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  *
  31  * lustre/mdt/mdt_handler.c
  32  *
  33  * Lustre Metadata Target (mdt) request handler
  34  *
  35  * Author: Peter Braam <braam@clusterfs.com>
  36  * Author: Andreas Dilger <adilger@clusterfs.com>
  37  * Author: Phil Schwan <phil@clusterfs.com>
  38  * Author: Mike Shaver <shaver@clusterfs.com>
  39  * Author: Nikita Danilov <nikita@clusterfs.com>
  40  * Author: Huang Hua <huanghua@clusterfs.com>
  41  * Author: Yury Umanets <umka@clusterfs.com>
  42  */
  43
  44 #define DEBUG_SUBSYSTEM S_MDS
  45
  46 #include <linux/module.h>
  47 #include <linux/pagemap.h>
  48
  49 #include <dt_object.h>
  50 #include <lustre_acl.h>
  51 #include <lustre_export.h>
  52 #include <uapi/linux/lustre/lustre_ioctl.h>
  53 #include <lustre_lfsck.h>
  54 #include <lustre_log.h>
  55 #include <lustre_nodemap.h>
  56 #include <lustre_mds.h>
  57 #include <uapi/linux/lustre/lustre_param.h>
  58 #include <lustre_quota.h>
  59 #include <lustre_swab.h>
  60 #include <lustre_lmv.h>
  61 #include <obd.h>
  62 #include <obd_support.h>
  63 #include <lustre_barrier.h>
  64 #include <obd_cksum.h>
  65 #include <llog_swab.h>
  66 #include <lustre_crypto.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_lh_reg_init(struct mdt_lock_handle *lh, struct ldlm_lock *lock)
 168 {
 169         mdt_lock_reg_init(lh, lock->l_req_mode);
 170         if (lock->l_req_mode == LCK_GROUP)
 171                 lh->mlh_gid = lock->l_policy_data.l_inodebits.li_gid;
 172 }
 173
 174 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode,
 175                        const struct lu_name *lname)
 176 {
 177         lh->mlh_reg_mode = lock_mode;
 178         lh->mlh_pdo_mode = LCK_MINMODE;
 179         lh->mlh_rreg_mode = lock_mode;
 180         lh->mlh_type = MDT_PDO_LOCK;
 181
 182         if (lu_name_is_valid(lname)) {
 183                 lh->mlh_pdo_hash = ll_full_name_hash(NULL, lname->ln_name,
 184                                                      lname->ln_namelen);
 185                 /* XXX Workaround for LU-2856
 186                  *
 187                  * Zero is a valid return value of full_name_hash, but
 188                  * several users of mlh_pdo_hash assume a non-zero
 189                  * hash value. We therefore map zero onto an
 190                  * arbitrary, but consistent value (1) to avoid
 191                  * problems further down the road. */
 192                 if (unlikely(lh->mlh_pdo_hash == 0))
 193                         lh->mlh_pdo_hash = 1;
 194         } else {
 195                 lh->mlh_pdo_hash = 0;
 196         }
 197 }
 198
 199 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 200                               struct mdt_lock_handle *lh)
 201 {
 202         mdl_mode_t mode;
 203         ENTRY;
 204
 205         /*
 206          * Any dir access needs couple of locks:
 207          *
 208          * 1) on part of dir we gonna take lookup/modify;
 209          *
 210          * 2) on whole dir to protect it from concurrent splitting and/or to
 211          * flush client's cache for readdir().
 212          *
 213          * so, for a given mode and object this routine decides what lock mode
 214          * to use for lock #2:
 215          *
 216          * 1) if caller's gonna lookup in dir then we need to protect dir from
 217          * being splitted only - LCK_CR
 218          *
 219          * 2) if caller's gonna modify dir then we need to protect dir from
 220          * being splitted and to flush cache - LCK_CW
 221          *
 222          * 3) if caller's gonna modify dir and that dir seems ready for
 223          * splitting then we need to protect it from any type of access
 224          * (lookup/modify/split) - LCK_EX --bzzz
 225          */
 226
 227         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 228         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 229
 230         /*
 231          * Ask underlaying level its opinion about preferable PDO lock mode
 232          * having access type passed as regular lock mode:
 233          *
 234          * - MDL_MINMODE means that lower layer does not want to specify lock
 235          * mode;
 236          *
 237          * - MDL_NL means that no PDO lock should be taken. This is used in some
 238          * cases. Say, for non-splittable directories no need to use PDO locks
 239          * at all.
 240          */
 241         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 242                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 243
 244         if (mode != MDL_MINMODE) {
 245                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 246         } else {
 247                 /*
 248                  * Lower layer does not want to specify locking mode. We do it
 249                  * our selves. No special protection is needed, just flush
 250                  * client's cache on modification and allow concurrent
 251                  * mondification.
 252                  */
 253                 switch (lh->mlh_reg_mode) {
 254                 case LCK_EX:
 255                         lh->mlh_pdo_mode = LCK_EX;
 256                         break;
 257                 case LCK_PR:
 258                         lh->mlh_pdo_mode = LCK_CR;
 259                         break;
 260                 case LCK_PW:
 261                         lh->mlh_pdo_mode = LCK_CW;
 262                         break;
 263                 default:
 264                         CERROR("Not expected lock type (0x%x)\n",
 265                                (int)lh->mlh_reg_mode);
 266                         LBUG();
 267                 }
 268         }
 269
 270         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 271         EXIT;
 272 }
 273
 274 /**
 275  * Check whether \a o is directory stripe object.
 276  *
 277  * \param[in]  info     thread environment
 278  * \param[in]  o        MDT object
 279  *
 280  * \retval 1    is directory stripe.
 281  * \retval 0    isn't directory stripe.
 282  * \retval < 1  error code
 283  */
 284 static int mdt_is_dir_stripe(struct mdt_thread_info *info,
 285                                 struct mdt_object *o)
 286 {
 287         struct md_attr *ma = &info->mti_attr;
 288         struct lmv_mds_md_v1 *lmv;
 289         int rc;
 290
 291         rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
 292         if (rc < 0)
 293                 return rc;
 294
 295         if (!(ma->ma_valid & MA_LMV))
 296                 return 0;
 297
 298         lmv = &ma->ma_lmv->lmv_md_v1;
 299
 300         if (!lmv_is_sane2(lmv))
 301                 return -EBADF;
 302
 303         if (le32_to_cpu(lmv->lmv_magic) == LMV_MAGIC_STRIPE)
 304                 return 1;
 305
 306         return 0;
 307 }
 308
 309 static int mdt_lookup_fileset(struct mdt_thread_info *info, const char *fileset,
 310                               struct lu_fid *fid)
 311 {
 312         struct mdt_device *mdt = info->mti_mdt;
 313         struct lu_name *lname = &info->mti_name;
 314         const char *start = fileset;
 315         char *filename = info->mti_filename;
 316         struct mdt_object *parent;
 317         u32 mode;
 318         int rc = 0;
 319
 320         LASSERT(!info->mti_cross_ref);
 321
 322         /*
 323          * We may want to allow this to mount a completely separate
 324          * fileset from the MDT in the future, but keeping it to
 325          * ROOT/ only for now avoid potential security issues.
 326          */
 327         *fid = mdt->mdt_md_root_fid;
 328
 329         while (rc == 0 && start != NULL && *start != '\0') {
 330                 const char *s1 = start;
 331                 const char *s2;
 332
 333                 while (*++s1 == '/')
 334                         ;
 335                 s2 = s1;
 336                 while (*s2 != '/' && *s2 != '\0')
 337                         s2++;
 338
 339                 if (s2 == s1)
 340                         break;
 341
 342                 start = s2;
 343
 344                 lname->ln_namelen = s2 - s1;
 345                 if (lname->ln_namelen > NAME_MAX) {
 346                         rc = -EINVAL;
 347                         break;
 348                 }
 349
 350                 /* reject .. as a path component */
 351                 if (lname->ln_namelen == 2 &&
 352                     strncmp(s1, "..", 2) == 0) {
 353                         rc = -EINVAL;
 354                         break;
 355                 }
 356
 357                 strncpy(filename, s1, lname->ln_namelen);
 358                 filename[lname->ln_namelen] = '\0';
 359                 lname->ln_name = filename;
 360
 361                 parent = mdt_object_find(info->mti_env, mdt, fid);
 362                 if (IS_ERR(parent)) {
 363                         rc = PTR_ERR(parent);
 364                         break;
 365                 }
 366                 /* Only got the fid of this obj by name */
 367                 fid_zero(fid);
 368                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
 369                                 fid, &info->mti_spec);
 370                 mdt_object_put(info->mti_env, parent);
 371         }
 372         if (!rc) {
 373                 parent = mdt_object_find(info->mti_env, mdt, fid);
 374                 if (IS_ERR(parent))
 375                         rc = PTR_ERR(parent);
 376                 else {
 377                         mode = lu_object_attr(&parent->mot_obj);
 378                         if (!S_ISDIR(mode)) {
 379                                 rc = -ENOTDIR;
 380                         } else if (mdt_is_remote_object(info, parent, parent)) {
 381                                 if (!mdt->mdt_enable_remote_subdir_mount) {
 382                                         rc = -EREMOTE;
 383                                         LCONSOLE_WARN("%s: subdir mount '%s' refused because 'enable_remote_subdir_mount=0': rc = %d\n",
 384                                                       mdt_obd_name(mdt),
 385                                                       fileset, rc);
 386                                 } else {
 387                                         LCONSOLE_INFO("%s: subdir mount '%s' is remote and may be slow\n",
 388                                                       mdt_obd_name(mdt),
 389                                                       fileset);
 390                                 }
 391                         }
 392                         mdt_object_put(info->mti_env, parent);
 393                 }
 394         }
 395
 396         return rc;
 397 }
 398
 399 static int mdt_get_root(struct tgt_session_info *tsi)
 400 {
 401         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
 402         struct mdt_device       *mdt = info->mti_mdt;
 403         struct mdt_body         *repbody;
 404         char                    *fileset = NULL, *buffer = NULL;
 405         int                      rc;
 406         struct obd_export       *exp = info->mti_exp;
 407         char                    *nodemap_fileset;
 408
 409         ENTRY;
 410
 411         rc = mdt_check_ucred(info);
 412         if (rc)
 413                 GOTO(out, rc = err_serious(rc));
 414
 415         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GET_ROOT_PACK))
 416                 GOTO(out, rc = err_serious(-ENOMEM));
 417
 418         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 419         if (req_capsule_get_size(info->mti_pill, &RMF_NAME, RCL_CLIENT) > 0) {
 420                 fileset = req_capsule_client_get(info->mti_pill, &RMF_NAME);
 421                 if (fileset == NULL)
 422                         GOTO(out, rc = err_serious(-EFAULT));
 423         }
 424
 425         nodemap_fileset = nodemap_get_fileset(exp->exp_target_data.ted_nodemap);
 426         if (nodemap_fileset && nodemap_fileset[0]) {
 427                 CDEBUG(D_INFO, "nodemap fileset is %s\n", nodemap_fileset);
 428                 if (fileset) {
 429                         /* consider fileset from client as a sub-fileset
 430                          * of the nodemap one */
 431                         OBD_ALLOC(buffer, PATH_MAX + 1);
 432                         if (buffer == NULL)
 433                                 GOTO(out, rc = err_serious(-ENOMEM));
 434                         if (snprintf(buffer, PATH_MAX + 1, "%s/%s",
 435                                      nodemap_fileset, fileset) >= PATH_MAX + 1)
 436                                 GOTO(out, rc = err_serious(-EINVAL));
 437                         fileset = buffer;
 438                 } else {
 439                         /* enforce fileset as specified in the nodemap */
 440                         fileset = nodemap_fileset;
 441                 }
 442         }
 443
 444         if (fileset) {
 445                 CDEBUG(D_INFO, "Getting fileset %s\n", fileset);
 446                 rc = mdt_lookup_fileset(info, fileset, &repbody->mbo_fid1);
 447                 if (rc < 0)
 448                         GOTO(out, rc = err_serious(rc));
 449         } else {
 450                 repbody->mbo_fid1 = mdt->mdt_md_root_fid;
 451         }
 452         repbody->mbo_valid |= OBD_MD_FLID;
 453
 454         EXIT;
 455 out:
 456         mdt_thread_info_fini(info);
 457         if (buffer)
 458                 OBD_FREE(buffer, PATH_MAX+1);
 459         return rc;
 460 }
 461
 462 static int mdt_statfs(struct tgt_session_info *tsi)
 463 {
 464         struct ptlrpc_request *req = tgt_ses_req(tsi);
 465         struct mdt_thread_info *info = tsi2mdt_info(tsi);
 466         struct mdt_device *mdt = info->mti_mdt;
 467         struct tg_grants_data *tgd = &mdt->mdt_lut.lut_tgd;
 468         struct md_device *next = mdt->mdt_child;
 469         struct ptlrpc_service_part *svcpt;
 470         struct obd_statfs *osfs;
 471         struct mdt_body *reqbody = NULL;
 472         struct mdt_statfs_cache *msf;
 473         ktime_t kstart = ktime_get();
 474         int current_blockbits;
 475         int rc;
 476
 477         ENTRY;
 478
 479         svcpt = req->rq_rqbd->rqbd_svcpt;
 480
 481         /* This will trigger a watchdog timeout */
 482         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 483                          (MDT_SERVICE_WATCHDOG_FACTOR *
 484                           at_get(&svcpt->scp_at_estimate)) + 1);
 485
 486         rc = mdt_check_ucred(info);
 487         if (rc)
 488                 GOTO(out, rc = err_serious(rc));
 489
 490         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
 491                 GOTO(out, rc = err_serious(-ENOMEM));
 492
 493         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 494         if (!osfs)
 495                 GOTO(out, rc = -EPROTO);
 496
 497         if (mdt_is_sum_statfs_client(req->rq_export) &&
 498                 lustre_packed_msg_size(req->rq_reqmsg) ==
 499                 req_capsule_fmt_size(req->rq_reqmsg->lm_magic,
 500                                      &RQF_MDS_STATFS_NEW, RCL_CLIENT)) {
 501                 req_capsule_extend(info->mti_pill, &RQF_MDS_STATFS_NEW);
 502                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
 503         }
 504
 505         if (reqbody && reqbody->mbo_valid & OBD_MD_FLAGSTATFS)
 506                 msf = &mdt->mdt_sum_osfs;
 507         else
 508                 msf = &mdt->mdt_osfs;
 509
 510         if (msf->msf_age + OBD_STATFS_CACHE_SECONDS <= ktime_get_seconds()) {
 511                         /** statfs data is too old, get up-to-date one */
 512                         if (reqbody && reqbody->mbo_valid & OBD_MD_FLAGSTATFS)
 513                                 rc = next->md_ops->mdo_statfs(info->mti_env,
 514                                                               next, osfs);
 515                         else
 516                                 rc = dt_statfs(info->mti_env, mdt->mdt_bottom,
 517                                                osfs);
 518                         if (rc)
 519                                 GOTO(out, rc);
 520                         spin_lock(&mdt->mdt_lock);
 521                         msf->msf_osfs = *osfs;
 522                         msf->msf_age = ktime_get_seconds();
 523                         spin_unlock(&mdt->mdt_lock);
 524         } else {
 525                         /** use cached statfs data */
 526                         spin_lock(&mdt->mdt_lock);
 527                         *osfs = msf->msf_osfs;
 528                         spin_unlock(&mdt->mdt_lock);
 529         }
 530
 531         /* tgd_blockbit is recordsize bits set during mkfs.
 532          * This once set does not change. However, 'zfs set'
 533          * can be used to change the MDT blocksize. Instead
 534          * of using cached value of 'tgd_blockbit' always
 535          * calculate the blocksize bits which may have
 536          * changed.
 537          */
 538         current_blockbits = fls64(osfs->os_bsize) - 1;
 539
 540         /* at least try to account for cached pages.  its still racy and
 541          * might be under-reporting if clients haven't announced their
 542          * caches with brw recently */
 543         CDEBUG(D_SUPER | D_CACHE, "blocks cached %llu granted %llu"
 544                " pending %llu free %llu avail %llu\n",
 545                tgd->tgd_tot_dirty, tgd->tgd_tot_granted,
 546                tgd->tgd_tot_pending,
 547                osfs->os_bfree << current_blockbits,
 548                osfs->os_bavail << current_blockbits);
 549
 550         osfs->os_bavail -= min_t(u64, osfs->os_bavail,
 551                                  ((tgd->tgd_tot_dirty + tgd->tgd_tot_pending +
 552                                    osfs->os_bsize - 1) >> current_blockbits));
 553
 554         tgt_grant_sanity_check(mdt->mdt_lu_dev.ld_obd, __func__);
 555         CDEBUG(D_CACHE, "%llu blocks: %llu free, %llu avail; "
 556                "%llu objects: %llu free; state %x\n",
 557                osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
 558                osfs->os_files, osfs->os_ffree, osfs->os_state);
 559
 560         if (!exp_grant_param_supp(tsi->tsi_exp) &&
 561             current_blockbits > COMPAT_BSIZE_SHIFT) {
 562                 /* clients which don't support OBD_CONNECT_GRANT_PARAM
 563                  * should not see a block size > page size, otherwise
 564                  * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
 565                  * block size which is the biggest block size known to work
 566                  * with all client's page size. */
 567                 osfs->os_blocks <<= current_blockbits - COMPAT_BSIZE_SHIFT;
 568                 osfs->os_bfree  <<= current_blockbits - COMPAT_BSIZE_SHIFT;
 569                 osfs->os_bavail <<= current_blockbits - COMPAT_BSIZE_SHIFT;
 570                 osfs->os_bsize = 1 << COMPAT_BSIZE_SHIFT;
 571         }
 572         if (rc == 0)
 573                 mdt_counter_incr(req, LPROC_MDT_STATFS,
 574                                  ktime_us_delta(ktime_get(), kstart));
 575 out:
 576         mdt_thread_info_fini(info);
 577         RETURN(rc);
 578 }
 579
 580 __u32 mdt_lmm_dom_entry_check(struct lov_mds_md *lmm, int *is_dom_only)
 581 {
 582         struct lov_comp_md_v1 *comp_v1;
 583         struct lov_mds_md *v1;
 584         __u32 off;
 585         __u32 dom_stripesize = 0;
 586         int i;
 587         bool has_ost_stripes = false;
 588
 589         ENTRY;
 590
 591         if (is_dom_only)
 592                 *is_dom_only = 0;
 593
 594         if (le32_to_cpu(lmm->lmm_magic) != LOV_MAGIC_COMP_V1)
 595                 RETURN(0);
 596
 597         comp_v1 = (struct lov_comp_md_v1 *)lmm;
 598         off = le32_to_cpu(comp_v1->lcm_entries[0].lcme_offset);
 599         v1 = (struct lov_mds_md *)((char *)comp_v1 + off);
 600
 601         /* Fast check for DoM entry with no mirroring, should be the first */
 602         if (le16_to_cpu(comp_v1->lcm_mirror_count) == 0 &&
 603             lov_pattern(le32_to_cpu(v1->lmm_pattern)) != LOV_PATTERN_MDT)
 604                 RETURN(0);
 605
 606         /* check all entries otherwise */
 607         for (i = 0; i < le16_to_cpu(comp_v1->lcm_entry_count); i++) {
 608                 struct lov_comp_md_entry_v1 *lcme;
 609
 610                 lcme = &comp_v1->lcm_entries[i];
 611                 if (!(le32_to_cpu(lcme->lcme_flags) & LCME_FL_INIT))
 612                         continue;
 613
 614                 off = le32_to_cpu(lcme->lcme_offset);
 615                 v1 = (struct lov_mds_md *)((char *)comp_v1 + off);
 616
 617                 if (lov_pattern(le32_to_cpu(v1->lmm_pattern)) ==
 618                     LOV_PATTERN_MDT)
 619                         dom_stripesize = le32_to_cpu(v1->lmm_stripe_size);
 620                 else
 621                         has_ost_stripes = true;
 622
 623                 if (dom_stripesize && has_ost_stripes)
 624                         RETURN(dom_stripesize);
 625         }
 626         /* DoM-only case exits here */
 627         if (is_dom_only && dom_stripesize)
 628                 *is_dom_only = 1;
 629         RETURN(dom_stripesize);
 630 }
 631
 632 /**
 633  * Pack size attributes into the reply.
 634  */
 635 int mdt_pack_size2body(struct mdt_thread_info *info,
 636                         const struct lu_fid *fid, struct lustre_handle *lh)
 637 {
 638         struct mdt_body *b;
 639         struct md_attr *ma = &info->mti_attr;
 640         __u32 dom_stripe;
 641         bool dom_lock = false;
 642
 643         ENTRY;
 644
 645         LASSERT(ma->ma_attr.la_valid & LA_MODE);
 646
 647         if (!S_ISREG(ma->ma_attr.la_mode) ||
 648             !(ma->ma_valid & MA_LOV && ma->ma_lmm != NULL))
 649                 RETURN(-ENODATA);
 650
 651         dom_stripe = mdt_lmm_dom_stripesize(ma->ma_lmm);
 652         /* no DoM stripe, no size in reply */
 653         if (!dom_stripe)
 654                 RETURN(-ENOENT);
 655
 656         if (lustre_handle_is_used(lh)) {
 657                 struct ldlm_lock *lock;
 658
 659                 lock = ldlm_handle2lock(lh);
 660                 if (lock != NULL) {
 661                         dom_lock = ldlm_has_dom(lock);
 662                         LDLM_LOCK_PUT(lock);
 663                 }
 664         }
 665
 666         /* no DoM lock, no size in reply */
 667         if (!dom_lock)
 668                 RETURN(0);
 669
 670         /* Either DoM lock exists or LMM has only DoM stripe then
 671          * return size on body. */
 672         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 673
 674         mdt_dom_object_size(info->mti_env, info->mti_mdt, fid, b, dom_lock);
 675         RETURN(0);
 676 }
 677
 678 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
 679 /*
 680  * Pack ACL data into the reply. UIDs/GIDs are mapped and filtered by nodemap.
 681  *
 682  * \param       info    thread info object
 683  * \param       repbody reply to pack ACLs into
 684  * \param       o       mdt object of file to examine
 685  * \param       nodemap nodemap of client to reply to
 686  * \retval      0       success
 687  * \retval      -errno  error getting or parsing ACL from disk
 688  */
 689 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
 690                       struct mdt_object *o, struct lu_nodemap *nodemap)
 691 {
 692         const struct lu_env     *env = info->mti_env;
 693         struct md_object        *next = mdt_object_child(o);
 694         struct lu_buf           *buf = &info->mti_buf;
 695         struct mdt_device       *mdt = info->mti_mdt;
 696         struct req_capsule *pill = info->mti_pill;
 697         int rc;
 698
 699         ENTRY;
 700
 701         buf->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
 702         buf->lb_len = req_capsule_get_size(pill, &RMF_ACL, RCL_SERVER);
 703         if (buf->lb_len == 0)
 704                 RETURN(0);
 705
 706         LASSERT(!info->mti_big_acl_used);
 707 again:
 708         rc = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_ACCESS);
 709         if (rc < 0) {
 710                 if (rc == -ENODATA) {
 711                         repbody->mbo_aclsize = 0;
 712                         repbody->mbo_valid |= OBD_MD_FLACL;
 713                         rc = 0;
 714                 } else if (rc == -EOPNOTSUPP) {
 715                         rc = 0;
 716                 } else if (rc == -ERANGE) {
 717                         if (exp_connect_large_acl(info->mti_exp) &&
 718                             !info->mti_big_acl_used) {
 719                                 if (info->mti_big_acl == NULL) {
 720                                         info->mti_big_aclsize =
 721                                                         min_t(unsigned int,
 722                                                               mdt->mdt_max_ea_size,
 723                                                               XATTR_SIZE_MAX);
 724                                         OBD_ALLOC_LARGE(info->mti_big_acl,
 725                                                         info->mti_big_aclsize);
 726                                         if (info->mti_big_acl == NULL) {
 727                                                 info->mti_big_aclsize = 0;
 728                                                 CERROR("%s: unable to grow "
 729                                                        DFID" ACL buffer\n",
 730                                                        mdt_obd_name(mdt),
 731                                                        PFID(mdt_object_fid(o)));
 732                                                 RETURN(-ENOMEM);
 733                                         }
 734                                 }
 735
 736                                 CDEBUG(D_INODE, "%s: grow the "DFID
 737                                        " ACL buffer to size %d\n",
 738                                        mdt_obd_name(mdt),
 739                                        PFID(mdt_object_fid(o)),
 740                                        info->mti_big_aclsize);
 741
 742                                 buf->lb_buf = info->mti_big_acl;
 743                                 buf->lb_len = info->mti_big_aclsize;
 744                                 info->mti_big_acl_used = 1;
 745                                 goto again;
 746                         }
 747                         /* FS has ACL bigger that our limits */
 748                         CDEBUG(D_INODE, "%s: "DFID" ACL can't fit into %d\n",
 749                                mdt_obd_name(mdt), PFID(mdt_object_fid(o)),
 750                                info->mti_big_aclsize);
 751                         rc = -E2BIG;
 752                 } else {
 753                         CERROR("%s: unable to read "DFID" ACL: rc = %d\n",
 754                                mdt_obd_name(mdt), PFID(mdt_object_fid(o)), rc);
 755                 }
 756         } else {
 757                 rc = nodemap_map_acl(nodemap, buf->lb_buf,
 758                                      rc, NODEMAP_FS_TO_CLIENT);
 759                 /* if all ACLs mapped out, rc is still >= 0 */
 760                 if (rc < 0) {
 761                         CERROR("%s: nodemap_map_acl unable to parse "DFID
 762                                " ACL: rc = %d\n", mdt_obd_name(mdt),
 763                                PFID(mdt_object_fid(o)), rc);
 764                         repbody->mbo_aclsize = 0;
 765                         repbody->mbo_valid &= ~OBD_MD_FLACL;
 766                 } else {
 767                         repbody->mbo_aclsize = rc;
 768                         repbody->mbo_valid |= OBD_MD_FLACL;
 769                         rc = 0;
 770                 }
 771         }
 772
 773         RETURN(rc);
 774 }
 775 #endif
 776
 777 /* XXX Look into layout in MDT layer. */
 778 static inline bool mdt_hsm_is_released(struct lov_mds_md *lmm)
 779 {
 780         struct lov_comp_md_v1   *comp_v1;
 781         struct lov_mds_md       *v1;
 782         int                      i;
 783
 784         if (lmm->lmm_magic == LOV_MAGIC_COMP_V1) {
 785                 comp_v1 = (struct lov_comp_md_v1 *)lmm;
 786
 787                 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
 788                         v1 = (struct lov_mds_md *)((char *)comp_v1 +
 789                                 comp_v1->lcm_entries[i].lcme_offset);
 790                         /* We don't support partial release for now */
 791                         if (!(v1->lmm_pattern & LOV_PATTERN_F_RELEASED))
 792                                 return false;
 793                 }
 794                 return true;
 795         } else {
 796                 return (lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) ?
 797                         true : false;
 798         }
 799 }
 800
 801 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 802                         const struct lu_attr *attr, const struct lu_fid *fid)
 803 {
 804         struct md_attr *ma = &info->mti_attr;
 805         struct obd_export *exp = info->mti_exp;
 806         struct lu_nodemap *nodemap = NULL;
 807
 808         LASSERT(ma->ma_valid & MA_INODE);
 809
 810         if (attr->la_valid & LA_ATIME) {
 811                 b->mbo_atime = attr->la_atime;
 812                 b->mbo_valid |= OBD_MD_FLATIME;
 813         }
 814         if (attr->la_valid & LA_MTIME) {
 815                 b->mbo_mtime = attr->la_mtime;
 816                 b->mbo_valid |= OBD_MD_FLMTIME;
 817         }
 818         if (attr->la_valid & LA_CTIME) {
 819                 b->mbo_ctime = attr->la_ctime;
 820                 b->mbo_valid |= OBD_MD_FLCTIME;
 821         }
 822         if (attr->la_valid & LA_BTIME) {
 823                 b->mbo_btime = attr->la_btime;
 824                 b->mbo_valid |= OBD_MD_FLBTIME;
 825         }
 826         if (attr->la_valid & LA_FLAGS) {
 827                 b->mbo_flags = attr->la_flags;
 828                 b->mbo_valid |= OBD_MD_FLFLAGS;
 829         }
 830         if (attr->la_valid & LA_NLINK) {
 831                 b->mbo_nlink = attr->la_nlink;
 832                 b->mbo_valid |= OBD_MD_FLNLINK;
 833         }
 834         if (attr->la_valid & (LA_UID|LA_GID|LA_PROJID)) {
 835                 nodemap = nodemap_get_from_exp(exp);
 836                 if (IS_ERR(nodemap))
 837                         goto out;
 838         }
 839         if (attr->la_valid & LA_UID) {
 840                 b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 841                                             NODEMAP_FS_TO_CLIENT,
 842                                             attr->la_uid);
 843                 b->mbo_valid |= OBD_MD_FLUID;
 844         }
 845         if (attr->la_valid & LA_GID) {
 846                 b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 847                                             NODEMAP_FS_TO_CLIENT,
 848                                             attr->la_gid);
 849                 b->mbo_valid |= OBD_MD_FLGID;
 850         }
 851
 852         if (attr->la_valid & LA_PROJID) {
 853                 b->mbo_projid = nodemap_map_id(nodemap, NODEMAP_PROJID,
 854                                                NODEMAP_FS_TO_CLIENT,
 855                                                attr->la_projid);
 856                 b->mbo_valid |= OBD_MD_FLPROJID;
 857         }
 858
 859         b->mbo_mode = attr->la_mode;
 860         if (attr->la_valid & LA_MODE)
 861                 b->mbo_valid |= OBD_MD_FLMODE;
 862         if (attr->la_valid & LA_TYPE)
 863                 b->mbo_valid |= OBD_MD_FLTYPE;
 864
 865         if (fid != NULL) {
 866                 b->mbo_fid1 = *fid;
 867                 b->mbo_valid |= OBD_MD_FLID;
 868                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid=%#llx\n",
 869                        PFID(fid), b->mbo_nlink, b->mbo_mode, b->mbo_valid);
 870         }
 871
 872         if (!(attr->la_valid & LA_TYPE))
 873                 return;
 874
 875         b->mbo_rdev   = attr->la_rdev;
 876         b->mbo_size   = attr->la_size;
 877         b->mbo_blocks = attr->la_blocks;
 878
 879         if (!S_ISREG(attr->la_mode)) {
 880                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 881         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
 882                 /* means no objects are allocated on osts. */
 883                 LASSERT(!(ma->ma_valid & MA_LOV));
 884                 /* just ignore blocks occupied by extend attributes on MDS */
 885                 b->mbo_blocks = 0;
 886                 /* if no object is allocated on osts, the size on mds is valid.
