ddp_inodespace - 1);
6227                 /* ocd_grant_tax_kb is in 1K byte blocks */
6228                 data->ocd_grant_tax_kb = ddp->ddp_extent_tax >> 10;
6229                 data->ocd_grant_max_blks = ddp->ddp_max_extent_blks;
6230         }
6231
6232         /* Save connect_data we have so far because tgt_grant_connect()
6233          * uses it to calculate grant, and we want to save the client
6234          * version before it is overwritten by LUSTRE_VERSION_CODE. */
6235         exp->exp_connect_data = *data;
6236         if (OCD_HAS_FLAG(data, GRANT))
6237                 tgt_grant_connect(env, exp, data, !reconnect);
6238
6239         if (OCD_HAS_FLAG(data, MAXBYTES))
6240                 data->ocd_maxbytes = mdt->mdt_lut.lut_dt_conf.ddp_maxbytes;
6241
6242         /* NB: Disregard the rule against updating
6243          * exp_connect_data.ocd_connect_flags in this case, since
6244          * tgt_client_new() needs to know if this is a lightweight
6245          * connection, and it is safe to expose this flag before
6246          * connection processing completes. */
6247         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
6248                 spin_lock(&exp->exp_lock);
6249                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
6250                 spin_unlock(&exp->exp_lock);
6251         }
6252
6253         data->ocd_version = LUSTRE_VERSION_CODE;
6254
6255         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
6256                 CWARN("%s: MDS requires FID support, but client not\n",
6257                       obd_name);
6258                 return -EBADE;
6259         }
6260
6261         if (OCD_HAS_FLAG(data, PINGLESS)) {
6262                 if (ptlrpc_pinger_suppress_pings()) {
6263                         spin_lock(&exp->exp_obd->obd_dev_lock);
6264                         list_del_init(&exp->exp_obd_chain_timed);
6265                         spin_unlock(&exp->exp_obd->obd_dev_lock);
6266                 } else {
6267                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
6268                 }
6269         }
6270
6271         data->ocd_max_easize = mdt->mdt_max_ea_size;
6272
6273         /* NB: Disregard the rule against updating
6274          * exp_connect_data.ocd_connect_flags in this case, since
6275          * tgt_client_new() needs to know if this is client supports
6276          * multiple modify RPCs, and it is safe to expose this flag before
6277          * connection processing completes. */
6278         if (data->ocd_connect_flags & OBD_CONNECT_MULTIMODRPCS) {
6279                 data->ocd_maxmodrpcs = max_mod_rpcs_per_client;
6280                 spin_lock(&exp->exp_lock);
6281                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_MULTIMODRPCS;
6282                 spin_unlock(&exp->exp_lock);
6283         }
6284
6285         if (OCD_HAS_FLAG(data, CKSUM)) {
6286                 __u32 cksum_types = data->ocd_cksum_types;
6287
6288                 /* The client set in ocd_cksum_types the checksum types it
6289                  * supports. We have to mask off the algorithms that we don't
6290                  * support */
6291                 data->ocd_cksum_types &=
6292                         obd_cksum_types_supported_server(obd_name);
6293
6294                 if (unlikely(data->ocd_cksum_types == 0)) {
6295                         CERROR("%s: Connect with checksum support but no "
6296                                "ocd_cksum_types is set\n",
6297                                exp->exp_obd->obd_name);
6298                         RETURN(-EPROTO);
6299                 }
6300
6301                 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
6302                        "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
6303                        cksum_types, data->ocd_cksum_types);
6304         } else {
6305                 /* This client does not support OBD_CONNECT_CKSUM
6306                  * fall back to CRC32 */
6307                 CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
6308                        "OBD_CONNECT_CKSUM, CRC32 will be used\n",
6309                        exp->exp_obd->obd_name, obd_export_nid2str(exp));
6310         }
6311
6312         if ((data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
6313             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
6314                 atomic_inc(&mdt->mdt_mds_mds_conns);
6315                 mdt_enable_slc(mdt);
6316         }
6317
6318         return 0;
6319 }
6320
6321 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
6322                                    struct mdt_thread_info *info,
6323                                    struct mdt_file_data *mfd)
6324 {
6325         struct lu_context ses;
6326         int rc;
6327         ENTRY;
6328
6329         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
6330         if (rc)
6331                 RETURN(rc);
6332
6333         env->le_ses = &ses;
6334         lu_context_enter(&ses);
6335
6336         mdt_ucred(info)->uc_valid = UCRED_OLD;
6337         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
6338
6339         lu_context_exit(&ses);
6340         lu_context_fini(&ses);
6341         env->le_ses = NULL;
6342
6343         RETURN(rc);
6344 }
6345
6346 static int mdt_export_cleanup(struct obd_export *exp)
6347 {
6348         LIST_HEAD(closing_list);
6349         struct mdt_export_data  *med = &exp->exp_mdt_data;
6350         struct obd_device       *obd = exp->exp_obd;
6351         struct mdt_device       *mdt;
6352         struct mdt_thread_info  *info;
6353         struct lu_env            env;
6354         struct mdt_file_data    *mfd, *n;
6355         int rc = 0;
6356         ENTRY;
6357
6358         spin_lock(&med->med_open_lock);
6359         while (!list_empty(&med->med_open_head)) {
6360                 struct list_head *tmp = med->med_open_head.next;
6361                 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
6362
6363                 /* Remove mfd handle so it can't be found again.
