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 mdt_device *mdt = info->mti_mdt;
 805         struct obd_export *exp = info->mti_exp;
 806         struct md_attr *ma = &info->mti_attr;
 807         struct lu_nodemap *nodemap = NULL;
 808
 809         LASSERT(ma->ma_valid & MA_INODE);
 810
 811         if (attr->la_valid & LA_ATIME) {
 812                 b->mbo_atime = attr->la_atime;
 813                 b->mbo_valid |= OBD_MD_FLATIME;
 814         }
 815         if (attr->la_valid & LA_MTIME) {
 816                 b->mbo_mtime = attr->la_mtime;
 817                 b->mbo_valid |= OBD_MD_FLMTIME;
 818         }
 819         if (attr->la_valid & LA_CTIME) {
 820                 b->mbo_ctime = attr->la_ctime;
 821                 b->mbo_valid |= OBD_MD_FLCTIME;
 822         }
 823         if (attr->la_valid & LA_BTIME) {
 824                 b->mbo_btime = attr->la_btime;
 825                 b->mbo_valid |= OBD_MD_FLBTIME;
 826         }
 827         if (attr->la_valid & LA_FLAGS) {
 828                 b->mbo_flags = attr->la_flags;
 829                 b->mbo_valid |= OBD_MD_FLFLAGS;
 830         }
 831         if (attr->la_valid & LA_NLINK) {
 832                 b->mbo_nlink = attr->la_nlink;
 833                 b->mbo_valid |= OBD_MD_FLNLINK;
 834         }
 835         if (attr->la_valid & (LA_UID|LA_GID|LA_PROJID)) {
 836                 nodemap = nodemap_get_from_exp(exp);
 837                 if (IS_ERR(nodemap))
 838                         goto out;
 839         }
 840         if (attr->la_valid & LA_UID) {
 841                 b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 842                                             NODEMAP_FS_TO_CLIENT,
 843                                             attr->la_uid);
 844                 b->mbo_valid |= OBD_MD_FLUID;
 845         }
 846         if (attr->la_valid & LA_GID) {
 847                 b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 848                                             NODEMAP_FS_TO_CLIENT,
 849                                             attr->la_gid);
 850                 b->mbo_valid |= OBD_MD_FLGID;
 851         }
 852
 853         if (attr->la_valid & LA_PROJID) {
 854                 b->mbo_projid = nodemap_map_id(nodemap, NODEMAP_PROJID,
 855                                                NODEMAP_FS_TO_CLIENT,
 856                                                attr->la_projid);
 857                 b->mbo_valid |= OBD_MD_FLPROJID;
 858         }
 859
 860         b->mbo_mode = attr->la_mode;
 861         if (attr->la_valid & LA_MODE)
 862                 b->mbo_valid |= OBD_MD_FLMODE;
 863         if (attr->la_valid & LA_TYPE)
 864                 b->mbo_valid |= OBD_MD_FLTYPE;
 865
 866         if (fid != NULL) {
 867                 b->mbo_fid1 = *fid;
 868                 b->mbo_valid |= OBD_MD_FLID;
 869                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid=%#llx\n",
 870                        PFID(fid), b->mbo_nlink, b->mbo_mode, b->mbo_valid);
 871         }
 872
 873         if (!(attr->la_valid & LA_TYPE))
 874                 return;
 875
 876         b->mbo_rdev   = attr->la_rdev;
 877         b->mbo_size   = attr->la_size;
 878         b->mbo_blocks = attr->la_blocks;
 879
 880         if (!S_ISREG(attr->la_mode)) {
 881                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 882         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
 883                 /* means no objects are allocated on osts. */
 884                 LASSERT(!(ma->ma_valid & MA_LOV));
 885                 /* just ignore blocks occupied by extend attributes on MDS */
 886                 b->mbo_blocks = 0;
 887                 /* if no object is allocated on osts, the size on mds is valid.
 888                  * b=22272 */
 889                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 890         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL) {
 891                 if (mdt_hsm_is_released(ma->ma_lmm)) {
 892                         /* A released file stores its size on MDS. */
 893                         /* But return 1 block for released file, unless tools
 894                          * like tar will consider it fully sparse. (LU-3864)
 895                          */
 896                         if (unlikely(b->mbo_size == 0))
 897                                 b->mbo_blocks = 0;
 898                         else
 899                                 b->mbo_blocks = 1;
 900                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 901                 } else if (info->mti_som_strict &&
 902                            mdt->mdt_opts.mo_enable_strict_som) {
 903                         /* use SOM for size*/
 904                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 905                 } else if (ma->ma_valid & MA_SOM) { /* lsom is valid */
 906                         b->mbo_valid |= OBD_MD_FLLAZYSIZE | OBD_MD_FLLAZYBLOCKS;
 907                         b->mbo_size = ma->ma_som.ms_size;
 908                         b->mbo_blocks = ma->ma_som.ms_blocks;
 909                 }
 910         }
 911
 912         if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE ||
 913                             b->mbo_valid & OBD_MD_FLLAZYSIZE))
 914                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 915                        PFID(fid), (unsigned long long)b->mbo_size);
 916
 917 out:
 918         if (!IS_ERR_OR_NULL(nodemap))
 919                 nodemap_putref(nodemap);
 920 }
 921
 922 static inline int mdt_body_has_lov(const struct lu_attr *la,
 923                                    const struct mdt_body *body)
 924 {
 925         return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
 926                (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
 927 }
 928
 929 void mdt_client_compatibility(struct mdt_thread_info *info)
 930 {
 931         struct mdt_body       *body;
 932         struct ptlrpc_request *req = mdt_info_req(info);
 933         struct obd_export     *exp = req->rq_export;
 934         struct md_attr        *ma = &info->mti_attr;
 935         struct lu_attr        *la = &ma->ma_attr;
 936         ENTRY;
 937
 938         if (exp_connect_layout(exp))
 939                 /* the client can deal with 16-bit lmm_stripe_count */
 940                 RETURN_EXIT;
 941
 942         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 943
 944         if (!mdt_body_has_lov(la, body))
 945                 RETURN_EXIT;
 946
 947         /* now we have a reply with a lov for a client not compatible with the
 948          * layout lock so we have to clean the layout generation number */
 949         if (S_ISREG(la->la_mode))
 950                 ma->ma_lmm->lmm_layout_gen = 0;
 951         EXIT;
 952 }
 953
 954 static int mdt_attr_get_eabuf_size(struct mdt_thread_info *info,
 955                                    struct mdt_object *o)
 956 {
 957         const struct lu_env *env = info->mti_env;
 958         int rc, rc2;
 959
 960         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 961                           XATTR_NAME_LOV);
 962
 963         if (rc == -ENODATA)
 964                 rc = 0;
 965
 966         if (rc < 0)
 967                 goto out;
 968
 969         /* Is it a directory? Let's check for the LMV as well */
 970         if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
 971                 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 972                                    XATTR_NAME_LMV);
 973
 974                 if (rc2 == -ENODATA)
 975                         rc2 = mo_xattr_get(env, mdt_object_child(o),
 976                                            &LU_BUF_NULL,
 977                                            XATTR_NAME_DEFAULT_LMV);
 978
 979                 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
 980                         rc = rc2;
 981         }
 982
 983 out:
 984         return rc;
 985 }
 986
 987 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
 988                       const char *name)
 989 {
 990         const struct lu_env *env = info->mti_env;
 991         int rc;
 992         ENTRY;
 993
 994         LASSERT(info->mti_big_lmm_used == 0);
 995         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
 996         if (rc < 0)
 997                 RETURN(rc);
 998
 999         /* big_lmm may need to be grown */
1000         if (info->mti_big_lmmsize < rc) {
1001                 int size = size_roundup_power2(rc);
1002
1003                 if (info->mti_big_lmmsize > 0) {
1004                         /* free old buffer */
1005                         LASSERT(info->mti_big_lmm);
1006                         OBD_FREE_LARGE(info->mti_big_lmm,
1007                                        info->mti_big_lmmsize);
1008                         info->mti_big_lmm = NULL;
1009                         info->mti_big_lmmsize = 0;
1010                 }
1011
1012                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
1013                 if (info->mti_big_lmm == NULL)
1014                         RETURN(-ENOMEM);
1015                 info->mti_big_lmmsize = size;
1016         }
1017         LASSERT(info->mti_big_lmmsize >= rc);
1018
1019         info->mti_buf.lb_buf = info->mti_big_lmm;
1020         info->mti_buf.lb_len = info->mti_big_lmmsize;
1021         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
1022
1023         RETURN(rc);
1024 }
1025
1026 int __mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
1027                      struct md_attr *ma, const char *name)
1028 {
1029         struct md_object *next = mdt_object_child(o);
1030         struct lu_buf    *buf = &info->mti_buf;
1031         int rc;
1032
1033         if (strcmp(name, XATTR_NAME_LOV) == 0) {
1034                 buf->lb_buf = ma->ma_lmm;
1035                 buf->lb_len = ma->ma_lmm_size;
1036                 LASSERT(!(ma->ma_valid & MA_LOV));