 887                  * b=22272 */
 888                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 889         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL) {
 890                 if (mdt_hsm_is_released(ma->ma_lmm)) {
 891                         /* A released file stores its size on MDS. */
 892                         /* But return 1 block for released file, unless tools
 893                          * like tar will consider it fully sparse. (LU-3864)
 894                          */
 895                         if (unlikely(b->mbo_size == 0))
 896                                 b->mbo_blocks = 0;
 897                         else
 898                                 b->mbo_blocks = 1;
 899                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 900                 } else if (info->mti_som_valid) { /* som is valid */
 901                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 902                 } else if (ma->ma_valid & MA_SOM) { /* lsom is valid */
 903                         b->mbo_valid |= OBD_MD_FLLAZYSIZE | OBD_MD_FLLAZYBLOCKS;
 904                         b->mbo_size = ma->ma_som.ms_size;
 905                         b->mbo_blocks = ma->ma_som.ms_blocks;
 906                 }
 907         }
 908
 909         if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE ||
 910                             b->mbo_valid & OBD_MD_FLLAZYSIZE))
 911                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 912                        PFID(fid), (unsigned long long)b->mbo_size);
 913
 914 out:
 915         if (!IS_ERR_OR_NULL(nodemap))
 916                 nodemap_putref(nodemap);
 917 }
 918
 919 static inline int mdt_body_has_lov(const struct lu_attr *la,
 920                                    const struct mdt_body *body)
 921 {
 922         return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
 923                (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
 924 }
 925
 926 void mdt_client_compatibility(struct mdt_thread_info *info)
 927 {
 928         struct mdt_body       *body;
 929         struct ptlrpc_request *req = mdt_info_req(info);
 930         struct obd_export     *exp = req->rq_export;
 931         struct md_attr        *ma = &info->mti_attr;
 932         struct lu_attr        *la = &ma->ma_attr;
 933         ENTRY;
 934
 935         if (exp_connect_layout(exp))
 936                 /* the client can deal with 16-bit lmm_stripe_count */
 937                 RETURN_EXIT;
 938
 939         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 940
 941         if (!mdt_body_has_lov(la, body))
 942                 RETURN_EXIT;
 943
 944         /* now we have a reply with a lov for a client not compatible with the
 945          * layout lock so we have to clean the layout generation number */
 946         if (S_ISREG(la->la_mode))
 947                 ma->ma_lmm->lmm_layout_gen = 0;
 948         EXIT;
 949 }
 950
 951 static int mdt_attr_get_eabuf_size(struct mdt_thread_info *info,
 952                                    struct mdt_object *o)
 953 {
 954         const struct lu_env *env = info->mti_env;
 955         int rc, rc2;
 956
 957         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 958                           XATTR_NAME_LOV);
 959
 960         if (rc == -ENODATA)
 961                 rc = 0;
 962
 963         if (rc < 0)
 964                 goto out;
 965
 966         /* Is it a directory? Let's check for the LMV as well */
 967         if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
 968                 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 969                                    XATTR_NAME_LMV);
 970
 971                 if (rc2 == -ENODATA)
 972                         rc2 = mo_xattr_get(env, mdt_object_child(o),
 973                                            &LU_BUF_NULL,
 974                                            XATTR_NAME_DEFAULT_LMV);
 975
 976                 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
 977                         rc = rc2;
 978         }
 979
 980 out:
 981         return rc;
 982 }
 983
 984 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
 985                       const char *name)
 986 {
 987         const struct lu_env *env = info->mti_env;
 988         int rc;
 989         ENTRY;
 990
 991         LASSERT(info->mti_big_lmm_used == 0);
 992         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
 993         if (rc < 0)
 994                 RETURN(rc);
 995
 996         /* big_lmm may need to be grown */
 997         if (info->mti_big_lmmsize < rc) {
 998                 int size = size_roundup_power2(rc);
 999
1000                 if (info->mti_big_lmmsize > 0) {
1001                         /* free old buffer */
1002                         LASSERT(info->mti_big_lmm);
1003                         OBD_FREE_LARGE(info->mti_big_lmm,
1004                                        info->mti_big_lmmsize);
1005                         info->mti_big_lmm = NULL;
1006                         info->mti_big_lmmsize = 0;
1007                 }
1008
1009                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
1010                 if (info->mti_big_lmm == NULL)
1011                         RETURN(-ENOMEM);
1012                 info->mti_big_lmmsize = size;
1013         }
1014         LASSERT(info->mti_big_lmmsize >= rc);
1015
1016         info->mti_buf.lb_buf = info->mti_big_lmm;
1017         info->mti_buf.lb_len = info->mti_big_lmmsize;
1018         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
1019
1020         RETURN(rc);
1021 }
1022
1023 int __mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
1024                      struct md_attr *ma, const char *name)
1025 {
1026         struct md_object *next = mdt_object_child(o);
1027         struct lu_buf    *buf = &info->mti_buf;
1028         int rc;
1029
1030         if (strcmp(name, XATTR_NAME_LOV) == 0) {
1031                 buf->lb_buf = ma->ma_lmm;
1032                 buf->lb_len = ma->ma_lmm_size;
1033                 LASSERT(!(ma->ma_valid & MA_LOV));
1034         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
1035                 buf->lb_buf = ma->ma_lmv;
1036                 buf->lb_len = ma->ma_lmv_size;
1037                 LASSERT(!(ma->ma_valid & MA_LMV));
1038         } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
1039                 buf->lb_buf = ma->ma_default_lmv;
1040                 buf->lb_len = ma->ma_default_lmv_size;
1041                 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
1042         } else {
1043                 return -EINVAL;
1044         }
1045
1046         LASSERT(buf->lb_buf);
1047
1048         rc = mo_xattr_get(info->mti_env, next, buf, name);
1049         if (rc > 0) {
1050
1051 got:
1052                 if (strcmp(name, XATTR_NAME_LOV) == 0) {
1053                         if (info->mti_big_lmm_used)
1054                                 ma->ma_lmm = info->mti_big_lmm;
1055
1056                         /* NOT return LOV EA with hole to old client. */
1057                         if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
1058                                      LOV_PATTERN_F_HOLE) &&
1059                             !(exp_connect_flags(info->mti_exp) &
1060                               OBD_CONNECT_LFSCK)) {
1061                                 return -EIO;
1062                         } else {
1063                                 ma->ma_lmm_size = rc;
1064                                 ma->ma_valid |= MA_LOV;
1065                         }
1066                 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
1067                         if (info->mti_big_lmm_used)
1068                                 ma->ma_lmv = info->mti_big_lmm;
1069
1070                         ma->ma_lmv_size = rc;
1071                         ma->ma_valid |= MA_LMV;
1072                 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
1073                         ma->ma_default_lmv_size = rc;
1074                         ma->ma_valid |= MA_LMV_DEF;
1075                 }
1076
1077                 /* Update mdt_max_mdsize so all clients will be aware that */
1078                 if (info->mti_mdt->mdt_max_mdsize < rc)
1079                         info->mti_mdt->mdt_max_mdsize = rc;
1080
1081                 rc = 0;
1082         } else if (rc == -ENODATA) {
1083                 /* no LOV EA */
1084                 rc = 0;
1085         } else if (rc == -ERANGE) {
1086                 /* Default LMV has fixed size, so it must be able to fit
1087                  * in the original buffer */
1088                 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
1089                         return rc;
1090                 rc = mdt_big_xattr_get(info, o, name);
1091                 if (rc > 0) {
1092                         info->mti_big_lmm_used = 1;
1093                         goto got;
1094                 }
1095         }
1096
1097         return rc;
1098 }
1099
1100 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
1101                    struct md_attr *ma, const char *name)
1102 {
1103         int rc;
1104
1105         if (!info->mti_big_lmm) {
1106                 OBD_ALLOC(info->mti_big_lmm, PAGE_SIZE);
1107                 if (!info->mti_big_lmm)
1108                         return -ENOMEM;
1109                 info->mti_big_lmmsize = PAGE_SIZE;
1110         }
1111
1112         if (strcmp(name, XATTR_NAME_LOV) == 0) {
1113                 ma->ma_lmm = info->mti_big_lmm;
1114                 ma->ma_lmm_size = info->mti_big_lmmsize;
1115                 ma->ma_valid &= ~MA_LOV;
1116         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
1117                 ma->ma_lmv = info->mti_big_lmm;
1118                 ma->ma_lmv_size = info->mti_big_lmmsize;
1119                 ma->ma_valid &= ~MA_LMV;
1120         } else {
1121                 LBUG();
1122         }
1123
1124         LASSERT(!info->mti_big_lmm_used);
1125         rc = __mdt_stripe_get(info, o, ma, name);
1126         /* since big_lmm is always used here, clear 'used' flag to avoid
1127          * assertion in mdt_big_xattr_get().
1128          */
1129         info->mti_big_lmm_used = 0;
1130
1131         return rc;
1132 }
1133
1134 int mdt_attr_get_pfid(struct mdt_thread_info *info, struct mdt_object *o,
1135                       struct lu_fid *pfid)
1136 {
1137         struct lu_buf           *buf = &info->mti_buf;
1138         struct link_ea_header   *leh;
1139         struct link_ea_entry    *lee;
1140         int                      rc;
1141         ENTRY;
1142
1143         buf->lb_buf = info->mti_big_lmm;
1144         buf->lb_len = info->mti_big_lmmsize;
1145         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
1146                           buf, XATTR_NAME_LINK);
1147         /* ignore errors, MA_PFID won't be set and it is
1148          * up to the caller to treat this as an error */
1149         if (rc == -ERANGE || buf->lb_len == 0) {
1150                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
1151                 buf->lb_buf = info->mti_big_lmm;
1152                 buf->lb_len = info->mti_big_lmmsize;
1153         }
1154
1155         if (rc < 0)
1156                 RETURN(rc);
1157         if (rc < sizeof(*leh)) {
1158                 CERROR("short LinkEA on "DFID": rc = %d\n",
1159                        PFID(mdt_object_fid(o)), rc);
1160                 RETURN(-ENODATA);
1161         }
1162
1163         leh = (struct link_ea_header *) buf->lb_buf;
1164         lee = (struct link_ea_entry *)(leh + 1);
1165         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
1166                 leh->leh_magic = LINK_EA_MAGIC;
1167                 leh->leh_reccount = __swab32(leh->leh_reccount);
1168                 leh->leh_len = __swab64(leh->leh_len);
1169         }
1170         if (leh->leh_magic != LINK_EA_MAGIC)
1171                 RETURN(-EINVAL);
1172         if (leh->leh_reccount == 0)
1173                 RETURN(-ENODATA);
1174
1175         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
1176         fid_be_to_cpu(pfid, pfid);
1177
1178         RETURN(0);
1179 }
1180
1181 int mdt_attr_get_pfid_name(struct mdt_thread_info *info, struct mdt_object *o,
1182                            struct lu_fid *pfid, struct lu_name *lname)
1183 {
1184         struct lu_buf *buf = &info->mti_buf;
1185         struct link_ea_header *leh;
1186         struct link_ea_entry *lee;
1187         int reclen;
1188         int rc;
1189
1190         buf->lb_buf = info->mti_xattr_buf;
1191         buf->lb_len = sizeof(info->mti_xattr_buf);
1192         rc = mo_xattr_get(info->mti_env, mdt_object_child(o), buf,
1193                           XATTR_NAME_LINK);
1194         if (rc == -ERANGE) {
1195                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
1196                 buf->lb_buf = info->mti_big_lmm;
1197                 buf->lb_len = info->mti_big_lmmsize;
1198         }
1199         if (rc < 0)
1200                 return rc;
1201
1202         if (rc < sizeof(*leh)) {
1203                 CERROR("short LinkEA on "DFID": rc = %d\n",
1204                        PFID(mdt_object_fid(o)), rc);
1205                 return -ENODATA;
1206         }
1207
1208         leh = (struct link_ea_header *)buf->lb_buf;
1209         lee = (struct link_ea_entry *)(leh + 1);
1210         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
1211                 leh->leh_magic = LINK_EA_MAGIC;
1212                 leh->leh_reccount = __swab32(leh->leh_reccount);
1213                 leh->leh_len = __swab64(leh->leh_len);
1214         }
1215         if (leh->leh_magic != LINK_EA_MAGIC)
1216                 return -EINVAL;
1217
1218         if (leh->leh_reccount == 0)
1219                 return -ENODATA;
1220
1221         linkea_entry_unpack(lee, &reclen, lname, pfid);
1222
1223         return 0;
1224 }
1225
1226 int mdt_attr_get_complex(struct mdt_thread_info *info,
1227                          struct mdt_object *o, struct md_attr *ma)
1228 {
1229         const struct lu_env *env = info->mti_env;
1230         struct md_object    *next = mdt_object_child(o);
1231         struct lu_buf       *buf = &info->mti_buf;
1232         int                  need = ma->ma_need;
1233         int                  rc = 0, rc2;
1234         u32                  mode;
1235         ENTRY;
1236
1237         ma->ma_valid = 0;
1238
1239         if (mdt_object_exists(o) == 0)
1240                 GOTO(out, rc = -ENOENT);
1241         mode = lu_object_attr(&next->mo_lu);
1242
1243         if (need & MA_INODE) {
1244                 ma->ma_need = MA_INODE;
1245                 rc = mo_attr_get(env, next, ma);
1246                 if (rc)
1247                         GOTO(out, rc);
1248
1249                 if (S_ISREG(mode))
1250                         (void) mdt_get_som(info, o, ma);
1251                 ma->ma_valid |= MA_INODE;
1252         }
1253
1254         if (need & MA_PFID) {
1255                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
1256                 if (rc == 0)
1257                         ma->ma_valid |= MA_PFID;
1258                 /* ignore this error, parent fid is not mandatory */
1259                 rc = 0;
1260         }
1261
1262         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
1263                 rc = __mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
1264                 if (rc)
1265                         GOTO(out, rc);
1266         }
1267
1268         if (need & MA_LMV && S_ISDIR(mode)) {
1269                 rc = __mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
1270                 if (rc != 0)
1271                         GOTO(out, rc);
1272         }
1273
1274         if (need & MA_LMV_DEF && S_ISDIR(mode)) {
1275                 rc = __mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
1276                 if (rc != 0)
1277                         GOTO(out, rc);
1278         }
1279
1280         /*
1281          * In the handle of MA_INODE, we may already get the SOM attr.
1282          */
1283         if (need & MA_SOM && S_ISREG(mode) && !(ma->ma_valid & MA_SOM)) {
1284                 rc = mdt_get_som(info, o, ma);
1285                 if (rc != 0)
1286                         GOTO(out, rc);
1287         }
1288
1289         if (need & MA_HSM && S_ISREG(mode)) {
1290                 buf->lb_buf = info->mti_xattr_buf;
1291                 buf->lb_len = sizeof(info->mti_xattr_buf);
1292                 BUILD_BUG_ON(sizeof(struct hsm_attrs) >
1293                              sizeof(info->mti_xattr_buf));
1294                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
1295                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
1296                 if (rc2 == 0)
1297                         ma->ma_valid |= MA_HSM;
1298                 else if (rc2 < 0 && rc2 != -ENODATA)
1299                         GOTO(out, rc = rc2);
1300         }
1301
1302 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1303         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
1304                 buf->lb_buf = ma->ma_acl;
1305                 buf->lb_len = ma->ma_acl_size;
1306                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
1307                 if (rc2 > 0) {
1308                         ma->ma_acl_size = rc2;
1309                         ma->ma_valid |= MA_ACL_DEF;
1310                 } else if (rc2 == -ENODATA) {
1311                         /* no ACLs */
1312                         ma->ma_acl_size = 0;
1313                 } else
1314                         GOTO(out, rc = rc2);
1315         }
1316 #endif
1317 out:
1318         ma->ma_need = need;
1319         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = %#llx ma_lmm=%p\n",
1320                rc, ma->ma_valid, ma->ma_lmm);
1321         RETURN(rc);
1322 }
1323
1324 static int mdt_getattr_internal(struct mdt_thread_info *info,
1325                                 struct mdt_object *o, int ma_need)
1326 {
1327         struct mdt_device *mdt = info->mti_mdt;
1328         struct md_object *next = mdt_object_child(o);
1329         const struct mdt_body *reqbody = info->mti_body;
1330         struct ptlrpc_request *req = mdt_info_req(info);
1331         struct md_attr *ma = &info->mti_attr;
1332         struct lu_attr *la = &ma->ma_attr;
1333         struct req_capsule *pill = info->mti_pill;
1334         const struct lu_env *env = info->mti_env;
1335         struct mdt_body *repbody;
1336         struct lu_buf *buffer = &info->mti_buf;
1337         struct obd_export *exp = info->mti_exp;
1338         ktime_t kstart = ktime_get();
1339         int rc;
1340
1341         ENTRY;
1342
1343         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
1344                 RETURN(err_serious(-ENOMEM));
1345
1346         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1347
1348         ma->ma_valid = 0;
1349
1350         if (mdt_object_remote(o)) {
1351                 /* This object is located on remote node.*/
1352                 /* Return -ENOTSUPP for old client */
1353                 if (!mdt_is_dne_client(req->rq_export))
1354                         GOTO(out, rc = -ENOTSUPP);
1355
1356                 repbody->mbo_fid1 = *mdt_object_fid(o);
1357                 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
1358                 GOTO(out, rc = 0);
1359         }
1360
1361         if (reqbody->mbo_eadatasize > 0) {
1362                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
1363                 if (buffer->lb_buf == NULL)
1364                         GOTO(out, rc = -EPROTO);
1365                 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
1366                                                       RCL_SERVER);
1367         } else {
1368                 buffer->lb_buf = NULL;
1369                 buffer->lb_len = 0;
1370                 ma_need &= ~(MA_LOV | MA_LMV);
1371                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
1372                        mdt_obd_name(info->mti_mdt),
1373                        req->rq_export->exp_client_uuid.uuid);
1374         }
1375
1376         /* from 2.12.58 intent_getattr pack default LMV in reply */
1377         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1378             ((reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA)) ==
1379                     (OBD_MD_MEA | OBD_MD_DEFAULT_MEA)) &&
1380             req_capsule_has_field(&req->rq_pill, &RMF_DEFAULT_MDT_MD,
1381                                   RCL_SERVER)) {
1382                 ma->ma_lmv = buffer->lb_buf;
1383                 ma->ma_lmv_size = buffer->lb_len;
1384                 ma->ma_default_lmv = req_capsule_server_get(pill,
1385                                                 &RMF_DEFAULT_MDT_MD);
1386                 ma->ma_default_lmv_size = req_capsule_get_size(pill,
1387                                                 &RMF_DEFAULT_MDT_MD,
1388                                                 RCL_SERVER);
1389                 ma->ma_need = MA_INODE;
1390                 if (ma->ma_lmv_size > 0)
1391                         ma->ma_need |= MA_LMV;
1392                 if (ma->ma_default_lmv_size > 0)
1393                         ma->ma_need |= MA_LMV_DEF;
1394         } else if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1395                    (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
1396                 /* If it is dir and client require MEA, then we got MEA */
1397                 /* Assumption: MDT_MD size is enough for lmv size. */
1398                 ma->ma_lmv = buffer->lb_buf;
1399                 ma->ma_lmv_size = buffer->lb_len;
1400                 ma->ma_need = MA_INODE;
1401                 if (ma->ma_lmv_size > 0) {
1402                         if (reqbody->mbo_valid & OBD_MD_MEA) {
1403                                 ma->ma_need |= MA_LMV;
1404                         } else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA) {
1405                                 ma->ma_need |= MA_LMV_DEF;
1406                                 ma->ma_default_lmv = buffer->lb_buf;
1407                                 ma->ma_lmv = NULL;
1408                                 ma->ma_default_lmv_size = buffer->lb_len;
1409                                 ma->ma_lmv_size = 0;
1410                         }
1411                 }
1412         } else {
1413                 ma->ma_lmm = buffer->lb_buf;
1414                 ma->ma_lmm_size = buffer->lb_len;
1415                 ma->ma_need = MA_INODE | MA_HSM;
1416                 if (ma->ma_lmm_size > 0) {
1417                         ma->ma_need |= MA_LOV;
1418                         /* Older clients may crash if they getattr overstriped
1419                          * files
1420                          */
1421                         if (!exp_connect_overstriping(exp) &&
1422                             mdt_lmm_is_overstriping(ma->ma_lmm))
1423                                 RETURN(-EOPNOTSUPP);
1424                 }
1425         }
1426
1427         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1428             reqbody->mbo_valid & OBD_MD_FLDIREA  &&
1429             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
1430                 /* get default stripe info for this dir. */
1431                 ma->ma_need |= MA_LOV_DEF;
1432         }
1433         ma->ma_need |= ma_need;
1434
1435         rc = mdt_attr_get_complex(info, o, ma);
1436         if (unlikely(rc)) {
1437                 CDEBUG_LIMIT(rc == -ENOENT ? D_OTHER : D_ERROR,
1438                              "%s: getattr error for "DFID": rc = %d\n",
1439                              mdt_obd_name(info->mti_mdt),
1440                              PFID(mdt_object_fid(o)), rc);
1441                 RETURN(rc);
1442         }
1443
1444         /* if file is released, check if a restore is running */
1445         if (ma->ma_valid & MA_HSM) {
1446                 repbody->mbo_valid |= OBD_MD_TSTATE;
1447                 if ((ma->ma_hsm.mh_flags & HS_RELEASED) &&
1448                     mdt_hsm_restore_is_running(info, mdt_object_fid(o)))
1449                         repbody->mbo_t_state = MS_RESTORE;
1450         }
1451
1452         if (unlikely(!(ma->ma_valid & MA_INODE)))
1453                 RETURN(-EFAULT);
1454
1455         mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
1456
1457         if (mdt_body_has_lov(la, reqbody)) {
1458                 u32 stripe_count = 1;
1459                 bool fixed_layout = false;
1460
1461                 if (ma->ma_valid & MA_LOV) {
1462                         LASSERT(ma->ma_lmm_size);
1463                         repbody->mbo_eadatasize = ma->ma_lmm_size;
1464                         if (S_ISDIR(la->la_mode))
1465                                 repbody->mbo_valid |= OBD_MD_FLDIREA;
1466                         else
1467                                 repbody->mbo_valid |= OBD_MD_FLEASIZE;
1468                         mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
1469                 }
1470                 if (ma->ma_valid & MA_LMV) {
1471                         struct lmv_mds_md_v1 *lmv = &ma->ma_lmv->lmv_md_v1;
1472                         u32 magic = le32_to_cpu(lmv->lmv_magic);
1473
1474                         /* Return -ENOTSUPP for old client */
1475                         if (!mdt_is_striped_client(req->rq_export))
1476                                 RETURN(-ENOTSUPP);
1477
1478                         LASSERT(S_ISDIR(la->la_mode));
1479                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1480                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1481                         repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
1482
1483                         stripe_count = le32_to_cpu(lmv->lmv_stripe_count);
1484                         fixed_layout = lmv_is_fixed(lmv);
1485                         if (magic == LMV_MAGIC_STRIPE && lmv_is_restriping(lmv))
1486                                 mdt_restripe_migrate_add(info, o);
1487                         else if (magic == LMV_MAGIC_V1 &&
1488                                  lmv_is_restriping(lmv))
1489                                 mdt_restripe_update_add(info, o);
1490                 }
1491                 if (ma->ma_valid & MA_LMV_DEF) {
1492                         /* Return -ENOTSUPP for old client */
1493                         if (!mdt_is_striped_client(req->rq_export))
1494                                 RETURN(-ENOTSUPP);
1495                         LASSERT(S_ISDIR(la->la_mode));
1496                         /*
1497                          * when ll_dir_getstripe() gets default LMV, it
1498                          * checks mbo_eadatasize.
1499                          */
1500                         if (!(ma->ma_valid & MA_LMV))
1501                                 repbody->mbo_eadatasize =
1502                                         ma->ma_default_lmv_size;
1503                         repbody->mbo_valid |= (OBD_MD_FLDIREA |
1504                                                OBD_MD_DEFAULT_MEA);
1505                 }
1506                 CDEBUG(D_VFSTRACE,
1507                        "dirent count %llu stripe count %u MDT count %d\n",
1508                        ma->ma_attr.la_dirent_count, stripe_count,
1509                        atomic_read(&mdt->mdt_mds_mds_conns) + 1);
1510                 if (ma->ma_attr.la_dirent_count != LU_DIRENT_COUNT_UNSET &&
1511                     ma->ma_attr.la_dirent_count >
1512                         mdt->mdt_restriper.mdr_dir_split_count &&
1513                     !fid_is_root(mdt_object_fid(o)) &&
1514                     mdt->mdt_enable_dir_auto_split &&
1515                     !o->mot_restriping &&
1516                     stripe_count < atomic_read(&mdt->mdt_mds_mds_conns) + 1 &&
1517                     !fixed_layout)
1518                         mdt_auto_split_add(info, o);
1519         } else if (S_ISLNK(la->la_mode) &&
1520                    reqbody->mbo_valid & OBD_MD_LINKNAME) {
1521                 buffer->lb_buf = ma->ma_lmm;
1522                 /* eadatasize from client includes NULL-terminator, so
1523                  * there is no need to read it */
1524                 buffer->lb_len = reqbody->mbo_eadatasize - 1;
1525                 rc = mo_readlink(env, next, buffer);
1526                 if (unlikely(rc <= 0)) {
1527                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
1528                                mdt_obd_name(info->mti_mdt),
1529                                PFID(mdt_object_fid(o)), rc);
1530                         rc = -EFAULT;
1531                 } else {
1532                         int print_limit = min_t(int, PAGE_SIZE - 128, rc);
1533
1534                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
1535                                 rc -= 2;
1536                         repbody->mbo_valid |= OBD_MD_LINKNAME;
1537                         /* we need to report back size with NULL-terminator
1538                          * because client expects that */
1539                         repbody->mbo_eadatasize = rc + 1;
1540                         if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
1541                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
1542                                        "on "DFID ", expected %d\n",
1543                                        mdt_obd_name(info->mti_mdt),
1544                                        rc, PFID(mdt_object_fid(o)),
1545                                        reqbody->mbo_eadatasize - 1);
1546                         /* NULL terminate */
1547                         ((char *)ma->ma_lmm)[rc] = 0;
1548
1549                         /* If the total CDEBUG() size is larger than a page, it
1550                          * will print a warning to the console, avoid this by
1551                          * printing just the last part of the symlink. */
1552                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
1553                                print_limit < rc ? "..." : "", print_limit,
1554                                (char *)ma->ma_lmm + rc - print_limit, rc);
1555                         rc = 0;
1556                 }
1557         }
1558
1559         if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
1560                 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1561                 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1562                 CDEBUG(D_INODE, "changing the max MD size to %u\n",
1563                        repbody->mbo_max_mdsize);
1564         }
1565
1566 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1567         if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1568                  (reqbody->mbo_valid & OBD_MD_FLACL)) {
1569                 struct lu_nodemap *nodemap = nodemap_get_from_exp(exp);
1570                 if (IS_ERR(nodemap))
1571                         RETURN(PTR_ERR(nodemap));
1572
1573                 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
1574                 nodemap_putref(nodemap);
1575         }
1576 #endif
1577
1578 out:
1579         if (rc == 0)
1580                 mdt_counter_incr(req, LPROC_MDT_GETATTR,
1581                                  ktime_us_delta(ktime_get(), kstart));
1582
1583         RETURN(rc);
1584 }
1585
1586 static int mdt_getattr(struct tgt_session_info *tsi)
1587 {
1588         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1589         struct mdt_object       *obj = info->mti_object;
1590         struct req_capsule      *pill = info->mti_pill;
1591         struct mdt_body         *reqbody;
1592         struct mdt_body         *repbody;
1593         int rc, rc2;
1594         ENTRY;
1595
1596         if (unlikely(info->mti_object == NULL))
1597                 RETURN(-EPROTO);
1598
1599         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1600         LASSERT(reqbody);
1601         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1602
1603         /* Special case for Data-on-MDT files to get data version */
1604         if (unlikely(reqbody->mbo_valid & OBD_MD_FLDATAVERSION)) {
1605                 rc = mdt_data_version_get(tsi);
1606                 GOTO(out, rc);
1607         }
1608
1609         /* Unlike intent case where we need to pre-fill out buffers early on
1610          * in intent policy for ldlm reasons, here we can have a much better
1611          * guess at EA size by just reading it from disk.
1612          * Exceptions are readdir and (missing) directory striping */
1613         /* Readlink */
1614         if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1615                 /* No easy way to know how long is the symlink, but it cannot
1616                  * be more than PATH_MAX, so we allocate +1 */
1617                 rc = PATH_MAX + 1;
1618         /* A special case for fs ROOT: getattr there might fetch
1619          * default EA for entire fs, not just for this dir!
1620          */
1621         } else if (lu_fid_eq(mdt_object_fid(obj),
1622                              &info->mti_mdt->mdt_md_root_fid) &&
1623                    (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1624                    (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1625                                                                  MDS_GETATTR)) {
1626                 /* Should the default strping be bigger, mdt_fix_reply
1627                  * will reallocate */
1628                 rc = DEF_REP_MD_SIZE;
1629         } else {
1630                 /* Read the actual EA size from disk */
1631                 rc = mdt_attr_get_eabuf_size(info, obj);
1632         }
1633
1634         if (rc < 0)
1635                 GOTO(out, rc = err_serious(rc));
1636
1637         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1638
1639         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
1640          * by default. If the target object has more ACL entries, then
1641          * enlarge the buffer when necessary. */
1642         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
1643                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
1644
1645         rc = req_capsule_server_pack(pill);
1646         if (unlikely(rc != 0))
1647                 GOTO(out, rc = err_serious(rc));
1648
1649         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1650         LASSERT(repbody != NULL);
1651         repbody->mbo_eadatasize = 0;
1652         repbody->mbo_aclsize = 0;
1653
1654         rc = mdt_check_ucred(info);
1655         if (unlikely(rc))
1656                 GOTO(out_shrink, rc);
1657
1658         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1659
1660         rc = mdt_getattr_internal(info, obj, 0);
1661         EXIT;
1662 out_shrink:
1663         mdt_client_compatibility(info);
1664         rc2 = mdt_fix_reply(info);
1665         if (rc == 0)
1666                 rc = rc2;
1667 out:
1668         mdt_thread_info_fini(info);
1669         return rc;
1670 }
1671
1672 /**
1673  * Handler of layout intent RPC requiring the layout modification
1674  *
1675  * \param[in]  info     thread environment
1676  * \param[in]  obj      object
1677  * \param[out] lhc      object ldlm lock handle
1678  * \param[in]  layout   layout change descriptor
1679  *
1680  * \retval 0    on success
1681  * \retval < 0  error code
1682  */
1683 int mdt_layout_change(struct mdt_thread_info *info, struct mdt_object *obj,
1684                       struct mdt_lock_handle *lhc,
1685                       struct md_layout_change *layout)
1686 {
1687         int rc;
1688
1689         ENTRY;
1690
1691         if (!mdt_object_exists(obj))
1692                 RETURN(-ENOENT);
1693
1694         if (!S_ISREG(lu_object_attr(&obj->mot_obj)))
1695                 RETURN(-EINVAL);
1696
1697         rc = mo_permission(info->mti_env, NULL, mdt_object_child(obj), NULL,
1698                            MAY_WRITE);
1699         if (rc)
1700                 RETURN(rc);
1701
1702         rc = mdt_check_resent_lock(info, obj, lhc);
1703         if (rc < 0)
1704                 RETURN(rc);
1705
1706         if (rc > 0) {
1707                 /* not resent */
1708                 __u64 lockpart = MDS_INODELOCK_LAYOUT;
1709
1710                 /* take layout lock to prepare layout change */
1711                 if (layout->mlc_opc == MD_LAYOUT_WRITE)
1712                         lockpart |= MDS_INODELOCK_UPDATE;
1713
1714                 mdt_lock_handle_init(lhc);
1715                 mdt_lock_reg_init(lhc, LCK_EX);
1716                 rc = mdt_reint_object_lock(info, obj, lhc, lockpart, false);
1717                 if (rc)
1718                         RETURN(rc);
1719         }
1720
1721         mutex_lock(&obj->mot_som_mutex);
1722         rc = mo_layout_change(info->mti_env, mdt_object_child(obj), layout);
1723         mutex_unlock(&obj->mot_som_mutex);
1724
1725         if (rc)
1726                 mdt_object_unlock(info, obj, lhc, 1);
1727
1728         RETURN(rc);
1729 }
1730
1731 /**
1732  * Exchange MOF_LOV_CREATED flags between two objects after a
1733  * layout swap. No assumption is made on whether o1 or o2 have
1734  * created objects or not.
1735  *
1736  * \param[in,out] o1    First swap layout object
1737  * \param[in,out] o2    Second swap layout object
1738  */
1739 static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2)
1740 {
1741         unsigned int o1_lov_created = o1->mot_lov_created;
1742
1743         mutex_lock(&o1->mot_lov_mutex);
1744         mutex_lock(&o2->mot_lov_mutex);
1745
1746         o1->mot_lov_created = o2->mot_lov_created;
1747         o2->mot_lov_created = o1_lov_created;
1748
1749         mutex_unlock(&o2->mot_lov_mutex);
1750         mutex_unlock(&o1->mot_lov_mutex);
1751 }
1752
1753 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1754 {
1755         struct mdt_thread_info  *info;
1756         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1757         struct obd_export       *exp = req->rq_export;
1758         struct mdt_object       *o1, *o2, *o;
1759         struct mdt_lock_handle  *lh1, *lh2;
1760         struct mdc_swap_layouts *msl;
1761         int                      rc;
1762         ENTRY;
1763
1764         /* client does not support layout lock, so layout swaping
1765          * is disabled.
1766          * FIXME: there is a problem for old clients which don't support
1767          * layout lock yet. If those clients have already opened the file
1768          * they won't be notified at all so that old layout may still be
1769          * used to do IO. This can be fixed after file release is landed by
1770          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1771         if (!exp_connect_layout(exp))
1772                 RETURN(-EOPNOTSUPP);
1773
1774         info = tsi2mdt_info(tsi);
1775         if (unlikely(info->mti_object == NULL))
1776                 RETURN(-EPROTO);
1777
1778         if (info->mti_dlm_req != NULL)
1779                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1780
1781         o1 = info->mti_object;
1782         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1783                                 &info->mti_body->mbo_fid2);
1784         if (IS_ERR(o))
1785                 GOTO(out, rc = PTR_ERR(o));
1786
1787         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1788                 GOTO(put, rc = -ENOENT);
1789
1790         rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1791         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1792                 GOTO(put, rc);
1793
1794         if (rc < 0)
1795                 swap(o1, o2);
1796
1797         /* permission check. Make sure the calling process having permission
1798          * to write both files. */
1799         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1800                            MAY_WRITE);
1801         if (rc < 0)
1802                 GOTO(put, rc);
1803
1804         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1805                            MAY_WRITE);
1806         if (rc < 0)
1807                 GOTO(put, rc);
1808
1809         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1810         if (msl == NULL)
1811                 GOTO(put, rc = -EPROTO);
1812
1813         lh1 = &info->mti_lh[MDT_LH_NEW];
1814         mdt_lock_reg_init(lh1, LCK_EX);
1815         lh2 = &info->mti_lh[MDT_LH_OLD];
1816         mdt_lock_reg_init(lh2, LCK_EX);
1817
1818         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1819                              MDS_INODELOCK_XATTR);
1820         if (rc < 0)
1821                 GOTO(put, rc);
1822
1823         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1824                              MDS_INODELOCK_XATTR);
1825         if (rc < 0)
1826                 GOTO(unlock1, rc);
1827
1828         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1829                              mdt_object_child(o2), msl->msl_flags);
1830         if (rc < 0)
1831                 GOTO(unlock2, rc);
1832
1833         mdt_swap_lov_flag(o1, o2);
1834
1835 unlock2:
1836         mdt_object_unlock(info, o2, lh2, rc);
1837 unlock1:
1838         mdt_object_unlock(info, o1, lh1, rc);
1839 put:
1840         mdt_object_put(info->mti_env, o);
1841 out:
1842         mdt_thread_info_fini(info);
1843         RETURN(rc);
1844 }
1845
1846 static int mdt_raw_lookup(struct mdt_thread_info *info,
1847                           struct mdt_object *parent,
1848                           const struct lu_name *lname)
1849 {
1850         struct lu_fid *fid = &info->mti_tmp_fid1;
1851         struct mdt_body *repbody;
1852         bool is_dotdot = false;
1853         bool is_old_parent_stripe = false;
1854         bool is_new_parent_checked = false;
1855         int rc;
1856
1857         ENTRY;
1858
1859         LASSERT(!info->mti_cross_ref);
1860         /* Always allow to lookup ".." */
1861         if (lname->ln_namelen == 2 &&
1862             lname->ln_name[0] == '.' && lname->ln_name[1] == '.') {
1863                 info->mti_spec.sp_permitted = 1;
1864                 is_dotdot = true;
1865                 if (mdt_is_dir_stripe(info, parent) == 1)
1866                         is_old_parent_stripe = true;
1867         }
1868
1869         mdt_object_get(info->mti_env, parent);
1870 lookup:
1871         /* Only got the fid of this obj by name */
1872         fid_zero(fid);
1873         rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname, fid,
1874                         &info->mti_spec);
1875         mdt_object_put(info->mti_env, parent);
1876         if (rc)
1877                 RETURN(rc);
1878
1879         /* getattr_name("..") should return master object FID for striped dir */
1880         if (is_dotdot && (is_old_parent_stripe || !is_new_parent_checked)) {
1881                 parent = mdt_object_find(info->mti_env, info->mti_mdt, fid);
1882                 if (IS_ERR(parent))
1883                         RETURN(PTR_ERR(parent));
1884
1885                 /* old client getattr_name("..") with stripe FID */
1886                 if (unlikely(is_old_parent_stripe)) {
1887                         is_old_parent_stripe = false;
1888                         goto lookup;
1889                 }
1890
1891                 /* ".." may be a stripe */
1892                 if (unlikely(mdt_is_dir_stripe(info, parent) == 1)) {
1893                         is_new_parent_checked = true;
1894                         goto lookup;
1895                 }
1896
1897                 mdt_object_put(info->mti_env, parent);
1898         }
1899
1900         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1901         repbody->mbo_fid1 = *fid;
1902         repbody->mbo_valid = OBD_MD_FLID;
1903
1904         RETURN(rc);
1905 }
1906
1907 /**
1908  * Find name matching hash
1909  *
1910  * We search \a child LinkEA for a name whose hash matches \a lname
1911  * (it contains an encoded hash).