6364                  * We are consuming the mfd_list reference here. */
6365                 class_handle_unhash(&mfd->mfd_open_handle);
6366                 list_move_tail(&mfd->mfd_list, &closing_list);
6367         }
6368         spin_unlock(&med->med_open_lock);
6369         mdt = mdt_dev(obd->obd_lu_dev);
6370         LASSERT(mdt != NULL);
6371
6372         rc = lu_env_init(&env, LCT_MD_THREAD);
6373         if (rc)
6374                 RETURN(rc);
6375
6376         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6377         LASSERT(info != NULL);
6378         memset(info, 0, sizeof *info);
6379         info->mti_env = &env;
6380         info->mti_mdt = mdt;
6381         info->mti_exp = exp;
6382
6383         if (!list_empty(&closing_list)) {
6384                 struct md_attr *ma = &info->mti_attr;
6385
6386                 /* Close any open files (which may also cause orphan
6387                  * unlinking). */
6388                 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
6389                         list_del_init(&mfd->mfd_list);
6390                         ma->ma_need = ma->ma_valid = 0;
6391
6392                         /* This file is being closed due to an eviction, it
6393                          * could have been modified and now dirty regarding to
6394                          * HSM archive, check this!
6395                          * The logic here is to mark a file dirty if there's a
6396                          * chance it was dirtied before the client was evicted,
6397                          * so that we don't have to wait for a release attempt
6398                          * before finding out the file was actually dirty and
6399                          * fail the release. Aggressively marking it dirty here
6400                          * will cause the policy engine to attempt to
6401                          * re-archive it; when rearchiving, we can compare the
6402                          * current version to the HSM data_version and make the
6403                          * archive request into a noop if it's not actually
6404                          * dirty.
6405                          */
6406                         if (mfd->mfd_open_flags & MDS_FMODE_WRITE)
6407                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
6408
6409                         /* Don't unlink orphan on failover umount, LU-184 */
6410                         if (exp->exp_flags & OBD_OPT_FAILOVER) {
6411                                 ma->ma_valid = MA_FLAGS;
6412                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
6413                         }
6414                         ma->ma_valid |= MA_FORCE_LOG;
6415                         mdt_mfd_close(info, mfd);
6416                 }
6417         }
6418         info->mti_mdt = NULL;
6419         /* cleanup client slot early */
6420         /* Do not erase record for recoverable client. */
6421         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
6422                 tgt_client_del(&env, exp);
6423         lu_env_fini(&env);
6424
6425         RETURN(rc);
6426 }
6427
6428 static int mdt_obd_disconnect(struct obd_export *exp)
6429 {
6430         int rc;
6431
6432         ENTRY;
6433
6434         LASSERT(exp);
6435         class_export_get(exp);
6436
6437         if (!(exp->exp_flags & OBD_OPT_FORCE))
6438                 tgt_grant_sanity_check(exp->exp_obd, __func__);
6439
6440         if ((exp_connect_flags(exp) & OBD_CONNECT_MDS_MDS) &&
6441             !(exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT)) {
6442                 struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
6443
6444                 if (atomic_dec_and_test(&mdt->mdt_mds_mds_conns))
6445                         mdt_disable_slc(mdt);
6446         }
6447
6448         rc = server_disconnect_export(exp);
6449         if (rc != 0)
6450                 CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
6451
6452         tgt_grant_discard(exp);
6453
6454         rc = mdt_export_cleanup(exp);
6455         nodemap_del_member(exp);
6456         class_export_put(exp);
6457         RETURN(rc);
6458 }
6459
6460 /* mds_connect copy */
6461 static int mdt_obd_connect(const struct lu_env *env,
6462                            struct obd_export **exp, struct obd_device *obd,
6463                            struct obd_uuid *cluuid,
6464                            struct obd_connect_data *data,
6465                            void *localdata)
6466 {
6467         struct obd_export       *lexp;
6468         struct lustre_handle    conn = { 0 };
6469         struct mdt_device       *mdt;
6470         int                      rc;
6471         lnet_nid_t              *client_nid = localdata;
6472         ENTRY;
6473
6474         LASSERT(env != NULL);
6475         LASSERT(data != NULL);
6476
6477         if (!exp || !obd || !cluuid)
6478                 RETURN(-EINVAL);
6479
6480         mdt = mdt_dev(obd->obd_lu_dev);
6481
6482         /*
6483          * first, check whether the stack is ready to handle requests
6484          * XXX: probably not very appropriate method is used now
6485          *      at some point we should find a better one
6486          */
6487         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) &&
6488             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) &&
6489             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
6490                 rc = obd_get_info(env, mdt->mdt_child_exp,
6491                                   sizeof(KEY_OSP_CONNECTED),
6492                                   KEY_OSP_CONNECTED, NULL, NULL);
6493                 if (rc)
6494                         RETURN(-EAGAIN);
6495                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
6496         }
6497
6498         rc = class_connect(&conn, obd, cluuid);
6499         if (rc)
6500                 RETURN(rc);
6501
6502         lexp = class_conn2export(&conn);
6503         LASSERT(lexp != NULL);
6504
6505         rc = nodemap_add_member(*client_nid, lexp);
6506         if (rc != 0 && rc != -EEXIST)
6507                 GOTO(out, rc);
6508
6509         rc = mdt_connect_internal(env, lexp, mdt, data, false);
6510         if (rc == 0) {
6511                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