1037         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
1038                 buf->lb_buf = ma->ma_lmv;
1039                 buf->lb_len = ma->ma_lmv_size;
1040                 LASSERT(!(ma->ma_valid & MA_LMV));
1041         } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
1042                 buf->lb_buf = ma->ma_default_lmv;
1043                 buf->lb_len = ma->ma_default_lmv_size;
1044                 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
1045         } else {
1046                 return -EINVAL;
1047         }
1048
1049         LASSERT(buf->lb_buf);
1050
1051         rc = mo_xattr_get(info->mti_env, next, buf, name);
1052         if (rc > 0) {
1053
1054 got:
1055                 if (strcmp(name, XATTR_NAME_LOV) == 0) {
1056                         if (info->mti_big_lmm_used)
1057                                 ma->ma_lmm = info->mti_big_lmm;
1058
1059                         /* NOT return LOV EA with hole to old client. */
1060                         if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
1061                                      LOV_PATTERN_F_HOLE) &&
1062                             !(exp_connect_flags(info->mti_exp) &
1063                               OBD_CONNECT_LFSCK)) {
1064                                 return -EIO;
1065                         } else {
1066                                 ma->ma_lmm_size = rc;
1067                                 ma->ma_valid |= MA_LOV;
1068                         }
1069                 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
1070                         if (info->mti_big_lmm_used)
1071                                 ma->ma_lmv = info->mti_big_lmm;
1072
1073                         ma->ma_lmv_size = rc;
1074                         ma->ma_valid |= MA_LMV;
1075                 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
1076                         ma->ma_default_lmv_size = rc;
1077                         ma->ma_valid |= MA_LMV_DEF;
1078                 }
1079
1080                 /* Update mdt_max_mdsize so all clients will be aware that */
1081                 if (info->mti_mdt->mdt_max_mdsize < rc)
1082                         info->mti_mdt->mdt_max_mdsize = rc;
1083
1084                 rc = 0;
1085         } else if (rc == -ENODATA) {
1086                 /* no LOV EA */
1087                 rc = 0;
1088         } else if (rc == -ERANGE) {
1089                 /* Default LMV has fixed size, so it must be able to fit
1090                  * in the original buffer */
1091                 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
1092                         return rc;
1093                 rc = mdt_big_xattr_get(info, o, name);
1094                 if (rc > 0) {
1095                         info->mti_big_lmm_used = 1;
1096                         goto got;
1097                 }
1098         }
1099
1100         return rc;
1101 }
1102
1103 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
1104                    struct md_attr *ma, const char *name)
1105 {
1106         int rc;
1107
1108         if (!info->mti_big_lmm) {
1109                 OBD_ALLOC(info->mti_big_lmm, PAGE_SIZE);
1110                 if (!info->mti_big_lmm)
1111                         return -ENOMEM;
1112                 info->mti_big_lmmsize = PAGE_SIZE;
1113         }
1114
1115         if (strcmp(name, XATTR_NAME_LOV) == 0) {
1116                 ma->ma_lmm = info->mti_big_lmm;
1117                 ma->ma_lmm_size = info->mti_big_lmmsize;
1118                 ma->ma_valid &= ~MA_LOV;
1119         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
1120                 ma->ma_lmv = info->mti_big_lmm;
1121                 ma->ma_lmv_size = info->mti_big_lmmsize;
1122                 ma->ma_valid &= ~MA_LMV;
1123         } else {
1124                 LBUG();
1125         }
1126
1127         LASSERT(!info->mti_big_lmm_used);
1128         rc = __mdt_stripe_get(info, o, ma, name);
1129         /* since big_lmm is always used here, clear 'used' flag to avoid
1130          * assertion in mdt_big_xattr_get().
1131          */
1132         info->mti_big_lmm_used = 0;
1133
1134         return rc;
1135 }
1136
1137 int mdt_attr_get_pfid(struct mdt_thread_info *info, struct mdt_object *o,
1138                       struct lu_fid *pfid)
1139 {
1140         struct lu_buf           *buf = &info->mti_buf;
1141         struct link_ea_header   *leh;
1142         struct link_ea_entry    *lee;
1143         int                      rc;
1144         ENTRY;
1145
1146         buf->lb_buf = info->mti_big_lmm;
1147         buf->lb_len = info->mti_big_lmmsize;
1148         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
1149                           buf, XATTR_NAME_LINK);
1150         /* ignore errors, MA_PFID won't be set and it is
1151          * up to the caller to treat this as an error */
1152         if (rc == -ERANGE || buf->lb_len == 0) {
1153                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
1154                 buf->lb_buf = info->mti_big_lmm;
1155                 buf->lb_len = info->mti_big_lmmsize;
1156         }
1157
1158         if (rc < 0)
1159                 RETURN(rc);
1160         if (rc < sizeof(*leh)) {
1161                 CERROR("short LinkEA on "DFID": rc = %d\n",
1162                        PFID(mdt_object_fid(o)), rc);
1163                 RETURN(-ENODATA);
1164         }
1165
1166         leh = (struct link_ea_header *) buf->lb_buf;
1167         lee = (struct link_ea_entry *)(leh + 1);
1168         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
1169                 leh->leh_magic = LINK_EA_MAGIC;
1170                 leh->leh_reccount = __swab32(leh->leh_reccount);
1171                 leh->leh_len = __swab64(leh->leh_len);
1172         }
1173         if (leh->leh_magic != LINK_EA_MAGIC)
1174                 RETURN(-EINVAL);
1175         if (leh->leh_reccount == 0)
1176                 RETURN(-ENODATA);
1177
1178         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
1179         fid_be_to_cpu(pfid, pfid);
1180
1181         RETURN(0);
1182 }
1183
1184 int mdt_attr_get_pfid_name(struct mdt_thread_info *info, struct mdt_object *o,
1185                            struct lu_fid *pfid, struct lu_name *lname)
1186 {
1187         struct lu_buf *buf = &info->mti_buf;
1188         struct link_ea_header *leh;
1189         struct link_ea_entry *lee;
1190         int reclen;
1191         int rc;
1192
1193         buf->lb_buf = info->mti_xattr_buf;
1194         buf->lb_len = sizeof(info->mti_xattr_buf);
1195         rc = mo_xattr_get(info->mti_env, mdt_object_child(o), buf,
1196                           XATTR_NAME_LINK);
1197         if (rc == -ERANGE) {
1198                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
1199                 buf->lb_buf = info->mti_big_lmm;
1200                 buf->lb_len = info->mti_big_lmmsize;
1201         }
1202         if (rc < 0)
1203                 return rc;
1204
1205         if (rc < sizeof(*leh)) {
1206                 CERROR("short LinkEA on "DFID": rc = %d\n",
1207                        PFID(mdt_object_fid(o)), rc);
1208                 return -ENODATA;
1209         }
1210
1211         leh = (struct link_ea_header *)buf->lb_buf;
1212         lee = (struct link_ea_entry *)(leh + 1);
1213         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
1214                 leh->leh_magic = LINK_EA_MAGIC;
1215                 leh->leh_reccount = __swab32(leh->leh_reccount);
1216                 leh->leh_len = __swab64(leh->leh_len);
1217         }
1218         if (leh->leh_magic != LINK_EA_MAGIC)
1219                 return -EINVAL;
1220
1221         if (leh->leh_reccount == 0)
1222                 return -ENODATA;
1223
1224         linkea_entry_unpack(lee, &reclen, lname, pfid);
1225
1226         return 0;
1227 }
1228
1229 int mdt_attr_get_complex(struct mdt_thread_info *info,
1230                          struct mdt_object *o, struct md_attr *ma)
1231 {
1232         const struct lu_env *env = info->mti_env;
1233         struct md_object    *next = mdt_object_child(o);
1234         struct lu_buf       *buf = &info->mti_buf;
1235         int                  need = ma->ma_need;
1236         int                  rc = 0, rc2;
1237         u32                  mode;
1238         ENTRY;
1239
1240         ma->ma_valid = 0;
1241
1242         if (mdt_object_exists(o) == 0)
1243                 GOTO(out, rc = -ENOENT);
1244         mode = lu_object_attr(&next->mo_lu);
1245
1246         if (need & MA_INODE) {
1247                 ma->ma_need = MA_INODE;
1248                 rc = mo_attr_get(env, next, ma);
1249                 if (rc)
1250                         GOTO(out, rc);
1251
1252                 if (S_ISREG(mode))
1253                         (void) mdt_get_som(info, o, ma);
1254                 ma->ma_valid |= MA_INODE;
1255         }
1256
1257         if (need & MA_PFID) {
1258                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
1259                 if (rc == 0)
1260                         ma->ma_valid |= MA_PFID;
1261                 /* ignore this error, parent fid is not mandatory */
1262                 rc = 0;
1263         }
1264
1265         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
1266                 rc = __mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
1267                 if (rc)
1268                         GOTO(out, rc);
1269         }
1270
1271         if (need & MA_LMV && S_ISDIR(mode)) {
1272                 rc = __mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
1273                 if (rc != 0)
1274                         GOTO(out, rc);
1275         }
1276
1277         if (need & MA_LMV_DEF && S_ISDIR(mode)) {
1278                 rc = __mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
1279                 if (rc != 0)
1280                         GOTO(out, rc);
1281         }
1282
1283         /*
1284          * In the handle of MA_INODE, we may already get the SOM attr.