1912  *
1913  * \param info mdt thread info
1914  * \param lname encoded hash to find
1915  * \param parent parent object
1916  * \param child object to search with LinkEA
1917  * \param force_check true to check hash even if LinkEA has only one entry
1918  *
1919  * \retval 1 match found
1920  * \retval 0 no match found
1921  * \retval -ev negative errno upon error
1922  */
1923 int find_name_matching_hash(struct mdt_thread_info *info, struct lu_name *lname,
1924                             struct mdt_object *parent, struct mdt_object *child,
1925                             bool force_check)
1926 {
1927         /* Here, lname is an encoded hash of on-disk name, and
1928          * client is doing access without encryption key.
1929          * So we need to get LinkEA, check parent fid is correct and
1930          * compare name hash with the one in the request.
1931          */
1932         struct lu_buf *buf = &info->mti_big_buf;
1933         struct lu_name name;
1934         struct lu_fid pfid;
1935         struct linkea_data ldata = { NULL };
1936         struct link_ea_header *leh;
1937         struct link_ea_entry *lee;
1938         struct lu_buf link = { 0 };
1939         char *hash = NULL;
1940         int reclen, count, rc;
1941
1942         ENTRY;
1943
1944         if (lname->ln_namelen < LLCRYPT_FNAME_DIGEST_SIZE)
1945                 RETURN(-EINVAL);
1946
1947         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
1948         if (!buf->lb_buf)
1949                 RETURN(-ENOMEM);
1950
1951         ldata.ld_buf = buf;
1952         rc = mdt_links_read(info, child, &ldata);
1953         if (rc < 0)
1954                 RETURN(rc);
1955
1956         leh = buf->lb_buf;
1957         if (force_check || leh->leh_reccount > 1) {
1958                 hash = kmalloc(lname->ln_namelen, GFP_NOFS);
1959                 if (!hash)
1960                         RETURN(-ENOMEM);
1961                 rc = critical_decode(lname->ln_name, lname->ln_namelen, hash);
1962         }
1963         lee = (struct link_ea_entry *)(leh + 1);
1964         for (count = 0; count < leh->leh_reccount; count++) {
1965                 linkea_entry_unpack(lee, &reclen, &name, &pfid);
1966                 if (!force_check && leh->leh_reccount == 1) {
1967                         /* if there is only one rec, it has to be it */
1968                         *lname = name;
1969                         break;
1970                 }
1971                 if (!parent || lu_fid_eq(&pfid, mdt_object_fid(parent))) {
1972                         lu_buf_check_and_alloc(&link, name.ln_namelen);
1973                         if (!link.lb_buf)
1974                                 GOTO(out_match, rc = -ENOMEM);
1975                         rc = critical_decode(name.ln_name, name.ln_namelen,
1976                                              link.lb_buf);
1977
1978                         if (memcmp(LLCRYPT_FNAME_DIGEST(link.lb_buf, rc),
1979                                    hash, LLCRYPT_FNAME_DIGEST_SIZE) == 0) {
1980                                 *lname = name;
1981                                 break;
1982                         }
1983                 }
1984                 lee = (struct link_ea_entry *) ((char *)lee + reclen);
1985         }
1986         if (count == leh->leh_reccount)
1987                 rc = 0;
1988         else
1989                 rc = 1;
1990
1991 out_match:
1992         lu_buf_free(&link);
1993         kfree(hash);
1994
1995         RETURN(rc);
1996 }
1997
1998 /*
1999  * UPDATE lock should be taken against parent, and be released before exit;
2000  * child_bits lock should be taken against child, and be returned back:
2001  *            (1)normal request should release the child lock;
2002  *            (2)intent request will grant the lock to client.
2003  */
2004 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
2005                                  struct mdt_lock_handle *lhc,
2006                                  __u64 child_bits,
2007                                  struct ldlm_reply *ldlm_rep)
2008 {
2009         struct ptlrpc_request *req = mdt_info_req(info);
2010         struct mdt_body *reqbody = NULL;
2011         struct mdt_object *parent = info->mti_object;
2012         struct mdt_object *child = NULL;
2013         struct lu_fid *child_fid = &info->mti_tmp_fid1;
2014         struct lu_name *lname = NULL;
2015         struct mdt_lock_handle *lhp = NULL;
2016         struct ldlm_lock *lock;
2017         struct req_capsule *pill = info->mti_pill;
2018         __u64 try_bits = 0;
2019         bool is_resent;
2020         int ma_need = 0;
2021         int rc;
2022
2023         ENTRY;
2024
2025         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
2026         LASSERT(ergo(is_resent,
2027                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
2028
2029         if (parent == NULL)
2030                 RETURN(-ENOENT);
2031
2032         if (info->mti_cross_ref) {
2033                 /* Only getattr on the child. Parent is on another node. */
2034                 mdt_set_disposition(info, ldlm_rep,
2035                                     DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
2036                 child = parent;
2037                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
2038                        "ldlm_rep = %p\n",
2039                        PFID(mdt_object_fid(child)), ldlm_rep);
2040
2041                 rc = mdt_check_resent_lock(info, child, lhc);
2042                 if (rc < 0) {
2043                         RETURN(rc);
2044                 } else if (rc > 0) {
2045                         mdt_lock_handle_init(lhc);
2046                         mdt_lock_reg_init(lhc, LCK_PR);
2047
2048                         /*
2049                          * Object's name entry is on another MDS, it will
2050                          * request PERM lock only because LOOKUP lock is owned
2051                          * by the MDS where name entry resides.
2052                          *
2053                          * TODO: it should try layout lock too. - Jinshan
2054                          */
2055                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
2056                                         MDS_INODELOCK_LAYOUT);
2057                         child_bits |= MDS_INODELOCK_PERM;
2058
2059                         rc = mdt_object_lock(info, child, lhc, child_bits);
2060                         if (rc < 0)
2061                                 RETURN(rc);
2062                 }
2063
2064                 /* Finally, we can get attr for child. */
2065                 if (!mdt_object_exists(child)) {
2066                         LU_OBJECT_DEBUG(D_INFO, info->mti_env,
2067                                         &child->mot_obj,
2068                                         "remote object doesn't exist.");
2069                         mdt_object_unlock(info, child, lhc, 1);
2070                         RETURN(-ENOENT);
2071                 }
2072
2073                 rc = mdt_getattr_internal(info, child, 0);
2074                 if (unlikely(rc != 0)) {
2075                         mdt_object_unlock(info, child, lhc, 1);
2076                         RETURN(rc);
2077                 }
2078
2079                 rc = mdt_pack_secctx_in_reply(info, child);
2080                 if (unlikely(rc)) {
2081                         mdt_object_unlock(info, child, lhc, 1);
2082                         RETURN(rc);
2083                 }
2084
2085                 rc = mdt_pack_encctx_in_reply(info, child);
2086                 if (unlikely(rc))
2087                         mdt_object_unlock(info, child, lhc, 1);
2088                 RETURN(rc);
2089         }
2090
2091         lname = &info->mti_name;
2092         mdt_name_unpack(pill, &RMF_NAME, lname, MNF_FIX_ANON);
2093
2094         if (info->mti_body->mbo_valid & OBD_MD_NAMEHASH) {
2095                 reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
2096                 if (unlikely(reqbody == NULL))
2097                         RETURN(err_serious(-EPROTO));
2098
2099                 *child_fid = reqbody->mbo_fid2;
2100                 if (unlikely(!fid_is_sane(child_fid)))
2101                         RETURN(err_serious(-EINVAL));
2102
2103                 if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
2104                         mdt_object_get(info->mti_env, parent);
2105                         child = parent;
2106                 } else {
2107                         child = mdt_object_find(info->mti_env, info->mti_mdt,
2108                                                 child_fid);
2109                         if (IS_ERR(child))
2110                                 RETURN(PTR_ERR(child));
2111                 }
2112
2113                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
2114                        "ldlm_rep = %p\n",
2115                        PFID(mdt_object_fid(parent)),
2116                        PFID(&reqbody->mbo_fid2), ldlm_rep);
2117         } else if (lu_name_is_valid(lname)) {
2118                 if (mdt_object_remote(parent)) {
2119                         CERROR("%s: parent "DFID" is on remote target\n",
2120                                mdt_obd_name(info->mti_mdt),
2121                                PFID(mdt_object_fid(parent)));
2122                         RETURN(-EPROTO);
2123                 }
2124
2125                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
2126                        "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
2127                        PNAME(lname), ldlm_rep);
2128         } else {
2129                 reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
2130                 if (unlikely(reqbody == NULL))
2131                         RETURN(err_serious(-EPROTO));
2132
2133                 *child_fid = reqbody->mbo_fid2;
2134                 if (unlikely(!fid_is_sane(child_fid)))
2135                         RETURN(err_serious(-EINVAL));
2136
2137                 if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
2138                         mdt_object_get(info->mti_env, parent);
2139                         child = parent;
2140                 } else {
2141                         child = mdt_object_find(info->mti_env, info->mti_mdt,
2142                                                 child_fid);
2143                         if (IS_ERR(child))
2144                                 RETURN(PTR_ERR(child));
2145                 }
2146
2147                 if (mdt_object_remote(child)) {
2148                         CERROR("%s: child "DFID" is on remote target\n",
2149                                mdt_obd_name(info->mti_mdt),
2150                                PFID(mdt_object_fid(child)));
2151                         GOTO(out_child, rc = -EPROTO);
2152                 }
2153
2154                 /* don't fetch LOOKUP lock if it's remote object */
2155                 rc = mdt_is_remote_object(info, parent, child);
2156                 if (rc < 0)
2157                         GOTO(out_child, rc);
2158                 if (rc)
2159                         child_bits &= ~MDS_INODELOCK_LOOKUP;
2160
2161                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
2162                        "ldlm_rep = %p\n",
2163                        PFID(mdt_object_fid(parent)),
2164                        PFID(&reqbody->mbo_fid2), ldlm_rep);
2165         }
2166
2167         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
2168
2169         if (unlikely(!mdt_object_exists(parent)) &&
2170             !(info->mti_body->mbo_valid & OBD_MD_NAMEHASH) &&
2171             lu_name_is_valid(lname)) {
2172                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
2173                                 &parent->mot_obj,
2174                                 "Parent doesn't exist!");
2175                 GOTO(out_child, rc = -ESTALE);
2176         }
2177
2178         if (!(info->mti_body->mbo_valid & OBD_MD_NAMEHASH) &&
2179             lu_name_is_valid(lname)) {
2180                 if (info->mti_body->mbo_valid == OBD_MD_FLID) {
2181                         rc = mdt_raw_lookup(info, parent, lname);
2182
2183                         RETURN(rc);
2184                 }
2185
2186                 /* step 1: lock parent only if parent is a directory */
2187                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
2188                         lhp = &info->mti_lh[MDT_LH_PARENT];
2189                         mdt_lock_pdo_init(lhp, LCK_PR, lname);
2190                         rc = mdt_object_lock(info, parent, lhp,
2191                                              MDS_INODELOCK_UPDATE);
2192                         if (unlikely(rc != 0))
2193                                 RETURN(rc);
2194                 }
2195
2196                 /* step 2: lookup child's fid by name */
2197                 fid_zero(child_fid);
2198                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
2199                                 child_fid, &info->mti_spec);
2200                 if (rc == -ENOENT)
2201                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
2202
2203                 if (rc != 0)
2204                         GOTO(unlock_parent, rc);
2205
2206                 child = mdt_object_find(info->mti_env, info->mti_mdt,
2207                                         child_fid);
2208                 if (unlikely(IS_ERR(child)))
2209                         GOTO(unlock_parent, rc = PTR_ERR(child));
2210         }
2211
2212         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
2213
2214         /* step 3: lock child regardless if it is local or remote. */
2215         LASSERT(child);
2216
2217         if (info->mti_body->mbo_valid & OBD_MD_NAMEHASH) {
2218                 /* Here, lname is an encoded hash of on-disk name, and
2219                  * client is doing access without encryption key.
2220                  * So we need to compare name hash with the one in the request.
2221                  */
2222                 if (!find_name_matching_hash(info, lname, parent,
2223                                              child, true)) {
2224                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
2225                         mdt_clear_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
2226                         GOTO(out_child, rc = -ENOENT);
2227                 }
2228         }
2229
2230         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
2231         if (!mdt_object_exists(child)) {
2232                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
2233                                 &child->mot_obj,
2234                                 "Object doesn't exist!");
2235                 GOTO(out_child, rc = -ENOENT);
2236         }
2237
2238         rc = mdt_check_resent_lock(info, child, lhc);
2239         if (rc < 0) {
2240                 GOTO(out_child, rc);
2241         } else if (rc > 0) {
2242                 mdt_lock_handle_init(lhc);
2243                 mdt_lock_reg_init(lhc, LCK_PR);
2244
2245                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
2246                     !mdt_object_remote(child)) {
2247                         struct md_attr *ma = &info->mti_attr;
2248
2249                         ma->ma_valid = 0;
2250                         ma->ma_need = MA_INODE;
2251                         rc = mdt_attr_get_complex(info, child, ma);
2252                         if (unlikely(rc != 0))
2253                                 GOTO(out_child, rc);
2254
2255                         /* If the file has not been changed for some time, we
2256                          * return not only a LOOKUP lock, but also an UPDATE
2257                          * lock and this might save us RPC on later STAT. For
2258                          * directories, it also let negative dentry cache start
2259                          * working for this dir. */
2260                         if (ma->ma_valid & MA_INODE &&
2261                             ma->ma_attr.la_valid & LA_CTIME &&
2262                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
2263                             ma->ma_attr.la_ctime < ktime_get_real_seconds())
2264                                 child_bits |= MDS_INODELOCK_UPDATE;
2265                 }
2266
2267                 /* layout lock must be granted in a best-effort way
2268                  * for IT operations */
2269                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
2270                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
2271                     !mdt_object_remote(child) && ldlm_rep != NULL) {
2272                         if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
2273                             exp_connect_layout(info->mti_exp)) {
2274                                 /* try to grant layout lock for regular file. */
2275                                 try_bits = MDS_INODELOCK_LAYOUT;
2276                         }
2277                         /* Acquire DOM lock in advance for data-on-mdt file */
2278                         if (child != parent)
2279                                 try_bits |= MDS_INODELOCK_DOM;
2280                 }
2281
2282                 if (try_bits != 0) {
2283                         /* try layout lock, it may fail to be granted due to
2284                          * contention at LOOKUP or UPDATE */
2285                         rc = mdt_object_lock_try(info, child, lhc, &child_bits,
2286                                                  try_bits, false);
2287                         if (child_bits & MDS_INODELOCK_LAYOUT)
2288                                 ma_need |= MA_LOV;
2289                 } else {
2290                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
2291                          * client will enqueue the lock to the remote MDT */
2292                         if (mdt_object_remote(child))
2293                                 child_bits &= ~MDS_INODELOCK_UPDATE;
2294                         rc = mdt_object_lock(info, child, lhc, child_bits);
2295                 }
2296                 if (unlikely(rc != 0))
2297                         GOTO(out_child, rc);
2298         }
2299
2300         /* finally, we can get attr for child. */
2301         rc = mdt_getattr_internal(info, child, ma_need);
2302         if (unlikely(rc != 0)) {
2303                 mdt_object_unlock(info, child, lhc, 1);
2304                 GOTO(out_child, rc);
2305         }
2306
2307         rc = mdt_pack_secctx_in_reply(info, child);
2308         if (unlikely(rc)) {
2309                 mdt_object_unlock(info, child, lhc, 1);
2310                 GOTO(out_child, rc);
2311         }
2312
2313         rc = mdt_pack_encctx_in_reply(info, child);
2314         if (unlikely(rc)) {
2315                 mdt_object_unlock(info, child, lhc, 1);
2316                 GOTO(out_child, rc);
2317         }
2318
2319         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2320         if (lock) {
2321                 /* Debugging code. */
2322                 LDLM_DEBUG(lock, "Returning lock to client");
2323                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
2324                                          &lock->l_resource->lr_name),
2325                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
2326                          PLDLMRES(lock->l_resource),
2327                          PFID(mdt_object_fid(child)));
2328
2329                 if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_PTLRPC_ENQ_RESEND))) {
2330                         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
2331                                 OBD_FAIL_TIMEOUT(OBD_FAIL_PTLRPC_ENQ_RESEND,
2332                                                  req->rq_deadline -
2333                                                  req->rq_arrival_time.tv_sec +
2334                                                  cfs_fail_val ?: 3);
2335                         /* Put the lock to the waiting list and force the cancel */
2336                         ldlm_set_ast_sent(lock);
2337                 }
2338
2339                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
2340                     !mdt_object_remote(child) && child != parent) {
2341                         mdt_object_put(info->mti_env, child);
2342                         rc = mdt_pack_size2body(info, child_fid,
2343                                                 &lhc->mlh_reg_lh);
2344                         if (rc != 0 && child_bits & MDS_INODELOCK_DOM) {
2345                                 /* DOM lock was taken in advance but this is
2346                                  * not DoM file. Drop the lock.
2347                                  */
2348                                 lock_res_and_lock(lock);
2349                                 ldlm_inodebits_drop(lock, MDS_INODELOCK_DOM);
2350                                 unlock_res_and_lock(lock);
2351                         }
2352                         LDLM_LOCK_PUT(lock);
2353                         GOTO(unlock_parent, rc = 0);
2354                 }
2355                 LDLM_LOCK_PUT(lock);
2356         }
2357
2358         EXIT;
2359 out_child:
2360         if (child)
2361                 mdt_object_put(info->mti_env, child);
2362 unlock_parent:
2363         if (lhp)
2364                 mdt_object_unlock(info, parent, lhp, 1);
2365         return rc;
2366 }
2367
2368 /* normal handler: should release the child lock */
2369 static int mdt_getattr_name(struct tgt_session_info *tsi)
2370 {
2371         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
2372         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
2373         struct mdt_body *reqbody;
2374         struct mdt_body *repbody;
2375         int rc, rc2;
2376
2377         ENTRY;
2378
2379         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
2380         LASSERT(reqbody != NULL);
2381         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
2382         LASSERT(repbody != NULL);
2383
2384         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
2385         repbody->mbo_eadatasize = 0;
2386         repbody->mbo_aclsize = 0;
2387
2388         rc = mdt_init_ucred(info, reqbody);
2389         if (unlikely(rc))
2390                 GOTO(out_shrink, rc);
2391
2392         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
2393         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
2394                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
2395                 lhc->mlh_reg_lh.cookie = 0;
2396         }
2397         mdt_exit_ucred(info);
2398         EXIT;
2399 out_shrink:
2400         mdt_client_compatibility(info);
2401         rc2 = mdt_fix_reply(info);
2402         if (rc == 0)
2403                 rc = rc2;
2404         mdt_thread_info_fini(info);
2405         return rc;
2406 }
2407
2408 static int mdt_rmfid_unlink(struct mdt_thread_info *info,
2409                             const struct lu_fid *pfid,
2410                             const struct lu_name *name,
2411                             struct mdt_object *obj, s64 ctime)
2412 {
2413         struct lu_fid *child_fid = &info->mti_tmp_fid1;
2414         struct ldlm_enqueue_info *einfo = &info->mti_einfo[0];
2415         struct mdt_device *mdt = info->mti_mdt;
2416         struct md_attr *ma = &info->mti_attr;
2417         struct mdt_lock_handle *parent_lh;
2418         struct mdt_lock_handle *child_lh;
2419         struct mdt_object *pobj;
2420         bool cos_incompat = false;
2421         int rc;
2422         ENTRY;
2423
2424         pobj = mdt_object_find(info->mti_env, mdt, pfid);
2425         if (IS_ERR(pobj))
2426                 GOTO(out, rc = PTR_ERR(pobj));
2427
2428         parent_lh = &info->mti_lh[MDT_LH_PARENT];
2429         mdt_lock_pdo_init(parent_lh, LCK_PW, name);
2430         rc = mdt_object_lock(info, pobj, parent_lh, MDS_INODELOCK_UPDATE);
2431         if (rc != 0)
2432                 GOTO(put_parent, rc);
2433
2434         if (mdt_object_remote(pobj))
2435                 cos_incompat = true;
2436
2437         rc = mdo_lookup(info->mti_env, mdt_object_child(pobj),
2438                         name, child_fid, &info->mti_spec);
2439         if (rc != 0)
2440                 GOTO(unlock_parent, rc);
2441
2442         if (!lu_fid_eq(child_fid, mdt_object_fid(obj)))
2443                 GOTO(unlock_parent, rc = -EREMCHG);
2444
2445         child_lh = &info->mti_lh[MDT_LH_CHILD];
2446         mdt_lock_reg_init(child_lh, LCK_EX);
2447         rc = mdt_reint_striped_lock(info, obj, child_lh,
2448                                     MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE,
2449                                     einfo, cos_incompat);
2450         if (rc != 0)
2451                 GOTO(unlock_parent, rc);
2452
2453         if (atomic_read(&obj->mot_open_count)) {
2454                 CDEBUG(D_OTHER, "object "DFID" open, skip\n",
2455                        PFID(mdt_object_fid(obj)));
2456                 GOTO(unlock_child, rc = -EBUSY);
2457         }
2458
2459         ma->ma_need = 0;
2460         ma->ma_valid = MA_INODE;
2461         ma->ma_attr.la_valid = LA_CTIME;
2462         ma->ma_attr.la_ctime = ctime;
2463
2464         mutex_lock(&obj->mot_lov_mutex);
2465
2466         rc = mdo_unlink(info->mti_env, mdt_object_child(pobj),
2467                         mdt_object_child(obj), name, ma, 0);
2468
2469         mutex_unlock(&obj->mot_lov_mutex);
2470
2471 unlock_child:
2472         mdt_reint_striped_unlock(info, obj, child_lh, einfo, 1);
2473 unlock_parent:
2474         mdt_object_unlock(info, pobj, parent_lh, 1);
2475 put_parent:
2476         mdt_object_put(info->mti_env, pobj);
2477 out:
2478         RETURN(rc);
2479 }
2480
2481 static int mdt_rmfid_check_permission(struct mdt_thread_info *info,
2482                                         struct mdt_object *obj)
2483 {
2484         struct lu_ucred *uc = lu_ucred(info->mti_env);
2485         struct md_attr *ma = &info->mti_attr;
2486         struct lu_attr *la = &ma->ma_attr;
2487         int rc = 0;
2488         ENTRY;
2489
2490         ma->ma_need = MA_INODE;
2491         rc = mo_attr_get(info->mti_env, mdt_object_child(obj), ma);
2492         if (rc)
2493                 GOTO(out, rc);
2494
2495         if (la->la_flags & LUSTRE_IMMUTABLE_FL)
2496                         rc = -EACCES;
2497
2498         if (cap_raised(uc->uc_cap, CAP_DAC_OVERRIDE))
2499                 RETURN(0);
2500         if (uc->uc_fsuid == la->la_uid) {
2501                 if ((la->la_mode & S_IWUSR) == 0)
2502                         rc = -EACCES;
2503         } else if (uc->uc_fsgid == la->la_gid) {
2504                 if ((la->la_mode & S_IWGRP) == 0)
2505                         rc = -EACCES;
2506         } else if ((la->la_mode & S_IWOTH) == 0) {
2507                         rc = -EACCES;
2508         }
2509
2510 out:
2511         RETURN(rc);
2512 }
2513
2514 static int mdt_rmfid_one(struct mdt_thread_info *info, struct lu_fid *fid,
2515                          s64 ctime)
2516 {
2517         struct mdt_device *mdt = info->mti_mdt;
2518         struct mdt_object *obj = NULL;
2519         struct linkea_data ldata = { NULL };
2520         struct lu_buf *buf = &info->mti_big_buf;
2521         struct lu_name *name = &info->mti_name;
2522         struct lu_fid *pfid = &info->mti_tmp_fid1;
2523         struct link_ea_header *leh;
2524         struct link_ea_entry *lee;
2525         int reclen, count, rc = 0;
2526         ENTRY;
2527
2528         if (!fid_is_sane(fid))
2529                 GOTO(out, rc = -EINVAL);
2530
2531         if (!fid_is_namespace_visible(fid))
2532                 GOTO(out, rc = -EINVAL);
2533
2534         obj = mdt_object_find(info->mti_env, mdt, fid);
2535         if (IS_ERR(obj))
2536                 GOTO(out, rc = PTR_ERR(obj));
2537
2538         if (mdt_object_remote(obj))
2539                 GOTO(out, rc = -EREMOTE);
2540         if (!mdt_object_exists(obj) || lu_object_is_dying(&obj->mot_header))
2541                 GOTO(out, rc = -ENOENT);
2542
2543         rc = mdt_rmfid_check_permission(info, obj);
2544         if (rc)
2545                 GOTO(out, rc);
2546
2547         /* take LinkEA */
2548         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
2549         if (!buf->lb_buf)
2550                 GOTO(out, rc = -ENOMEM);
2551
2552         ldata.ld_buf = buf;
2553         rc = mdt_links_read(info, obj, &ldata);
2554         if (rc)
2555                 GOTO(out, rc);
2556
2557         leh = buf->lb_buf;
2558         lee = (struct link_ea_entry *)(leh + 1);
2559         for (count = 0; count < leh->leh_reccount; count++) {
2560                 /* remove every hardlink */
2561                 linkea_entry_unpack(lee, &reclen, name, pfid);
2562                 lee = (struct link_ea_entry *) ((char *)lee + reclen);
2563                 rc = mdt_rmfid_unlink(info, pfid, name, obj, ctime);
2564                 if (rc)
2565                         break;
2566         }
2567
2568 out:
2569         if (obj && !IS_ERR(obj))
2570                 mdt_object_put(info->mti_env, obj);
2571         if (info->mti_big_buf.lb_buf)
2572                 lu_buf_free(&info->mti_big_buf);
2573
2574         RETURN(rc);
2575 }
2576
2577 static int mdt_rmfid(struct tgt_session_info *tsi)
2578 {
2579         struct mdt_thread_info *mti = tsi2mdt_info(tsi);
2580         struct mdt_body *reqbody;
2581         struct lu_fid *fids, *rfids;
2582         int bufsize, rc;
2583         __u32 *rcs;
2584         int i, nr;
2585         ENTRY;
2586
2587         reqbody = req_capsule_client_get(tsi->tsi_pill, &RMF_MDT_BODY);
2588         if (reqbody == NULL)
2589                 RETURN(-EPROTO);
2590         bufsize = req_capsule_get_size(tsi->tsi_pill, &RMF_FID_ARRAY,
2591                                        RCL_CLIENT);
2592         nr = bufsize / sizeof(struct lu_fid);
2593         if (nr * sizeof(struct lu_fid) != bufsize)
2594                 RETURN(-EINVAL);
2595         req_capsule_set_size(tsi->tsi_pill, &RMF_RCS,
2596                              RCL_SERVER, nr * sizeof(__u32));
2597         req_capsule_set_size(tsi->tsi_pill, &RMF_FID_ARRAY,
2598                              RCL_SERVER, nr * sizeof(struct lu_fid));
2599         rc = req_capsule_server_pack(tsi->tsi_pill);
2600         if (rc)
2601                 GOTO(out, rc = err_serious(rc));
2602         fids = req_capsule_client_get(tsi->tsi_pill, &RMF_FID_ARRAY);
2603         if (fids == NULL)
2604                 RETURN(-EPROTO);
2605         rcs = req_capsule_server_get(tsi->tsi_pill, &RMF_RCS);
2606         LASSERT(rcs);
2607         rfids = req_capsule_server_get(tsi->tsi_pill, &RMF_FID_ARRAY);
2608         LASSERT(rfids);
2609
2610         mdt_init_ucred(mti, reqbody);
2611         for (i = 0; i < nr; i++) {
2612                 rfids[i] = fids[i];
2613                 rcs[i] = mdt_rmfid_one(mti, fids + i, reqbody->mbo_ctime);
2614         }
2615         mdt_exit_ucred(mti);
2616
2617 out:
2618         RETURN(rc);
2619 }
2620
2621 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2622                          void *karg, void __user *uarg);
2623
2624 static int mdt_set_info(struct tgt_session_info *tsi)
2625 {
2626         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2627         char                    *key;
2628         void                    *val;
2629         int                      keylen, vallen, rc = 0;
2630
2631         ENTRY;
2632
2633         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
2634         if (key == NULL) {
2635                 DEBUG_REQ(D_HA, req, "no set_info key");
2636                 RETURN(err_serious(-EFAULT));
2637         }
2638
2639         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
2640                                       RCL_CLIENT);
2641
2642         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
2643         if (val == NULL) {
2644                 DEBUG_REQ(D_HA, req, "no set_info val");
2645                 RETURN(err_serious(-EFAULT));
2646         }
2647
2648         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
2649                                       RCL_CLIENT);
2650
2651         /* Swab any part of val you need to here */
2652         if (KEY_IS(KEY_READ_ONLY)) {
2653                 spin_lock(&req->rq_export->exp_lock);
2654                 if (*(__u32 *)val)
2655                         *exp_connect_flags_ptr(req->rq_export) |=
2656                                 OBD_CONNECT_RDONLY;
2657                 else
2658                         *exp_connect_flags_ptr(req->rq_export) &=
2659                                 ~OBD_CONNECT_RDONLY;
2660                 spin_unlock(&req->rq_export->exp_lock);
2661         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2662                 struct changelog_setinfo *cs = val;
2663
2664                 if (vallen != sizeof(*cs)) {
2665                         CERROR("%s: bad changelog_clear setinfo size %d\n",
2666                                tgt_name(tsi->tsi_tgt), vallen);
2667                         RETURN(-EINVAL);
2668                 }
2669                 if (req_capsule_req_need_swab(&req->rq_pill)) {
2670                         __swab64s(&cs->cs_recno);
2671                         __swab32s(&cs->cs_id);
2672                 }
2673
2674                 if (!mdt_is_rootadmin(tsi2mdt_info(tsi)))
2675                         RETURN(-EACCES);
2676                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
2677                                    vallen, val, NULL);
2678         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2679                 if (vallen > 0)
2680                         obd_export_evict_by_nid(req->rq_export->exp_obd, val);
2681         } else {
2682                 RETURN(-EINVAL);
2683         }
2684         RETURN(rc);
2685 }
2686
2687 static int mdt_readpage(struct tgt_session_info *tsi)
2688 {
2689         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
2690         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
2691         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
2692         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
2693         struct mdt_body         *repbody;
2694         int                      rc;
2695         int                      i;
2696
2697         ENTRY;
2698
2699         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
2700                 RETURN(err_serious(-ENOMEM));
2701
2702         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
2703         if (repbody == NULL || reqbody == NULL)
2704                 RETURN(err_serious(-EFAULT));
2705
2706         /*
2707          * prepare @rdpg before calling lower layers and transfer itself. Here
2708          * reqbody->size contains offset of where to start to read and
2709          * reqbody->nlink contains number bytes to read.