6512
6513                 LASSERT(lcd);
6514                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
6515                 rc = tgt_client_new(env, lexp);
6516                 if (rc == 0)
6517                         mdt_export_stats_init(obd, lexp, localdata);
6518         }
6519 out:
6520         if (rc != 0) {
6521                 class_disconnect(lexp);
6522                 nodemap_del_member(lexp);
6523                 *exp = NULL;
6524         } else {
6525                 *exp = lexp;
6526                 /* Because we do not want this export to be evicted by pinger,
6527                  * let's not add this export to the timed chain list. */
6528                 if (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) {
6529                         spin_lock(&lexp->exp_obd->obd_dev_lock);
6530                         list_del_init(&lexp->exp_obd_chain_timed);
6531                         spin_unlock(&lexp->exp_obd->obd_dev_lock);
6532                 }
6533         }
6534
6535         RETURN(rc);
6536 }
6537
6538 static int mdt_obd_reconnect(const struct lu_env *env,
6539                              struct obd_export *exp, struct obd_device *obd,
6540                              struct obd_uuid *cluuid,
6541                              struct obd_connect_data *data,
6542                              void *localdata)
6543 {
6544         lnet_nid_t             *client_nid = localdata;
6545         int                     rc;
6546         ENTRY;
6547
6548         if (exp == NULL || obd == NULL || cluuid == NULL)
6549                 RETURN(-EINVAL);
6550
6551         rc = nodemap_add_member(*client_nid, exp);
6552         if (rc != 0 && rc != -EEXIST)
6553                 RETURN(rc);
6554
6555         rc = mdt_connect_internal(env, exp, mdt_dev(obd->obd_lu_dev), data,
6556                                   true);
6557         if (rc == 0)
6558                 mdt_export_stats_init(obd, exp, localdata);
6559         else
6560                 nodemap_del_member(exp);
6561
6562         RETURN(rc);
6563 }
6564
6565 /* FIXME: Can we avoid using these two interfaces? */
6566 static int mdt_init_export(struct obd_export *exp)
6567 {
6568         struct mdt_export_data *med = &exp->exp_mdt_data;
6569         int                     rc;
6570         ENTRY;
6571
6572         INIT_LIST_HEAD(&med->med_open_head);
6573         spin_lock_init(&med->med_open_lock);
6574         spin_lock(&exp->exp_lock);
6575         exp->exp_connecting = 1;
6576         spin_unlock(&exp->exp_lock);
6577
6578         OBD_ALLOC(exp->exp_used_slots,
6579                   BITS_TO_LONGS(OBD_MAX_RIF_MAX) * sizeof(long));
6580         if (exp->exp_used_slots == NULL)
6581                 RETURN(-ENOMEM);
6582
6583         /* self-export doesn't need client data and ldlm initialization */
6584         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
6585                                      &exp->exp_client_uuid)))
6586                 RETURN(0);
6587
6588         rc = tgt_client_alloc(exp);
6589         if (rc)
6590                 GOTO(err, rc);
6591
6592         rc = ldlm_init_export(exp);
6593         if (rc)
6594                 GOTO(err_free, rc);
6595
6596         RETURN(rc);
6597
6598 err_free:
6599         tgt_client_free(exp);
6600 err:
6601         OBD_FREE(exp->exp_used_slots,
6602                  BITS_TO_LONGS(OBD_MAX_RIF_MAX) * sizeof(long));
6603         exp->exp_used_slots = NULL;
6604
6605         CERROR("%s: Failed to initialize export: rc = %d\n",
6606                exp->exp_obd->obd_name, rc);
6607         return rc;
6608 }
6609
6610 static int mdt_destroy_export(struct obd_export *exp)
6611 {
6612         ENTRY;
6613
6614         target_destroy_export(exp);
6615         if (exp->exp_used_slots)
6616                 OBD_FREE(exp->exp_used_slots,
6617                          BITS_TO_LONGS(OBD_MAX_RIF_MAX) * sizeof(long));
6618
6619         /* destroy can be called from failed obd_setup, so
6620          * checking uuid is safer than obd_self_export */
6621         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
6622                                      &exp->exp_client_uuid)))
6623                 RETURN(0);
6624
6625         ldlm_destroy_export(exp);
6626         tgt_client_free(exp);
6627
6628         LASSERT(list_empty(&exp->exp_outstanding_replies));
6629         LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
6630
6631         /*
6632          * discard grants once we're sure no more
6633          * interaction with the client is possible
6634          */
6635         tgt_grant_discard(exp);
6636         if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
6637                 exp->exp_obd->u.obt.obt_lut->lut_tgd.tgd_tot_granted_clients--;
6638
6639         if (!(exp->exp_flags & OBD_OPT_FORCE))
6640                 tgt_grant_sanity_check(exp->exp_obd, __func__);
6641
6642         RETURN(0);
6643 }
6644
6645 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
6646                    struct linkea_data *ldata)
6647 {
6648         int rc;
6649
6650         LASSERT(ldata->ld_buf->lb_buf != NULL);
6651
6652         if (!mdt_object_exists(mdt_obj))
6653                 return -ENODATA;
6654
6655         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6656                           ldata->ld_buf, XATTR_NAME_LINK);
6657         if (rc == -ERANGE) {
6658                 /* Buf was too small, figure out what we need. */
6659                 lu_buf_free(ldata->ld_buf);
6660                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6661                                   ldata->ld_buf, XATTR_NAME_LINK);
6662                 if (rc < 0)
6663                         return rc;
6664                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
6665                 if (ldata->ld_buf->lb_buf == NULL)
6666                         return -ENOMEM;
6667                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6668                                   ldata->ld_buf, XATTR_NAME_LINK);
6669         }
6670         if (rc < 0)
6671                 return rc;
6672
6673         return linkea_init_with_rec(ldata);
6674 }
6675
6676 /**
6677  * Given an MDT object, try to look up the full path to the object.