1285          */
1286         if (need & MA_SOM && S_ISREG(mode) && !(ma->ma_valid & MA_SOM)) {
1287                 rc = mdt_get_som(info, o, ma);
1288                 if (rc != 0)
1289                         GOTO(out, rc);
1290         }
1291
1292         if (need & MA_HSM && S_ISREG(mode)) {
1293                 buf->lb_buf = info->mti_xattr_buf;
1294                 buf->lb_len = sizeof(info->mti_xattr_buf);
1295                 BUILD_BUG_ON(sizeof(struct hsm_attrs) >
1296                              sizeof(info->mti_xattr_buf));
1297                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
1298                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
1299                 if (rc2 == 0)
1300                         ma->ma_valid |= MA_HSM;
1301                 else if (rc2 < 0 && rc2 != -ENODATA)
1302                         GOTO(out, rc = rc2);
1303         }
1304
1305 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1306         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
1307                 buf->lb_buf = ma->ma_acl;
1308                 buf->lb_len = ma->ma_acl_size;
1309                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
1310                 if (rc2 > 0) {
1311                         ma->ma_acl_size = rc2;
1312                         ma->ma_valid |= MA_ACL_DEF;
1313                 } else if (rc2 == -ENODATA) {
1314                         /* no ACLs */
1315                         ma->ma_acl_size = 0;
1316                 } else
1317                         GOTO(out, rc = rc2);
1318         }
1319 #endif
1320 out:
1321         ma->ma_need = need;
1322         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = %#llx ma_lmm=%p\n",
1323                rc, ma->ma_valid, ma->ma_lmm);
1324         RETURN(rc);
1325 }
1326
1327 static void mdt_preset_encctx_size(struct mdt_thread_info *info)
1328 {
1329         struct req_capsule *pill = info->mti_pill;
1330
1331         ENTRY;
1332         if (req_capsule_has_field(pill, &RMF_FILE_ENCCTX,
1333                                   RCL_SERVER))
1334                 /* pre-set size in server part with max size */
1335                 req_capsule_set_size(pill, &RMF_FILE_ENCCTX,
1336                                      RCL_SERVER,
1337                                      info->mti_mdt->mdt_max_mdsize);
1338         EXIT;
1339 }
1340
1341 static int mdt_getattr_internal(struct mdt_thread_info *info,
1342                                 struct mdt_object *o, int ma_need)
1343 {
1344         struct mdt_device *mdt = info->mti_mdt;
1345         struct md_object *next = mdt_object_child(o);
1346         const struct mdt_body *reqbody = info->mti_body;
1347         struct ptlrpc_request *req = mdt_info_req(info);
1348         struct md_attr *ma = &info->mti_attr;
1349         struct lu_attr *la = &ma->ma_attr;
1350         struct req_capsule *pill = info->mti_pill;
1351         const struct lu_env *env = info->mti_env;
1352         struct mdt_body *repbody;
1353         struct lu_buf *buffer = &info->mti_buf;
1354         struct obd_export *exp = info->mti_exp;
1355         ktime_t kstart = ktime_get();
1356         int rc;
1357
1358         ENTRY;
1359
1360         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
1361                 RETURN(err_serious(-ENOMEM));
1362
1363         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1364
1365         ma->ma_valid = 0;
1366
1367         if (mdt_object_remote(o)) {
1368                 /* This object is located on remote node.*/
1369                 /* Return -ENOTSUPP for old client */
1370                 if (!mdt_is_dne_client(req->rq_export))
1371                         GOTO(out, rc = -ENOTSUPP);
1372
1373                 repbody->mbo_fid1 = *mdt_object_fid(o);
1374                 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
1375                 GOTO(out, rc = 0);
1376         }
1377
1378         if (reqbody->mbo_eadatasize > 0) {
1379                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
1380                 if (buffer->lb_buf == NULL)
1381                         GOTO(out, rc = -EPROTO);
1382                 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
1383                                                       RCL_SERVER);
1384         } else {
1385                 buffer->lb_buf = NULL;
1386                 buffer->lb_len = 0;
1387                 ma_need &= ~(MA_LOV | MA_LMV);
1388                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
1389                        mdt_obd_name(info->mti_mdt),
1390                        req->rq_export->exp_client_uuid.uuid);
1391         }
1392
1393         /* from 2.12.58 intent_getattr pack default LMV in reply */
1394         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1395             ((reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA)) ==
1396                     (OBD_MD_MEA | OBD_MD_DEFAULT_MEA)) &&
1397             req_capsule_has_field(&req->rq_pill, &RMF_DEFAULT_MDT_MD,
1398                                   RCL_SERVER)) {
1399                 ma->ma_lmv = buffer->lb_buf;
1400                 ma->ma_lmv_size = buffer->lb_len;
1401                 ma->ma_default_lmv = req_capsule_server_get(pill,
1402                                                 &RMF_DEFAULT_MDT_MD);
1403                 ma->ma_default_lmv_size = req_capsule_get_size(pill,
1404                                                 &RMF_DEFAULT_MDT_MD,
1405                                                 RCL_SERVER);
1406                 ma->ma_need = MA_INODE;
1407                 if (ma->ma_lmv_size > 0)
1408                         ma->ma_need |= MA_LMV;
1409                 if (ma->ma_default_lmv_size > 0)
1410                         ma->ma_need |= MA_LMV_DEF;
1411         } else if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1412                    (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
1413                 /* If it is dir and client require MEA, then we got MEA */
1414                 /* Assumption: MDT_MD size is enough for lmv size. */
1415                 ma->ma_lmv = buffer->lb_buf;
1416                 ma->ma_lmv_size = buffer->lb_len;
1417                 ma->ma_need = MA_INODE;
1418                 if (ma->ma_lmv_size > 0) {
1419                         if (reqbody->mbo_valid & OBD_MD_MEA) {
1420                                 ma->ma_need |= MA_LMV;
1421                         } else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA) {
1422                                 ma->ma_need |= MA_LMV_DEF;
1423                                 ma->ma_default_lmv = buffer->lb_buf;
1424                                 ma->ma_lmv = NULL;
1425                                 ma->ma_default_lmv_size = buffer->lb_len;
1426                                 ma->ma_lmv_size = 0;
1427                         }
1428                 }
1429         } else {
1430                 ma->ma_lmm = buffer->lb_buf;
1431                 ma->ma_lmm_size = buffer->lb_len;
1432                 ma->ma_need = MA_INODE | MA_HSM;
1433                 if (ma->ma_lmm_size > 0) {
1434                         ma->ma_need |= MA_LOV;
1435                         /* Older clients may crash if they getattr overstriped
1436                          * files
1437                          */
1438                         if (!exp_connect_overstriping(exp) &&
1439                             mdt_lmm_is_overstriping(ma->ma_lmm))
1440                                 RETURN(-EOPNOTSUPP);
1441                 }
1442         }
1443
1444         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1445             reqbody->mbo_valid & OBD_MD_FLDIREA  &&
1446             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
1447                 /* get default stripe info for this dir. */
1448                 ma->ma_need |= MA_LOV_DEF;
1449         }
1450         ma->ma_need |= ma_need;
1451
1452         rc = mdt_attr_get_complex(info, o, ma);
1453         if (unlikely(rc)) {
1454                 CDEBUG_LIMIT(rc == -ENOENT ? D_OTHER : D_ERROR,
1455                              "%s: getattr error for "DFID": rc = %d\n",
1456                              mdt_obd_name(info->mti_mdt),
1457                              PFID(mdt_object_fid(o)), rc);
1458                 RETURN(rc);
1459         }
1460
1461         /* if file is released, check if a restore is running */
1462         if (ma->ma_valid & MA_HSM) {
1463                 repbody->mbo_valid |= OBD_MD_TSTATE;