2710          */
2711         rdpg->rp_hash = reqbody->mbo_size;
2712         if (rdpg->rp_hash != reqbody->mbo_size) {
2713                 CERROR("Invalid hash: %#llx != %#llx\n",
2714                        rdpg->rp_hash, reqbody->mbo_size);
2715                 RETURN(-EFAULT);
2716         }
2717
2718         rdpg->rp_attrs = reqbody->mbo_mode;
2719         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
2720                 rdpg->rp_attrs |= LUDA_64BITHASH;
2721         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
2722                                 exp_max_brw_size(tsi->tsi_exp));
2723         rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >>
2724                           PAGE_SHIFT;
2725         OBD_ALLOC_PTR_ARRAY_LARGE(rdpg->rp_pages, rdpg->rp_npages);
2726         if (rdpg->rp_pages == NULL)
2727                 RETURN(-ENOMEM);
2728
2729         for (i = 0; i < rdpg->rp_npages; ++i) {
2730                 rdpg->rp_pages[i] = alloc_page(GFP_NOFS);
2731                 if (rdpg->rp_pages[i] == NULL)
2732                         GOTO(free_rdpg, rc = -ENOMEM);
2733         }
2734
2735         /* call lower layers to fill allocated pages with directory data */
2736         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
2737         if (rc < 0)
2738                 GOTO(free_rdpg, rc);
2739
2740         /* send pages to client */
2741         rc = tgt_sendpage(tsi, rdpg, rc);
2742
2743         EXIT;
2744 free_rdpg:
2745
2746         for (i = 0; i < rdpg->rp_npages; i++)
2747                 if (rdpg->rp_pages[i] != NULL)
2748                         __free_page(rdpg->rp_pages[i]);
2749         OBD_FREE_PTR_ARRAY_LARGE(rdpg->rp_pages, rdpg->rp_npages);
2750
2751         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
2752                 RETURN(0);
2753
2754         return rc;
2755 }
2756
2757 static int mdt_fix_attr_ucred(struct mdt_thread_info *info, __u32 op)
2758 {
2759         struct lu_ucred *uc = mdt_ucred_check(info);
2760         struct lu_attr *attr = &info->mti_attr.ma_attr;
2761
2762         if (uc == NULL)
2763                 return -EINVAL;
2764
2765         if (op != REINT_SETATTR) {
2766                 if ((attr->la_valid & LA_UID) && (attr->la_uid != -1))
2767                         attr->la_uid = uc->uc_fsuid;
2768                 /* for S_ISGID, inherit gid from his parent, such work will be
2769                  * done in cmm/mdd layer, here set all cases as uc->uc_fsgid. */
2770                 if ((attr->la_valid & LA_GID) && (attr->la_gid != -1))
2771                         attr->la_gid = uc->uc_fsgid;
2772         }
2773
2774         return 0;
2775 }
2776
2777 static inline bool mdt_is_readonly_open(struct mdt_thread_info *info, __u32 op)
2778 {
2779         return op == REINT_OPEN &&
2780              !(info->mti_spec.sp_cr_flags & (MDS_FMODE_WRITE | MDS_OPEN_CREAT));
2781 }
2782
2783 static void mdt_preset_secctx_size(struct mdt_thread_info *info)
2784 {
2785         struct req_capsule *pill = info->mti_pill;
2786
2787         if (req_capsule_has_field(pill, &RMF_FILE_SECCTX,
2788                                   RCL_SERVER) &&
2789             req_capsule_has_field(pill, &RMF_FILE_SECCTX_NAME,
2790                                   RCL_CLIENT)) {
2791                 if (req_capsule_get_size(pill, &RMF_FILE_SECCTX_NAME,
2792                                          RCL_CLIENT) != 0)
2793                         /* pre-set size in server part with max size */
2794                         req_capsule_set_size(pill, &RMF_FILE_SECCTX,
2795                                              RCL_SERVER,
2796                                              OBD_MAX_DEFAULT_EA_SIZE);
2797                 else
2798                         req_capsule_set_size(pill, &RMF_FILE_SECCTX,
2799                                              RCL_SERVER, 0);
2800         }
2801 }
2802
2803 static void mdt_preset_encctx_size(struct mdt_thread_info *info)
2804 {
2805         struct req_capsule *pill = info->mti_pill;
2806
2807         if (req_capsule_has_field(pill, &RMF_FILE_ENCCTX,
2808                                   RCL_SERVER))
2809                 /* pre-set size in server part with max size */
2810                 req_capsule_set_size(pill, &RMF_FILE_ENCCTX,
2811                                      RCL_SERVER,
2812                                      info->mti_mdt->mdt_max_mdsize);
2813 }
2814
2815 static int mdt_reint_internal(struct mdt_thread_info *info,
2816                               struct mdt_lock_handle *lhc,
2817                               __u32 op)
2818 {
2819         struct req_capsule      *pill = info->mti_pill;
2820         struct mdt_body         *repbody;
2821         int                      rc = 0, rc2;
2822
2823         ENTRY;
2824
2825         rc = mdt_reint_unpack(info, op);
2826         if (rc != 0) {
2827                 CERROR("Can't unpack reint, rc %d\n", rc);
2828                 RETURN(err_serious(rc));
2829         }
2830
2831
2832         /* check if the file system is set to readonly. O_RDONLY open
2833          * is still allowed even the file system is set to readonly mode */
2834         if (mdt_rdonly(info->mti_exp) && !mdt_is_readonly_open(info, op))
2835                 RETURN(err_serious(-EROFS));
2836
2837         /* for replay (no_create) lmm is not needed, client has it already */
2838         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2839                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2840                                      DEF_REP_MD_SIZE);
2841
2842         /* llog cookies are always 0, the field is kept for compatibility */
2843         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2844                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
2845
2846         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
2847          * by default. If the target object has more ACL entries, then
2848          * enlarge the buffer when necessary. */
2849         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
2850                 req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
2851                                      LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
2852
2853         mdt_preset_secctx_size(info);
2854         mdt_preset_encctx_size(info);
2855
2856         rc = req_capsule_server_pack(pill);
2857         if (rc != 0) {
2858                 CERROR("Can't pack response, rc %d\n", rc);
2859                 RETURN(err_serious(rc));
2860         }
2861
2862         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
2863                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
2864                 LASSERT(repbody);
2865                 repbody->mbo_eadatasize = 0;
2866                 repbody->mbo_aclsize = 0;
2867         }
2868
2869         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
2870
2871         /* for replay no cookkie / lmm need, because client have this already */
2872         if (info->mti_spec.no_create)
2873                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2874                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
2875
2876         rc = mdt_init_ucred_reint(info);
2877         if (rc)
2878                 GOTO(out_shrink, rc);
2879
2880         rc = mdt_fix_attr_ucred(info, op);
2881         if (rc != 0)
2882                 GOTO(out_ucred, rc = err_serious(rc));
2883
2884         rc = mdt_check_resent(info, mdt_reconstruct, lhc);
2885         if (rc < 0) {
2886                 GOTO(out_ucred, rc);
2887         } else if (rc == 1) {
2888                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt");
2889                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
2890                 GOTO(out_ucred, rc);
2891         }
2892         rc = mdt_reint_rec(info, lhc);
2893         EXIT;
2894 out_ucred:
2895         mdt_exit_ucred(info);
2896 out_shrink:
2897         mdt_client_compatibility(info);
2898
2899         rc2 = mdt_fix_reply(info);
2900         if (rc == 0)
2901                 rc = rc2;
2902
2903         /*
2904          * Data-on-MDT optimization - read data along with OPEN and return it
2905          * in reply when possible.
2906          */
2907         if (rc == 0 && op == REINT_OPEN && !req_is_replay(pill->rc_req))
2908                 rc = mdt_dom_read_on_open(info, info->mti_mdt,
2909                                           &lhc->mlh_reg_lh);
2910
2911         return rc;
2912 }
2913
2914 static long mdt_reint_opcode(struct ptlrpc_request *req,
2915                              const struct req_format **fmt)
2916 {
2917         struct mdt_device       *mdt;
2918         struct mdt_rec_reint    *rec;
2919         long                     opc;
2920
2921         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
2922         if (rec != NULL) {
2923                 opc = rec->rr_opcode;
2924                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
2925                 if (opc < REINT_MAX && fmt[opc] != NULL)
2926                         req_capsule_extend(&req->rq_pill, fmt[opc]);
2927                 else {
2928                         mdt = mdt_exp2dev(req->rq_export);
2929                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
2930                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
2931                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
2932                         opc = err_serious(-EFAULT);
2933                 }
2934         } else {
2935                 opc = err_serious(-EFAULT);
2936         }
2937         return opc;
2938 }
2939
2940 static int mdt_reint(struct tgt_session_info *tsi)
2941 {
2942         long opc;
2943         int  rc;
2944         static const struct req_format *reint_fmts[REINT_MAX] = {
2945                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
2946                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
2947                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
2948                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
2949                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
2950                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
2951                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
2952                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
2953                 [REINT_MIGRATE]  = &RQF_MDS_REINT_MIGRATE,
2954                 [REINT_RESYNC]   = &RQF_MDS_REINT_RESYNC,
2955         };
2956
2957         ENTRY;
2958
2959         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
2960         if (opc >= 0) {
2961                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2962                 /*
2963                  * No lock possible here from client to pass it to reint code
2964                  * path.
2965                  */
2966                 rc = mdt_reint_internal(info, NULL, opc);
2967                 mdt_thread_info_fini(info);
2968         } else {
2969                 rc = opc;
2970         }
2971
2972         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
2973         RETURN(rc);
2974 }
2975
2976 /* this should sync the whole device */
2977 int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
2978 {
2979         struct dt_device *dt = mdt->mdt_bottom;
2980         int rc;
2981         ENTRY;
2982
2983         rc = dt->dd_ops->dt_sync(env, dt);
2984         RETURN(rc);
2985 }
2986
2987 /* this should sync this object */
2988 static int mdt_object_sync(const struct lu_env *env, struct obd_export *exp,
2989                            struct mdt_object *mo)
2990 {
2991         int rc = 0;
2992
2993         ENTRY;
2994
2995         if (!mdt_object_exists(mo)) {
2996                 CWARN("%s: non existing object "DFID": rc = %d\n",
2997                       exp->exp_obd->obd_name, PFID(mdt_object_fid(mo)),
2998                       -ESTALE);
2999                 RETURN(-ESTALE);
3000         }
3001
3002         if (S_ISREG(lu_object_attr(&mo->mot_obj))) {
3003                 struct lu_target *tgt = tgt_ses_info(env)->tsi_tgt;
3004                 dt_obj_version_t version;
3005
3006                 version = dt_version_get(env, mdt_obj2dt(mo));
3007                 if (version > tgt->lut_obd->obd_last_committed)
3008                         rc = mo_object_sync(env, mdt_object_child(mo));
3009         } else {
3010                 rc = mo_object_sync(env, mdt_object_child(mo));
3011         }
3012
3013         RETURN(rc);
3014 }
3015
3016 static int mdt_sync(struct tgt_session_info *tsi)
3017 {
3018         struct ptlrpc_request   *req = tgt_ses_req(tsi);
3019         struct req_capsule      *pill = tsi->tsi_pill;
3020         struct mdt_body         *body;
3021         ktime_t                  kstart = ktime_get();
3022         int                      rc;
3023
3024         ENTRY;
3025
3026         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
3027                 RETURN(err_serious(-ENOMEM));
3028
3029         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
3030                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
3031         } else {
3032                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
3033
3034                 if (unlikely(info->mti_object == NULL))
3035                         RETURN(-EPROTO);
3036
3037                 /* sync an object */
3038                 rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp,
3039                                      info->mti_object);
3040                 if (rc == 0) {
3041                         const struct lu_fid *fid;
3042                         struct lu_attr *la = &info->mti_attr.ma_attr;
3043
3044                         info->mti_attr.ma_need = MA_INODE;
3045                         info->mti_attr.ma_valid = 0;
3046                         rc = mdt_attr_get_complex(info, info->mti_object,
3047                                                   &info->mti_attr);
3048                         if (rc == 0) {
3049                                 body = req_capsule_server_get(pill,
3050                                                               &RMF_MDT_BODY);
3051                                 fid = mdt_object_fid(info->mti_object);
3052                                 mdt_pack_attr2body(info, body, la, fid);
3053                         }
3054                 }
3055                 mdt_thread_info_fini(info);
3056         }
3057         if (rc == 0)
3058                 mdt_counter_incr(req, LPROC_MDT_SYNC,
3059                                  ktime_us_delta(ktime_get(), kstart));
3060
3061         RETURN(rc);
3062 }
3063
3064 static int mdt_data_sync(struct tgt_session_info *tsi)
3065 {
3066         struct mdt_thread_info *info;
3067         struct mdt_device *mdt = mdt_exp2dev(tsi->tsi_exp);
3068         struct ost_body *body = tsi->tsi_ost_body;
3069         struct ost_body *repbody;
3070         struct mdt_object *mo = NULL;
3071         struct md_attr *ma;
3072         int rc = 0;
3073
3074         ENTRY;
3075
3076         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
3077
3078         /* if no fid is specified then do nothing,
3079          * device sync is done via MDS_SYNC */
3080         if (fid_is_zero(&tsi->tsi_fid))
3081                 RETURN(0);
3082
3083         mo = mdt_object_find(tsi->tsi_env, mdt, &tsi->tsi_fid);
3084         if (IS_ERR(mo))
3085                 RETURN(PTR_ERR(mo));
3086
3087         rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp, mo);
3088         if (rc)
3089                 GOTO(put, rc);
3090
3091         repbody->oa.o_oi = body->oa.o_oi;
3092         repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
3093
3094         info = tsi2mdt_info(tsi);
3095         ma = &info->mti_attr;
3096         ma->ma_need = MA_INODE;
3097         ma->ma_valid = 0;
3098         rc = mdt_attr_get_complex(info, mo, ma);
3099         if (rc == 0)
3100                 obdo_from_la(&repbody->oa, &ma->ma_attr, VALID_FLAGS);
3101         else
3102                 rc = 0;
3103         mdt_thread_info_fini(info);
3104
3105         EXIT;
3106 put:
3107         if (mo != NULL)
3108                 mdt_object_put(tsi->tsi_env, mo);
3109         return rc;
3110 }
3111
3112 /*
3113  * Handle quota control requests to consult current usage/limit, but also
3114  * to configure quota enforcement
3115  */
3116 static int mdt_quotactl(struct tgt_session_info *tsi)
3117 {
3118         struct obd_export *exp  = tsi->tsi_exp;
3119         struct req_capsule *pill = tsi->tsi_pill;
3120         struct obd_quotactl *oqctl, *repoqc;
3121         int id, rc;
3122         struct mdt_device *mdt = mdt_exp2dev(exp);
3123         struct lu_device *qmt = mdt->mdt_qmt_dev;
3124         struct lu_nodemap *nodemap;
3125         ENTRY;
3126
3127         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
3128         if (!oqctl)
3129                 RETURN(err_serious(-EPROTO));
3130
3131         rc = req_capsule_server_pack(pill);
3132         if (rc)
3133                 RETURN(err_serious(rc));
3134
3135         nodemap = nodemap_get_from_exp(exp);
3136         if (IS_ERR(nodemap))
3137                 RETURN(PTR_ERR(nodemap));
3138
3139         switch (oqctl->qc_cmd) {
3140                 /* master quotactl */
3141         case Q_SETINFO:
3142         case Q_SETQUOTA:
3143         case LUSTRE_Q_SETDEFAULT:
3144         case LUSTRE_Q_SETQUOTAPOOL:
3145         case LUSTRE_Q_SETINFOPOOL:
3146         case LUSTRE_Q_SETDEFAULT_POOL:
3147                 if (!nodemap_can_setquota(nodemap, oqctl->qc_type,
3148                                           oqctl->qc_id))
3149                         GOTO(out_nodemap, rc = -EPERM);
3150                 /* fallthrough */
3151         case Q_GETINFO:
3152         case Q_GETQUOTA:
3153         case LUSTRE_Q_GETDEFAULT:
3154         case LUSTRE_Q_GETQUOTAPOOL:
3155         case LUSTRE_Q_GETINFOPOOL:
3156         case LUSTRE_Q_GETDEFAULT_POOL:
3157                 if (qmt == NULL)
3158                         GOTO(out_nodemap, rc = -EOPNOTSUPP);
3159                 /* slave quotactl */
3160                 /* fallthrough */
3161         case Q_GETOINFO:
3162         case Q_GETOQUOTA:
3163                 break;
3164         default:
3165                 rc = -EFAULT;
3166                 CERROR("%s: unsupported quotactl command %d: rc = %d\n",
3167                        mdt_obd_name(mdt), oqctl->qc_cmd, rc);
3168                 GOTO(out_nodemap, rc);
3169         }
3170
3171         id = oqctl->qc_id;
3172         switch (oqctl->qc_type) {
3173         case USRQUOTA:
3174                 id = nodemap_map_id(nodemap, NODEMAP_UID,
3175                                     NODEMAP_CLIENT_TO_FS, id);
3176                 break;
3177         case GRPQUOTA:
3178                 id = nodemap_map_id(nodemap, NODEMAP_GID,
3179                                     NODEMAP_CLIENT_TO_FS, id);
3180                 break;
3181         case PRJQUOTA:
3182                 id = nodemap_map_id(nodemap, NODEMAP_PROJID,
3183                                     NODEMAP_CLIENT_TO_FS, id);
3184                 break;
3185         default:
3186                 GOTO(out_nodemap, rc = -EOPNOTSUPP);
3187         }
3188         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
3189         if (repoqc == NULL)
3190                 GOTO(out_nodemap, rc = err_serious(-EFAULT));
3191
3192         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA)
3193                 barrier_exit(tsi->tsi_tgt->lut_bottom);
3194
3195         if (oqctl->qc_id != id)
3196                 swap(oqctl->qc_id, id);
3197
3198         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA) {
3199                 if (unlikely(!barrier_entry(tsi->tsi_tgt->lut_bottom)))
3200                         RETURN(-EINPROGRESS);
3201         }
3202
3203         switch (oqctl->qc_cmd) {
3204
3205         case Q_GETINFO:
3206         case Q_SETINFO:
3207         case Q_SETQUOTA:
3208         case Q_GETQUOTA:
3209         case LUSTRE_Q_SETDEFAULT:
3210         case LUSTRE_Q_GETDEFAULT:
3211         case LUSTRE_Q_SETQUOTAPOOL:
3212         case LUSTRE_Q_GETQUOTAPOOL:
3213         case LUSTRE_Q_SETINFOPOOL:
3214         case LUSTRE_Q_GETINFOPOOL:
3215         case LUSTRE_Q_SETDEFAULT_POOL:
3216         case LUSTRE_Q_GETDEFAULT_POOL:
3217                 /* forward quotactl request to QMT */
3218                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
3219                 break;
3220
3221         case Q_GETOINFO:
3222         case Q_GETOQUOTA:
3223                 /* slave quotactl */
3224                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
3225                                    oqctl);
3226                 break;
3227
3228         default:
3229                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
3230                 GOTO(out_nodemap, rc = -EFAULT);
3231         }
3232
3233         if (oqctl->qc_id != id)
3234                 swap(oqctl->qc_id, id);
3235
3236         QCTL_COPY(repoqc, oqctl);
3237         EXIT;
3238
3239 out_nodemap:
3240         nodemap_putref(nodemap);
3241
3242         return rc;
3243 }
3244
3245 /** clone llog ctxt from child (mdd)
3246  * This allows remote llog (replicator) access.
3247  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
3248  * context was originally set up, or we can handle them directly.
3249  * I choose the latter, but that means I need any llog
3250  * contexts set up by child to be accessable by the mdt.  So we clone the
3251  * context into our context list here.
3252  */
3253 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
3254                                int idx)
3255 {
3256         struct md_device  *next = mdt->mdt_child;
3257         struct llog_ctxt *ctxt;
3258         int rc;
3259
3260         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
3261                 return 0;
3262
3263         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
3264         if (rc || ctxt == NULL) {
3265                 return 0;
3266         }
3267
3268         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
3269         if (rc)
3270                 CERROR("Can't set mdt ctxt %d\n", rc);
3271
3272         return rc;
3273 }
3274
3275 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
3276                                  struct mdt_device *mdt, int idx)
3277 {
3278         struct llog_ctxt *ctxt;
3279
3280         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
3281         if (ctxt == NULL)
3282                 return 0;
3283         /* Put once for the get we just did, and once for the clone */
3284         llog_ctxt_put(ctxt);
3285         llog_ctxt_put(ctxt);
3286         return 0;
3287 }
3288
3289 /*
3290  * sec context handlers
3291  */
3292 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
3293 {
3294         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
3295
3296         return 0;
3297 }
3298
3299 /*
3300  * quota request handlers
3301  */
3302 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
3303 {
3304         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
3305         struct lu_device        *qmt = mdt->mdt_qmt_dev;
3306         int                      rc;
3307         ENTRY;
3308
3309         if (qmt == NULL)
3310                 RETURN(err_serious(-EOPNOTSUPP));
3311
3312         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
3313         RETURN(rc);
3314 }
3315
3316 struct mdt_object *mdt_object_new(const struct lu_env *env,
3317                                   struct mdt_device *d,
3318                                   const struct lu_fid *f)
3319 {
3320         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
3321         struct lu_object *o;
3322         struct mdt_object *m;
3323         ENTRY;
3324
3325         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
3326         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
3327         if (unlikely(IS_ERR(o)))
3328                 m = (struct mdt_object *)o;
3329         else
3330                 m = mdt_obj(o);
3331         RETURN(m);
3332 }
3333
3334 struct mdt_object *mdt_object_find(const struct lu_env *env,
3335                                    struct mdt_device *d,
3336                                    const struct lu_fid *f)
3337 {
3338         struct lu_object *o;
3339         struct mdt_object *m;
3340         ENTRY;
3341
3342         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
3343         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
3344         if (unlikely(IS_ERR(o)))
3345                 m = (struct mdt_object *)o;
3346         else
3347                 m = mdt_obj(o);
3348
3349         RETURN(m);
3350 }
3351
3352 /**
3353  * Asyncronous commit for mdt device.
3354  *
3355  * Pass asynchonous commit call down the MDS stack.
3356  *
3357  * \param env environment
3358  * \param mdt the mdt device
3359  */
3360 static void mdt_device_commit_async(const struct lu_env *env,
3361                                     struct mdt_device *mdt)
3362 {
3363         struct dt_device *dt = mdt->mdt_bottom;
3364         int rc;
3365         ENTRY;
3366
3367         rc = dt->dd_ops->dt_commit_async(env, dt);
3368         if (unlikely(rc != 0))
3369                 CWARN("%s: async commit start failed: rc = %d\n",
3370                       mdt_obd_name(mdt), rc);
3371         atomic_inc(&mdt->mdt_async_commit_count);
3372         EXIT;
3373 }
3374
3375 /**
3376  * Mark the lock as "synchonous".
3377  *
3378  * Mark the lock to deffer transaction commit to the unlock time.
3379  *
3380  * \param lock the lock to mark as "synchonous"
3381  *
3382  * \see mdt_is_lock_sync
3383  * \see mdt_save_lock
3384  */
3385 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
3386 {
3387         lock->l_ast_data = (void*)1;
3388 }
3389
3390 /**
3391  * Check whehter the lock "synchonous" or not.
3392  *
3393  * \param lock the lock to check
3394  * \retval 1 the lock is "synchonous"
3395  * \retval 0 the lock isn't "synchronous"
3396  *
3397  * \see mdt_set_lock_sync
3398  * \see mdt_save_lock
3399  */
3400 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
3401 {
3402         return lock->l_ast_data != NULL;
3403 }
3404
3405 /**
3406  * Blocking AST for mdt locks.
3407  *
3408  * Starts transaction commit if in case of COS lock conflict or
3409  * deffers such a commit to the mdt_save_lock.
3410  *
3411  * \param lock the lock which blocks a request or cancelling lock
3412  * \param desc unused
3413  * \param data unused
3414  * \param flag indicates whether this cancelling or blocking callback
3415  * \retval 0
3416  * \see ldlm_blocking_ast_nocheck
3417  */
3418 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
3419                      void *data, int flag)
3420 {
3421         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
3422         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
3423         struct ldlm_cb_set_arg *arg = data;
3424         bool commit_async = false;
3425         int rc;
3426         ENTRY;
3427
3428         if (flag == LDLM_CB_CANCELING)
3429                 RETURN(0);
3430
3431         lock_res_and_lock(lock);
3432         if (lock->l_blocking_ast != mdt_blocking_ast) {
3433                 unlock_res_and_lock(lock);
3434                 RETURN(0);
3435         }
3436
3437         /* A blocking ast may be sent from ldlm_lock_decref_internal
3438          * when the last reference to a local lock was released and
3439          * during blocking event from ldlm_work_bl_ast_lock().
3440          * The 'data' parameter is l_ast_data in the first case and
3441          * callback arguments in the second one. Distinguish them by that.
3442          */
3443         if (!data || data == lock->l_ast_data || !arg->bl_desc)
3444                 goto skip_cos_checks;
3445
3446         if (lock->l_req_mode & (LCK_PW | LCK_EX)) {
3447                 if (mdt_cos_is_enabled(mdt)) {
3448                         if (!arg->bl_desc->bl_same_client)
3449                                 mdt_set_lock_sync(lock);
3450                 } else if (mdt_slc_is_enabled(mdt) &&
3451                            arg->bl_desc->bl_cos_incompat) {
3452                         mdt_set_lock_sync(lock);
3453                         /*
3454                          * we may do extra commit here, but there is a small
3455                          * window to miss a commit: lock was unlocked (saved),
3456                          * then a conflict lock queued and we come here, but
3457                          * REP-ACK not received, so lock was not converted to
3458                          * COS mode yet.
3459                          * Fortunately this window is quite small, so the
3460                          * extra commit should be rare (not to say distributed
3461                          * operation is rare too).
3462                          */
3463                         commit_async = true;
3464                 }
3465         } else if (lock->l_req_mode == LCK_COS) {
3466                 commit_async = true;
3467         }
3468
3469 skip_cos_checks:
3470         rc = ldlm_blocking_ast_nocheck(lock);
3471
3472         if (commit_async) {
3473                 struct lu_env env;
3474
3475                 rc = lu_env_init(&env, LCT_LOCAL);
3476                 if (unlikely(rc != 0))
3477                         CWARN("%s: lu_env initialization failed, cannot "
3478                               "start asynchronous commit: rc = %d\n",
3479                               obd->obd_name, rc);
3480                 else
3481                         mdt_device_commit_async(&env, mdt);
3482                 lu_env_fini(&env);
3483         }
3484         RETURN(rc);
3485 }
3486
3487 /*
3488  * Blocking AST for cross-MDT lock
3489  *
3490  * Discard lock from uncommitted_slc_locks and cancel it.
3491  *
3492  * \param lock  the lock which blocks a request or cancelling lock
3493  * \param desc  unused
3494  * \param data  unused
3495  * \param flag  indicates whether this cancelling or blocking callback
3496  * \retval      0 on success
3497  * \retval      negative number on error
3498  */
3499 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
3500                             void *data, int flag)
3501 {
3502         int rc = 0;
3503         ENTRY;
3504
3505         switch (flag) {
3506         case LDLM_CB_BLOCKING: {
3507                 struct lustre_handle lockh;
3508
3509                 ldlm_lock2handle(lock, &lockh);
3510                 rc = ldlm_cli_cancel(&lockh,
3511                         ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
3512                 if (rc < 0) {
3513                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
3514                         RETURN(rc);
3515                 }
3516                 break;
3517         }
3518         case LDLM_CB_CANCELING: {
3519                 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
3520                 struct mdt_device *mdt =
3521                                 mdt_dev(obd->obd_lu_dev->ld_site->ls_top_dev);
3522
3523                 LDLM_DEBUG(lock, "Revoke remote lock\n");
3524
3525                 /* discard slc lock here so that it can be cleaned anytime,
3526                  * especially for cleanup_resource() */
3527                 tgt_discard_slc_lock(&mdt->mdt_lut, lock);
3528
3529                 /* once we cache lock, l_ast_data is set to mdt_object */
3530                 if (lock->l_ast_data != NULL) {
3531                         struct mdt_object *mo = lock->l_ast_data;
3532                         struct lu_env env;
3533
3534                         rc = lu_env_init(&env, LCT_MD_THREAD);
3535                         if (unlikely(rc != 0)) {
3536                                 CWARN("%s: lu_env initialization failed, object %p "DFID" is leaked!: rc = %d\n",
3537                                       obd->obd_name, mo,
3538                                       PFID(mdt_object_fid(mo)), rc);
3539                                 RETURN(rc);
3540                         }
3541
3542                         if (lock->l_policy_data.l_inodebits.bits &
3543                             (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)) {
3544                                 rc = mo_invalidate(&env, mdt_object_child(mo));
3545                                 mo->mot_cache_attr = 0;
3546                         }
3547                         mdt_object_put(&env, mo);
3548                         lu_env_fini(&env);
3549                 }
3550                 break;
3551         }
3552         default:
3553                 LBUG();
3554         }
3555
3556         RETURN(rc);
3557 }
3558
3559 int mdt_check_resent_lock(struct mdt_thread_info *info,
3560                           struct mdt_object *mo,
3561                           struct mdt_lock_handle *lhc)
3562 {
3563         /* the lock might already be gotten in ldlm_handle_enqueue() */
3564         if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
3565                 struct ptlrpc_request *req = mdt_info_req(info);
3566                 struct ldlm_lock      *lock;
3567
3568                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
3569                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
3570                 if (lock == NULL) {
3571                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
3572                          * a resend case, however, it could be already destroyed
3573                          * due to client eviction or a raced cancel RPC. */
3574                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx",
3575                                           lhc->mlh_reg_lh.cookie);
3576                         RETURN(-ESTALE);
3577                 }
3578
3579                 if (!fid_res_name_eq(mdt_object_fid(mo),
3580                                      &lock->l_resource->lr_name)) {
3581                         CWARN("%s: Although resent, but still not "
3582                               "get child lock:"DFID"\n",
3583                               info->mti_exp->exp_obd->obd_name,
3584                               PFID(mdt_object_fid(mo)));
3585                         LDLM_LOCK_PUT(lock);
3586                         RETURN(-EPROTO);
3587                 }
3588                 LDLM_LOCK_PUT(lock);
3589                 return 0;
3590         }
3591         return 1;
3592 }
3593
3594 static void mdt_remote_object_lock_created_cb(struct ldlm_lock *lock)
3595 {
3596         mdt_object_get(NULL, lock->l_ast_data);
3597 }
3598
3599 int mdt_remote_object_lock_try(struct mdt_thread_info *mti,
3600                                struct mdt_object *o, const struct lu_fid *fid,
3601                                struct lustre_handle *lh, enum ldlm_mode mode,
3602                                __u64 *ibits, __u64 trybits, bool cache)
3603 {
3604         struct ldlm_enqueue_info *einfo = &mti->mti_remote_einfo;
3605         union ldlm_policy_data *policy = &mti->mti_policy;
3606         struct ldlm_res_id *res_id = &mti->mti_res_id;
3607         int rc = 0;
3608         ENTRY;
3609
3610         LASSERT(mdt_object_remote(o));
3611
3612         fid_build_reg_res_name(fid, res_id);
3613
3614         memset(einfo, 0, sizeof(*einfo));
3615         einfo->ei_type = LDLM_IBITS;
3616         einfo->ei_mode = mode;
3617         einfo->ei_cb_bl = mdt_remote_blocking_ast;
3618         einfo->ei_cb_cp = ldlm_completion_ast;
3619         einfo->ei_enq_slave = 0;
3620         einfo->ei_res_id = res_id;
3621
3622         if (cache) {
3623                 /*
3624                  * if we cache lock, couple lock with mdt_object, so that object
3625                  * can be easily found in lock ASTs.
3626                  */
3627                 einfo->ei_cbdata = o;
3628                 einfo->ei_cb_created = mdt_remote_object_lock_created_cb;
3629         }
3630
3631         memset(policy, 0, sizeof(*policy));
3632         policy->l_inodebits.bits = *ibits;
3633         policy->l_inodebits.try_bits = trybits;
3634
3635         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
3636                             policy);
3637
3638         /* Return successfully acquired bits to a caller */
3639         if (rc == 0) {
3640                 struct ldlm_lock *lock = ldlm_handle2lock(lh);
3641
3642                 LASSERT(lock);
3643                 *ibits = lock->l_policy_data.l_inodebits.bits;
3644                 LDLM_LOCK_PUT(lock);
3645         }
3646         RETURN(rc);
3647 }
3648
3649 int mdt_remote_object_lock(struct mdt_thread_info *mti, struct mdt_object *o,
3650                            const struct lu_fid *fid, struct lustre_handle *lh,
3651                            enum ldlm_mode mode, __u64 ibits, bool cache)
3652 {
3653         return mdt_remote_object_lock_try(mti, o, fid, lh, mode, &ibits, 0,
3654                                           cache);
3655 }
3656
3657 int mdt_object_local_lock(struct mdt_thread_info *info, struct mdt_object *o,
3658                           struct mdt_lock_handle *lh, __u64 *ibits,
3659                           __u64 trybits, bool cos_incompat)
3660 {
3661         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
3662         union ldlm_policy_data *policy = &info->mti_policy;
3663         struct ldlm_res_id *res_id = &info->mti_res_id;
3664         __u64 dlmflags = 0, *cookie = NULL;
3665         int rc;
3666         ENTRY;
3667
3668         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
3669         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
3670         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
3671         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
3672
3673         if (cos_incompat) {
3674                 LASSERT(lh->mlh_reg_mode == LCK_PW ||
3675                         lh->mlh_reg_mode == LCK_EX);
3676                 dlmflags |= LDLM_FL_COS_INCOMPAT;
3677         } else if (mdt_cos_is_enabled(info->mti_mdt)) {
3678                 dlmflags |= LDLM_FL_COS_ENABLED;
3679         }
3680
3681         /* Only enqueue LOOKUP lock for remote object */
3682         LASSERT(ergo(mdt_object_remote(o), *ibits == MDS_INODELOCK_LOOKUP));
3683
3684         /* Lease lock are granted with LDLM_FL_CANCEL_ON_BLOCK */
3685         if (lh->mlh_type == MDT_REG_LOCK && lh->mlh_reg_mode == LCK_EX &&
3686             *ibits == MDS_INODELOCK_OPEN)
3687                 dlmflags |= LDLM_FL_CANCEL_ON_BLOCK;
3688
3689         if (lh->mlh_type == MDT_PDO_LOCK) {
3690                 /* check for exists after object is locked */
3691                 if (mdt_object_exists(o) == 0) {
3692                         /* Non-existent object shouldn't have PDO lock */
3693                         RETURN(-ESTALE);
3694                 } else {
3695                         /* Non-dir object shouldn't have PDO lock */
3696                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
3697                                 RETURN(-ENOTDIR);
3698                 }
3699         }
3700
3701         fid_build_reg_res_name(mdt_object_fid(o), res_id);
3702         dlmflags |= LDLM_FL_ATOMIC_CB;
3703
3704         if (info->mti_exp)
3705                 cookie = &info->mti_exp->exp_handle.h_cookie;
3706
3707         /*
3708          * Take PDO lock on whole directory and build correct @res_id for lock
3709          * on part of directory.
3710          */
3711         if (lh->mlh_pdo_hash != 0) {
3712                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
3713                 mdt_lock_pdo_mode(info, o, lh);
3714                 if (lh->mlh_pdo_mode != LCK_NL) {
3715                         /*
3716                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
3717                          * is never going to be sent to client and we do not
3718                          * want it slowed down due to possible cancels.
3719                          */
3720                         policy->l_inodebits.bits =
3721                                 *ibits & MDS_INODELOCK_UPDATE;
3722                         policy->l_inodebits.try_bits =
3723                                 trybits & MDS_INODELOCK_UPDATE;
3724                         /* at least one of them should be set */
3725                         LASSERT(policy->l_inodebits.bits |
3726                                 policy->l_inodebits.try_bits);
3727                         rc = mdt_fid_lock(info->mti_env, ns, &lh->mlh_pdo_lh,
3728                                           lh->mlh_pdo_mode, policy, res_id,
3729                                           dlmflags, cookie);
3730                         if (unlikely(rc != 0))
3731                                 GOTO(out_unlock, rc);
3732                 }
3733
3734                 /*
3735                  * Finish res_id initializing by name hash marking part of
3736                  * directory which is taking modification.
3737                  */
3738                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
3739         }
3740
3741         policy->l_inodebits.bits = *ibits;
3742         policy->l_inodebits.try_bits = trybits;
3743         policy->l_inodebits.li_gid = lh->mlh_gid;
3744
3745         /*
3746          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
3747          * going to be sent to client. If it is - mdt_intent_policy() path will
3748          * fix it up and turn FL_LOCAL flag off.