6678  * Part of the MDT layer implementation of lfs fid2path.
6679  *
6680  * \param[in]     info  Per-thread common data shared by MDT level handlers.
6681  * \param[in]     obj   Object to do path lookup of
6682  * \param[in,out] fp    User-provided struct to store path information
6683  * \param[in]     root_fid Root FID of current path should reach
6684  *
6685  * \retval 0 Lookup successful, path information stored in fp
6686  * \retval -EAGAIN Lookup failed, usually because object is being moved
6687  * \retval negative errno if there was a problem
6688  */
6689 static int mdt_path_current(struct mdt_thread_info *info,
6690                             struct mdt_object *obj,
6691                             struct getinfo_fid2path *fp,
6692                             struct lu_fid *root_fid)
6693 {
6694         struct mdt_device       *mdt = info->mti_mdt;
6695         struct mdt_object       *mdt_obj;
6696         struct link_ea_header   *leh;
6697         struct link_ea_entry    *lee;
6698         struct lu_name          *tmpname = &info->mti_name;
6699         struct lu_fid           *tmpfid = &info->mti_tmp_fid1;
6700         struct lu_buf           *buf = &info->mti_big_buf;
6701         char                    *ptr;
6702         int                     reclen;
6703         struct linkea_data      ldata = { NULL };
6704         int                     rc = 0;
6705         bool                    first = true;
6706         ENTRY;
6707
6708         /* temp buffer for path element, the buffer will be finally freed
6709          * in mdt_thread_info_fini */
6710         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
6711         if (buf->lb_buf == NULL)
6712                 RETURN(-ENOMEM);
6713
6714         ldata.ld_buf = buf;
6715         ptr = fp->gf_u.gf_path + fp->gf_pathlen - 1;
6716         *ptr = 0;
6717         --ptr;
6718         *tmpfid = fp->gf_fid = *mdt_object_fid(obj);
6719
6720         while (!lu_fid_eq(root_fid, &fp->gf_fid)) {
6721                 struct lu_buf           lmv_buf;
6722
6723                 if (!lu_fid_eq(root_fid, &mdt->mdt_md_root_fid) &&
6724                     lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid))
6725                         GOTO(out, rc = -ENOENT);
6726
6727                 if (lu_fid_eq(mdt_object_fid(obj), tmpfid)) {
6728                         mdt_obj = obj;
6729                         mdt_object_get(info->mti_env, mdt_obj);
6730                 } else {
6731                         mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
6732                         if (IS_ERR(mdt_obj))
6733                                 GOTO(out, rc = PTR_ERR(mdt_obj));
6734                 }
6735
6736                 if (!mdt_object_exists(mdt_obj)) {
6737                         mdt_object_put(info->mti_env, mdt_obj);
6738                         GOTO(out, rc = -ENOENT);
6739                 }
6740
6741                 if (mdt_object_remote(mdt_obj)) {
6742                         mdt_object_put(info->mti_env, mdt_obj);
6743                         GOTO(remote_out, rc = -EREMOTE);
6744                 }
6745
6746                 rc = mdt_links_read(info, mdt_obj, &ldata);
6747                 if (rc != 0) {
6748                         mdt_object_put(info->mti_env, mdt_obj);
6749                         GOTO(out, rc);
6750                 }
6751
6752                 leh = buf->lb_buf;
6753                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
6754                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
6755                 /* If set, use link #linkno for path lookup, otherwise use
6756                    link #0.  Only do this for the final path element. */
6757                 if (first && fp->gf_linkno < leh->leh_reccount) {
6758                         int count;
6759                         for (count = 0; count < fp->gf_linkno; count++) {
6760                                 lee = (struct link_ea_entry *)
6761                                      ((char *)lee + reclen);
6762                                 linkea_entry_unpack(lee, &reclen, tmpname,
6763                                                     tmpfid);
6764                         }
6765                         if (fp->gf_linkno < leh->leh_reccount - 1)
6766                                 /* indicate to user there are more links */
6767                                 fp->gf_linkno++;
6768                 }
6769
6770                 lmv_buf.lb_buf = info->mti_xattr_buf;
6771                 lmv_buf.lb_len = sizeof(info->mti_xattr_buf);
6772                 /* Check if it is slave stripes */
6773                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6774                                   &lmv_buf, XATTR_NAME_LMV);
6775                 mdt_object_put(info->mti_env, mdt_obj);
6776                 if (rc > 0) {
6777                         union lmv_mds_md *lmm = lmv_buf.lb_buf;
6778
6779                         /* For slave stripes, get its master */
6780                         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE) {
6781                                 fp->gf_fid = *tmpfid;
6782                                 continue;
6783                         }
6784                 } else if (rc < 0 && rc != -ENODATA) {
6785                         GOTO(out, rc);
6786                 }
6787
6788                 rc = 0;
6789
6790                 /* Pack the name in the end of the buffer */