1464                 if ((ma->ma_hsm.mh_flags & HS_RELEASED) &&
1465                     mdt_hsm_restore_is_running(info, mdt_object_fid(o)))
1466                         repbody->mbo_t_state = MS_RESTORE;
1467         }
1468
1469         if (unlikely(!(ma->ma_valid & MA_INODE)))
1470                 RETURN(-EFAULT);
1471
1472         mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
1473
1474         if (mdt_body_has_lov(la, reqbody)) {
1475                 u32 stripe_count = 1;
1476                 bool fixed_layout = false;
1477
1478                 if (ma->ma_valid & MA_LOV) {
1479                         LASSERT(ma->ma_lmm_size);
1480                         repbody->mbo_eadatasize = ma->ma_lmm_size;
1481                         if (S_ISDIR(la->la_mode))
1482                                 repbody->mbo_valid |= OBD_MD_FLDIREA;
1483                         else
1484                                 repbody->mbo_valid |= OBD_MD_FLEASIZE;
1485                         mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
1486                 }
1487                 if (ma->ma_valid & MA_LMV) {
1488                         struct lmv_mds_md_v1 *lmv = &ma->ma_lmv->lmv_md_v1;
1489                         u32 magic = le32_to_cpu(lmv->lmv_magic);
1490
1491                         /* Return -ENOTSUPP for old client */
1492                         if (!mdt_is_striped_client(req->rq_export))
1493                                 RETURN(-ENOTSUPP);
1494
1495                         LASSERT(S_ISDIR(la->la_mode));
1496                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1497                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1498                         repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
1499
1500                         stripe_count = le32_to_cpu(lmv->lmv_stripe_count);
1501                         fixed_layout = lmv_is_fixed(lmv);
1502                         if (magic == LMV_MAGIC_STRIPE && lmv_is_restriping(lmv))
1503                                 mdt_restripe_migrate_add(info, o);
1504                         else if (magic == LMV_MAGIC_V1 &&
1505                                  lmv_is_restriping(lmv))
1506                                 mdt_restripe_update_add(info, o);
1507                 }
1508                 if (ma->ma_valid & MA_LMV_DEF) {
1509                         /* Return -ENOTSUPP for old client */
1510                         if (!mdt_is_striped_client(req->rq_export))
1511                                 RETURN(-ENOTSUPP);
1512                         LASSERT(S_ISDIR(la->la_mode));
1513                         /*
1514                          * when ll_dir_getstripe() gets default LMV, it
1515                          * checks mbo_eadatasize.
1516                          */
1517                         if (!(ma->ma_valid & MA_LMV))
1518                                 repbody->mbo_eadatasize =
1519                                         ma->ma_default_lmv_size;
1520                         repbody->mbo_valid |= (OBD_MD_FLDIREA |
1521                                                OBD_MD_DEFAULT_MEA);
1522                 }
1523                 CDEBUG(D_VFSTRACE,
1524                        "dirent count %llu stripe count %u MDT count %d\n",
1525                        ma->ma_attr.la_dirent_count, stripe_count,
1526                        atomic_read(&mdt->mdt_mds_mds_conns) + 1);
1527                 if (ma->ma_attr.la_dirent_count != LU_DIRENT_COUNT_UNSET &&
1528                     ma->ma_attr.la_dirent_count >
1529                         mdt->mdt_restriper.mdr_dir_split_count &&
1530                     !fid_is_root(mdt_object_fid(o)) &&
1531                     mdt->mdt_enable_dir_auto_split &&
1532                     !o->mot_restriping &&
1533                     stripe_count < atomic_read(&mdt->mdt_mds_mds_conns) + 1 &&
1534                     !fixed_layout)
1535                         mdt_auto_split_add(info, o);
1536         } else if (S_ISLNK(la->la_mode) &&
1537                    reqbody->mbo_valid & OBD_MD_LINKNAME) {
1538                 buffer->lb_buf = ma->ma_lmm;
1539                 /* eadatasize from client includes NULL-terminator, so
1540                  * there is no need to read it */
1541                 buffer->lb_len = reqbody->mbo_eadatasize - 1;
1542                 rc = mo_readlink(env, next, buffer);
1543                 if (unlikely(rc <= 0)) {
1544                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
1545                                mdt_obd_name(info->mti_mdt),
1546                                PFID(mdt_object_fid(o)), rc);
1547                         rc = -EFAULT;
1548                 } else {
1549                         int print_limit = min_t(int, PAGE_SIZE - 128, rc);
1550
1551                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
1552                                 rc -= 2;
1553                         repbody->mbo_valid |= OBD_MD_LINKNAME;
1554                         /* we need to report back size with NULL-terminator
1555                          * because client expects that */
1556                         repbody->mbo_eadatasize = rc + 1;
1557                         if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
1558                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
1559                                        "on "DFID ", expected %d\n",
1560                                        mdt_obd_name(info->mti_mdt),
1561                                        rc, PFID(mdt_object_fid(o)),
1562                                        reqbody->mbo_eadatasize - 1);
1563                         /* NULL terminate */
1564                         ((char *)ma->ma_lmm)[rc] = 0;
1565
1566                         /* If the total CDEBUG() size is larger than a page, it
1567                          * will print a warning to the console, avoid this by
1568                          * printing just the last part of the symlink. */
1569                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
1570                                print_limit < rc ? "..." : "", print_limit,
1571                                (char *)ma->ma_lmm + rc - print_limit, rc);
1572                         rc = 0;
1573                 }
1574         }
1575
1576         if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
1577                 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1578                 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1579                 CDEBUG(D_INODE, "changing the max MD size to %u\n",
1580                        repbody->mbo_max_mdsize);
1581         }
1582
1583 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1584         if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1585                  (reqbody->mbo_valid & OBD_MD_FLACL)) {
1586                 struct lu_nodemap *nodemap = nodemap_get_from_exp(exp);
1587                 if (IS_ERR(nodemap))
1588                         RETURN(PTR_ERR(nodemap));
1589
1590                 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
1591                 nodemap_putref(nodemap);
1592         }
1593 #endif
1594
1595 out:
1596         if (rc == 0)
1597                 mdt_counter_incr(req, LPROC_MDT_GETATTR,
1598                                  ktime_us_delta(ktime_get(), kstart));
1599
1600         RETURN(rc);
1601 }
1602
1603 static int mdt_getattr(struct tgt_session_info *tsi)
1604 {
1605         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1606         struct mdt_object       *obj = info->mti_object;
1607         struct req_capsule      *pill = info->mti_pill;
1608         struct mdt_body         *reqbody;
1609         struct mdt_body         *repbody;
1610         int rc, rc2;
1611         ENTRY;
1612
1613         if (unlikely(info->mti_object == NULL))
1614                 RETURN(-EPROTO);
1615
1616         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1617         LASSERT(reqbody);
1618         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1619
1620         /* Special case for Data-on-MDT files to get data version */
1621         if (unlikely(reqbody->mbo_valid & OBD_MD_FLDATAVERSION)) {
1622                 rc = mdt_data_version_get(tsi);
1623                 GOTO(out, rc);
1624         }
1625
1626         /* Unlike intent case where we need to pre-fill out buffers early on
1627          * in intent policy for ldlm reasons, here we can have a much better
1628          * guess at EA size by just reading it from disk.