3749          */
3750         rc = mdt_fid_lock(info->mti_env, ns, &lh->mlh_reg_lh, lh->mlh_reg_mode,
3751                           policy, res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
3752                           cookie);
3753 out_unlock:
3754         if (rc != 0)
3755                 mdt_object_unlock(info, o, lh, 1);
3756         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
3757                    lh->mlh_pdo_hash != 0 &&
3758                    (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
3759                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
3760
3761         /* Return successfully acquired bits to a caller */
3762         if (rc == 0) {
3763                 struct ldlm_lock *lock = ldlm_handle2lock(&lh->mlh_reg_lh);
3764
3765                 LASSERT(lock);
3766                 *ibits = lock->l_policy_data.l_inodebits.bits;
3767                 LDLM_LOCK_PUT(lock);
3768         }
3769         RETURN(rc);
3770 }
3771
3772 static int
3773 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
3774                          struct mdt_lock_handle *lh, __u64 *ibits,
3775                          __u64 trybits, bool cos_incompat)
3776 {
3777         struct mdt_lock_handle *local_lh = NULL;
3778         int rc;
3779         ENTRY;
3780
3781         if (!mdt_object_remote(o)) {
3782                 rc = mdt_object_local_lock(info, o, lh, ibits, trybits,
3783                                            cos_incompat);
3784                 RETURN(rc);
3785         }
3786
3787         /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
3788         *ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
3789                     MDS_INODELOCK_XATTR);
3790
3791         /* Only enqueue LOOKUP lock for remote object */
3792         if (*ibits & MDS_INODELOCK_LOOKUP) {
3793                 __u64 local = MDS_INODELOCK_LOOKUP;
3794
3795                 rc = mdt_object_local_lock(info, o, lh, &local, 0,
3796                                            cos_incompat);
3797                 if (rc != ELDLM_OK)
3798                         RETURN(rc);
3799
3800                 local_lh = lh;
3801         }
3802
3803         if ((*ibits | trybits) & MDS_INODELOCK_UPDATE) {
3804                 /* Sigh, PDO needs to enqueue 2 locks right now, but
3805                  * enqueue RPC can only request 1 lock, to avoid extra
3806                  * RPC, so it will instead enqueue EX lock for remote
3807                  * object anyway XXX*/
3808                 if (lh->mlh_type == MDT_PDO_LOCK &&
3809                     lh->mlh_pdo_hash != 0) {
3810                         CDEBUG(D_INFO,
3811                                "%s: "DFID" convert PDO lock to EX lock.\n",
3812                                mdt_obd_name(info->mti_mdt),
3813                                PFID(mdt_object_fid(o)));
3814                         lh->mlh_pdo_hash = 0;
3815                         lh->mlh_rreg_mode = LCK_EX;
3816                         lh->mlh_type = MDT_REG_LOCK;
3817                 }
3818
3819                 rc = mdt_remote_object_lock_try(info, o, mdt_object_fid(o),
3820                                                 &lh->mlh_rreg_lh,
3821                                                 lh->mlh_rreg_mode,
3822                                                 ibits, trybits, false);
3823                 if (rc != ELDLM_OK) {
3824                         if (local_lh != NULL)
3825                                 mdt_object_unlock(info, o, local_lh, rc);
3826                         RETURN(rc);
3827                 }
3828         }
3829
3830         /* other components like LFSCK can use lockless access
3831          * and populate cache, so we better invalidate it */
3832         mo_invalidate(info->mti_env, mdt_object_child(o));
3833
3834         RETURN(0);
3835 }
3836
3837 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
3838                     struct mdt_lock_handle *lh, __u64 ibits)
3839 {
3840         return mdt_object_lock_internal(info, o, lh, &ibits, 0, false);
3841 }
3842
3843 int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
3844                           struct mdt_lock_handle *lh, __u64 ibits,
3845                           bool cos_incompat)
3846 {
3847         LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
3848         return mdt_object_lock_internal(info, o, lh, &ibits, 0,
3849                                         cos_incompat);
3850 }
3851
3852 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
3853                         struct mdt_lock_handle *lh, __u64 *ibits,
3854                         __u64 trybits, bool cos_incompat)
3855 {
3856         bool trylock_only = *ibits == 0;
3857         int rc;
3858
3859         LASSERT(!(*ibits & trybits));
3860         rc = mdt_object_lock_internal(info, o, lh, ibits, trybits,
3861                                       cos_incompat);
3862         if (rc && trylock_only) { /* clear error for try ibits lock only */
3863                 LASSERT(*ibits == 0);
3864                 rc = 0;
3865         }
3866         return rc;
3867 }
3868
3869 /**
3870  * Save a lock within request object.
3871  *
3872  * Keep the lock referenced until whether client ACK or transaction
3873  * commit happens or release the lock immediately depending on input
3874  * parameters. If COS is ON, a write lock is converted to COS lock
3875  * before saving.
3876  *
3877  * \param info thead info object
3878  * \param h lock handle
3879  * \param mode lock mode
3880  * \param decref force immediate lock releasing
3881  */
3882 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
3883                    enum ldlm_mode mode, int decref)
3884 {
3885         ENTRY;
3886
3887         if (lustre_handle_is_used(h)) {
3888                 if (decref || !info->mti_has_trans ||
3889                     !(mode & (LCK_PW | LCK_EX))) {
3890                         mdt_fid_unlock(h, mode);
3891                 } else {
3892                         struct mdt_device *mdt = info->mti_mdt;
3893                         struct ldlm_lock *lock = ldlm_handle2lock(h);
3894                         struct ptlrpc_request *req = mdt_info_req(info);
3895                         bool cos = mdt_cos_is_enabled(mdt);
3896                         bool convert_lock = !cos && mdt_slc_is_enabled(mdt);
3897
3898                         LASSERTF(lock != NULL, "no lock for cookie %#llx\n",
3899                                  h->cookie);
3900
3901                         /* there is no request if mdt_object_unlock() is called
3902                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
3903                         if (likely(req != NULL)) {
3904                                 LDLM_DEBUG(lock, "save lock request %p reply "
3905                                         "state %p transno %lld\n", req,
3906                                         req->rq_reply_state, req->rq_transno);
3907                                 if (cos) {
3908                                         ldlm_lock_mode_downgrade(lock, LCK_COS);
3909                                         mode = LCK_COS;
3910                                 }
3911                                 if (req->rq_export->exp_disconnected)
3912                                         mdt_fid_unlock(h, mode);
3913                                 else
3914                                         ptlrpc_save_lock(req, h, mode, cos,
3915                                                          convert_lock);
3916                         } else {
3917                                 mdt_fid_unlock(h, mode);
3918                         }
3919                         if (mdt_is_lock_sync(lock)) {
3920                                 CDEBUG(D_HA, "found sync-lock,"
3921                                        " async commit started\n");
3922                                 mdt_device_commit_async(info->mti_env,
3923                                                         mdt);
3924                         }
3925                         LDLM_LOCK_PUT(lock);
3926                 }
3927                 h->cookie = 0ull;
3928         }
3929
3930         EXIT;
3931 }
3932
3933 /**
3934  * Save cross-MDT lock in uncommitted_slc_locks
3935  *
3936  * Keep the lock referenced until transaction commit happens or release the lock
3937  * immediately depending on input parameters.
3938  *
3939  * \param info thead info object
3940  * \param h lock handle
3941  * \param mode lock mode
3942  * \param decref force immediate lock releasing
3943  */
3944 static void mdt_save_remote_lock(struct mdt_thread_info *info,
3945                                  struct mdt_object *o, struct lustre_handle *h,
3946                                  enum ldlm_mode mode, int decref)
3947 {
3948         ENTRY;
3949
3950         if (lustre_handle_is_used(h)) {
3951                 struct ldlm_lock *lock = ldlm_handle2lock(h);
3952                 struct ptlrpc_request *req = mdt_info_req(info);
3953
3954                 if (o != NULL &&
3955                     (lock->l_policy_data.l_inodebits.bits &
3956                      (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)))
3957                         mo_invalidate(info->mti_env, mdt_object_child(o));
3958
3959                 if (decref || !info->mti_has_trans || !req ||
3960                     !(mode & (LCK_PW | LCK_EX))) {
3961                         ldlm_lock_decref_and_cancel(h, mode);
3962                         LDLM_LOCK_PUT(lock);
3963                 } else {
3964                         tgt_save_slc_lock(&info->mti_mdt->mdt_lut, lock,
3965                                           req->rq_transno);
3966                         ldlm_lock_decref(h, mode);
3967                 }
3968                 h->cookie = 0ull;
3969         }
3970
3971         EXIT;
3972 }
3973
3974 /**
3975  * Unlock mdt object.
3976  *
3977  * Immeditely release the regular lock and the PDO lock or save the
3978  * lock in request and keep them referenced until client ACK or
3979  * transaction commit.
3980  *
3981  * \param info thread info object
3982  * \param o mdt object
3983  * \param lh mdt lock handle referencing regular and PDO locks
3984  * \param decref force immediate lock releasing
3985  *
3986  * XXX o is not used and may be NULL, see hsm_cdt_request_completed().
3987  */
3988 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
3989                        struct mdt_lock_handle *lh, int decref)
3990 {
3991         ENTRY;
3992
3993         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
3994         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
3995         mdt_save_remote_lock(info, o, &lh->mlh_rreg_lh, lh->mlh_rreg_mode,
3996                              decref);
3997
3998         EXIT;
3999 }
4000
4001 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
4002                                         const struct lu_fid *f,
4003                                         struct mdt_lock_handle *lh,
4004                                         __u64 ibits)
4005 {
4006         struct mdt_object *o;
4007
4008         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
4009         if (!IS_ERR(o)) {
4010                 int rc;
4011
4012                 rc = mdt_object_lock(info, o, lh, ibits);
4013                 if (rc != 0) {
4014                         mdt_object_put(info->mti_env, o);
4015                         o = ERR_PTR(rc);
4016                 }
4017         }
4018         return o;
4019 }
4020
4021 void mdt_object_unlock_put(struct mdt_thread_info * info,
4022                            struct mdt_object * o,
4023                            struct mdt_lock_handle *lh,
4024                            int decref)
4025 {
4026         mdt_object_unlock(info, o, lh, decref);
4027         mdt_object_put(info->mti_env, o);
4028 }
4029
4030 /*
4031  * Generic code handling requests that have struct mdt_body passed in:
4032  *
4033  *  - extract mdt_body from request and save it in @info, if present;
4034  *
4035  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
4036  *  @info;
4037  *
4038  *  - if HAS_BODY flag is set for this request type check whether object
4039  *  actually exists on storage (lu_object_exists()).
4040  *
4041  */
4042 static int mdt_body_unpack(struct mdt_thread_info *info,
4043                            enum tgt_handler_flags flags)
4044 {
4045         const struct mdt_body    *body;
4046         struct mdt_object        *obj;
4047         const struct lu_env      *env;
4048         struct req_capsule       *pill;
4049         int                       rc;
4050         ENTRY;
4051
4052         env = info->mti_env;
4053         pill = info->mti_pill;
4054
4055         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
4056         if (body == NULL)
4057                 RETURN(-EFAULT);
4058
4059         if (!(body->mbo_valid & OBD_MD_FLID))
4060                 RETURN(0);
4061
4062         if (!fid_is_sane(&body->mbo_fid1)) {
4063                 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
4064                 RETURN(-EINVAL);
4065         }
4066
4067         obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
4068         if (!IS_ERR(obj)) {
4069                 if ((flags & HAS_BODY) && !mdt_object_exists(obj)) {
4070                         mdt_object_put(env, obj);
4071                         rc = -ENOENT;
4072                 } else {
4073                         info->mti_object = obj;
4074                         rc = 0;
4075                 }
4076         } else
4077                 rc = PTR_ERR(obj);
4078
4079         RETURN(rc);
4080 }
4081
4082 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info,
4083                                    enum tgt_handler_flags flags)
4084 {
4085         struct req_capsule *pill = info->mti_pill;
4086         int rc;
4087
4088         ENTRY;
4089
4090         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
4091                 rc = mdt_body_unpack(info, flags);
4092         else
4093                 rc = 0;
4094
4095         if (rc == 0 && (flags & HAS_REPLY)) {
4096                 /* Pack reply. */
4097                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
4098                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
4099                                              DEF_REP_MD_SIZE);
4100                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
4101                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
4102                                              RCL_SERVER, 0);
4103
4104                 /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
4105                  * by default. If the target object has more ACL entries, then
4106                  * enlarge the buffer when necessary. */
4107                 if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
4108                         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
4109                                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
4110
4111                 mdt_preset_secctx_size(info);
4112                 mdt_preset_encctx_size(info);
4113
4114                 rc = req_capsule_server_pack(pill);
4115                 if (rc)
4116                         CWARN("%s: cannot pack response: rc = %d\n",
4117                                       mdt_obd_name(info->mti_mdt), rc);
4118         }
4119         RETURN(rc);
4120 }
4121
4122 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
4123 {
4124         lh->mlh_type = MDT_NUL_LOCK;
4125         lh->mlh_reg_lh.cookie = 0ull;
4126         lh->mlh_reg_mode = LCK_MINMODE;
4127         lh->mlh_pdo_lh.cookie = 0ull;
4128         lh->mlh_pdo_mode = LCK_MINMODE;
4129         lh->mlh_rreg_lh.cookie = 0ull;
4130         lh->mlh_rreg_mode = LCK_MINMODE;
4131 }
4132
4133 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
4134 {
4135         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
4136         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
4137 }
4138
4139 /*
4140  * Initialize fields of struct mdt_thread_info. Other fields are left in
4141  * uninitialized state, because it's too expensive to zero out whole
4142  * mdt_thread_info (> 1K) on each request arrival.
4143  */
4144 void mdt_thread_info_init(struct ptlrpc_request *req,
4145                           struct mdt_thread_info *info)
4146 {
4147         int i;
4148
4149         info->mti_pill = &req->rq_pill;
4150
4151         /* lock handle */
4152         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
4153                 mdt_lock_handle_init(&info->mti_lh[i]);
4154
4155         /* mdt device: it can be NULL while CONNECT */
4156         if (req->rq_export) {
4157                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
4158                 info->mti_exp = req->rq_export;
4159         } else
4160                 info->mti_mdt = NULL;
4161         info->mti_env = req->rq_svc_thread->t_env;
4162         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
4163
4164         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
4165         info->mti_big_buf = LU_BUF_NULL;
4166         info->mti_body = NULL;
4167         info->mti_object = NULL;
4168         info->mti_dlm_req = NULL;
4169         info->mti_has_trans = 0;
4170         info->mti_cross_ref = 0;
4171         info->mti_opdata = 0;
4172         info->mti_big_lmm_used = 0;
4173         info->mti_big_acl_used = 0;
4174         info->mti_som_valid = 0;
4175
4176         info->mti_spec.no_create = 0;
4177         info->mti_spec.sp_rm_entry = 0;
4178         info->mti_spec.sp_permitted = 0;
4179
4180         info->mti_spec.u.sp_ea.eadata = NULL;
4181         info->mti_spec.u.sp_ea.eadatalen = 0;
4182 }
4183
4184 void mdt_thread_info_fini(struct mdt_thread_info *info)
4185 {
4186         int i;
4187
4188         if (info->mti_object != NULL) {
4189                 mdt_object_put(info->mti_env, info->mti_object);
4190                 info->mti_object = NULL;
4191         }
4192
4193         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
4194                 mdt_lock_handle_fini(&info->mti_lh[i]);
4195         info->mti_env = NULL;
4196         info->mti_pill = NULL;
4197         info->mti_exp = NULL;
4198         info->mti_mdt = NULL;
4199
4200         if (unlikely(info->mti_big_buf.lb_buf != NULL))
4201                 lu_buf_free(&info->mti_big_buf);
4202 }
4203
4204 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
4205 {
4206         struct mdt_thread_info  *mti;
4207
4208         mti = mdt_th_info(tsi->tsi_env);
4209         LASSERT(mti != NULL);
4210
4211         mdt_thread_info_init(tgt_ses_req(tsi), mti);
4212         if (tsi->tsi_corpus != NULL) {
4213                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
4214                 lu_object_get(tsi->tsi_corpus);
4215         }
4216         mti->mti_body = tsi->tsi_mdt_body;
4217         mti->mti_dlm_req = tsi->tsi_dlm_req;
4218
4219         return mti;
4220 }
4221
4222 static int mdt_tgt_connect(struct tgt_session_info *tsi)
4223 {
4224         if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
4225             cfs_fail_val ==
4226             tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id)
4227                 schedule_timeout_uninterruptible(cfs_time_seconds(3));
4228
4229         return tgt_connect(tsi);
4230 }
4231
4232 static int mdt_intent_glimpse(enum ldlm_intent_flags it_opc,
4233                               struct mdt_thread_info *info,
4234                               struct ldlm_lock **lockp, __u64 flags)
4235 {
4236         return mdt_glimpse_enqueue(info, info->mti_mdt->mdt_namespace,
4237                                    lockp, flags);
4238 }
4239 static int mdt_intent_brw(enum ldlm_intent_flags it_opc,
4240                           struct mdt_thread_info *info,
4241                           struct ldlm_lock **lockp, __u64 flags)
4242 {
4243         return mdt_brw_enqueue(info, info->mti_mdt->mdt_namespace,
4244                                lockp, flags);
4245 }
4246
4247 int mdt_intent_lock_replace(struct mdt_thread_info *info,
4248                             struct ldlm_lock **lockp,
4249                             struct mdt_lock_handle *lh,
4250                             __u64 flags, int result)
4251 {
4252         struct ptlrpc_request  *req = mdt_info_req(info);
4253         struct ldlm_lock       *lock = *lockp;
4254         struct ldlm_lock       *new_lock;
4255
4256         /* If possible resent found a lock, @lh is set to its handle */
4257         new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
4258
4259         if (new_lock == NULL) {
4260                 if (flags & LDLM_FL_INTENT_ONLY) {
4261                         result = 0;
4262                 } else if (flags & LDLM_FL_RESENT) {
4263                         /* Lock is pinned by ldlm_handle_enqueue0() as it is a
4264                          * resend case, however, it could be already destroyed
4265                          * due to client eviction or a raced cancel RPC.
4266                          */
4267                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx\n",
4268                                           lh->mlh_reg_lh.cookie);
4269                         result = -ESTALE;
4270                 } else {
4271                         CERROR("%s: Invalid lockh=%#llx flags=%#llx fid1="DFID" fid2="DFID": rc = %d\n",
4272                                mdt_obd_name(info->mti_mdt),
4273                                lh->mlh_reg_lh.cookie, flags,
4274                                PFID(&info->mti_tmp_fid1),
4275                                PFID(&info->mti_tmp_fid2), result);
4276                         result = -ESTALE;
4277                 }
4278                 lh->mlh_reg_lh.cookie = 0;
4279                 RETURN(result);
4280         }
4281
4282         /*
4283          * If we've already given this lock to a client once, then we should
4284          * have no readers or writers.  Otherwise, we should have one reader
4285          * _or_ writer ref (which will be zeroed below) before returning the
4286          * lock to a client.
4287          */
4288         if (new_lock->l_export == req->rq_export) {
4289                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
4290         } else {
4291                 LASSERT(new_lock->l_export == NULL);
4292                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
4293         }
4294
4295         *lockp = new_lock;
4296
4297         if (new_lock->l_export == req->rq_export) {
4298                 /*
4299                  * Already gave this to the client, which means that we
4300                  * reconstructed a reply.
4301                  */
4302                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
4303                         MSG_RESENT);
4304
4305                 LDLM_LOCK_RELEASE(new_lock);
4306                 lh->mlh_reg_lh.cookie = 0;
4307                 RETURN(ELDLM_LOCK_REPLACED);
4308         }
4309
4310         /*
4311          * Fixup the lock to be given to the client.
4312          */
4313         lock_res_and_lock(new_lock);
4314         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
4315          * possible blocking AST. */
4316         while (new_lock->l_readers > 0) {
4317                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
4318                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
4319                 new_lock->l_readers--;
4320         }
4321         while (new_lock->l_writers > 0) {
4322                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
4323                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
4324                 new_lock->l_writers--;
4325         }
4326
4327         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
4328         new_lock->l_blocking_ast = lock->l_blocking_ast;
4329         new_lock->l_completion_ast = lock->l_completion_ast;
4330         if (ldlm_has_dom(new_lock))
4331                 new_lock->l_glimpse_ast = ldlm_server_glimpse_ast;
4332         new_lock->l_remote_handle = lock->l_remote_handle;
4333         new_lock->l_flags &= ~LDLM_FL_LOCAL;
4334
4335         unlock_res_and_lock(new_lock);
4336
4337         cfs_hash_add(new_lock->l_export->exp_lock_hash,
4338                      &new_lock->l_remote_handle,
4339                      &new_lock->l_exp_hash);
4340
4341         LDLM_LOCK_RELEASE(new_lock);
4342         lh->mlh_reg_lh.cookie = 0;
4343
4344         RETURN(ELDLM_LOCK_REPLACED);
4345 }
4346
4347 void mdt_intent_fixup_resent(struct mdt_thread_info *info,
4348                              struct ldlm_lock *new_lock,
4349                              struct mdt_lock_handle *lh, __u64 flags)
4350 {
4351         struct ptlrpc_request  *req = mdt_info_req(info);
4352         struct ldlm_request    *dlmreq;
4353
4354         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
4355                 return;
4356
4357         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
4358
4359         /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
4360          * lock was found by ldlm_handle_enqueue(); if so @lh must be
4361          * initialized. */
4362         if (flags & LDLM_FL_RESENT) {
4363                 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
4364                 lh->mlh_reg_mode = new_lock->l_granted_mode;
4365
4366                 LDLM_DEBUG(new_lock, "Restoring lock cookie");
4367                 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie %#llx",
4368                           lh->mlh_reg_lh.cookie);
4369                 return;
4370         }
4371
4372         /*
4373          * If the xid matches, then we know this is a resent request, and allow
4374          * it. (It's probably an OPEN, for which we don't send a lock.
4375          */
4376         if (req_can_reconstruct(req, NULL) != 0)
4377                 return;
4378
4379         /*
4380          * This remote handle isn't enqueued, so we never received or processed
4381          * this request.  Clear MSG_RESENT, because it can be handled like any
4382          * normal request now.
4383          */
4384         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
4385
4386         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle %#llx",
4387                   dlmreq->lock_handle[0].cookie);
4388 }
4389
4390 static int mdt_intent_getxattr(enum ldlm_intent_flags it_opc,
4391                                struct mdt_thread_info *info,
4392                                struct ldlm_lock **lockp,
4393                                __u64 flags)
4394 {
4395         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4396         struct ldlm_reply      *ldlm_rep = NULL;
4397         int rc;
4398         ENTRY;
4399
4400         /*
4401          * Initialize lhc->mlh_reg_lh either from a previously granted lock
4402          * (for the resend case) or a new lock. Below we will use it to
4403          * replace the original lock.
4404          */
4405         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4406         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
4407                 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
4408                 rc = mdt_object_lock(info, info->mti_object, lhc,
4409                                      MDS_INODELOCK_XATTR);
4410                 if (rc)
4411                         return rc;
4412         }
4413
4414         rc = mdt_getxattr(info);
4415
4416         if (mdt_info_req(info)->rq_repmsg != NULL)
4417                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4418
4419         if (ldlm_rep == NULL ||
4420             OBD_FAIL_CHECK(OBD_FAIL_MDS_XATTR_REP)) {
4421                 mdt_object_unlock(info,  info->mti_object, lhc, 1);
4422                 if (is_serious(rc))
4423                         RETURN(rc);
4424                 else
4425                         RETURN(err_serious(-EFAULT));
4426         }
4427
4428         ldlm_rep->lock_policy_res2 = clear_serious(rc);
4429
4430         /* This is left for interop instead of adding a new interop flag.
4431          * LU-7433 */
4432 #if LUSTRE_VERSION_CODE > OBD_OCD_VERSION(3, 0, 0, 0)
4433         if (ldlm_rep->lock_policy_res2) {
4434                 mdt_object_unlock(info, info->mti_object, lhc, 1);
4435                 RETURN(ELDLM_LOCK_ABORTED);
4436         }
4437 #endif
4438
4439         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4440         RETURN(rc);
4441 }
4442
4443 static int mdt_intent_getattr(enum ldlm_intent_flags it_opc,
4444                               struct mdt_thread_info *info,
4445                               struct ldlm_lock **lockp,
4446                               __u64 flags)
4447 {
4448         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4449         __u64                   child_bits;
4450         struct ldlm_reply      *ldlm_rep;
4451         struct mdt_body        *reqbody;
4452         struct mdt_body        *repbody;
4453         int                     rc, rc2;
4454         ENTRY;
4455
4456         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
4457         LASSERT(reqbody);
4458
4459         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
4460         LASSERT(repbody);
4461
4462         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
4463         repbody->mbo_eadatasize = 0;
4464         repbody->mbo_aclsize = 0;
4465
4466         switch (it_opc) {
4467         case IT_LOOKUP:
4468                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
4469                 break;
4470         case IT_GETATTR:
4471                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
4472                              MDS_INODELOCK_PERM;
4473                 break;
4474         default:
4475                 CERROR("%s: unsupported intent %#x\n",
4476                        mdt_obd_name(info->mti_mdt), (unsigned int)it_opc);
4477                 GOTO(out_shrink, rc = -EINVAL);
4478         }
4479
4480         rc = mdt_init_ucred(info, reqbody);
4481         if (rc)
4482                 GOTO(out_shrink, rc);
4483
4484         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4485         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
4486
4487         /* Get lock from request for possible resent case. */
4488         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4489
4490         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
4491         ldlm_rep->lock_policy_res2 = clear_serious(rc);
4492
4493         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
4494                 ldlm_rep->lock_policy_res2 = 0;
4495         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
4496             ldlm_rep->lock_policy_res2) {
4497                 lhc->mlh_reg_lh.cookie = 0ull;
4498                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
4499         }
4500
4501         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4502         EXIT;
4503 out_ucred:
4504         mdt_exit_ucred(info);
4505 out_shrink:
4506         mdt_client_compatibility(info);
4507         rc2 = mdt_fix_reply(info);
4508         if (rc == 0)
4509                 rc = rc2;
4510         return rc;
4511 }
4512
4513 static int mdt_intent_layout(enum ldlm_intent_flags it_opc,
4514                              struct mdt_thread_info *info,
4515                              struct ldlm_lock **lockp,
4516                              __u64 flags)
4517 {
4518         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4519         struct md_layout_change layout = { .mlc_opc = MD_LAYOUT_NOP };
4520         struct layout_intent *intent;
4521         struct ldlm_reply *ldlm_rep;
4522         struct lu_fid *fid = &info->mti_tmp_fid2;
4523         struct mdt_object *obj = NULL;
4524         int layout_size = 0;
4525         struct lu_buf *buf = &layout.mlc_buf;
4526         int rc = 0;
4527
4528         ENTRY;
4529
4530         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
4531
4532         intent = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
4533         if (intent == NULL)
4534                 RETURN(-EPROTO);
4535
4536         CDEBUG(D_INFO, DFID "got layout change request from client: "
4537                "opc:%u flags:%#x extent "DEXT"\n",
4538                PFID(fid), intent->li_opc, intent->li_flags,
4539                PEXT(&intent->li_extent));
4540
4541         switch (intent->li_opc) {
4542         case LAYOUT_INTENT_TRUNC:
4543         case LAYOUT_INTENT_WRITE:
4544                 layout.mlc_opc = MD_LAYOUT_WRITE;
4545                 layout.mlc_intent = intent;
4546                 break;
4547         case LAYOUT_INTENT_ACCESS:
4548                 break;
4549         case LAYOUT_INTENT_READ:
4550         case LAYOUT_INTENT_GLIMPSE:
4551         case LAYOUT_INTENT_RELEASE:
4552         case LAYOUT_INTENT_RESTORE:
4553                 CERROR("%s: Unsupported layout intent opc %d\n",
4554                        mdt_obd_name(info->mti_mdt), intent->li_opc);
4555                 RETURN(-ENOTSUPP);
4556         default:
4557                 CERROR("%s: Unknown layout intent opc %d\n",
4558                        mdt_obd_name(info->mti_mdt), intent->li_opc);
4559                 RETURN(-EINVAL);
4560         }
4561
4562         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
4563         if (IS_ERR(obj))
4564                 RETURN(PTR_ERR(obj));
4565
4566         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
4567                 /* if layout is going to be changed don't use the current EA
4568                  * size but the maximum one. That buffer will be shrinked
4569                  * to the actual size in req_capsule_shrink() before reply.
4570                  */
4571                 if (layout.mlc_opc == MD_LAYOUT_WRITE) {
4572                         layout_size = info->mti_mdt->mdt_max_mdsize;
4573                 } else {
4574                         layout_size = mdt_attr_get_eabuf_size(info, obj);
4575                         if (layout_size < 0)
4576                                 GOTO(out, rc = layout_size);
4577
4578                         if (layout_size > info->mti_mdt->mdt_max_mdsize)
4579                                 info->mti_mdt->mdt_max_mdsize = layout_size;
4580                 }
4581                 CDEBUG(D_INFO, "%s: layout_size %d\n",
4582                        mdt_obd_name(info->mti_mdt), layout_size);
4583         }
4584
4585         /*
4586          * set reply buffer size, so that ldlm_handle_enqueue0()->
4587          * ldlm_lvbo_fill() will fill the reply buffer with lovea.
4588          */
4589         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
4590                              layout_size);
4591         rc = req_capsule_server_pack(info->mti_pill);
4592         if (rc)
4593                 GOTO(out, rc);
4594
4595         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4596         if (!ldlm_rep)
4597                 GOTO(out, rc = -EPROTO);
4598
4599         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
4600
4601         /* take lock in ldlm_lock_enqueue() for LAYOUT_INTENT_ACCESS */
4602         if (layout.mlc_opc == MD_LAYOUT_NOP)
4603                 GOTO(out, rc = 0);
4604
4605         rc = mdt_check_resent(info, mdt_reconstruct_generic, lhc);
4606         if (rc < 0)
4607                 GOTO(out, rc);
4608         if (rc == 1) {
4609                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
4610                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
4611                 GOTO(out, rc);
4612         }
4613
4614         buf->lb_buf = NULL;
4615         buf->lb_len = 0;
4616         if (unlikely(req_is_replay(mdt_info_req(info)))) {
4617                 buf->lb_buf = req_capsule_client_get(info->mti_pill,
4618                                                      &RMF_EADATA);
4619                 buf->lb_len = req_capsule_get_size(info->mti_pill,
4620                                                      &RMF_EADATA, RCL_CLIENT);
4621                 /*
4622                  * If it's a replay of layout write intent RPC, the client has
4623                  * saved the extended lovea when it get reply then.
4624                  */
4625                 if (buf->lb_len > 0)
4626                         mdt_fix_lov_magic(info, buf->lb_buf);
4627         }
4628
4629         /* Get lock from request for possible resent case. */
4630         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4631         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
4632
4633         /*
4634          * Instantiate some layout components, if @buf contains lovea, then it's
4635          * a replay of the layout intent write RPC.
4636          */
4637         rc = mdt_layout_change(info, obj, lhc, &layout);
4638         ldlm_rep->lock_policy_res2 = clear_serious(rc);
4639
4640         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
4641                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4642                 if (rc == ELDLM_LOCK_REPLACED &&
4643                     (*lockp)->l_granted_mode == LCK_EX)
4644                         ldlm_lock_mode_downgrade(*lockp, LCK_CR);
4645         }
4646
4647         EXIT;
4648 out:
4649         mdt_object_put(info->mti_env, obj);
4650         return rc;
4651 }
4652
4653 static int mdt_intent_open(enum ldlm_intent_flags it_opc,
4654                            struct mdt_thread_info *info,
4655                            struct ldlm_lock **lockp,
4656                            __u64 flags)
4657 {
4658         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4659         struct ldlm_reply      *rep = NULL;
4660         long                    opc;
4661         int                     rc;
4662         struct ptlrpc_request  *req = mdt_info_req(info);
4663
4664         static const struct req_format *intent_fmts[REINT_MAX] = {
4665                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
4666                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
4667         };
4668
4669         ENTRY;
4670
4671         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
4672         if (opc < 0)
4673                 RETURN(opc);
4674
4675         /* Get lock from request for possible resent case. */
4676         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4677
4678         rc = mdt_reint_internal(info, lhc, opc);
4679
4680         if (rc < 0 && lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
4681                 DEBUG_REQ(D_ERROR, req, "Replay open failed with %d", rc);
4682
4683         /* Check whether the reply has been packed successfully. */
4684         if (mdt_info_req(info)->rq_repmsg != NULL)
4685                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4686         if (rep == NULL) {
4687                 if (is_serious(rc))
4688                         RETURN(rc);
4689                 else
4690                         RETURN(err_serious(-EFAULT));
4691         }
4692
4693         /* MDC expects this in any case */
4694         if (rc != 0)
4695                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
4696
4697         /* the open lock or the lock for cross-ref object should be
4698          * returned to the client */
4699         if (lustre_handle_is_used(&lhc->mlh_reg_lh) &&
4700             (rc == 0 || rc == -MDT_EREMOTE_OPEN)) {
4701                 rep->lock_policy_res2 = 0;
4702                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4703                 RETURN(rc);
4704         }
4705
4706         rep->lock_policy_res2 = clear_serious(rc);
4707
4708         if (rep->lock_policy_res2 == -ENOENT &&
4709             mdt_get_disposition(rep, DISP_LOOKUP_NEG) &&
4710             !mdt_get_disposition(rep, DISP_OPEN_CREATE))
4711                 rep->lock_policy_res2 = 0;
4712
4713         lhc->mlh_reg_lh.cookie = 0ull;
4714         if (rc == -ENOTCONN || rc == -ENODEV ||
4715             rc == -EOVERFLOW) { /**< if VBR failure then return error */
4716                 /*
4717                  * If it is the disconnect error (ENODEV & ENOCONN), the error
4718                  * will be returned by rq_status, and client at ptlrpc layer
4719                  * will detect this, then disconnect, reconnect the import
4720                  * immediately, instead of impacting the following the rpc.
4721                  */
4722                 RETURN(rc);
4723         }
4724         /*
4725          * For other cases, the error will be returned by intent, and client
4726          * will retrieve the result from intent.
4727          */
4728         RETURN(ELDLM_LOCK_ABORTED);
4729 }
4730
4731 static int mdt_intent_opc(enum ldlm_intent_flags it_opc,
4732                           struct mdt_thread_info *info,
4733                           struct ldlm_lock **lockp,
4734                           u64 flags /* LDLM_FL_* */)
4735 {
4736         struct req_capsule *pill = info->mti_pill;
4737         struct ptlrpc_request *req = mdt_info_req(info);
4738         const struct req_format *it_format;
4739         int (*it_handler)(enum ldlm_intent_flags,
4740                           struct mdt_thread_info *,
4741                           struct ldlm_lock **,
4742                           u64);
4743         enum tgt_handler_flags it_handler_flags = 0;
4744         struct ldlm_reply *rep;
4745         bool check_mdt_object = false;
4746         int rc;
4747         ENTRY;
4748
4749         switch (it_opc) {
4750         case IT_OPEN:
4751         case IT_OPEN|IT_CREAT:
4752                 /*
4753                  * OCREAT is not a IS_MUTABLE request since the file may
4754                  * already exist. We do the extra check of
4755                  * OBD_CONNECT_RDONLY in mdt_reint_open() when we
4756                  * really need to create the object.