6791                 ptr -= tmpname->ln_namelen;
6792                 if (ptr - 1 <= fp->gf_u.gf_path)
6793                         GOTO(out, rc = -EOVERFLOW);
6794                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
6795                 *(--ptr) = '/';
6796
6797                 /* keep the last resolved fid to the client, so the
6798                  * client will build the left path on another MDT for
6799                  * remote object */
6800                 fp->gf_fid = *tmpfid;
6801
6802                 first = false;
6803         }
6804
6805 remote_out:
6806         ptr++; /* skip leading / */
6807         memmove(fp->gf_u.gf_path, ptr,
6808                 fp->gf_u.gf_path + fp->gf_pathlen - ptr);
6809
6810 out:
6811         RETURN(rc);
6812 }
6813
6814 /**
6815  * Given an MDT object, use mdt_path_current to get the path.
6816  * Essentially a wrapper to retry mdt_path_current a set number of times
6817  * if -EAGAIN is returned (usually because an object is being moved).
6818  *
6819  * Part of the MDT layer implementation of lfs fid2path.
6820  *
6821  * \param[in]     info  Per-thread common data shared by mdt level handlers.
6822  * \param[in]     obj   Object to do path lookup of
6823  * \param[in,out] fp    User-provided struct for arguments and to store path
6824  *                      information
6825  *
6826  * \retval 0 Lookup successful, path information stored in fp
6827  * \retval negative errno if there was a problem
6828  */
6829 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
6830                     struct getinfo_fid2path *fp, struct lu_fid *root_fid)
6831 {
6832         struct mdt_device       *mdt = info->mti_mdt;
6833         int                     tries = 3;
6834         int                     rc = -EAGAIN;
6835         ENTRY;
6836
6837         if (fp->gf_pathlen < 3)
6838                 RETURN(-EOVERFLOW);
6839
6840         if (root_fid == NULL)
6841                 root_fid = &mdt->mdt_md_root_fid;
6842
6843         if (lu_fid_eq(root_fid, mdt_object_fid(obj))) {
6844                 fp->gf_u.gf_path[0] = '\0';
6845                 RETURN(0);
6846         }
6847
6848         /* Retry multiple times in case file is being moved */
6849         while (tries-- && rc == -EAGAIN)
6850                 rc = mdt_path_current(info, obj, fp, root_fid);
6851
6852         RETURN(rc);
6853 }
6854
6855 /**
6856  * Get the full path of the provided FID, as of changelog record recno.
6857  *
6858  * This checks sanity and looks up object for user provided FID
6859  * before calling the actual path lookup code.
6860  *
6861  * Part of the MDT layer implementation of lfs fid2path.
6862  *
6863  * \param[in]     info  Per-thread common data shared by mdt level handlers.
6864  * \param[in,out] fp    User-provided struct for arguments and to store path
6865  *                      information
6866  *
6867  * \retval 0 Lookup successful, path information and recno stored in fp
6868  * \retval -ENOENT, object does not exist
6869  * \retval negative errno if there was a problem
6870  */
6871 static int mdt_fid2path(struct mdt_thread_info *info,
6872                         struct lu_fid *root_fid,
6873                         struct getinfo_fid2path *fp)
6874 {
6875         struct mdt_device *mdt = info->mti_mdt;
6876         struct mdt_object *obj;
6877         int    rc;
6878         ENTRY;
6879
6880         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
6881                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
6882
6883         if (!fid_is_sane(&fp->gf_fid))
6884                 RETURN(-EINVAL);
6885
6886         if (!fid_is_namespace_visible(&fp->gf_fid)) {
6887                 CDEBUG(D_INFO, "%s: "DFID" is invalid, f_seq should be >= %#llx"
6888                        ", or f_oid != 0, or f_ver == 0\n", mdt_obd_name(mdt),
6889                        PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
6890                 RETURN(-EINVAL);
6891         }
6892
6893         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
6894         if (IS_ERR(obj)) {
6895                 rc = PTR_ERR(obj);
6896                 CDEBUG(D_IOCTL, "cannot find "DFID": rc = %d\n",
6897                        PFID(&fp->gf_fid), rc);
6898                 RETURN(rc);
6899         }
6900
6901         if (mdt_object_remote(obj))
6902                 rc = -EREMOTE;
6903         else if (!mdt_object_exists(obj))
6904                 rc = -ENOENT;
6905         else
6906                 rc = 0;
6907
6908         if (rc < 0) {
6909                 mdt_object_put(info->mti_env, obj);
6910                 CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n",
6911                        PFID(&fp->gf_fid), rc);
6912                 RETURN(rc);
6913         }
6914
6915         rc = mdt_path(info, obj, fp, root_fid);
6916
6917         CDEBUG(D_INFO, "fid "DFID", path %s recno %#llx linkno %u\n",
6918                PFID(&fp->gf_fid), fp->gf_u.gf_path,
6919                fp->gf_recno, fp->gf_linkno);
6920
6921         mdt_object_put(info->mti_env, obj);
6922
6923         RETURN(rc);
6924 }
6925
6926 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, int keylen,
6927                             void *val, int vallen)
6928 {
6929         struct getinfo_fid2path *fpout, *fpin;
6930         struct lu_fid *root_fid = NULL;
6931         int rc = 0;
6932
6933         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