1629          * Exceptions are readdir and (missing) directory striping */
1630         /* Readlink */
1631         if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1632                 /* No easy way to know how long is the symlink, but it cannot
1633                  * be more than PATH_MAX, so we allocate +1 */
1634                 rc = PATH_MAX + 1;
1635         /* A special case for fs ROOT: getattr there might fetch
1636          * default EA for entire fs, not just for this dir!
1637          */
1638         } else if (lu_fid_eq(mdt_object_fid(obj),
1639                              &info->mti_mdt->mdt_md_root_fid) &&
1640                    (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1641                    (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1642                                                                  MDS_GETATTR)) {
1643                 /* Should the default strping be bigger, mdt_fix_reply
1644                  * will reallocate */
1645                 rc = DEF_REP_MD_SIZE;
1646         } else {
1647                 /* Read the actual EA size from disk */
1648                 rc = mdt_attr_get_eabuf_size(info, obj);
1649         }
1650
1651         if (rc < 0)
1652                 GOTO(out, rc = err_serious(rc));
1653
1654         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1655
1656         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
1657          * by default. If the target object has more ACL entries, then
1658          * enlarge the buffer when necessary. */
1659         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
1660                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
1661         mdt_preset_encctx_size(info);
1662
1663         rc = req_capsule_server_pack(pill);
1664         if (unlikely(rc != 0))
1665                 GOTO(out, rc = err_serious(rc));
1666
1667         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1668         LASSERT(repbody != NULL);
1669         repbody->mbo_eadatasize = 0;
1670         repbody->mbo_aclsize = 0;
1671
1672         rc = mdt_check_ucred(info);
1673         if (unlikely(rc))
1674                 GOTO(out_shrink, rc);
1675
1676         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1677
1678         rc = mdt_getattr_internal(info, obj, 0);
1679         if (unlikely(rc))
1680                 GOTO(out_shrink, rc);
1681
1682         rc = mdt_pack_encctx_in_reply(info, obj);
1683         EXIT;
1684 out_shrink:
1685         mdt_client_compatibility(info);
1686         rc2 = mdt_fix_reply(info);
1687         if (rc == 0)
1688                 rc = rc2;
1689 out:
1690         mdt_thread_info_fini(info);
1691         return rc;
1692 }
1693
1694 /**
1695  * Handler of layout intent RPC requiring the layout modification
1696  *
1697  * \param[in]  info     thread environment
1698  * \param[in]  obj      object
1699  * \param[out] lhc      object ldlm lock handle
1700  * \param[in]  layout   layout change descriptor
1701  *
1702  * \retval 0    on success
1703  * \retval < 0  error code
1704  */
1705 int mdt_layout_change(struct mdt_thread_info *info, struct mdt_object *obj,
1706                       struct mdt_lock_handle *lhc,
1707                       struct md_layout_change *layout)
1708 {
1709         int rc;
1710
1711         ENTRY;
1712
1713         if (!mdt_object_exists(obj))
1714                 RETURN(-ENOENT);
1715
1716         if (!S_ISREG(lu_object_attr(&obj->mot_obj)))
1717                 RETURN(-EINVAL);
1718
1719         rc = mo_permission(info->mti_env, NULL, mdt_object_child(obj), NULL,
1720                            MAY_WRITE);
1721         if (rc)
1722                 RETURN(rc);
1723
1724         rc = mdt_check_resent_lock(info, obj, lhc);
1725         if (rc < 0)
1726                 RETURN(rc);
1727
1728         if (rc > 0) {
1729                 /* not resent */
1730                 __u64 lockpart = MDS_INODELOCK_LAYOUT;
1731
1732                 /* take layout lock to prepare layout change */
1733                 if (layout->mlc_opc == MD_LAYOUT_WRITE)
1734                         lockpart |= MDS_INODELOCK_UPDATE;
1735
1736                 mdt_lock_handle_init(lhc);
1737                 mdt_lock_reg_init(lhc, LCK_EX);
1738                 rc = mdt_reint_object_lock(info, obj, lhc, lockpart, false);
1739                 if (rc)
1740                         RETURN(rc);
1741         }
1742
1743         mutex_lock(&obj->mot_som_mutex);
1744         rc = mo_layout_change(info->mti_env, mdt_object_child(obj), layout);
1745         mutex_unlock(&obj->mot_som_mutex);
1746
1747         if (rc)
1748                 mdt_object_unlock(info, obj, lhc, 1);
1749
1750         RETURN(rc);
1751 }
1752
1753 /**
1754  * Exchange MOF_LOV_CREATED flags between two objects after a
1755  * layout swap. No assumption is made on whether o1 or o2 have
1756  * created objects or not.
1757  *
1758  * \param[in,out] o1    First swap layout object
1759  * \param[in,out] o2    Second swap layout object
1760  */
1761 static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2)
1762 {
1763         unsigned int o1_lov_created = o1->mot_lov_created;
1764
1765         mutex_lock(&o1->mot_lov_mutex);
1766         mutex_lock(&o2->mot_lov_mutex);
1767
1768         o1->mot_lov_created = o2->mot_lov_created;
1769         o2->mot_lov_created = o1_lov_created;
1770
1771         mutex_unlock(&o2->mot_lov_mutex);
1772         mutex_unlock(&o1->mot_lov_mutex);
1773 }
1774
1775 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1776 {
1777         struct mdt_thread_info  *info;
1778         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1779         struct obd_export       *exp = req->rq_export;
1780         struct mdt_object       *o1, *o2, *o;
1781         struct mdt_lock_handle  *lh1, *lh2;
1782         struct mdc_swap_layouts *msl;
1783         int                      rc;
1784         ENTRY;
1785
1786         /* client does not support layout lock, so layout swaping
1787          * is disabled.