4757                  */
4758                 it_format = &RQF_LDLM_INTENT;
4759                 it_handler = &mdt_intent_open;
4760                 break;
4761         case IT_GETATTR:
4762                 check_mdt_object = true;
4763                 /* fallthrough */
4764         case IT_LOOKUP:
4765                 it_format = &RQF_LDLM_INTENT_GETATTR;
4766                 it_handler = &mdt_intent_getattr;
4767                 it_handler_flags = HAS_REPLY;
4768                 break;
4769         case IT_GETXATTR:
4770                 check_mdt_object = true;
4771                 it_format = &RQF_LDLM_INTENT_GETXATTR;
4772                 it_handler = &mdt_intent_getxattr;
4773                 it_handler_flags = HAS_BODY;
4774                 break;
4775         case IT_LAYOUT:
4776                 it_format = &RQF_LDLM_INTENT_LAYOUT;
4777                 it_handler = &mdt_intent_layout;
4778                 break;
4779         case IT_GLIMPSE:
4780                 it_format = &RQF_LDLM_INTENT;
4781                 it_handler = &mdt_intent_glimpse;
4782                 break;
4783         case IT_BRW:
4784                 it_format = &RQF_LDLM_INTENT;
4785                 it_handler = &mdt_intent_brw;
4786                 break;
4787         case IT_QUOTA_DQACQ:
4788         case IT_QUOTA_CONN: {
4789                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
4790
4791                 if (qmt == NULL)
4792                         RETURN(-EOPNOTSUPP);
4793
4794                 if (mdt_rdonly(req->rq_export))
4795                         RETURN(-EROFS);
4796
4797                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
4798                 /* pass the request to quota master */
4799                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
4800                                                  mdt_info_req(info), lockp,
4801                                                  flags);
4802                 RETURN(rc);
4803         }
4804         default:
4805                 CERROR("%s: unknown intent code %#x\n",
4806                        mdt_obd_name(info->mti_mdt), it_opc);
4807                 RETURN(-EPROTO);
4808         }
4809
4810         req_capsule_extend(pill, it_format);
4811
4812         rc = mdt_unpack_req_pack_rep(info, it_handler_flags);
4813         if (rc < 0)
4814                 RETURN(rc);
4815
4816         if (unlikely(info->mti_object == NULL && check_mdt_object))
4817                 RETURN(-EPROTO);
4818
4819         if (it_handler_flags & IS_MUTABLE && mdt_rdonly(req->rq_export))
4820                 RETURN(-EROFS);
4821
4822         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_INTENT_DELAY, 10);
4823
4824         /* execute policy */
4825         rc = (*it_handler)(it_opc, info, lockp, flags);
4826
4827         /* Check whether the reply has been packed successfully. */
4828         if (req->rq_repmsg != NULL) {
4829                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4830                 rep->lock_policy_res2 =
4831                         ptlrpc_status_hton(rep->lock_policy_res2);
4832         }
4833
4834         RETURN(rc);
4835 }
4836
4837 static void mdt_ptlrpc_stats_update(struct ptlrpc_request *req,
4838                                     enum ldlm_intent_flags it_opc)
4839 {
4840         struct lprocfs_stats *srv_stats = ptlrpc_req2svc(req)->srv_stats;
4841
4842         /* update stats when IT code is known */
4843         if (srv_stats != NULL)
4844                 lprocfs_counter_incr(srv_stats,
4845                                 PTLRPC_LAST_CNTR + (it_opc == IT_GLIMPSE ?
4846                                 LDLM_GLIMPSE_ENQUEUE : LDLM_IBITS_ENQUEUE));
4847 }
4848
4849 static int mdt_intent_policy(const struct lu_env *env,
4850                              struct ldlm_namespace *ns,
4851                              struct ldlm_lock **lockp,
4852                              void *req_cookie,
4853                              enum ldlm_mode mode,
4854                              __u64 flags, void *data)
4855 {
4856         struct tgt_session_info *tsi;
4857         struct mdt_thread_info  *info;
4858         struct ptlrpc_request   *req  =  req_cookie;
4859         struct ldlm_intent      *it;
4860         struct req_capsule      *pill;
4861         const struct ldlm_lock_desc *ldesc;
4862         int rc;
4863
4864         ENTRY;
4865
4866         LASSERT(req != NULL);
4867
4868         tsi = tgt_ses_info(env);
4869
4870         info = tsi2mdt_info(tsi);
4871         LASSERT(info != NULL);
4872         pill = info->mti_pill;
4873         LASSERT(pill->rc_req == req);
4874         ldesc = &info->mti_dlm_req->lock_desc;
4875
4876         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
4877                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
4878                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
4879                 if (it != NULL) {
4880                         mdt_ptlrpc_stats_update(req, it->opc);
4881                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
4882                         if (rc == 0)
4883                                 rc = ELDLM_OK;
4884
4885                         /* Lock without inodebits makes no sense and will oops
4886                          * later in ldlm. Let's check it now to see if we have
4887                          * ibits corrupted somewhere in mdt_intent_opc().
4888                          * The case for client miss to set ibits has been
4889                          * processed by others. */
4890                         LASSERT(ergo(ldesc->l_resource.lr_type == LDLM_IBITS,
4891                                 ldesc->l_policy_data.l_inodebits.bits != 0));
4892                 } else {
4893                         rc = err_serious(-EFAULT);
4894                 }
4895         } else if (ldesc->l_resource.lr_type == LDLM_IBITS &&
4896                    ldesc->l_policy_data.l_inodebits.bits == MDS_INODELOCK_DOM) {
4897                 struct ldlm_reply *rep;
4898
4899                 /* No intent was provided but INTENT flag is set along with
4900                  * DOM bit, this is considered as GLIMPSE request.
4901                  * This logic is common for MDT and OST glimpse
4902                  */
4903                 mdt_ptlrpc_stats_update(req, IT_GLIMPSE);
4904                 rc = mdt_glimpse_enqueue(info, ns, lockp, flags);
4905                 /* Check whether the reply has been packed successfully. */
4906                 if (req->rq_repmsg != NULL) {
4907                         rep = req_capsule_server_get(info->mti_pill,
4908                                                      &RMF_DLM_REP);
4909                         rep->lock_policy_res2 =
4910                                 ptlrpc_status_hton(rep->lock_policy_res2);
4911                 }
4912         } else {
4913                 /* No intent was provided */
4914                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
4915                 rc = req_capsule_server_pack(pill);
4916                 if (rc)
4917                         rc = err_serious(rc);
4918         }
4919         mdt_thread_info_fini(info);
4920         RETURN(rc);
4921 }
4922
4923 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
4924 {
4925         struct seq_server_site  *ss = mdt_seq_site(mdt);
4926
4927         if (ss->ss_node_id == 0)
4928                 return;
4929
4930         if (ss->ss_client_seq != NULL) {
4931                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
4932                 ss->ss_client_seq->lcs_exp = NULL;
4933         }
4934
4935         if (ss->ss_server_fld != NULL) {
4936                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
4937                 ss->ss_server_fld->lsf_control_exp = NULL;
4938         }
4939 }
4940
4941 static void mdt_seq_fini_cli(struct mdt_device *mdt)
4942 {
4943         struct seq_server_site *ss = mdt_seq_site(mdt);
4944
4945         if (ss == NULL)
4946                 return;
4947
4948         if (ss->ss_server_seq != NULL)
4949                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
4950 }
4951
4952 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
4953 {
4954         mdt_seq_fini_cli(mdt);
4955         mdt_deregister_seq_exp(mdt);
4956
4957         return seq_site_fini(env, mdt_seq_site(mdt));
4958 }
4959
4960 /**
4961  * It will retrieve its FLDB entries from MDT0, and it only happens
4962  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
4963  * and it needs to refresh FLDB from the MDT0.
4964  **/
4965 static int mdt_register_lwp_callback(void *data)
4966 {
4967         struct lu_env           env;
4968         struct mdt_device       *mdt = data;
4969         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
4970         int                     rc;
4971         ENTRY;
4972
4973         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
4974
4975         rc = lu_env_init(&env, LCT_MD_THREAD);
4976         if (rc < 0) {
4977                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
4978                 RETURN(rc);
4979         }
4980
4981         /* Allocate new sequence now to avoid creating local transaction
4982          * in the normal transaction process */
4983         rc = seq_server_check_and_alloc_super(&env,
4984                                               mdt_seq_site(mdt)->ss_server_seq);
4985         if (rc < 0)
4986                 GOTO(out, rc);
4987
4988         if (fld->lsf_new) {
4989                 rc = fld_update_from_controller(&env, fld);
4990                 if (rc != 0) {
4991                         CERROR("%s: cannot update controller: rc = %d\n",
4992                                mdt_obd_name(mdt), rc);
4993                         GOTO(out, rc);
4994                 }
4995         }
4996 out:
4997         lu_env_fini(&env);
4998         RETURN(rc);
4999 }
5000
5001 static int mdt_register_seq_exp(struct mdt_device *mdt)
5002 {
5003         struct seq_server_site  *ss = mdt_seq_site(mdt);
5004         char                    *lwp_name = NULL;
5005         int                     rc;
5006
5007         if (ss->ss_node_id == 0)
5008                 return 0;
5009
5010         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
5011         if (lwp_name == NULL)
5012                 GOTO(out_free, rc = -ENOMEM);
5013
5014         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
5015         if (rc != 0)
5016                 GOTO(out_free, rc);
5017
5018         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
5019                                       NULL, NULL);
5020         if (rc != 0)
5021                 GOTO(out_free, rc);
5022
5023         rc = lustre_register_lwp_item(lwp_name,
5024                                       &ss->ss_server_fld->lsf_control_exp,
5025                                       mdt_register_lwp_callback, mdt);
5026         if (rc != 0) {
5027                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
5028                 ss->ss_client_seq->lcs_exp = NULL;
5029                 GOTO(out_free, rc);
5030         }
5031 out_free:
5032         if (lwp_name != NULL)
5033                 OBD_FREE(lwp_name, MAX_OBD_NAME);
5034
5035         return rc;
5036 }
5037
5038 /*
5039  * Init client sequence manager which is used by local MDS to talk to sequence
5040  * controller on remote node.
5041  */
5042 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
5043 {
5044         struct seq_server_site *ss = mdt_seq_site(mdt);
5045         char *prefix;
5046         ENTRY;
5047
5048         /* check if this is adding the first MDC and controller is not yet
5049          * initialized. */
5050         OBD_ALLOC_PTR(ss->ss_client_seq);
5051         if (ss->ss_client_seq == NULL)
5052                 RETURN(-ENOMEM);
5053
5054         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
5055         if (prefix == NULL) {
5056                 OBD_FREE_PTR(ss->ss_client_seq);
5057                 ss->ss_client_seq = NULL;
5058                 RETURN(-ENOMEM);
5059         }
5060
5061         /* Note: seq_client_fini will be called in seq_site_fini */
5062         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
5063         seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
5064                         prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
5065                                                             NULL);
5066         OBD_FREE(prefix, MAX_OBD_NAME + 5);
5067
5068         RETURN(seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq));
5069 }
5070
5071 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
5072 {
5073         struct seq_server_site  *ss;
5074         int                     rc;
5075         ENTRY;
5076
5077         ss = mdt_seq_site(mdt);
5078         /* init sequence controller server(MDT0) */
5079         if (ss->ss_node_id == 0) {
5080                 OBD_ALLOC_PTR(ss->ss_control_seq);
5081                 if (ss->ss_control_seq == NULL)
5082                         RETURN(-ENOMEM);
5083
5084                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
5085                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
5086                                      ss);
5087                 if (rc)
5088                         GOTO(out_seq_fini, rc);
5089         }
5090
5091         /* Init normal sequence server */
5092         OBD_ALLOC_PTR(ss->ss_server_seq);
5093         if (ss->ss_server_seq == NULL)
5094                 GOTO(out_seq_fini, rc = -ENOMEM);
5095
5096         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
5097                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
5098         if (rc)
5099                 GOTO(out_seq_fini, rc);
5100
5101         /* init seq client for seq server to talk to seq controller(MDT0) */
5102         rc = mdt_seq_init_cli(env, mdt);
5103         if (rc != 0)
5104                 GOTO(out_seq_fini, rc);
5105
5106         if (ss->ss_node_id != 0)
5107                 /* register controller export through lwp */
5108                 rc = mdt_register_seq_exp(mdt);
5109
5110         EXIT;
5111 out_seq_fini:
5112         if (rc)
5113                 mdt_seq_fini(env, mdt);
5114
5115         return rc;
5116 }
5117
5118 /*
5119  * FLD wrappers
5120  */
5121 static int mdt_fld_fini(const struct lu_env *env,
5122                         struct mdt_device *m)
5123 {
5124         struct seq_server_site *ss = mdt_seq_site(m);
5125         ENTRY;
5126
5127         if (ss && ss->ss_server_fld) {
5128                 fld_server_fini(env, ss->ss_server_fld);
5129                 OBD_FREE_PTR(ss->ss_server_fld);
5130                 ss->ss_server_fld = NULL;
5131         }
5132
5133         RETURN(0);
5134 }
5135
5136 static int mdt_fld_init(const struct lu_env *env,
5137                         const char *uuid,
5138                         struct mdt_device *m)
5139 {
5140         struct seq_server_site *ss;
5141         int rc;
5142         ENTRY;
5143
5144         ss = mdt_seq_site(m);
5145
5146         OBD_ALLOC_PTR(ss->ss_server_fld);
5147         if (ss->ss_server_fld == NULL)
5148                 RETURN(rc = -ENOMEM);
5149
5150         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
5151                              LU_SEQ_RANGE_MDT);
5152         if (rc) {
5153                 OBD_FREE_PTR(ss->ss_server_fld);
5154                 ss->ss_server_fld = NULL;
5155                 RETURN(rc);
5156         }
5157
5158         RETURN(0);
5159 }
5160
5161 static void mdt_stack_pre_fini(const struct lu_env *env,
5162                            struct mdt_device *m, struct lu_device *top)
5163 {
5164         struct lustre_cfg_bufs  *bufs;
5165         struct lustre_cfg       *lcfg;
5166         struct mdt_thread_info  *info;
5167         ENTRY;
5168
5169         LASSERT(top);
5170
5171         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5172         LASSERT(info != NULL);
5173
5174         bufs = &info->mti_u.bufs;
5175
5176         LASSERT(m->mdt_child_exp);
5177         LASSERT(m->mdt_child_exp->exp_obd);
5178
5179         /* process cleanup, pass mdt obd name to get obd umount flags */
5180         /* XXX: this is needed because all layers are referenced by
5181          * objects (some of them are pinned by osd, for example *
5182          * the proper solution should be a model where object used
5183          * by osd only doesn't have mdt/mdd slices -bzzz */
5184         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
5185         lustre_cfg_bufs_set_string(bufs, 1, NULL);
5186         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5187         if (!lcfg)
5188                 RETURN_EXIT;
5189         lustre_cfg_init(lcfg, LCFG_PRE_CLEANUP, bufs);
5190
5191         top->ld_ops->ldo_process_config(env, top, lcfg);
5192         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5193         EXIT;
5194 }
5195
5196 static void mdt_stack_fini(const struct lu_env *env,
5197                            struct mdt_device *m, struct lu_device *top)
5198 {
5199         struct obd_device       *obd = mdt2obd_dev(m);
5200         struct lustre_cfg_bufs  *bufs;
5201         struct lustre_cfg       *lcfg;
5202         struct mdt_thread_info  *info;
5203         char                     flags[3] = "";
5204         ENTRY;
5205
5206         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5207         LASSERT(info != NULL);
5208
5209         lu_dev_del_linkage(top->ld_site, top);
5210
5211         lu_site_purge(env, top->ld_site, -1);
5212
5213         bufs = &info->mti_u.bufs;
5214         /* process cleanup, pass mdt obd name to get obd umount flags */
5215         /* another purpose is to let all layers to release their objects */
5216         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
5217         if (obd->obd_force)
5218                 strcat(flags, "F");
5219         if (obd->obd_fail)
5220                 strcat(flags, "A");
5221         lustre_cfg_bufs_set_string(bufs, 1, flags);
5222         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5223         if (!lcfg)
5224                 RETURN_EXIT;
5225         lustre_cfg_init(lcfg, LCFG_CLEANUP, bufs);
5226
5227         LASSERT(top);
5228         top->ld_ops->ldo_process_config(env, top, lcfg);
5229         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5230
5231         lu_site_purge(env, top->ld_site, -1);
5232
5233         m->mdt_child = NULL;
5234         m->mdt_bottom = NULL;
5235
5236         obd_disconnect(m->mdt_child_exp);
5237         m->mdt_child_exp = NULL;
5238
5239         obd_disconnect(m->mdt_bottom_exp);
5240         m->mdt_child_exp = NULL;
5241 }
5242
5243 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
5244                                const char *next, struct obd_export **exp)
5245 {
5246         struct obd_connect_data *data = NULL;
5247         struct obd_device       *obd;
5248         int                      rc;
5249         ENTRY;
5250
5251         OBD_ALLOC_PTR(data);
5252         if (data == NULL)
5253                 GOTO(out, rc = -ENOMEM);
5254
5255         obd = class_name2obd(next);
5256         if (obd == NULL) {
5257                 CERROR("%s: can't locate next device: %s\n",
5258                        mdt_obd_name(m), next);
5259                 GOTO(out, rc = -ENOTCONN);
5260         }
5261
5262         data->ocd_connect_flags = OBD_CONNECT_VERSION;
5263         data->ocd_version = LUSTRE_VERSION_CODE;
5264
5265         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
5266         if (rc) {
5267                 CERROR("%s: cannot connect to next dev %s (%d)\n",
5268                        mdt_obd_name(m), next, rc);
5269                 GOTO(out, rc);
5270         }
5271
5272 out:
5273         if (data)
5274                 OBD_FREE_PTR(data);
5275         RETURN(rc);
5276 }
5277
5278 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
5279                           struct lustre_cfg *cfg)
5280 {
5281         char                   *dev = lustre_cfg_string(cfg, 0);
5282         int                     rc, name_size, uuid_size;
5283         char                   *name, *uuid, *p;
5284         struct lustre_cfg_bufs *bufs;
5285         struct lustre_cfg      *lcfg;
5286         struct obd_device      *obd;
5287         struct lustre_profile  *lprof;
5288         struct lu_site         *site;
5289         ENTRY;
5290
5291         /* in 1.8 we had the only device in the stack - MDS.
5292          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
5293          * in 2.3 OSD is instantiated by obd_mount.c, so we need
5294          * to generate names and setup MDT, MDD. MDT will be using
5295          * generated name to connect to MDD. for MDD the next device
5296          * will be LOD with name taken from so called "profile" which
5297          * is generated by mount_option line
5298          *
5299          * 1.8 MGS generates config. commands like this:
5300          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
5301          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
5302          * 2.0 MGS generates config. commands like this:
5303          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
5304          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
5305          *                    3:lustre-MDT0000-mdtlov  4:f
5306          *
5307          * we generate MDD name from MDT one, just replacing T with D
5308          *
5309          * after all the preparations, the logical equivalent will be
5310          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
5311          *                    3:lustre-MDT0000-mdtlov  4:f
5312          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
5313          *                    3:lustre-MDD0000  4:f
5314          *
5315          *  notice we build the stack from down to top: MDD first, then MDT */
5316
5317         name_size = MAX_OBD_NAME;
5318         uuid_size = MAX_OBD_NAME;
5319
5320         OBD_ALLOC(name, name_size);
5321         OBD_ALLOC(uuid, uuid_size);
5322         if (name == NULL || uuid == NULL)
5323                 GOTO(cleanup_mem, rc = -ENOMEM);
5324
5325         OBD_ALLOC_PTR(bufs);
5326         if (!bufs)
5327                 GOTO(cleanup_mem, rc = -ENOMEM);
5328
5329         strcpy(name, dev);
5330         p = strstr(name, "-MDT");
5331         if (p == NULL)
5332                 GOTO(free_bufs, rc = -ENOMEM);
5333         p[3] = 'D';
5334
5335         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
5336
5337         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
5338         if (lprof == NULL || lprof->lp_dt == NULL) {
5339                 CERROR("can't find the profile: %s\n",
5340                        lustre_cfg_string(cfg, 0));
5341                 GOTO(free_bufs, rc = -EINVAL);
5342         }
5343
5344         lustre_cfg_bufs_reset(bufs, name);
5345         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
5346         lustre_cfg_bufs_set_string(bufs, 2, uuid);
5347         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
5348
5349         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5350         if (!lcfg)
5351                 GOTO(put_profile, rc = -ENOMEM);
5352         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
5353
5354         rc = class_attach(lcfg);
5355         if (rc)
5356                 GOTO(lcfg_cleanup, rc);
5357
5358         obd = class_name2obd(name);
5359         if (!obd) {
5360                 CERROR("Can not find obd %s (%s in config)\n",
5361                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
5362                 GOTO(lcfg_cleanup, rc = -EINVAL);
5363         }
5364
5365         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5366
5367         lustre_cfg_bufs_reset(bufs, name);
5368         lustre_cfg_bufs_set_string(bufs, 1, uuid);
5369         lustre_cfg_bufs_set_string(bufs, 2, dev);
5370         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
5371
5372         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5373         if (!lcfg)
5374                 GOTO(class_detach, rc = -ENOMEM);
5375         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
5376
5377         rc = class_setup(obd, lcfg);
5378         if (rc)
5379                 GOTO(class_detach, rc);
5380
5381         /* connect to MDD we just setup */
5382         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
5383         if (rc)
5384                 GOTO(class_detach, rc);
5385
5386         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
5387         LASSERT(site);
5388         LASSERT(mdt_lu_site(mdt) == NULL);
5389         mdt->mdt_lu_dev.ld_site = site;
5390         site->ls_top_dev = &mdt->mdt_lu_dev;
5391         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
5392
5393         /* now connect to bottom OSD */
5394         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
5395         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
5396         if (rc)
5397                 GOTO(class_detach, rc);
5398         mdt->mdt_bottom =
5399                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
5400
5401         rc = lu_env_refill((struct lu_env *)env);
5402         if (rc != 0)
5403                 CERROR("Failure to refill session: '%d'\n", rc);
5404
5405         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
5406
5407         EXIT;
5408 class_detach:
5409         if (rc)
5410                 class_detach(obd, lcfg);
5411 lcfg_cleanup:
5412         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5413 put_profile:
5414         class_put_profile(lprof);
5415 free_bufs:
5416         OBD_FREE_PTR(bufs);
5417 cleanup_mem:
5418         if (name)
5419                 OBD_FREE(name, name_size);
5420         if (uuid)
5421                 OBD_FREE(uuid, uuid_size);
5422         RETURN(rc);
5423 }
5424
5425 /* setup quota master target on MDT0 */
5426 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
5427                           struct lustre_cfg *cfg)
5428 {
5429         struct obd_device       *obd;
5430         char                    *dev = lustre_cfg_string(cfg, 0);
5431         char                    *qmtname, *uuid, *p;
5432         struct lustre_cfg_bufs  *bufs;
5433         struct lustre_cfg       *lcfg;
5434         struct lustre_profile   *lprof;
5435         struct obd_connect_data *data;
5436         int                      rc;
5437         ENTRY;
5438
5439         LASSERT(mdt->mdt_qmt_exp == NULL);
5440         LASSERT(mdt->mdt_qmt_dev == NULL);
5441
5442         /* quota master is on MDT0 only for now */
5443         if (mdt->mdt_seq_site.ss_node_id != 0)
5444                 RETURN(0);
5445
5446         /* MGS generates config commands which look as follows:
5447          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
5448          *                    3:lustre-MDT0000-mdtlov  4:f
5449          *
5450          * We generate the QMT name from the MDT one, just replacing MD with QM
5451          * after all the preparations, the logical equivalent will be:
5452          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
5453          *                    3:lustre-MDT0000-osd  4:f */
5454         OBD_ALLOC(qmtname, MAX_OBD_NAME);
5455         OBD_ALLOC(uuid, UUID_MAX);
5456         OBD_ALLOC_PTR(bufs);
5457         OBD_ALLOC_PTR(data);
5458         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
5459                 GOTO(cleanup_mem, rc = -ENOMEM);
5460
5461         strcpy(qmtname, dev);
5462         p = strstr(qmtname, "-MDT");
5463         if (p == NULL)
5464                 GOTO(cleanup_mem, rc = -ENOMEM);
5465         /* replace MD with QM */
5466         p[1] = 'Q';
5467         p[2] = 'M';
5468
5469         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
5470
5471         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
5472         if (lprof == NULL || lprof->lp_dt == NULL) {
5473                 CERROR("can't find profile for %s\n",
5474                        lustre_cfg_string(cfg, 0));
5475                 GOTO(cleanup_mem, rc = -EINVAL);
5476         }
5477
5478         lustre_cfg_bufs_reset(bufs, qmtname);
5479         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
5480         lustre_cfg_bufs_set_string(bufs, 2, uuid);
5481         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
5482
5483         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5484         if (!lcfg)
5485                 GOTO(put_profile, rc = -ENOMEM);
5486         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
5487
5488         rc = class_attach(lcfg);
5489         if (rc)
5490                 GOTO(lcfg_cleanup, rc);
5491
5492         obd = class_name2obd(qmtname);
5493         if (!obd) {
5494                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
5495                        lustre_cfg_string(cfg, 0));
5496                 GOTO(lcfg_cleanup, rc = -EINVAL);
5497         }
5498
5499         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5500
5501         lustre_cfg_bufs_reset(bufs, qmtname);
5502         lustre_cfg_bufs_set_string(bufs, 1, uuid);
5503         lustre_cfg_bufs_set_string(bufs, 2, dev);
5504
5505         /* for quota, the next device should be the OSD device */
5506         lustre_cfg_bufs_set_string(bufs, 3,
5507                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
5508
5509         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5510         if (!lcfg)
5511                 GOTO(class_detach, rc = -ENOMEM);
5512         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
5513
5514         rc = class_setup(obd, lcfg);
5515         if (rc)
5516                 GOTO(class_detach, rc);
5517
5518         mdt->mdt_qmt_dev = obd->obd_lu_dev;
5519
5520         /* configure local quota objects */
5521         if (OBD_FAIL_CHECK(OBD_FAIL_QUOTA_INIT))
5522                 rc = -EBADF;
5523         else
5524                 rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
5525                                                            &mdt->mdt_lu_dev,
5526                                                            mdt->mdt_qmt_dev);
5527         if (rc)
5528                 GOTO(class_cleanup, rc);
5529
5530         /* connect to quota master target */
5531         data->ocd_connect_flags = OBD_CONNECT_VERSION;
5532         data->ocd_version = LUSTRE_VERSION_CODE;
5533         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
5534                          data, NULL);
5535         if (rc) {
5536                 CERROR("cannot connect to quota master device %s (%d)\n",
5537                        qmtname, rc);
5538                 GOTO(class_cleanup, rc);
5539         }
5540
5541         EXIT;
5542 class_cleanup:
5543         if (rc) {
5544                 class_manual_cleanup(obd);
5545                 mdt->mdt_qmt_dev = NULL;
5546                 GOTO(lcfg_cleanup, rc);
5547         }
5548 class_detach:
5549         if (rc)
5550                 class_detach(obd, lcfg);
5551 lcfg_cleanup:
5552         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5553 put_profile:
5554         class_put_profile(lprof);
5555 cleanup_mem:
5556         if (bufs)
5557                 OBD_FREE_PTR(bufs);
5558         if (qmtname)
5559                 OBD_FREE(qmtname, MAX_OBD_NAME);
5560         if (uuid)
5561                 OBD_FREE(uuid, UUID_MAX);
5562         if (data)
5563                 OBD_FREE_PTR(data);
5564         return rc;
5565 }
5566
5567 /* Shutdown quota master target associated with mdt */
5568 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
5569 {
5570         ENTRY;
5571
5572         if (mdt->mdt_qmt_exp == NULL)
5573                 RETURN_EXIT;
5574         LASSERT(mdt->mdt_qmt_dev != NULL);
5575
5576         /* the qmt automatically shuts down when the mdt disconnects */
5577         obd_disconnect(mdt->mdt_qmt_exp);
5578         mdt->mdt_qmt_exp = NULL;
5579         mdt->mdt_qmt_dev = NULL;
5580         EXIT;
5581 }
5582
5583 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
5584  * for now. This will be removed along with converting rest of MDT code
5585  * to use tgt_session_info */
5586 static int mdt_tgt_getxattr(struct tgt_session_info *tsi)
5587 {
5588         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
5589         int                      rc;
5590
5591         if (unlikely(info->mti_object == NULL))
5592                 return -EPROTO;
5593
5594         rc = mdt_getxattr(info);
5595
5596         mdt_thread_info_fini(info);
5597         return rc;
5598 }
5599
5600 static int mdt_llog_open(struct tgt_session_info *tsi)
5601 {
5602         ENTRY;
5603
5604         if (!mdt_is_rootadmin(tsi2mdt_info(tsi)))
5605                 RETURN(err_serious(-EACCES));
5606
5607         RETURN(tgt_llog_open(tsi));
5608 }
5609
5610 #define OBD_FAIL_OST_READ_NET   OBD_FAIL_OST_BRW_NET
5611 #define OBD_FAIL_OST_WRITE_NET  OBD_FAIL_OST_BRW_NET
5612 #define OST_BRW_READ    OST_READ
5613 #define OST_BRW_WRITE   OST_WRITE
5614
5615 static struct tgt_handler mdt_tgt_handlers[] = {
5616 TGT_RPC_HANDLER(MDS_FIRST_OPC,
5617                 0,                      MDS_CONNECT,    mdt_tgt_connect,
5618                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
5619 TGT_RPC_HANDLER(MDS_FIRST_OPC,
5620                 0,                      MDS_DISCONNECT, tgt_disconnect,
5621                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
5622 TGT_RPC_HANDLER(MDS_FIRST_OPC,
5623                 HAS_REPLY,              MDS_SET_INFO,   mdt_set_info,
5624                 &RQF_MDT_SET_INFO, LUSTRE_MDS_VERSION),
5625 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
5626 TGT_MDT_HDL(HAS_REPLY,          MDS_GET_ROOT,   mdt_get_root),
5627 TGT_MDT_HDL(HAS_BODY,           MDS_GETATTR,    mdt_getattr),
5628 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_GETATTR_NAME,
5629                                                         mdt_getattr_name),
5630 TGT_MDT_HDL(HAS_BODY,           MDS_GETXATTR,   mdt_tgt_getxattr),
5631 TGT_MDT_HDL(HAS_REPLY,          MDS_STATFS,     mdt_statfs),
5632 TGT_MDT_HDL(IS_MUTABLE,         MDS_REINT,      mdt_reint),
5633 TGT_MDT_HDL(HAS_BODY,           MDS_CLOSE,      mdt_close),
5634 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_READPAGE,   mdt_readpage),
5635 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_SYNC,       mdt_sync),
5636 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
5637 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_PROGRESS,
5638                                                         mdt_hsm_progress),
5639 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_CT_REGISTER,
5640                                                         mdt_hsm_ct_register),
5641 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_CT_UNREGISTER,
5642                                                         mdt_hsm_ct_unregister),
5643 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_STATE_GET,
5644                                                         mdt_hsm_state_get),
5645 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_STATE_SET,
5646                                                         mdt_hsm_state_set),
5647 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_ACTION,       mdt_hsm_action),
5648 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_REQUEST,
5649                                                         mdt_hsm_request),
5650 TGT_MDT_HDL(HAS_KEY | HAS_BODY | HAS_REPLY | IS_MUTABLE,
5651             MDS_SWAP_LAYOUTS,
5652             mdt_swap_layouts),
5653 TGT_MDT_HDL(IS_MUTABLE,         MDS_RMFID,      mdt_rmfid),
5654 };
5655
5656 static struct tgt_handler mdt_io_ops[] = {
5657 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY, OST_BRW_READ, tgt_brw_read,
5658                                                         mdt_hp_brw),
5659 TGT_OST_HDL_HP(HAS_BODY | IS_MUTABLE,    OST_BRW_WRITE, tgt_brw_write,
5660                                                         mdt_hp_brw),
5661 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY | IS_MUTABLE,
5662                                          OST_PUNCH,     mdt_punch_hdl,
5663                                                         mdt_hp_punch),
5664 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SYNC,     mdt_data_sync),
5665 TGT_OST_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, OST_FALLOCATE,
5666                                                         mdt_fallocate_hdl),
5667 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SEEK, tgt_lseek),
5668 };
5669
5670 static struct tgt_handler mdt_sec_ctx_ops[] = {
5671 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
5672 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
5673 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
5674 };
5675
5676 static struct tgt_handler mdt_quota_ops[] = {
5677 TGT_QUOTA_HDL(HAS_REPLY,                QUOTA_DQACQ,      mdt_quota_dqacq),
5678 };
5679
5680 static struct tgt_handler mdt_llog_handlers[] = {
5681         TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_CREATE,      mdt_llog_open),
5682         TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_NEXT_BLOCK,  tgt_llog_next_block),
5683         TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
5684         TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_PREV_BLOCK,  tgt_llog_prev_block),
5685 };
5686
5687 static struct tgt_opc_slice mdt_common_slice[] = {
5688         {
5689                 .tos_opc_start  = MDS_FIRST_OPC,
5690                 .tos_opc_end    = MDS_LAST_OPC,
5691                 .tos_hs         = mdt_tgt_handlers
5692         },
5693         {
5694                 .tos_opc_start  = OBD_FIRST_OPC,
5695                 .tos_opc_end    = OBD_LAST_OPC,
5696                 .tos_hs         = tgt_obd_handlers
5697         },
5698         {
5699                 .tos_opc_start  = LDLM_FIRST_OPC,
5700                 .tos_opc_end    = LDLM_LAST_OPC,
5701                 .tos_hs         = tgt_dlm_handlers
5702         },
5703         {
5704                 .tos_opc_start  = SEC_FIRST_OPC,
5705                 .tos_opc_end    = SEC_LAST_OPC,
5706                 .tos_hs         = mdt_sec_ctx_ops
5707         },
5708         {
5709                 .tos_opc_start  = OUT_UPDATE_FIRST_OPC,
5710                 .tos_opc_end    = OUT_UPDATE_LAST_OPC,
5711                 .tos_hs         = tgt_out_handlers
5712         },
5713         {
5714                 .tos_opc_start  = FLD_FIRST_OPC,
5715                 .tos_opc_end    = FLD_LAST_OPC,
5716                 .tos_hs         = fld_handlers
5717         },
5718         {
5719                 .tos_opc_start  = SEQ_FIRST_OPC,
5720                 .tos_opc_end    = SEQ_LAST_OPC,
5721                 .tos_hs         = seq_handlers
5722         },
5723         {
5724                 .tos_opc_start  = QUOTA_DQACQ,
5725                 .tos_opc_end    = QUOTA_LAST_OPC,
5726                 .tos_hs         = mdt_quota_ops
5727         },
5728         {
5729                 .tos_opc_start  = LLOG_FIRST_OPC,
5730                 .tos_opc_end    = LLOG_LAST_OPC,
5731                 .tos_hs         = mdt_llog_handlers
5732         },
5733         {
5734                 .tos_opc_start  = LFSCK_FIRST_OPC,
5735                 .tos_opc_end    = LFSCK_LAST_OPC,
5736                 .tos_hs         = tgt_lfsck_handlers
5737         },
5738         {
5739                 .tos_opc_start  = OST_FIRST_OPC,
5740                 .tos_opc_end    = OST_LAST_OPC,
5741                 .tos_hs         = mdt_io_ops
5742         },
5743         {
5744                 .tos_hs         = NULL
5745         }
5746 };
5747
5748 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
5749 {
5750         struct md_device *next = m->mdt_child;
5751         struct lu_device *d = &m->mdt_lu_dev;
5752         struct obd_device *obd = mdt2obd_dev(m);
5753         struct lfsck_stop stop;
5754
5755         ENTRY;
5756         stop.ls_status = LS_PAUSED;
5757         stop.ls_flags = 0;
5758         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
5759
5760         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
5761
5762         mdt_restriper_stop(m);
5763         ping_evictor_stop();
5764
5765         /* Remove the HSM /proc entry so the coordinator cannot be
5766          * restarted by a user while it's shutting down.