6934         fpout = val;
6935
6936         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
6937                 lustre_swab_fid2path(fpin);
6938
6939         memcpy(fpout, fpin, sizeof(*fpin));
6940         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
6941                 RETURN(-EINVAL);
6942
6943         if (keylen >= cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*fpin) +
6944                       sizeof(struct lu_fid)) {
6945                 /* client sent its root FID, which is normally fileset FID */
6946                 root_fid = fpin->gf_u.gf_root_fid;
6947                 if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
6948                         lustre_swab_lu_fid(root_fid);
6949
6950                 if (root_fid != NULL && !fid_is_sane(root_fid))
6951                         RETURN(-EINVAL);
6952         }
6953
6954         rc = mdt_fid2path(info, root_fid, fpout);
6955         RETURN(rc);
6956 }
6957
6958 int mdt_get_info(struct tgt_session_info *tsi)
6959 {
6960         char    *key;
6961         int      keylen;
6962         __u32   *vallen;
6963         void    *valout;
6964         int      rc;
6965
6966         ENTRY;
6967
6968         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
6969         if (key == NULL) {
6970                 CDEBUG(D_IOCTL, "No GETINFO key\n");
6971                 RETURN(err_serious(-EFAULT));
6972         }
6973         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
6974                                       RCL_CLIENT);
6975
6976         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
6977         if (vallen == NULL) {
6978                 CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n",
6979                                 tgt_name(tsi->tsi_tgt));
6980                 RETURN(err_serious(-EFAULT));
6981         }
6982
6983         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
6984                              *vallen);
6985         rc = req_capsule_server_pack(tsi->tsi_pill);
6986         if (rc)
6987                 RETURN(err_serious(rc));
6988
6989         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
6990         if (valout == NULL) {
6991                 CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n",
6992                                 tgt_name(tsi->tsi_tgt));
6993                 RETURN(err_serious(-EFAULT));
6994         }
6995
6996         if (KEY_IS(KEY_FID2PATH)) {
6997                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
6998
6999                 rc = mdt_rpc_fid2path(info, key, keylen, valout, *vallen);
7000                 mdt_thread_info_fini(info);
7001         } else {
7002                 rc = -EINVAL;
7003         }
7004         RETURN(rc);
7005 }
7006
7007 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
7008 {
7009         struct obd_ioctl_data *data = karg;
7010         struct lu_fid *fid;
7011         __u64 version;
7012         struct mdt_object *obj;
7013         struct mdt_lock_handle  *lh;
7014         int rc;
7015         ENTRY;
7016
7017         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
7018             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
7019                 RETURN(-EINVAL);
7020
7021         fid = (struct lu_fid *)data->ioc_inlbuf1;
7022
7023         if (!fid_is_sane(fid))
7024                 RETURN(-EINVAL);
7025
7026         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
7027
7028         lh = &mti->mti_lh[MDT_LH_PARENT];
7029         mdt_lock_reg_init(lh, LCK_CR);
7030
7031         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
7032         if (IS_ERR(obj))
7033                 RETURN(PTR_ERR(obj));
7034
7035         if (mdt_object_remote(obj)) {
7036                 rc = -EREMOTE;
7037                 /**
7038                  * before calling version get the correct MDS should be
7039                  * fid, this is error to find remote object here
7040                  */
7041                 CERROR("nonlocal object "DFID"\n", PFID(fid));
7042         } else if (!mdt_object_exists(obj)) {
7043                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
7044                 rc = -ENOENT;
7045         } else {
7046                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
7047                *(__u64 *)data->ioc_inlbuf2 = version;
7048                 rc = 0;
7049         }
7050         mdt_object_unlock_put(mti, obj, lh, 1);
7051         RETURN(rc);
7052 }
7053
7054 /* ioctls on obd dev */
7055 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
7056                          void *karg, void __user *uarg)
7057 {
7058         struct lu_env      env;
7059         struct obd_device *obd = exp->exp_obd;
7060         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
7061         struct dt_device  *dt = mdt->mdt_bottom;
7062         int rc;
7063
7064         ENTRY;
7065         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
7066         rc = lu_env_init(&env, LCT_MD_THREAD);
7067         if (rc)
7068                 RETURN(rc);
7069
7070         switch (cmd) {
7071         case OBD_IOC_SYNC:
7072                 rc = mdt_device_sync(&env, mdt);
7073                 break;
7074         case OBD_IOC_SET_READONLY:
7075                 rc = dt_sync(&env, dt);
7076                 if (rc == 0)
7077                         rc = dt_ro(&env, dt);
7078                 break;