1788          * FIXME: there is a problem for old clients which don't support
1789          * layout lock yet. If those clients have already opened the file
1790          * they won't be notified at all so that old layout may still be
1791          * used to do IO. This can be fixed after file release is landed by
1792          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1793         if (!exp_connect_layout(exp))
1794                 RETURN(-EOPNOTSUPP);
1795
1796         info = tsi2mdt_info(tsi);
1797         if (unlikely(info->mti_object == NULL))
1798                 RETURN(-EPROTO);
1799
1800         if (info->mti_dlm_req != NULL)
1801                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1802
1803         o1 = info->mti_object;
1804         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1805                                 &info->mti_body->mbo_fid2);
1806         if (IS_ERR(o))
1807                 GOTO(out, rc = PTR_ERR(o));
1808
1809         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1810                 GOTO(put, rc = -ENOENT);
1811
1812         rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1813         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1814                 GOTO(put, rc);
1815
1816         if (rc < 0)
1817                 swap(o1, o2);
1818
1819         /* permission check. Make sure the calling process having permission
1820          * to write both files. */
1821         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1822                            MAY_WRITE);
1823         if (rc < 0)
1824                 GOTO(put, rc);
1825
1826         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1827                            MAY_WRITE);
1828         if (rc < 0)
1829                 GOTO(put, rc);
1830
1831         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1832         if (msl == NULL)
1833                 GOTO(put, rc = -EPROTO);
1834
1835         lh1 = &info->mti_lh[MDT_LH_NEW];
1836         mdt_lock_reg_init(lh1, LCK_EX);
1837         lh2 = &info->mti_lh[MDT_LH_OLD];
1838         mdt_lock_reg_init(lh2, LCK_EX);
1839
1840         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1841                              MDS_INODELOCK_XATTR);
1842         if (rc < 0)
1843                 GOTO(put, rc);
1844
1845         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1846                              MDS_INODELOCK_XATTR);
1847         if (rc < 0)
1848                 GOTO(unlock1, rc);
1849
1850         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1851                              mdt_object_child(o2), msl->msl_flags);
1852         if (rc < 0)
1853                 GOTO(unlock2, rc);
1854
1855         mdt_swap_lov_flag(o1, o2);
1856
1857 unlock2:
1858         mdt_object_unlock(info, o2, lh2, rc);
1859 unlock1:
1860         mdt_object_unlock(info, o1, lh1, rc);
1861 put:
1862         mdt_object_put(info->mti_env, o);
1863 out:
1864         mdt_thread_info_fini(info);
1865         RETURN(rc);
1866 }
1867
1868 static int mdt_raw_lookup(struct mdt_thread_info *info,
1869                           struct mdt_object *parent,
1870                           const struct lu_name *lname)
1871 {
1872         struct lu_fid *fid = &info->mti_tmp_fid1;
1873         struct mdt_body *repbody;
1874         bool is_dotdot = false;
1875         bool is_old_parent_stripe = false;
1876         bool is_new_parent_checked = false;
1877         int rc;
1878
1879         ENTRY;
1880
1881         LASSERT(!info->mti_cross_ref);
1882         /* Always allow to lookup ".." */
1883         if (lname->ln_namelen == 2 &&
1884             lname->ln_name[0] == '.' && lname->ln_name[1] == '.') {
1885                 info->mti_spec.sp_permitted = 1;
1886                 is_dotdot = true;
1887                 if (mdt_is_dir_stripe(info, parent) == 1)
1888                         is_old_parent_stripe = true;
1889         }
1890
1891         mdt_object_get(info->mti_env, parent);
1892 lookup:
1893         /* Only got the fid of this obj by name */
1894         fid_zero(fid);
1895         rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname, fid,
1896                         &info->mti_spec);
1897         mdt_object_put(info->mti_env, parent);
1898         if (rc)
1899                 RETURN(rc);
1900
1901         /* getattr_name("..") should return master object FID for striped dir */
1902         if (is_dotdot && (is_old_parent_stripe || !is_new_parent_checked)) {
1903                 parent = mdt_object_find(info->mti_env, info->mti_mdt, fid);
1904                 if (IS_ERR(parent))
1905                         RETURN(PTR_ERR(parent));
1906
1907                 /* old client getattr_name("..") with stripe FID */
1908                 if (unlikely(is_old_parent_stripe)) {
1909                         is_old_parent_stripe = false;
1910                         goto lookup;
1911                 }
1912
1913                 /* ".." may be a stripe */
1914                 if (unlikely(mdt_is_dir_stripe(info, parent) == 1)) {
1915                         is_new_parent_checked = true;
1916                         goto lookup;
1917                 }
1918
1919                 mdt_object_put(info->mti_env, parent);
1920         }
1921
1922         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1923         repbody->mbo_fid1 = *fid;
1924         repbody->mbo_valid = OBD_MD_FLID;
1925
1926         RETURN(rc);
1927 }
1928
1929 /**
1930  * Find name matching hash
1931  *
1932  * We search \a child LinkEA for a name whose hash matches \a lname
1933  * (it contains an encoded hash).
1934  *
1935  * \param info mdt thread info
1936  * \param lname encoded hash to find
1937  * \param parent parent object
1938  * \param child object to search with LinkEA
1939  *
1940  * \retval 1 match found
1941  * \retval 0 no match found
1942  * \retval -ev negative errno upon error
1943  */
1944 int find_name_matching_hash(struct mdt_thread_info *info, struct lu_name *lname,
1945                             struct mdt_object *parent, struct mdt_object *child)
1946 {
1947         /* Here, lname is an encoded hash of on-disk name, and
1948          * client is doing access without encryption key.
1949          * So we need to get LinkEA, check parent fid is correct and
1950          * compare name hash with the one in the request.
1951          */
1952         struct lu_buf *buf = &info->mti_big_buf;
1953         struct lu_name name;
1954         struct lu_fid pfid;
1955         struct linkea_data ldata = { NULL };
1956         struct link_ea_header *leh;
1957         struct link_ea_entry *lee;
1958         struct lu_buf link = { 0 };
1959         char *hash;
1960         int reclen, count, rc;
1961
1962         ENTRY;
1963         if (lname->ln_namelen < LL_CRYPTO_BLOCK_SIZE)
1964                 RETURN(-EINVAL);
1965
1966         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
1967         if (!buf->lb_buf)
1968                 RETURN(-ENOMEM);
1969
1970         ldata.ld_buf = buf;
1971         rc = mdt_links_read(info, child, &ldata);
1972         if (rc < 0)
1973                 RETURN(rc);
1974
1975         hash = kmalloc(lname->ln_namelen, GFP_NOFS);
1976         if (!hash)
1977                 RETURN(-ENOMEM);
1978         rc = critical_decode(lname->ln_name, lname->ln_namelen, hash);
1979
1980         leh = buf->lb_buf;
1981         lee = (struct link_ea_entry *)(leh + 1);
1982         for (count = 0; count < leh->leh_reccount; count++) {
1983                 linkea_entry_unpack(lee, &reclen, &name, &pfid);
1984                 if (!parent || lu_fid_eq(&pfid, mdt_object_fid(parent))) {
1985                         lu_buf_check_and_alloc(&link, name.ln_namelen);
1986                         if (!link.lb_buf)
1987                                 GOTO(out_match, rc = -ENOMEM);
1988                         rc = critical_decode(name.ln_name, name.ln_namelen,
1989                                              link.lb_buf);
1990
1991                         if (memcmp(LLCRYPT_EXTRACT_DIGEST(link.lb_buf, rc),
1992                                    hash, LL_CRYPTO_BLOCK_SIZE) == 0) {
1993                                 *lname = name;
1994                                 break;
1995                         }
1996                 }
1997                 lee = (struct link_ea_entry *) ((char *)lee + reclen);
1998         }
1999         if (count == leh->leh_reccount)
2000                 rc = 0;
2001         else
2002                 rc = 1;
2003
2004 out_match:
2005         lu_buf_free(&link);
2006         kfree(hash);
2007
2008         RETURN(rc);
2009 }
2010
2011 /*
2012  * UPDATE lock should be taken against parent, and be released before exit;
2013  * child_bits lock should be taken against child, and be returned back:
2014  *            (1)normal request should release the child lock;
2015  *            (2)intent request will grant the lock to client.
2016  */
2017 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
2018                                  struct mdt_lock_handle *lhc,
2019                                  __u64 child_bits,
2020                                  struct ldlm_reply *ldlm_rep)
2021 {
2022         struct ptlrpc_request *req = mdt_info_req(info);
2023         struct mdt_body *reqbody = NULL;
2024         struct mdt_object *parent = info->mti_object;
2025         struct mdt_object *child = NULL;
2026         struct lu_fid *child_fid = &info->mti_tmp_fid1;
2027         struct lu_name *lname = NULL;
2028         struct mdt_lock_handle *lhp = NULL;
2029         struct ldlm_lock *lock;
2030         struct req_capsule *pill = info->mti_pill;
2031         __u64 try_bits = 0;
2032         bool is_resent;
2033         int ma_need = 0;
2034         int rc;
2035
2036         ENTRY;
2037
2038         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
2039         LASSERT(ergo(is_resent,
2040                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
2041
2042         if (parent == NULL)
2043                 RETURN(-ENOENT);
2044
2045         if (info->mti_cross_ref) {
2046                 /* Only getattr on the child. Parent is on another node. */
2047                 mdt_set_disposition(info, ldlm_rep,
2048                                     DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
2049                 child = parent;
2050                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
2051                        "ldlm_rep = %p\n",
2052                        PFID(mdt_object_fid(child)), ldlm_rep);
2053
2054                 rc = mdt_check_resent_lock(info, child, lhc);
2055                 if (rc < 0) {
2056                         RETURN(rc);
2057                 } else if (rc > 0) {
2058                         mdt_lock_handle_init(lhc);
2059                         mdt_lock_reg_init(lhc, LCK_PR);
2060
2061                         /*
2062                          * Object's name entry is on another MDS, it will
2063                          * request PERM lock only because LOOKUP lock is owned
2064                          * by the MDS where name entry resides.