5767          */
5768         mdt_hsm_cdt_stop(m);
5769
5770         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
5771         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
5772
5773         if (m->mdt_namespace != NULL)
5774                 ldlm_namespace_free_prior(m->mdt_namespace, NULL,
5775                                           d->ld_obd->obd_force);
5776
5777         obd_exports_barrier(obd);
5778         obd_zombie_barrier();
5779
5780         mdt_quota_fini(env, m);
5781
5782         cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
5783
5784         /* Calling the cleanup functions in the same order as in the mdt_init0
5785          * error path
5786          */
5787         mdt_tunables_fini(m);
5788
5789         target_recovery_fini(obd);
5790         upcall_cache_cleanup(m->mdt_identity_cache);
5791         m->mdt_identity_cache = NULL;
5792
5793         mdt_fs_cleanup(env, m);
5794
5795         tgt_fini(env, &m->mdt_lut);
5796
5797         mdt_hsm_cdt_fini(m);
5798
5799         if (m->mdt_los != NULL) {
5800                 local_oid_storage_fini(env, m->mdt_los);
5801                 m->mdt_los = NULL;
5802         }
5803
5804         if (m->mdt_namespace != NULL) {
5805                 ldlm_namespace_free_post(m->mdt_namespace);
5806                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
5807         }
5808
5809         if (m->mdt_md_root != NULL) {
5810                 mdt_object_put(env, m->mdt_md_root);
5811                 m->mdt_md_root = NULL;
5812         }
5813
5814         mdt_seq_fini(env, m);
5815
5816         mdt_fld_fini(env, m);
5817
5818         /*
5819          * Finish the stack
5820          */
5821         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
5822
5823         LASSERT(atomic_read(&d->ld_ref) == 0);
5824
5825         server_put_mount(mdt_obd_name(m), true);
5826
5827         EXIT;
5828 }
5829
5830 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
5831
5832 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
5833                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
5834 {
5835         const struct dt_device_param *dt_conf;
5836         struct mdt_thread_info *info;
5837         struct obd_device *obd;
5838         const char *dev = lustre_cfg_string(cfg, 0);
5839         const char *num = lustre_cfg_string(cfg, 2);
5840         struct tg_grants_data *tgd = &m->mdt_lut.lut_tgd;
5841         struct lustre_mount_info *lmi = NULL;
5842         struct lustre_sb_info *lsi;
5843         struct lu_site *s;
5844         struct seq_server_site *ss_site;
5845         const char *identity_upcall = "NONE";
5846         struct md_device *next;
5847         struct lu_fid fid;
5848         int rc;
5849         long node_id;
5850         mntopt_t mntopts;
5851         ENTRY;
5852
5853         lu_device_init(&m->mdt_lu_dev, ldt);
5854         /*
5855          * Environment (env) might be missing mdt_thread_key values at that
5856          * point, if device is allocated when mdt_thread_key is in QUIESCENT
5857          * mode.
5858          *
5859          * Usually device allocation path doesn't use module key values, but
5860          * mdt has to do a lot of work here, so allocate key value.
5861          */
5862         rc = lu_env_refill((struct lu_env *)env);
5863         if (rc != 0)
5864                 RETURN(rc);
5865
5866         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5867         LASSERT(info != NULL);
5868
5869         obd = class_name2obd(dev);
5870         LASSERT(obd != NULL);
5871
5872         m->mdt_max_mdsize = MAX_MD_SIZE_OLD;
5873         m->mdt_opts.mo_evict_tgt_nids = 1;
5874         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
5875
5876         lmi = server_get_mount(dev);
5877         if (lmi == NULL) {
5878                 CERROR("Cannot get mount info for %s!\n", dev);
5879                 RETURN(-EFAULT);
5880         } else {
5881                 lsi = s2lsi(lmi->lmi_sb);
5882                 LASSERT(lsi->lsi_lmd);
5883                 /* CMD is supported only in IAM mode */
5884                 LASSERT(num);
5885                 rc = kstrtol(num, 10, &node_id);
5886                 if (rc)
5887                         RETURN(rc);
5888
5889                 obd->u.obt.obt_magic = OBT_MAGIC;
5890                 if (lsi->lsi_lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
5891                         m->mdt_skip_lfsck = 1;
5892         }
5893
5894         /* Just try to get a DoM lock by default. Otherwise, having a group
5895          * lock granted, it may get blocked for a long time. */
5896         m->mdt_opts.mo_dom_lock = TRYLOCK_DOM_ON_OPEN;
5897         /* DoM files are read at open and data is packed in the reply */
5898         m->mdt_opts.mo_dom_read_open = 1;
5899
5900         m->mdt_squash.rsi_uid = 0;
5901         m->mdt_squash.rsi_gid = 0;
5902         INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
5903         spin_lock_init(&m->mdt_squash.rsi_lock);
5904         spin_lock_init(&m->mdt_lock);
5905         m->mdt_enable_remote_dir = 1;
5906         m->mdt_enable_striped_dir = 1;
5907         m->mdt_enable_dir_migration = 1;
5908         m->mdt_enable_dir_restripe = 0;
5909         m->mdt_enable_dir_auto_split = 0;
5910         m->mdt_enable_remote_dir_gid = 0;
5911         m->mdt_enable_chprojid_gid = 0;
5912         m->mdt_enable_remote_rename = 1;
5913         m->mdt_dir_restripe_nsonly = 1;
5914         m->mdt_enable_remote_subdir_mount = 1;
5915
5916         atomic_set(&m->mdt_mds_mds_conns, 0);
5917         atomic_set(&m->mdt_async_commit_count, 0);
5918
5919         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
5920         m->mdt_lu_dev.ld_obd = obd;
5921         /* Set this lu_device to obd for error handling purposes. */
5922         obd->obd_lu_dev = &m->mdt_lu_dev;
5923
5924         /* init the stack */
5925         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
5926         if (rc) {
5927                 CERROR("%s: Can't init device stack, rc %d\n",
5928                        mdt_obd_name(m), rc);
5929                 GOTO(err_lmi, rc);
5930         }
5931
5932         s = mdt_lu_site(m);
5933         ss_site = mdt_seq_site(m);
5934         s->ld_seq_site = ss_site;
5935         ss_site->ss_lu = s;
5936
5937         /* set server index */
5938         ss_site->ss_node_id = node_id;
5939
5940         /* failover is the default
5941          * FIXME: we do not failout mds0/mgs, which may cause some problems.
5942          * assumed whose ss_node_id == 0 XXX
5943          * */
5944         obd->obd_replayable = 1;
5945         /* No connection accepted until configurations will finish */
5946         obd->obd_no_conn = 1;
5947
5948         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
5949                 char *str = lustre_cfg_string(cfg, 4);
5950                 if (strchr(str, 'n')) {
5951                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
5952                         obd->obd_replayable = 0;
5953                 }
5954         }
5955
5956         rc = mdt_fld_init(env, mdt_obd_name(m), m);
5957         if (rc)
5958                 GOTO(err_fini_stack, rc);
5959
5960         rc = mdt_seq_init(env, m);
5961         if (rc)
5962                 GOTO(err_fini_fld, rc);
5963
5964         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
5965                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
5966         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
5967                                               LDLM_NAMESPACE_SERVER,
5968                                               LDLM_NAMESPACE_GREEDY,
5969                                               LDLM_NS_TYPE_MDT);
5970         if (IS_ERR(m->mdt_namespace)) {
5971                 rc = PTR_ERR(m->mdt_namespace);
5972                 CERROR("%s: unable to create server namespace: rc = %d\n",
5973                        obd->obd_name, rc);
5974                 m->mdt_namespace = NULL;
5975                 GOTO(err_fini_seq, rc);
5976         }
5977
5978         m->mdt_namespace->ns_lvbp = m;
5979         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
5980
5981         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
5982         /* set obd_namespace for compatibility with old code */
5983         obd->obd_namespace = m->mdt_namespace;
5984
5985         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
5986                       OBD_FAIL_MDS_ALL_REQUEST_NET,
5987                       OBD_FAIL_MDS_ALL_REPLY_NET);
5988         if (rc)
5989                 GOTO(err_free_ns, rc);
5990
5991         /* Amount of available space excluded from granting and reserved
5992          * for metadata. It is a percentage of the total MDT size. */
5993         tgd->tgd_reserved_pcnt = 10;
5994
5995         if (ONE_MB_BRW_SIZE < (1U << tgd->tgd_blockbits))
5996                 m->mdt_brw_size = 1U << tgd->tgd_blockbits;
5997         else
5998                 m->mdt_brw_size = ONE_MB_BRW_SIZE;
5999
6000         rc = mdt_fs_setup(env, m, obd, lsi);
6001         if (rc)
6002                 GOTO(err_tgt, rc);
6003
6004         fid.f_seq = FID_SEQ_LOCAL_NAME;
6005         fid.f_oid = 1;
6006         fid.f_ver = 0;
6007         rc = local_oid_storage_init(env, m->mdt_bottom, &fid, &m->mdt_los);
6008         if (rc != 0)
6009                 GOTO(err_fs_cleanup, rc);
6010
6011         rc = mdt_hsm_cdt_init(m);
6012         if (rc != 0) {
6013                 CERROR("%s: error initializing coordinator, rc %d\n",
6014                        mdt_obd_name(m), rc);
6015                 GOTO(err_los_fini, rc);
6016         }
6017
6018         tgt_adapt_sptlrpc_conf(&m->mdt_lut);
6019
6020         next = m->mdt_child;
6021         dt_conf = next->md_ops->mdo_dtconf_get(env, next);
6022
6023         mntopts = dt_conf->ddp_mntopts;
6024
6025         if (mntopts & MNTOPT_USERXATTR)
6026                 m->mdt_opts.mo_user_xattr = 1;
6027         else
6028                 m->mdt_opts.mo_user_xattr = 0;
6029
6030         m->mdt_max_ea_size = dt_conf->ddp_max_ea_size;
6031
6032         if (mntopts & MNTOPT_ACL)
6033                 m->mdt_opts.mo_acl = 1;
6034         else
6035                 m->mdt_opts.mo_acl = 0;
6036
6037         /* XXX: to support suppgid for ACL, we enable identity_upcall
6038          * by default, otherwise, maybe got unexpected -EACCESS. */
6039         if (m->mdt_opts.mo_acl)
6040                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
6041
6042         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
6043                                                 identity_upcall,
6044                                                 &mdt_identity_upcall_cache_ops);
6045         if (IS_ERR(m->mdt_identity_cache)) {
6046                 rc = PTR_ERR(m->mdt_identity_cache);
6047                 m->mdt_identity_cache = NULL;
6048                 GOTO(err_free_hsm, rc);
6049         }
6050
6051         rc = mdt_tunables_init(m, dev);
6052         if (rc) {
6053                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
6054                 GOTO(err_recovery, rc);
6055         }
6056
6057         rc = mdt_quota_init(env, m, cfg);
6058         if (rc)
6059                 GOTO(err_procfs, rc);
6060
6061         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
6062         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
6063                            "mdt_ldlm_client", m->mdt_ldlm_client);
6064
6065         ping_evictor_start();
6066
6067         /* recovery will be started upon mdt_prepare()
6068          * when the whole stack is complete and ready
6069          * to serve the requests */
6070
6071         /* Reduce the initial timeout on an MDS because it doesn't need such
6072          * a long timeout as an OST does. Adaptive timeouts will adjust this
6073          * value appropriately. */
6074         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
6075                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
6076
6077         if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_LOCAL_RECOV))
6078                 m->mdt_lut.lut_local_recovery = 1;
6079
6080         rc = mdt_restriper_start(m);
6081         if (rc)
6082                 GOTO(err_ping_evictor, rc);
6083
6084         RETURN(0);
6085
6086 err_ping_evictor:
6087         ping_evictor_stop();
6088 err_procfs:
6089         mdt_tunables_fini(m);
6090 err_recovery:
6091         upcall_cache_cleanup(m->mdt_identity_cache);
6092         m->mdt_identity_cache = NULL;
6093 err_free_hsm:
6094         mdt_hsm_cdt_fini(m);
6095 err_los_fini:
6096         local_oid_storage_fini(env, m->mdt_los);
6097         m->mdt_los = NULL;
6098 err_fs_cleanup:
6099         mdt_fs_cleanup(env, m);
6100 err_tgt:
6101         /* keep recoverable clients */
6102         obd->obd_fail = 1;
6103         target_recovery_fini(obd);
6104         obd_exports_barrier(obd);
6105         obd_zombie_barrier();
6106         tgt_fini(env, &m->mdt_lut);
6107 err_free_ns:
6108         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
6109         obd->obd_namespace = m->mdt_namespace = NULL;
6110 err_fini_seq:
6111         mdt_seq_fini(env, m);
6112 err_fini_fld:
6113         mdt_fld_fini(env, m);
6114 err_fini_stack:
6115         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
6116 err_lmi:
6117         if (lmi)
6118                 server_put_mount(dev, true);
6119         return(rc);
6120 }
6121
6122 /* For interoperability, the left element is old parameter, the right one
6123  * is the new version of the parameter, if some parameter is deprecated,
6124  * the new version should be set as NULL. */
6125 static struct cfg_interop_param mdt_interop_param[] = {
6126         { "mdt.group_upcall",   NULL },
6127         { "mdt.quota_type",     NULL },
6128         { "mdd.quota_type",     NULL },
6129         { "mdt.som",            NULL },
6130         { "mdt.rootsquash",     "mdt.root_squash" },
6131         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
6132         { NULL }
6133 };
6134
6135 /* used by MGS to process specific configurations */
6136 static int mdt_process_config(const struct lu_env *env,
6137                               struct lu_device *d, struct lustre_cfg *cfg)
6138 {
6139         struct mdt_device *m = mdt_dev(d);
6140         struct md_device *md_next = m->mdt_child;
6141         struct lu_device *next = md2lu_dev(md_next);
6142         int rc;
6143         ENTRY;
6144
6145         switch (cfg->lcfg_command) {
6146         case LCFG_PARAM: {
6147                 struct obd_device *obd = d->ld_obd;
6148                 /* For interoperability */
6149                 struct cfg_interop_param *ptr = NULL;
6150                 struct lustre_cfg *old_cfg = NULL;
6151                 char *param = NULL;
6152                 ssize_t count;
6153
6154                 param = lustre_cfg_string(cfg, 1);
6155                 if (param == NULL) {
6156                         CERROR("param is empty\n");
6157                         rc = -EINVAL;
6158                         break;
6159                 }
6160
6161                 ptr = class_find_old_param(param, mdt_interop_param);
6162                 if (ptr != NULL) {
6163                         if (ptr->new_param == NULL) {
6164                                 rc = 0;
6165                                 CWARN("For interoperability, skip this %s."
6166                                       " It is obsolete.\n", ptr->old_param);
6167                                 break;
6168                         }
6169
6170                         CWARN("Found old param %s, changed it to %s.\n",
6171                               ptr->old_param, ptr->new_param);
6172
6173                         old_cfg = cfg;
6174                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
6175                         if (IS_ERR(cfg)) {
6176                                 rc = PTR_ERR(cfg);
6177                                 break;
6178                         }
6179                 }
6180
6181                 count = class_modify_config(cfg, PARAM_MDT,
6182                                             &obd->obd_kset.kobj);
6183                 if (count < 0) {
6184                         struct coordinator *cdt = &m->mdt_coordinator;
6185
6186                         /* is it an HSM var ? */
6187                         count = class_modify_config(cfg, PARAM_HSM,
6188                                                     &cdt->cdt_hsm_kobj);
6189                         if (count < 0)
6190                                 /* we don't understand; pass it on */
6191                                 rc = next->ld_ops->ldo_process_config(env, next,
6192                                                                       cfg);
6193                         else
6194                                 rc = count > 0 ? 0 : count;
6195                 } else {
6196                         rc = count > 0 ? 0 : count;
6197                 }
6198
6199                 if (old_cfg)
6200                         OBD_FREE(cfg, lustre_cfg_len(cfg->lcfg_bufcount,
6201                                                      cfg->lcfg_buflens));
6202                 break;
6203         }
6204         default:
6205                 /* others are passed further */
6206                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
6207                 break;
6208         }
6209         RETURN(rc);
6210 }
6211
6212 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
6213                                           const struct lu_object_header *hdr,
6214                                           struct lu_device *d)
6215 {
6216         struct mdt_object *mo;
6217
6218         ENTRY;
6219
6220         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
6221         if (mo != NULL) {
6222                 struct lu_object *o;
6223                 struct lu_object_header *h;
6224
6225                 o = &mo->mot_obj;
6226                 h = &mo->mot_header;
6227                 lu_object_header_init(h);
6228                 lu_object_init(o, h, d);
6229                 lu_object_add_top(h, o);
6230                 o->lo_ops = &mdt_obj_ops;
6231                 spin_lock_init(&mo->mot_write_lock);
6232                 mutex_init(&mo->mot_som_mutex);
6233                 mutex_init(&mo->mot_lov_mutex);
6234                 init_rwsem(&mo->mot_dom_sem);
6235                 init_rwsem(&mo->mot_open_sem);
6236                 atomic_set(&mo->mot_open_count, 0);
6237                 mo->mot_restripe_offset = 0;
6238                 INIT_LIST_HEAD(&mo->mot_restripe_linkage);
6239                 RETURN(o);
6240         }
6241         RETURN(NULL);
6242 }
6243
6244 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
6245                            const struct lu_object_conf *unused)
6246 {
6247         struct mdt_device *d = mdt_dev(o->lo_dev);
6248         struct lu_device  *under;
6249         struct lu_object  *below;
6250         int                rc = 0;
6251         ENTRY;
6252
6253         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
6254                PFID(lu_object_fid(o)));
6255
6256         under = &d->mdt_child->md_lu_dev;
6257         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
6258         if (below != NULL) {
6259                 lu_object_add(o, below);
6260         } else
6261                 rc = -ENOMEM;
6262
6263         RETURN(rc);
6264 }
6265
6266 static void mdt_object_free_rcu(struct rcu_head *head)
6267 {
6268         struct mdt_object *mo = container_of(head, struct mdt_object,
6269                                              mot_header.loh_rcu);
6270
6271         kmem_cache_free(mdt_object_kmem, mo);
6272 }
6273
6274 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
6275 {
6276         struct mdt_object *mo = mdt_obj(o);
6277         struct lu_object_header *h;
6278         ENTRY;
6279
6280         h = o->lo_header;
6281         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
6282                PFID(lu_object_fid(o)));
6283
6284         LASSERT(atomic_read(&mo->mot_open_count) == 0);
6285         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
6286
6287         lu_object_fini(o);
6288         lu_object_header_fini(h);
6289         OBD_FREE_PRE(mo, sizeof(*mo), "slab-freed");
6290         call_rcu(&mo->mot_header.loh_rcu, mdt_object_free_rcu);
6291
6292         EXIT;
6293 }
6294
6295 static int mdt_object_print(const struct lu_env *env, void *cookie,
6296                             lu_printer_t p, const struct lu_object *o)
6297 {
6298         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
6299
6300         return (*p)(env, cookie,
6301                     LUSTRE_MDT_NAME"-object@%p(%s %s, writecount=%d)",
6302                     mdto, mdto->mot_lov_created ? "lov_created" : "",
6303                     mdto->mot_cache_attr ? "cache_attr" : "",
6304                     mdto->mot_write_count);
6305 }
6306
6307 static int mdt_prepare(const struct lu_env *env,
6308                 struct lu_device *pdev,
6309                 struct lu_device *cdev)
6310 {
6311         struct mdt_device *mdt = mdt_dev(cdev);
6312         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
6313         struct obd_device *obd = cdev->ld_obd;
6314         int rc;
6315
6316         ENTRY;
6317
6318         LASSERT(obd);
6319
6320         rc = next->ld_ops->ldo_prepare(env, cdev, next);
6321         if (rc)
6322                 RETURN(rc);
6323
6324         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
6325         if (rc)
6326                 RETURN(rc);
6327
6328         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
6329         if (rc)
6330                 RETURN(rc);
6331
6332         rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
6333         /* The LFSCK instance is registered just now, so it must be there when
6334          * register the namespace to such instance. */
6335         LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
6336
6337         if (mdt->mdt_seq_site.ss_node_id == 0) {
6338                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
6339                                                          &mdt->mdt_md_root_fid);
6340                 if (rc)
6341                         RETURN(rc);
6342         }
6343
6344         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
6345
6346         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
6347         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
6348         LASSERT(obd->obd_no_conn);
6349         spin_lock(&obd->obd_dev_lock);
6350         obd->obd_no_conn = 0;
6351         spin_unlock(&obd->obd_dev_lock);
6352
6353         if (obd->obd_recovering == 0)
6354                 mdt_postrecov(env, mdt);
6355
6356         RETURN(rc);
6357 }
6358
6359 const struct lu_device_operations mdt_lu_ops = {
6360         .ldo_object_alloc   = mdt_object_alloc,
6361         .ldo_process_config = mdt_process_config,
6362         .ldo_prepare        = mdt_prepare,
6363 };
6364
6365 static const struct lu_object_operations mdt_obj_ops = {
6366         .loo_object_init    = mdt_object_init,
6367         .loo_object_free    = mdt_object_free,
6368         .loo_object_print   = mdt_object_print
6369 };
6370
6371 static int mdt_obd_set_info_async(const struct lu_env *env,
6372                                   struct obd_export *exp,
6373                                   __u32 keylen, void *key,
6374                                   __u32 vallen, void *val,
6375                                   struct ptlrpc_request_set *set)
6376 {
6377         int rc;
6378
6379         ENTRY;
6380
6381         if (KEY_IS(KEY_SPTLRPC_CONF)) {
6382                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp));
6383                 RETURN(rc);
6384         }
6385
6386         RETURN(0);
6387 }
6388
6389 static inline void mdt_enable_slc(struct mdt_device *mdt)
6390 {
6391         if (mdt->mdt_lut.lut_sync_lock_cancel == SYNC_LOCK_CANCEL_NEVER)
6392                 mdt->mdt_lut.lut_sync_lock_cancel = SYNC_LOCK_CANCEL_BLOCKING;
6393 }
6394
6395 static inline void mdt_disable_slc(struct mdt_device *mdt)
6396 {
6397         if (mdt->mdt_lut.lut_sync_lock_cancel == SYNC_LOCK_CANCEL_BLOCKING)
6398                 mdt->mdt_lut.lut_sync_lock_cancel = SYNC_LOCK_CANCEL_NEVER;
6399 }
6400
6401 /**
6402  * Match client and server connection feature flags.
6403  *
6404  * Compute the compatibility flags for a connection request based on
6405  * features mutually supported by client and server.
6406  *
6407  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
6408  * must not be updated here, otherwise a partially initialized value may
6409  * be exposed. After the connection request is successfully processed,
6410  * the top-level MDT connect request handler atomically updates the export
6411  * connect flags from the obd_connect_data::ocd_connect_flags field of the
6412  * reply. \see mdt_connect().
6413  *
6414  * Before 2.7.50 clients will send a struct obd_connect_data_v1 rather than a
6415  * full struct obd_connect_data. So care must be taken when accessing fields
6416  * that are not present in struct obd_connect_data_v1. See LU-16.
6417  *
6418  * \param exp   the obd_export associated with this client/target pair
6419  * \param mdt   the target device for the connection
6420  * \param data  stores data for this connect request
6421  *
6422  * \retval 0       success
6423  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
6424  * \retval -EBADE  client and server feature requirements are incompatible
6425  */
6426 static int mdt_connect_internal(const struct lu_env *env,
6427                                 struct obd_export *exp,
6428                                 struct mdt_device *mdt,
6429                                 struct obd_connect_data *data, bool reconnect)
6430 {
6431         const char *obd_name = mdt_obd_name(mdt);
6432         LASSERT(data != NULL);
6433
6434         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
6435
6436         if (mdt->mdt_bottom->dd_rdonly &&
6437             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
6438             !(data->ocd_connect_flags & OBD_CONNECT_RDONLY))
6439                 RETURN(-EACCES);
6440
6441         if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
6442                 data->ocd_connect_flags2 &= MDT_CONNECT_SUPPORTED2;
6443
6444         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
6445
6446         if (!mdt->mdt_opts.mo_acl)
6447                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
6448
6449         if (!mdt->mdt_opts.mo_user_xattr)
6450                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
6451
6452         if (OCD_HAS_FLAG(data, BRW_SIZE)) {
6453                 data->ocd_brw_size = min(data->ocd_brw_size,
6454                                          mdt->mdt_brw_size);
6455                 if (data->ocd_brw_size == 0) {
6456                         CERROR("%s: cli %s/%p ocd_connect_flags: %#llx "
6457                                "ocd_version: %x ocd_grant: %d ocd_index: %u "
6458                                "ocd_brw_size unexpectedly zero, network data "
6459                                "corruption? Refusing to connect this client\n",
6460                                obd_name, exp->exp_client_uuid.uuid,
6461                                exp, data->ocd_connect_flags, data->ocd_version,
6462                                data->ocd_grant, data->ocd_index);
6463                         return -EPROTO;
6464                 }
6465         }
6466
6467         if (OCD_HAS_FLAG(data, GRANT_PARAM)) {
6468                 struct dt_device_param *ddp = &mdt->mdt_lut.lut_dt_conf;
6469
6470                 /* client is reporting its page size, for future use */
6471                 exp->exp_target_data.ted_pagebits = data->ocd_grant_blkbits;
6472                 data->ocd_grant_blkbits  = mdt->mdt_lut.lut_tgd.tgd_blockbits;
6473                 /* ddp_inodespace may not be power-of-two value, eg. for ldiskfs
6474                  * it's LDISKFS_DIR_REC_LEN(20) = 28. */
6475                 data->ocd_grant_inobits = fls(ddp->ddp_inodespace - 1);
6476                 /* ocd_grant_tax_kb is in 1K byte blocks */
6477                 data->ocd_grant_tax_kb = ddp->ddp_extent_tax >> 10;
6478                 data->ocd_grant_max_blks = ddp->ddp_max_extent_blks;
6479         }
6480
6481         /* Save connect_data we have so far because tgt_grant_connect()
6482          * uses it to calculate grant, and we want to save the client
6483          * version before it is overwritten by LUSTRE_VERSION_CODE. */
6484         exp->exp_connect_data = *data;
6485         if (OCD_HAS_FLAG(data, GRANT))
6486                 tgt_grant_connect(env, exp, data, !reconnect);
6487
6488         if (OCD_HAS_FLAG(data, MAXBYTES))
6489                 data->ocd_maxbytes = mdt->mdt_lut.lut_dt_conf.ddp_maxbytes;
6490
6491         /* NB: Disregard the rule against updating
6492          * exp_connect_data.ocd_connect_flags in this case, since
6493          * tgt_client_new() needs to know if this is a lightweight
6494          * connection, and it is safe to expose this flag before
6495          * connection processing completes. */
6496         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
6497                 spin_lock(&exp->exp_lock);
6498                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
6499                 spin_unlock(&exp->exp_lock);
6500         }
6501
6502         data->ocd_version = LUSTRE_VERSION_CODE;
6503
6504         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
6505                 CWARN("%s: MDS requires FID support, but client not\n",
6506                       obd_name);
6507                 return -EBADE;
6508         }
6509
6510         if (OCD_HAS_FLAG(data, PINGLESS)) {
6511                 if (ptlrpc_pinger_suppress_pings()) {
6512                         spin_lock(&exp->exp_obd->obd_dev_lock);
6513                         list_del_init(&exp->exp_obd_chain_timed);
6514                         spin_unlock(&exp->exp_obd->obd_dev_lock);
6515                 } else {
6516                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
6517                 }
6518         }
6519
6520         data->ocd_max_easize = mdt->mdt_max_ea_size;
6521
6522         /* NB: Disregard the rule against updating
6523          * exp_connect_data.ocd_connect_flags in this case, since
6524          * tgt_client_new() needs to know if this is client supports
6525          * multiple modify RPCs, and it is safe to expose this flag before
6526          * connection processing completes. */
6527         if (data->ocd_connect_flags & OBD_CONNECT_MULTIMODRPCS) {
6528                 data->ocd_maxmodrpcs = max_mod_rpcs_per_client;
6529                 spin_lock(&exp->exp_lock);
6530                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_MULTIMODRPCS;
6531                 spin_unlock(&exp->exp_lock);
6532         }
6533
6534         if (OCD_HAS_FLAG(data, CKSUM)) {
6535                 __u32 cksum_types = data->ocd_cksum_types;
6536
6537                 tgt_mask_cksum_types(&mdt->mdt_lut, &data->ocd_cksum_types);
6538
6539                 if (unlikely(data->ocd_cksum_types == 0)) {
6540                         CERROR("%s: Connect with checksum support but no "
6541                                "ocd_cksum_types is set\n",
6542                                exp->exp_obd->obd_name);
6543                         RETURN(-EPROTO);
6544                 }
6545
6546                 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
6547                        "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
6548                        cksum_types, data->ocd_cksum_types);
6549         } else {
6550                 /* This client does not support OBD_CONNECT_CKSUM
6551                  * fall back to CRC32 */
6552                 CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
6553                        "OBD_CONNECT_CKSUM, CRC32 will be used\n",
6554                        exp->exp_obd->obd_name, obd_export_nid2str(exp));
6555         }
6556
6557         if ((data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
6558             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
6559                 atomic_inc(&mdt->mdt_mds_mds_conns);
6560                 mdt_enable_slc(mdt);
6561         }
6562
6563         if (!mdt->mdt_lut.lut_dt_conf.ddp_has_lseek_data_hole)
6564                 data->ocd_connect_flags2 &= ~OBD_CONNECT2_LSEEK;
6565
6566         return 0;
6567 }
6568
6569 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
6570                                    struct mdt_thread_info *info,
6571                                    struct mdt_file_data *mfd)
6572 {
6573         struct lu_context ses;
6574         int rc;
6575         ENTRY;
6576
6577         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
6578         if (rc)
6579                 RETURN(rc);
6580
6581         env->le_ses = &ses;
6582         lu_context_enter(&ses);
6583
6584         mdt_ucred(info)->uc_valid = UCRED_OLD;
6585         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
6586
6587         lu_context_exit(&ses);
6588         lu_context_fini(&ses);
6589         env->le_ses = NULL;
6590
6591         RETURN(rc);
6592 }
6593
6594 static int mdt_export_cleanup(struct obd_export *exp)
6595 {
6596         LIST_HEAD(closing_list);
6597         struct mdt_export_data  *med = &exp->exp_mdt_data;
6598         struct obd_device       *obd = exp->exp_obd;
6599         struct mdt_device       *mdt;
6600         struct mdt_thread_info  *info;
6601         struct lu_env            env;
6602         struct mdt_file_data    *mfd, *n;
6603         int rc = 0;
6604         ENTRY;
6605
6606         spin_lock(&med->med_open_lock);
6607         while (!list_empty(&med->med_open_head)) {
6608                 struct list_head *tmp = med->med_open_head.next;
6609                 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
6610
6611                 /* Remove mfd handle so it can't be found again.
6612                  * We are consuming the mfd_list reference here. */
6613                 class_handle_unhash(&mfd->mfd_open_handle);
6614                 list_move_tail(&mfd->mfd_list, &closing_list);
6615         }
6616         spin_unlock(&med->med_open_lock);
6617         mdt = mdt_dev(obd->obd_lu_dev);
6618         LASSERT(mdt != NULL);
6619
6620         rc = lu_env_init(&env, LCT_MD_THREAD);
6621         if (rc)
6622                 RETURN(rc);
6623
6624         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6625         LASSERT(info != NULL);
6626         memset(info, 0, sizeof *info);
6627         info->mti_env = &env;
6628         info->mti_mdt = mdt;
6629         info->mti_exp = exp;
6630
6631         if (!list_empty(&closing_list)) {
6632                 struct md_attr *ma = &info->mti_attr;
6633
6634                 /* Close any open files (which may also cause orphan
6635                  * unlinking). */
6636                 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
6637                         list_del_init(&mfd->mfd_list);
6638                         ma->ma_need = ma->ma_valid = 0;
6639
6640                         /* This file is being closed due to an eviction, it
6641                          * could have been modified and now dirty regarding to
6642                          * HSM archive, check this!
6643                          * The logic here is to mark a file dirty if there's a
6644                          * chance it was dirtied before the client was evicted,
6645                          * so that we don't have to wait for a release attempt
6646                          * before finding out the file was actually dirty and
6647                          * fail the release. Aggressively marking it dirty here
6648                          * will cause the policy engine to attempt to
6649                          * re-archive it; when rearchiving, we can compare the
6650                          * current version to the HSM data_version and make the
6651                          * archive request into a noop if it's not actually
6652                          * dirty.