7079         case OBD_IOC_ABORT_RECOVERY:
7080                 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
7081                 obd->obd_abort_recovery = 1;
7082                 target_stop_recovery_thread(obd);
7083                 rc = 0;
7084                 break;
7085         case OBD_IOC_CHANGELOG_REG:
7086         case OBD_IOC_CHANGELOG_DEREG:
7087         case OBD_IOC_CHANGELOG_CLEAR:
7088                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(&env,
7089                                                            mdt->mdt_child,
7090                                                            cmd, len, karg);
7091                 break;
7092         case OBD_IOC_START_LFSCK: {
7093                 struct md_device *next = mdt->mdt_child;
7094                 struct obd_ioctl_data *data = karg;
7095                 struct lfsck_start_param lsp;
7096
7097                 if (unlikely(data == NULL)) {
7098                         rc = -EINVAL;
7099                         break;
7100                 }
7101
7102                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
7103                 lsp.lsp_index_valid = 0;
7104                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
7105                 break;
7106         }
7107         case OBD_IOC_STOP_LFSCK: {
7108                 struct md_device        *next = mdt->mdt_child;
7109                 struct obd_ioctl_data   *data = karg;
7110                 struct lfsck_stop        stop;
7111
7112                 stop.ls_status = LS_STOPPED;
7113                 /* Old lfsck utils may pass NULL @stop. */
7114                 if (data->ioc_inlbuf1 == NULL)
7115                         stop.ls_flags = 0;
7116                 else
7117                         stop.ls_flags =
7118                         ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
7119
7120                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
7121                 break;
7122         }
7123         case OBD_IOC_QUERY_LFSCK: {
7124                 struct md_device        *next = mdt->mdt_child;
7125                 struct obd_ioctl_data   *data = karg;
7126
7127                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0,
7128                                                  data->ioc_inlbuf1);
7129                 break;
7130         }
7131         case OBD_IOC_GET_OBJ_VERSION: {
7132                 struct mdt_thread_info *mti;
7133                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
7134                 memset(mti, 0, sizeof *mti);
7135                 mti->mti_env = &env;
7136                 mti->mti_mdt = mdt;
7137                 mti->mti_exp = exp;
7138
7139                 rc = mdt_ioc_version_get(mti, karg);
7140                 break;
7141         }
7142         case OBD_IOC_CATLOGLIST: {
7143                 struct mdt_thread_info *mti;
7144
7145                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
7146                 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
7147                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
7148                                        &mti->mti_tmp_fid1);
7149                 break;
7150          }
7151         default:
7152                 rc = -EOPNOTSUPP;
7153                 CERROR("%s: Not supported cmd = %d, rc = %d\n",
7154                         mdt_obd_name(mdt), cmd, rc);
7155         }
7156
7157         lu_env_fini(&env);
7158         RETURN(rc);
7159 }
7160
7161 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
7162 {
7163         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
7164         int rc;
7165         ENTRY;
7166
7167         if (!mdt->mdt_skip_lfsck && !mdt->mdt_bottom->dd_rdonly) {
7168                 struct lfsck_start_param lsp;
7169
7170                 lsp.lsp_start = NULL;
7171                 lsp.lsp_index_valid = 0;
7172                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
7173                                                            OBD_IOC_START_LFSCK,
7174                                                            0, &lsp);
7175                 if (rc != 0 && rc != -EALREADY)
7176                         CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
7177                               mdt_obd_name(mdt), rc);
7178         }
7179
7180         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
7181         RETURN(rc);
7182 }
7183
7184 static int mdt_obd_postrecov(struct obd_device *obd)
7185 {
7186         struct lu_env env;
7187         int rc;
7188
7189         rc = lu_env_init(&env, LCT_MD_THREAD);
7190         if (rc)
7191                 RETURN(rc);
7192         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
7193         lu_env_fini(&env);
7194         return rc;
7195 }
7196
7197 static const struct obd_ops mdt_obd_device_ops = {
7198         .o_owner          = THIS_MODULE,
7199         .o_set_info_async = mdt_obd_set_info_async,
7200         .o_connect        = mdt_obd_connect,
7201         .o_reconnect      = mdt_obd_reconnect,
7202         .o_disconnect     = mdt_obd_disconnect,
7203         .o_init_export    = mdt_init_export,
7204         .o_destroy_export = mdt_destroy_export,
7205         .o_iocontrol      = mdt_iocontrol,
7206         .o_postrecov      = mdt_obd_postrecov,
7207         /* Data-on-MDT IO methods */
7208         .o_preprw         = mdt_obd_preprw,