2065                          *
2066                          * TODO: it should try layout lock too. - Jinshan
2067                          */
2068                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
2069                                         MDS_INODELOCK_LAYOUT);
2070                         child_bits |= MDS_INODELOCK_PERM;
2071
2072                         rc = mdt_object_lock(info, child, lhc, child_bits);
2073                         if (rc < 0)
2074                                 RETURN(rc);
2075                 }
2076
2077                 /* Finally, we can get attr for child. */
2078                 if (!mdt_object_exists(child)) {
2079                         LU_OBJECT_DEBUG(D_INFO, info->mti_env,
2080                                         &child->mot_obj,
2081                                         "remote object doesn't exist.");
2082                         mdt_object_unlock(info, child, lhc, 1);
2083                         RETURN(-ENOENT);
2084                 }
2085
2086                 rc = mdt_getattr_internal(info, child, 0);
2087                 if (unlikely(rc != 0)) {
2088                         mdt_object_unlock(info, child, lhc, 1);
2089                         RETURN(rc);
2090                 }
2091
2092                 rc = mdt_pack_secctx_in_reply(info, child);
2093                 if (unlikely(rc)) {
2094                         mdt_object_unlock(info, child, lhc, 1);
2095                         RETURN(rc);
2096                 }
2097
2098                 rc = mdt_pack_encctx_in_reply(info, child);
2099                 if (unlikely(rc))
2100                         mdt_object_unlock(info, child, lhc, 1);
2101                 RETURN(rc);
2102         }
2103
2104         lname = &info->mti_name;
2105         mdt_name_unpack(pill, &RMF_NAME, lname, MNF_FIX_ANON);
2106
2107         if (info->mti_body->mbo_valid & OBD_MD_NAMEHASH) {
2108                 reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
2109                 if (unlikely(reqbody == NULL))
2110                         RETURN(err_serious(-EPROTO));
2111
2112                 *child_fid = reqbody->mbo_fid2;
2113                 if (unlikely(!fid_is_sane(child_fid)))
2114                         RETURN(err_serious(-EINVAL));
2115
2116                 if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
2117                         mdt_object_get(info->mti_env, parent);
2118                         child = parent;
2119                 } else {
2120                         child = mdt_object_find(info->mti_env, info->mti_mdt,
2121                                                 child_fid);
2122                         if (IS_ERR(child))
2123                                 RETURN(PTR_ERR(child));
2124                 }
2125
2126                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
2127                        "ldlm_rep = %p\n",
2128                        PFID(mdt_object_fid(parent)),
2129                        PFID(&reqbody->mbo_fid2), ldlm_rep);
2130         } else if (lu_name_is_valid(lname)) {
2131                 if (mdt_object_remote(parent)) {
2132                         CERROR("%s: parent "DFID" is on remote target\n",
2133                                mdt_obd_name(info->mti_mdt),
2134                                PFID(mdt_object_fid(parent)));
2135                         RETURN(-EPROTO);
2136                 }
2137
2138                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
2139                        "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
2140                        PNAME(lname), ldlm_rep);
2141         } else {
2142                 reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
2143                 if (unlikely(reqbody == NULL))
2144                         RETURN(err_serious(-EPROTO));
2145
2146                 *child_fid = reqbody->mbo_fid2;
2147                 if (unlikely(!fid_is_sane(child_fid)))
2148                         RETURN(err_serious(-EINVAL));
2149
2150                 if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
2151                         mdt_object_get(info->mti_env, parent);
2152                         child = parent;
2153                 } else {
2154                         child = mdt_object_find(info->mti_env, info->mti_mdt,
2155                                                 child_fid);
2156                         if (IS_ERR(child))
2157                                 RETURN(PTR_ERR(child));
2158                 }
2159
2160                 if (mdt_object_remote(child)) {
2161                         CERROR("%s: child "DFID" is on remote target\n",
2162                                mdt_obd_name(info->mti_mdt),
2163                                PFID(mdt_object_fid(child)));
2164                         GOTO(out_child, rc = -EPROTO);
2165                 }
2166
2167                 /* don't fetch LOOKUP lock if it's remote object */
2168                 rc = mdt_is_remote_object(info, parent, child);
2169                 if (rc < 0)
2170                         GOTO(out_child, rc);
2171                 if (rc)
2172                         child_bits &= ~MDS_INODELOCK_LOOKUP;
2173
2174                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
2175                        "ldlm_rep = %p\n",
2176                        PFID(mdt_object_fid(parent)),
2177                        PFID(&reqbody->mbo_fid2), ldlm_rep);
2178         }
2179
2180         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
2181
2182         if (unlikely(!mdt_object_exists(parent)) &&
2183             !(info->mti_body->mbo_valid & OBD_MD_NAMEHASH) &&
2184             lu_name_is_valid(lname)) {
2185                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
2186                                 &parent->mot_obj,
2187                                 "Parent doesn't exist!");
2188                 GOTO(out_child, rc = -ESTALE);
2189         }
2190
2191         if (!child && is_resent) {
2192                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2193                 if (lock == NULL) {
2194                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
2195                          * a resend case, however, it could be already destroyed
2196                          * due to client eviction or a raced cancel RPC.
2197                          */
2198                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx",
2199                                           lhc->mlh_reg_lh.cookie);
2200                         RETURN(-ESTALE);
2201                 }
2202                 fid_extract_from_res_name(child_fid,
2203                                           &lock->l_resource->lr_name);
2204                 LDLM_LOCK_PUT(lock);
2205                 child = mdt_object_find(info->mti_env, info->mti_mdt,
2206                                         child_fid);
2207                 if (IS_ERR(child))
2208                         RETURN(PTR_ERR(child));
2209         } else if (!(info->mti_body->mbo_valid & OBD_MD_NAMEHASH) &&
2210             lu_name_is_valid(lname)) {
2211                 if (info->mti_body->mbo_valid == OBD_MD_FLID) {
2212                         rc = mdt_raw_lookup(info, parent, lname);
2213
2214                         RETURN(rc);
2215                 }
2216
2217                 /* step 1: lock parent only if parent is a directory */
2218                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
2219                         lhp = &info->mti_lh[MDT_LH_PARENT];
2220                         mdt_lock_pdo_init(lhp, LCK_PR, lname);
2221                         rc = mdt_object_lock(info, parent, lhp,
2222                                              MDS_INODELOCK_UPDATE);
2223                         if (unlikely(rc != 0))
2224                                 RETURN(rc);
2225                 }
2226
2227                 /* step 2: lookup child's fid by name */
2228                 fid_zero(child_fid);
2229                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
2230                                 child_fid, &info->mti_spec);
2231                 if (rc == -ENOENT)
2232                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
2233
2234                 if (rc != 0)
2235                         GOTO(unlock_parent, rc);
2236
2237                 child = mdt_object_find(info->mti_env, info->mti_mdt,
2238                                         child_fid);
2239                 if (unlikely(IS_ERR(child)))
2240                         GOTO(unlock_parent, rc = PTR_ERR(child));
2241         }
2242
2243         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
2244
2245         /* step 3: lock child regardless if it is local or remote. */
2246         LASSERT(child);
2247
2248         if (info->mti_body->mbo_valid & OBD_MD_NAMEHASH) {
2249                 /* Here, lname is an encoded hash of on-disk name, and
2250                  * client is doing access without encryption key.
2251                  * So we need to compare name hash with the one in the request.