6653                          */
6654                         if (mfd->mfd_open_flags & MDS_FMODE_WRITE)
6655                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
6656
6657                         /* Don't unlink orphan on failover umount, LU-184 */
6658                         if (exp->exp_flags & OBD_OPT_FAILOVER ||
6659                             exp->exp_obd->obd_stopping) {
6660                                 ma->ma_valid = MA_FLAGS;
6661                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
6662                         }
6663                         ma->ma_valid |= MA_FORCE_LOG;
6664                         mdt_mfd_close(info, mfd);
6665                 }
6666         }
6667         info->mti_mdt = NULL;
6668         /* cleanup client slot early */
6669         /* Do not erase record for recoverable client. */
6670         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
6671                 tgt_client_del(&env, exp);
6672         lu_env_fini(&env);
6673
6674         RETURN(rc);
6675 }
6676
6677 static int mdt_obd_disconnect(struct obd_export *exp)
6678 {
6679         int rc;
6680
6681         ENTRY;
6682
6683         LASSERT(exp);
6684         class_export_get(exp);
6685
6686         if (!(exp->exp_flags & OBD_OPT_FORCE))
6687                 tgt_grant_sanity_check(exp->exp_obd, __func__);
6688
6689         if ((exp_connect_flags(exp) & OBD_CONNECT_MDS_MDS) &&
6690             !(exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT)) {
6691                 struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
6692
6693                 if (atomic_dec_and_test(&mdt->mdt_mds_mds_conns))
6694                         mdt_disable_slc(mdt);
6695         }
6696
6697         rc = server_disconnect_export(exp);
6698         if (rc != 0)
6699                 CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
6700
6701         tgt_grant_discard(exp);
6702
6703         rc = mdt_export_cleanup(exp);
6704         nodemap_del_member(exp);
6705         class_export_put(exp);
6706         RETURN(rc);
6707 }
6708
6709 /* mds_connect copy */
6710 static int mdt_obd_connect(const struct lu_env *env,
6711                            struct obd_export **exp, struct obd_device *obd,
6712                            struct obd_uuid *cluuid,
6713                            struct obd_connect_data *data,
6714                            void *localdata)
6715 {
6716         struct obd_export       *lexp;
6717         struct lustre_handle    conn = { 0 };
6718         struct mdt_device       *mdt;
6719         int                      rc;
6720         lnet_nid_t              *client_nid = localdata;
6721         ENTRY;
6722
6723         LASSERT(env != NULL);
6724         LASSERT(data != NULL);
6725
6726         if (!exp || !obd || !cluuid)
6727                 RETURN(-EINVAL);
6728
6729         mdt = mdt_dev(obd->obd_lu_dev);
6730
6731         /*
6732          * first, check whether the stack is ready to handle requests
6733          * XXX: probably not very appropriate method is used now
6734          *      at some point we should find a better one
6735          */
6736         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) &&
6737             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) &&
6738             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
6739                 rc = obd_get_info(env, mdt->mdt_child_exp,
6740                                   sizeof(KEY_OSP_CONNECTED),
6741                                   KEY_OSP_CONNECTED, NULL, NULL);
6742                 if (rc)
6743                         RETURN(-EAGAIN);
6744                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
6745         }
6746
6747         rc = class_connect(&conn, obd, cluuid);
6748         if (rc)
6749                 RETURN(rc);
6750
6751         lexp = class_conn2export(&conn);
6752         LASSERT(lexp != NULL);
6753
6754         rc = nodemap_add_member(*client_nid, lexp);
6755         if (rc != 0 && rc != -EEXIST)
6756                 GOTO(out, rc);
6757
6758         rc = mdt_connect_internal(env, lexp, mdt, data, false);
6759         if (rc == 0) {
6760                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
6761
6762                 LASSERT(lcd);
6763                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
6764                 rc = tgt_client_new(env, lexp);
6765                 if (rc == 0)
6766                         mdt_export_stats_init(obd, lexp, localdata);
6767         }
6768 out:
6769         if (rc != 0) {
6770                 class_disconnect(lexp);
6771                 nodemap_del_member(lexp);
6772                 *exp = NULL;
6773         } else {
6774                 *exp = lexp;
6775                 /* Because we do not want this export to be evicted by pinger,
6776                  * let's not add this export to the timed chain list. */
6777                 if (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) {
6778                         spin_lock(&lexp->exp_obd->obd_dev_lock);
6779                         list_del_init(&lexp->exp_obd_chain_timed);
6780                         spin_unlock(&lexp->exp_obd->obd_dev_lock);
6781                 }
6782         }
6783
6784         RETURN(rc);
6785 }
6786
6787 static int mdt_obd_reconnect(const struct lu_env *env,
6788                              struct obd_export *exp, struct obd_device *obd,
6789                              struct obd_uuid *cluuid,
6790                              struct obd_connect_data *data,
6791                              void *localdata)
6792 {
6793         lnet_nid_t             *client_nid = localdata;
6794         int                     rc;
6795         ENTRY;
6796
6797         if (exp == NULL || obd == NULL || cluuid == NULL)
6798                 RETURN(-EINVAL);
6799
6800         rc = nodemap_add_member(*client_nid, exp);
6801         if (rc != 0 && rc != -EEXIST)
6802                 RETURN(rc);
6803
6804         rc = mdt_connect_internal(env, exp, mdt_dev(obd->obd_lu_dev), data,
6805                                   true);
6806         if (rc == 0)
6807                 mdt_export_stats_init(obd, exp, localdata);
6808         else
6809                 nodemap_del_member(exp);
6810
6811         RETURN(rc);
6812 }
6813
6814 /* FIXME: Can we avoid using these two interfaces? */
6815 static int mdt_init_export(struct obd_export *exp)
6816 {
6817         struct mdt_export_data *med = &exp->exp_mdt_data;
6818         int                     rc;
6819         ENTRY;
6820
6821         INIT_LIST_HEAD(&med->med_open_head);
6822         spin_lock_init(&med->med_open_lock);
6823         spin_lock(&exp->exp_lock);
6824         exp->exp_connecting = 1;
6825         spin_unlock(&exp->exp_lock);
6826
6827         OBD_ALLOC(exp->exp_used_slots,
6828                   BITS_TO_LONGS(OBD_MAX_RIF_MAX) * sizeof(long));
6829         if (exp->exp_used_slots == NULL)
6830                 RETURN(-ENOMEM);
6831
6832         /* self-export doesn't need client data and ldlm initialization */
6833         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
6834                                      &exp->exp_client_uuid)))
6835                 RETURN(0);
6836
6837         rc = tgt_client_alloc(exp);
6838         if (rc)
6839                 GOTO(err, rc);
6840
6841         rc = ldlm_init_export(exp);
6842         if (rc)
6843                 GOTO(err_free, rc);
6844
6845         RETURN(rc);
6846
6847 err_free:
6848         tgt_client_free(exp);
6849 err:
6850         OBD_FREE(exp->exp_used_slots,
6851                  BITS_TO_LONGS(OBD_MAX_RIF_MAX) * sizeof(long));
6852         exp->exp_used_slots = NULL;
6853
6854         CERROR("%s: Failed to initialize export: rc = %d\n",
6855                exp->exp_obd->obd_name, rc);
6856         return rc;
6857 }
6858
6859 static int mdt_destroy_export(struct obd_export *exp)
6860 {
6861         ENTRY;
6862
6863         target_destroy_export(exp);
6864         if (exp->exp_used_slots)
6865                 OBD_FREE(exp->exp_used_slots,
6866                          BITS_TO_LONGS(OBD_MAX_RIF_MAX) * sizeof(long));
6867
6868         /* destroy can be called from failed obd_setup, so
6869          * checking uuid is safer than obd_self_export */
6870         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
6871                                      &exp->exp_client_uuid)))
6872                 RETURN(0);
6873
6874         ldlm_destroy_export(exp);
6875         tgt_client_free(exp);
6876
6877         LASSERT(list_empty(&exp->exp_outstanding_replies));
6878         LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
6879
6880         /*
6881          * discard grants once we're sure no more
6882          * interaction with the client is possible
6883          */
6884         tgt_grant_discard(exp);
6885         if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
6886                 exp->exp_obd->u.obt.obt_lut->lut_tgd.tgd_tot_granted_clients--;
6887
6888         if (!(exp->exp_flags & OBD_OPT_FORCE))
6889                 tgt_grant_sanity_check(exp->exp_obd, __func__);
6890
6891         RETURN(0);
6892 }
6893
6894 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
6895                    struct linkea_data *ldata)
6896 {
6897         int rc;
6898
6899         LASSERT(ldata->ld_buf->lb_buf != NULL);
6900
6901         if (!mdt_object_exists(mdt_obj))
6902                 return -ENODATA;
6903
6904         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6905                           ldata->ld_buf, XATTR_NAME_LINK);
6906         if (rc == -ERANGE) {
6907                 /* Buf was too small, figure out what we need. */
6908                 lu_buf_free(ldata->ld_buf);
6909                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6910                                   ldata->ld_buf, XATTR_NAME_LINK);
6911                 if (rc < 0)
6912                         return rc;
6913                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
6914                 if (ldata->ld_buf->lb_buf == NULL)
6915                         return -ENOMEM;
6916                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6917                                   ldata->ld_buf, XATTR_NAME_LINK);
6918         }
6919         if (rc < 0)
6920                 return rc;
6921
6922         return linkea_init_with_rec(ldata);
6923 }
6924
6925 /**
6926  * Given an MDT object, try to look up the full path to the object.
6927  * Part of the MDT layer implementation of lfs fid2path.
6928  *
6929  * \param[in]     info  Per-thread common data shared by MDT level handlers.
6930  * \param[in]     obj   Object to do path lookup of
6931  * \param[in,out] fp    User-provided struct to store path information
6932  * \param[in]     root_fid Root FID of current path should reach
6933  *
6934  * \retval 0 Lookup successful, path information stored in fp
6935  * \retval -EAGAIN Lookup failed, usually because object is being moved
6936  * \retval negative errno if there was a problem
6937  */
6938 static int mdt_path_current(struct mdt_thread_info *info,
6939                             struct mdt_object *obj,
6940                             struct getinfo_fid2path *fp,
6941                             struct lu_fid *root_fid)
6942 {
6943         struct mdt_device *mdt = info->mti_mdt;
6944         struct lu_name *tmpname = &info->mti_name;
6945         struct lu_fid *tmpfid = &info->mti_tmp_fid1;
6946         struct lu_buf *buf = &info->mti_big_buf;
6947         struct linkea_data ldata = { NULL };
6948         bool first = true;
6949         struct mdt_object *mdt_obj;
6950         struct link_ea_header *leh;
6951         struct link_ea_entry *lee;
6952         char *ptr;
6953         int reclen;
6954         int rc = 0;
6955
6956         ENTRY;
6957
6958         /* temp buffer for path element, the buffer will be finally freed
6959          * in mdt_thread_info_fini */
6960         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
6961         if (buf->lb_buf == NULL)
6962                 RETURN(-ENOMEM);
6963
6964         ldata.ld_buf = buf;
6965         ptr = fp->gf_u.gf_path + fp->gf_pathlen - 1;
6966         *ptr = 0;
6967         --ptr;
6968         *tmpfid = fp->gf_fid = *mdt_object_fid(obj);
6969
6970         while (!lu_fid_eq(root_fid, &fp->gf_fid)) {
6971                 if (!lu_fid_eq(root_fid, &mdt->mdt_md_root_fid) &&
6972                     lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid))
6973                         GOTO(out, rc = -ENOENT);
6974
6975                 if (lu_fid_eq(mdt_object_fid(obj), tmpfid)) {
6976                         mdt_obj = obj;
6977                         mdt_object_get(info->mti_env, mdt_obj);
6978                 } else {
6979                         mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
6980                         if (IS_ERR(mdt_obj))
6981                                 GOTO(out, rc = PTR_ERR(mdt_obj));
6982                 }
6983
6984                 if (!mdt_object_exists(mdt_obj)) {
6985                         mdt_object_put(info->mti_env, mdt_obj);
6986                         GOTO(out, rc = -ENOENT);
6987                 }
6988
6989                 if (mdt_object_remote(mdt_obj)) {
6990                         mdt_object_put(info->mti_env, mdt_obj);
6991                         GOTO(remote_out, rc = -EREMOTE);
6992                 }
6993
6994                 rc = mdt_links_read(info, mdt_obj, &ldata);
6995                 if (rc != 0) {
6996                         mdt_object_put(info->mti_env, mdt_obj);
6997                         GOTO(out, rc);
6998                 }
6999
7000                 leh = buf->lb_buf;
7001                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
7002                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
7003                 /* If set, use link #linkno for path lookup, otherwise use
7004                    link #0.  Only do this for the final path element. */
7005                 if (first && fp->gf_linkno < leh->leh_reccount) {
7006                         int count;
7007                         for (count = 0; count < fp->gf_linkno; count++) {
7008                                 lee = (struct link_ea_entry *)
7009                                      ((char *)lee + reclen);
7010                                 linkea_entry_unpack(lee, &reclen, tmpname,
7011                                                     tmpfid);
7012                         }
7013                         if (fp->gf_linkno < leh->leh_reccount - 1)
7014                                 /* indicate to user there are more links */
7015                                 fp->gf_linkno++;
7016                 }
7017
7018                 /* Check if it is slave stripes */
7019                 rc = mdt_is_dir_stripe(info, mdt_obj);
7020                 mdt_object_put(info->mti_env, mdt_obj);
7021                 if (rc < 0)
7022                         GOTO(out, rc);
7023                 if (rc == 1) {
7024                         fp->gf_fid = *tmpfid;
7025                         continue;
7026                 }
7027
7028                 /* Pack the name in the end of the buffer */
7029                 ptr -= tmpname->ln_namelen;
7030                 if (ptr - 1 <= fp->gf_u.gf_path)
7031                         GOTO(out, rc = -EOVERFLOW);
7032                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
7033                 *(--ptr) = '/';
7034
7035                 /* keep the last resolved fid to the client, so the
7036                  * client will build the left path on another MDT for
7037                  * remote object */
7038                 fp->gf_fid = *tmpfid;
7039
7040                 first = false;
7041         }
7042
7043         /* non-zero will be treated as an error */
7044         rc = 0;
7045
7046 remote_out:
7047         ptr++; /* skip leading / */
7048         memmove(fp->gf_u.gf_path, ptr,
7049                 fp->gf_u.gf_path + fp->gf_pathlen - ptr);
7050
7051 out:
7052         RETURN(rc);
7053 }
7054
7055 /**
7056  * Given an MDT object, use mdt_path_current to get the path.
7057  * Essentially a wrapper to retry mdt_path_current a set number of times
7058  * if -EAGAIN is returned (usually because an object is being moved).
7059  *
7060  * Part of the MDT layer implementation of lfs fid2path.
7061  *
7062  * \param[in]     info  Per-thread common data shared by mdt level handlers.
7063  * \param[in]     obj   Object to do path lookup of
7064  * \param[in,out] fp    User-provided struct for arguments and to store path
7065  *                      information
7066  *
7067  * \retval 0 Lookup successful, path information stored in fp
7068  * \retval negative errno if there was a problem
7069  */
7070 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
7071                     struct getinfo_fid2path *fp, struct lu_fid *root_fid)
7072 {
7073         struct mdt_device       *mdt = info->mti_mdt;
7074         int                     tries = 3;
7075         int                     rc = -EAGAIN;
7076         ENTRY;
7077
7078         if (fp->gf_pathlen < 3)
7079                 RETURN(-EOVERFLOW);
7080
7081         if (root_fid == NULL)
7082                 root_fid = &mdt->mdt_md_root_fid;
7083
7084         if (lu_fid_eq(root_fid, mdt_object_fid(obj))) {
7085                 fp->gf_u.gf_path[0] = '\0';
7086                 RETURN(0);
7087         }
7088
7089         /* Retry multiple times in case file is being moved */
7090         while (tries-- && rc == -EAGAIN)
7091                 rc = mdt_path_current(info, obj, fp, root_fid);
7092
7093         RETURN(rc);
7094 }
7095
7096 /**
7097  * Get the full path of the provided FID, as of changelog record recno.
7098  *
7099  * This checks sanity and looks up object for user provided FID
7100  * before calling the actual path lookup code.
7101  *
7102  * Part of the MDT layer implementation of lfs fid2path.
7103  *
7104  * \param[in]     info  Per-thread common data shared by mdt level handlers.
7105  * \param[in,out] fp    User-provided struct for arguments and to store path
7106  *                      information
7107  *
7108  * \retval 0 Lookup successful, path information and recno stored in fp
7109  * \retval -ENOENT, object does not exist
7110  * \retval negative errno if there was a problem
7111  */
7112 static int mdt_fid2path(struct mdt_thread_info *info,
7113                         struct lu_fid *root_fid,
7114                         struct getinfo_fid2path *fp)
7115 {
7116         struct mdt_device *mdt = info->mti_mdt;
7117         struct mdt_object *obj;
7118         int    rc;
7119         ENTRY;
7120
7121         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
7122                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
7123
7124         if (!fid_is_sane(&fp->gf_fid))
7125                 RETURN(-EINVAL);
7126
7127         if (!fid_is_namespace_visible(&fp->gf_fid)) {
7128                 CDEBUG(D_INFO, "%s: "DFID" is invalid, f_seq should be >= %#llx"
7129                        ", or f_oid != 0, or f_ver == 0\n", mdt_obd_name(mdt),
7130                        PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
7131                 RETURN(-EINVAL);
7132         }
7133
7134         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
7135         if (IS_ERR(obj)) {
7136                 rc = PTR_ERR(obj);
7137                 CDEBUG(D_IOCTL, "cannot find "DFID": rc = %d\n",
7138                        PFID(&fp->gf_fid), rc);
7139                 RETURN(rc);
7140         }
7141
7142         if (mdt_object_remote(obj)) {
7143                 rc = -EREMOTE;
7144         } else if (!mdt_object_exists(obj)) {
7145                 rc = -ENOENT;
7146         } else {
7147                 struct lu_attr la = { 0 };
7148                 struct dt_object *dt = mdt_obj2dt(obj);
7149
7150                 if (dt && dt->do_ops && dt->do_ops->do_attr_get)
7151                         dt_attr_get(info->mti_env, mdt_obj2dt(obj), &la);
7152                 if (la.la_valid & LA_FLAGS && la.la_flags & LUSTRE_ENCRYPT_FL)
7153                         /* path resolution cannot be carried out on server
7154                          * side for encrypted files
7155                          */
7156                         rc = -ENODATA;
7157                 else
7158                         rc = 0;
7159         }
7160
7161         if (rc < 0) {
7162                 mdt_object_put(info->mti_env, obj);
7163                 CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n",
7164                        PFID(&fp->gf_fid), rc);
7165                 RETURN(rc);
7166         }
7167
7168         rc = mdt_path(info, obj, fp, root_fid);
7169
7170         CDEBUG(D_INFO, "fid "DFID", path %s recno %#llx linkno %u\n",
7171                PFID(&fp->gf_fid), fp->gf_u.gf_path,
7172                fp->gf_recno, fp->gf_linkno);
7173
7174         mdt_object_put(info->mti_env, obj);
7175
7176         RETURN(rc);
7177 }
7178
7179 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, int keylen,
7180                             void *val, int vallen)
7181 {
7182         struct getinfo_fid2path *fpout, *fpin;
7183         struct lu_fid *root_fid = NULL;
7184         int rc = 0;
7185
7186         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
7187         fpout = val;
7188
7189         if (req_capsule_req_need_swab(info->mti_pill))
7190                 lustre_swab_fid2path(fpin);
7191
7192         memcpy(fpout, fpin, sizeof(*fpin));
7193         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
7194                 RETURN(-EINVAL);
7195
7196         if (keylen >= cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*fpin) +
7197                       sizeof(struct lu_fid)) {
7198                 /* client sent its root FID, which is normally fileset FID */
7199                 root_fid = fpin->gf_u.gf_root_fid;
7200                 if (req_capsule_req_need_swab(info->mti_pill))
7201                         lustre_swab_lu_fid(root_fid);
7202
7203                 if (root_fid != NULL && !fid_is_sane(root_fid))
7204                         RETURN(-EINVAL);
7205         }
7206
7207         rc = mdt_fid2path(info, root_fid, fpout);
7208         RETURN(rc);
7209 }
7210
7211 int mdt_get_info(struct tgt_session_info *tsi)
7212 {
7213         char    *key;
7214         int      keylen;
7215         __u32   *vallen;
7216         void    *valout;
7217         int      rc;
7218
7219         ENTRY;
7220
7221         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
7222         if (key == NULL) {
7223                 CDEBUG(D_IOCTL, "No GETINFO key\n");
7224                 RETURN(err_serious(-EFAULT));
7225         }
7226         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
7227                                       RCL_CLIENT);
7228
7229         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
7230         if (vallen == NULL) {
7231                 CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n",
7232                                 tgt_name(tsi->tsi_tgt));
7233                 RETURN(err_serious(-EFAULT));
7234         }
7235
7236         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
7237                              *vallen);
7238         rc = req_capsule_server_pack(tsi->tsi_pill);
7239         if (rc)
7240                 RETURN(err_serious(rc));
7241
7242         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
7243         if (valout == NULL) {
7244                 CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n",
7245                                 tgt_name(tsi->tsi_tgt));
7246                 RETURN(err_serious(-EFAULT));
7247         }
7248
7249         if (KEY_IS(KEY_FID2PATH)) {
7250                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
7251
7252                 rc = mdt_rpc_fid2path(info, key, keylen, valout, *vallen);
7253                 mdt_thread_info_fini(info);
7254         } else {
7255                 rc = -EINVAL;
7256         }
7257         RETURN(rc);
7258 }
7259
7260 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
7261 {
7262         struct obd_ioctl_data *data = karg;
7263         struct lu_fid *fid;
7264         __u64 version;
7265         struct mdt_object *obj;
7266         struct mdt_lock_handle  *lh;
7267         int rc;
7268         ENTRY;
7269
7270         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
7271             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
7272                 RETURN(-EINVAL);
7273
7274         fid = (struct lu_fid *)data->ioc_inlbuf1;
7275
7276         if (!fid_is_sane(fid))
7277                 RETURN(-EINVAL);
7278
7279         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
7280
7281         lh = &mti->mti_lh[MDT_LH_PARENT];
7282         mdt_lock_reg_init(lh, LCK_CR);
7283
7284         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
7285         if (IS_ERR(obj))
7286                 RETURN(PTR_ERR(obj));
7287
7288         if (mdt_object_remote(obj)) {
7289                 rc = -EREMOTE;
7290                 /**
7291                  * before calling version get the correct MDS should be
7292                  * fid, this is error to find remote object here
7293                  */
7294                 CERROR("nonlocal object "DFID"\n", PFID(fid));
7295         } else if (!mdt_object_exists(obj)) {
7296                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
7297                 rc = -ENOENT;
7298         } else {
7299                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
7300                *(__u64 *)data->ioc_inlbuf2 = version;
7301                 rc = 0;
7302         }
7303         mdt_object_unlock_put(mti, obj, lh, 1);
7304         RETURN(rc);
7305 }
7306
7307 /* ioctls on obd dev */
7308 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
7309                          void *karg, void __user *uarg)
7310 {
7311         struct lu_env      env;
7312         struct obd_device *obd = exp->exp_obd;
7313         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
7314         struct dt_device  *dt = mdt->mdt_bottom;
7315         int rc;
7316
7317         ENTRY;
7318         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
7319         rc = lu_env_init(&env, LCT_MD_THREAD);
7320         if (rc)
7321                 RETURN(rc);
7322
7323         switch (cmd) {
7324         case OBD_IOC_SYNC:
7325                 rc = mdt_device_sync(&env, mdt);
7326                 break;
7327         case OBD_IOC_SET_READONLY:
7328                 rc = dt_sync(&env, dt);
7329                 if (rc == 0)
7330                         rc = dt_ro(&env, dt);
7331                 break;
7332         case OBD_IOC_ABORT_RECOVERY: {
7333                 struct obd_ioctl_data *data = karg;
7334
7335                 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
7336                 if (data->ioc_type & OBD_FLG_ABORT_RECOV_MDT) {
7337                         obd->obd_abort_recov_mdt = 1;
7338                         wake_up(&obd->obd_next_transno_waitq);
7339                 } else { /* if (data->ioc_type & OBD_FLG_ABORT_RECOV_OST) */
7340                         /* lctl didn't set OBD_FLG_ABORT_RECOV_OST < 2.13.57 */
7341                         obd->obd_abort_recovery = 1;
7342                         target_stop_recovery_thread(obd);
7343                 }
7344                 rc = 0;
7345                 break;
7346         }
7347         case OBD_IOC_CHANGELOG_REG:
7348         case OBD_IOC_CHANGELOG_DEREG:
7349         case OBD_IOC_CHANGELOG_CLEAR:
7350                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(&env,
7351                                                            mdt->mdt_child,
7352                                                            cmd, len, karg);
7353                 break;
7354         case OBD_IOC_START_LFSCK: {
7355                 struct md_device *next = mdt->mdt_child;
7356                 struct obd_ioctl_data *data = karg;
7357                 struct lfsck_start_param lsp;
7358
7359                 if (unlikely(data == NULL)) {
7360                         rc = -EINVAL;
7361                         break;
7362                 }
7363
7364                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
7365                 lsp.lsp_index_valid = 0;
7366                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
7367                 break;
7368         }
7369         case OBD_IOC_STOP_LFSCK: {
7370                 struct md_device        *next = mdt->mdt_child;
7371                 struct obd_ioctl_data   *data = karg;
7372                 struct lfsck_stop        stop;
7373
7374                 stop.ls_status = LS_STOPPED;
7375                 /* Old lfsck utils may pass NULL @stop. */
7376                 if (data->ioc_inlbuf1 == NULL)
7377                         stop.ls_flags = 0;
7378                 else
7379                         stop.ls_flags =
7380                         ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
7381
7382                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
7383                 break;
7384         }
7385         case OBD_IOC_QUERY_LFSCK: {
7386                 struct md_device        *next = mdt->mdt_child;
7387                 struct obd_ioctl_data   *data = karg;
7388
7389                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0,
7390                                                  data->ioc_inlbuf1);
7391                 break;
7392         }
7393         case OBD_IOC_GET_OBJ_VERSION: {
7394                 struct mdt_thread_info *mti;
7395                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
7396                 memset(mti, 0, sizeof *mti);
7397                 mti->mti_env = &env;
7398                 mti->mti_mdt = mdt;
7399                 mti->mti_exp = exp;
7400
7401                 rc = mdt_ioc_version_get(mti, karg);
7402                 break;
7403         }
7404         case OBD_IOC_CATLOGLIST: {
7405                 struct mdt_thread_info *mti;
7406
7407                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
7408                 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
7409                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
7410                                        &mti->mti_tmp_fid1);
7411                 break;
7412          }
7413         default:
7414                 rc = -EOPNOTSUPP;
7415                 CERROR("%s: Not supported cmd = %d, rc = %d\n",
7416                         mdt_obd_name(mdt), cmd, rc);
7417         }
7418
7419         lu_env_fini(&env);
7420         RETURN(rc);
7421 }
7422
7423 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
7424 {
7425         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
7426         int rc;
7427         ENTRY;
7428
7429         if (!mdt->mdt_skip_lfsck && !mdt->mdt_bottom->dd_rdonly) {
7430                 struct lfsck_start_param lsp;
7431
7432                 lsp.lsp_start = NULL;
7433                 lsp.lsp_index_valid = 0;
7434                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
7435                                                            OBD_IOC_START_LFSCK,
7436                                                            0, &lsp);
7437                 if (rc != 0 && rc != -EALREADY)
7438                         CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
7439                               mdt_obd_name(mdt), rc);
7440         }
7441
7442         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
7443         RETURN(rc);
7444 }
7445
7446 static int mdt_obd_postrecov(struct obd_device *obd)
7447 {
7448         struct lu_env env;
7449         int rc;
7450
7451         rc = lu_env_init(&env, LCT_MD_THREAD);
7452         if (rc)
7453                 RETURN(rc);
7454         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
7455         lu_env_fini(&env);
7456         return rc;
7457 }
7458
7459 static const struct obd_ops mdt_obd_device_ops = {
7460         .o_owner          = THIS_MODULE,
7461         .o_set_info_async = mdt_obd_set_info_async,
7462         .o_connect        = mdt_obd_connect,
7463         .o_reconnect      = mdt_obd_reconnect,
7464         .o_disconnect     = mdt_obd_disconnect,
7465         .o_init_export    = mdt_init_export,
7466         .o_destroy_export = mdt_destroy_export,
7467         .o_iocontrol      = mdt_iocontrol,
7468         .o_postrecov      = mdt_obd_postrecov,
7469         /* Data-on-MDT IO methods */
7470         .o_preprw         = mdt_obd_preprw,
7471         .o_commitrw       = mdt_obd_commitrw,
7472 };
7473
7474 static struct lu_device* mdt_device_fini(const struct lu_env *env,
7475                                          struct lu_device *d)
7476 {
7477         struct mdt_device *m = mdt_dev(d);
7478         ENTRY;
7479
7480         mdt_fini(env, m);
7481         RETURN(NULL);
7482 }
7483
7484 static struct lu_device *mdt_device_free(const struct lu_env *env,
7485                                          struct lu_device *d)
7486 {
7487         struct mdt_device *m = mdt_dev(d);
7488         ENTRY;
7489
7490         lu_device_fini(&m->mdt_lu_dev);
7491         OBD_FREE_PTR(m);
7492
7493         RETURN(NULL);
7494 }
7495
7496 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
7497                                           struct lu_device_type *t,
7498                                           struct lustre_cfg *cfg)
7499 {
7500         struct lu_device  *l;
7501         struct mdt_device *m;
7502
7503         OBD_ALLOC_PTR(m);
7504         if (m != NULL) {
7505                 int rc;
7506
7507                 l = &m->mdt_lu_dev;
7508                 rc = mdt_init0(env, m, t, cfg);
7509                 if (rc != 0) {
7510                         mdt_device_free(env, l);
7511                         l = ERR_PTR(rc);
7512                         return l;
7513                 }
7514         } else
7515                 l = ERR_PTR(-ENOMEM);
7516         return l;
7517 }
7518
7519 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
7520 LU_KEY_INIT(mdt, struct mdt_thread_info);
7521
7522 static void mdt_key_fini(const struct lu_context *ctx,
7523                          struct lu_context_key *key, void* data)
7524 {
7525         struct mdt_thread_info *info = data;
7526
7527         if (info->mti_big_lmm) {
7528                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
7529                 info->mti_big_lmm = NULL;
7530                 info->mti_big_lmmsize = 0;
7531         }
7532
7533         if (info->mti_big_acl) {
7534                 OBD_FREE_LARGE(info->mti_big_acl, info->mti_big_aclsize);
7535                 info->mti_big_acl = NULL;
7536                 info->mti_big_aclsize = 0;
7537         }
7538
7539         OBD_FREE_PTR(info);
7540 }
7541
7542 /* context key: mdt_thread_key */
7543 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
7544
7545 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
7546 {
7547         return lu_ucred(info->mti_env);
7548 }
7549
7550 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
7551 {
7552         return lu_ucred_check(info->mti_env);
7553 }
7554
7555 /**
7556  * Enable/disable COS (Commit On Sharing).
7557  *
7558  * Set/Clear the COS flag in mdt options.
7559  *
7560  * \param mdt mdt device
7561  * \param val 0 disables COS, other values enable COS
7562  */
7563 void mdt_enable_cos(struct mdt_device *mdt, bool val)
7564 {
7565         struct lu_env env;
7566         int rc;
7567
7568         mdt->mdt_opts.mo_cos = val;
7569         rc = lu_env_init(&env, LCT_LOCAL);
7570         if (unlikely(rc != 0)) {
7571                 CWARN("%s: lu_env initialization failed, cannot "
7572                       "sync: rc = %d\n", mdt_obd_name(mdt), rc);
7573                 return;
7574         }
7575         mdt_device_sync(&env, mdt);
7576         lu_env_fini(&env);
7577 }
7578
7579 /**
7580  * Check COS (Commit On Sharing) status.
7581  *
7582  * Return COS flag status.
7583  *
7584  * \param mdt mdt device
7585  */
7586 int mdt_cos_is_enabled(struct mdt_device *mdt)
7587 {
7588         return mdt->mdt_opts.mo_cos != 0;
7589 }
7590
7591 static const struct lu_device_type_operations mdt_device_type_ops = {
7592         .ldto_device_alloc = mdt_device_alloc,
7593         .ldto_device_free  = mdt_device_free,
7594         .ldto_device_fini  = mdt_device_fini
7595 };
7596
7597 static struct lu_device_type mdt_device_type = {
7598         .ldt_tags     = LU_DEVICE_MD,
7599         .ldt_name     = LUSTRE_MDT_NAME,
7600         .ldt_ops      = &mdt_device_type_ops,
7601         .ldt_ctx_tags = LCT_MD_THREAD
7602 };
7603
7604 static int __init mdt_init(void)
7605 {
7606         int rc;
7607
7608         BUILD_BUG_ON(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") !=
7609                      FID_NOBRACE_LEN + 1);
7610         BUILD_BUG_ON(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") !=
7611                      FID_LEN + 1);
7612         rc = lu_kmem_init(mdt_caches);
7613         if (rc)
7614                 return rc;
7615
7616         rc = mds_mod_init();
7617         if (rc)
7618                 GOTO(lu_fini, rc);
7619
7620         rc = class_register_type(&mdt_obd_device_ops, NULL, true,
7621                                  LUSTRE_MDT_NAME, &mdt_device_type);
7622         if (rc)
7623                 GOTO(mds_fini, rc);
7624 lu_fini:
7625         if (rc)
7626                 lu_kmem_fini(mdt_caches);
7627 mds_fini:
7628         if (rc)
7629                 mds_mod_exit();
7630         return rc;
7631 }
7632
7633 static void __exit mdt_exit(void)
7634 {
7635         class_unregister_type(LUSTRE_MDT_NAME);
7636         mds_mod_exit();
7637         lu_kmem_fini(mdt_caches);
7638 }
7639
7640 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
7641 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
7642 MODULE_VERSION(LUSTRE_VERSION_STRING);
7643 MODULE_LICENSE("GPL");
7644
7645 module_init(mdt_init);
7646 module_exit(mdt_exit);