7209         .o_commitrw       = mdt_obd_commitrw,
7210 };
7211
7212 static struct lu_device* mdt_device_fini(const struct lu_env *env,
7213                                          struct lu_device *d)
7214 {
7215         struct mdt_device *m = mdt_dev(d);
7216         ENTRY;
7217
7218         mdt_fini(env, m);
7219         RETURN(NULL);
7220 }
7221
7222 static struct lu_device *mdt_device_free(const struct lu_env *env,
7223                                          struct lu_device *d)
7224 {
7225         struct mdt_device *m = mdt_dev(d);
7226         ENTRY;
7227
7228         lu_device_fini(&m->mdt_lu_dev);
7229         OBD_FREE_PTR(m);
7230
7231         RETURN(NULL);
7232 }
7233
7234 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
7235                                           struct lu_device_type *t,
7236                                           struct lustre_cfg *cfg)
7237 {
7238         struct lu_device  *l;
7239         struct mdt_device *m;
7240
7241         OBD_ALLOC_PTR(m);
7242         if (m != NULL) {
7243                 int rc;
7244
7245                 l = &m->mdt_lu_dev;
7246                 rc = mdt_init0(env, m, t, cfg);
7247                 if (rc != 0) {
7248                         mdt_device_free(env, l);
7249                         l = ERR_PTR(rc);
7250                         return l;
7251                 }
7252         } else
7253                 l = ERR_PTR(-ENOMEM);
7254         return l;
7255 }
7256
7257 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
7258 LU_KEY_INIT(mdt, struct mdt_thread_info);
7259
7260 static void mdt_key_fini(const struct lu_context *ctx,
7261                          struct lu_context_key *key, void* data)
7262 {
7263         struct mdt_thread_info *info = data;
7264
7265         if (info->mti_big_lmm) {
7266                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
7267                 info->mti_big_lmm = NULL;
7268                 info->mti_big_lmmsize = 0;
7269         }
7270
7271         if (info->mti_big_acl) {
7272                 OBD_FREE_LARGE(info->mti_big_acl, info->mti_big_aclsize);
7273                 info->mti_big_acl = NULL;
7274                 info->mti_big_aclsize = 0;
7275         }
7276
7277         OBD_FREE_PTR(info);
7278 }
7279
7280 /* context key: mdt_thread_key */
7281 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
7282
7283 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
7284 {
7285         return lu_ucred(info->mti_env);
7286 }
7287
7288 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
7289 {
7290         return lu_ucred_check(info->mti_env);
7291 }
7292
7293 /**
7294  * Enable/disable COS (Commit On Sharing).
7295  *
7296  * Set/Clear the COS flag in mdt options.
7297  *
7298  * \param mdt mdt device
7299  * \param val 0 disables COS, other values enable COS
7300  */
7301 void mdt_enable_cos(struct mdt_device *mdt, bool val)
7302 {
7303         struct lu_env env;
7304         int rc;
7305
7306         mdt->mdt_opts.mo_cos = val;
7307         rc = lu_env_init(&env, LCT_LOCAL);
7308         if (unlikely(rc != 0)) {
7309                 CWARN("%s: lu_env initialization failed, cannot "
7310                       "sync: rc = %d\n", mdt_obd_name(mdt), rc);
7311                 return;
7312         }
7313         mdt_device_sync(&env, mdt);
7314         lu_env_fini(&env);
7315 }
7316
7317 /**
7318  * Check COS (Commit On Sharing) status.
7319  *
7320  * Return COS flag status.
7321  *
7322  * \param mdt mdt device
7323  */
7324 int mdt_cos_is_enabled(struct mdt_device *mdt)
7325 {
7326         return mdt->mdt_opts.mo_cos != 0;
7327 }
7328
7329 static struct lu_device_type_operations mdt_device_type_ops = {
7330         .ldto_device_alloc = mdt_device_alloc,
7331         .ldto_device_free  = mdt_device_free,
7332         .ldto_device_fini  = mdt_device_fini
7333 };
7334
7335 static struct lu_device_type mdt_device_type = {
7336         .ldt_tags     = LU_DEVICE_MD,
7337         .ldt_name     = LUSTRE_MDT_NAME,
7338         .ldt_ops      = &mdt_device_type_ops,
7339         .ldt_ctx_tags = LCT_MD_THREAD
7340 };
7341
7342 static int __init mdt_init(void)
7343 {
7344         int rc;
7345
7346         BUILD_BUG_ON(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") !=
7347                      FID_NOBRACE_LEN + 1);
7348         BUILD_BUG_ON(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") !=
7349                      FID_LEN + 1);
7350         rc = lu_kmem_init(mdt_caches);
7351         if (rc)
7352                 return rc;
7353
7354         rc = mds_mod_init();
7355         if (rc)
7356                 GOTO(lu_fini, rc);
7357
7358         rc = class_register_type(&mdt_obd_device_ops, NULL, true, NULL,
7359                                  LUSTRE_MDT_NAME, &mdt_device_type);
7360         if (rc)
7361                 GOTO(mds_fini, rc);
7362 lu_fini:
7363         if (rc)
7364                 lu_kmem_fini(mdt_caches);
7365 mds_fini:
7366         if (rc)
7367                 mds_mod_exit();
7368         return rc;
7369 }
7370
7371 static void __exit mdt_exit(void)
7372 {
7373         class_unregister_type(LUSTRE_MDT_NAME);
7374         mds_mod_exit();
7375         lu_kmem_fini(mdt_caches);
7376 }
7377
7378 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
7379 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
7380 MODULE_VERSION(LUSTRE_VERSION_STRING);
7381 MODULE_LICENSE("GPL");
7382
7383 module_init(mdt_init);
7384 module_exit(mdt_exit);