2252                  */
2253                 if (!find_name_matching_hash(info, lname, parent,
2254                                              child)) {
2255                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
2256                         mdt_clear_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
2257                         GOTO(out_child, rc = -ENOENT);
2258                 }
2259         }
2260
2261         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
2262         if (!mdt_object_exists(child)) {
2263                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
2264                                 &child->mot_obj,
2265                                 "Object doesn't exist!");
2266                 GOTO(out_child, rc = -ENOENT);
2267         }
2268
2269         rc = mdt_check_resent_lock(info, child, lhc);
2270         if (rc < 0) {
2271                 GOTO(out_child, rc);
2272         } else if (rc > 0) {
2273                 mdt_lock_handle_init(lhc);
2274                 mdt_lock_reg_init(lhc, LCK_PR);
2275
2276                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
2277                     !mdt_object_remote(child)) {
2278                         struct md_attr *ma = &info->mti_attr;
2279
2280                         ma->ma_valid = 0;
2281                         ma->ma_need = MA_INODE;
2282                         rc = mdt_attr_get_complex(info, child, ma);
2283                         if (unlikely(rc != 0))
2284                                 GOTO(out_child, rc);
2285
2286                         /* If the file has not been changed for some time, we
2287                          * return not only a LOOKUP lock, but also an UPDATE
2288                          * lock and this might save us RPC on later STAT. For
2289                          * directories, it also let negative dentry cache start
2290                          * working for this dir. */
2291                         if (ma->ma_valid & MA_INODE &&
2292                             ma->ma_attr.la_valid & LA_CTIME &&
2293                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
2294                             ma->ma_attr.la_ctime < ktime_get_real_seconds())
2295                                 child_bits |= MDS_INODELOCK_UPDATE;
2296                 }
2297
2298                 /* layout lock must be granted in a best-effort way
2299                  * for IT operations */
2300                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
2301                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
2302                     !mdt_object_remote(child) && ldlm_rep != NULL) {
2303                         if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
2304                             exp_connect_layout(info->mti_exp)) {
2305                                 /* try to grant layout lock for regular file. */
2306                                 try_bits = MDS_INODELOCK_LAYOUT;
2307                         }
2308                         /* Acquire DOM lock in advance for data-on-mdt file */
2309                         if (child != parent)
2310                                 try_bits |= MDS_INODELOCK_DOM;
2311                 }
2312
2313                 if (try_bits != 0) {
2314                         /* try layout lock, it may fail to be granted due to
2315                          * contention at LOOKUP or UPDATE */
2316                         rc = mdt_object_lock_try(info, child, lhc, &child_bits,
2317                                                  try_bits, false);
2318                         if (child_bits & MDS_INODELOCK_LAYOUT)
2319                                 ma_need |= MA_LOV;
2320                 } else {
2321                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
2322                          * client will enqueue the lock to the remote MDT */
2323                         if (mdt_object_remote(child))
2324                                 child_bits &= ~MDS_INODELOCK_UPDATE;
2325                         rc = mdt_object_lock(info, child, lhc, child_bits);
2326                 }
2327                 if (unlikely(rc != 0))
2328                         GOTO(out_child, rc);
2329         }
2330
2331         /* finally, we can get attr for child. */
2332         rc = mdt_getattr_internal(info, child, ma_need);
2333         if (unlikely(rc != 0)) {
2334                 if (!is_resent)
2335                         mdt_object_unlock(info, child, lhc, 1);
2336                 GOTO(out_child, rc);
2337         }
2338
2339         rc = mdt_pack_secctx_in_reply(info, child);
2340         if (unlikely(rc)) {
2341                 if (!is_resent)
2342                         mdt_object_unlock(info, child, lhc, 1);
2343                 GOTO(out_child, rc);
2344         }
2345
2346         rc = mdt_pack_encctx_in_reply(info, child);
2347         if (unlikely(rc)) {
2348                 if (!is_resent)
2349                         mdt_object_unlock(info, child, lhc, 1);
2350                 GOTO(out_child, rc);
2351         }
2352
2353         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2354         if (lock) {
2355                 /* Debugging code. */
2356                 LDLM_DEBUG(lock, "Returning lock to client");
2357                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
2358                                          &lock->l_resource->lr_name),
2359                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
2360                          PLDLMRES(lock->l_resource),
2361                          PFID(mdt_object_fid(child)));
2362
2363                 if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_PTLRPC_ENQ_RESEND))) {
2364                         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
2365                                 OBD_FAIL_TIMEOUT(OBD_FAIL_PTLRPC_ENQ_RESEND,
2366                                                  req->rq_deadline -
2367                                                  req->rq_arrival_time.tv_sec +
2368                                                  cfs_fail_val ?: 3);
2369                         /* Put the lock to the waiting list and force the cancel */
2370                         ldlm_set_ast_sent(lock);
2371                 }
2372
2373                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
2374                     !mdt_object_remote(child) && child != parent) {
2375                         mdt_object_put(info->mti_env, child);
2376                         rc = mdt_pack_size2body(info, child_fid,
2377                                                 &lhc->mlh_reg_lh);
2378                         if (rc != 0 && child_bits & MDS_INODELOCK_DOM) {
2379                                 /* DOM lock was taken in advance but this is
2380                                  * not DoM file. Drop the lock.
2381                                  */
2382                                 lock_res_and_lock(lock);
2383                                 ldlm_inodebits_drop(lock, MDS_INODELOCK_DOM);
2384                                 unlock_res_and_lock(lock);
2385                         }
2386                         LDLM_LOCK_PUT(lock);
2387                         GOTO(unlock_parent, rc = 0);
2388                 }
2389                 LDLM_LOCK_PUT(lock);
2390         }
2391
2392         EXIT;
2393 out_child:
2394         if (child)
2395                 mdt_object_put(info->mti_env, child);
2396 unlock_parent:
2397         if (lhp)
2398                 mdt_object_unlock(info, parent, lhp, 1);
2399         return rc;
2400 }
2401
2402 /* normal handler: should release the child lock */
2403 static int mdt_getattr_name(struct tgt_session_info *tsi)
2404 {
2405         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
2406         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
2407         struct mdt_body *reqbody;
2408         struct mdt_body *repbody;
2409         int rc, rc2;
2410
2411         ENTRY;
2412
2413         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
2414         LASSERT(reqbody != NULL);
2415         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
2416         LASSERT(repbody != NULL);
2417
2418         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
2419         repbody->mbo_eadatasize = 0;
2420         repbody->mbo_aclsize = 0;
2421
2422         rc = mdt_init_ucred(info, reqbody);
2423         if (unlikely(rc))
2424                 GOTO(out_shrink, rc);
2425
2426         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
2427         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
2428                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
2429                 lhc->mlh_reg_lh.cookie = 0;
2430         }
2431         mdt_exit_ucred(info);
2432         EXIT;
2433 out_shrink:
2434         mdt_client_compatibility(info);
2435         rc2 = mdt_fix_reply(info);
2436         if (rc == 0)
2437                 rc = rc2;
2438         mdt_thread_info_fini(info);
2439         return rc;
2440 }
2441
2442 static int mdt_rmfid_unlink(struct mdt_thread_info *info,
2443                             const struct lu_fid *pfid,
2444                             const struct lu_name *name,
2445                             struct mdt_object *obj, s64 ctime)
2446 {
2447         struct lu_fid *child_fid = &info->mti_tmp_fid1;
2448         struct ldlm_enqueue_info *einfo = &info->mti_einfo[0];
2449         struct mdt_device *mdt = info->mti_mdt;
2450         struct md_attr *ma = &info->mti_attr;
2451         struct mdt_lock_handle *parent_lh;
2452         struct mdt_lock_handle *child_lh;
2453         struct mdt_object *pobj;
2454         bool cos_incompat = false;
2455         int rc;
2456         ENTRY;
2457
2458         pobj = mdt_object_find(info->mti_env, mdt, pfid);
2459         if (IS_ERR(pobj))
2460                 GOTO(out, rc = PTR_ERR(pobj));
2461
2462         parent_lh = &info->mti_lh[MDT_LH_PARENT];
2463         mdt_lock_pdo_init(parent_lh, LCK_PW, name);
2464         rc = mdt_object_lock(info, pobj, parent_lh, MDS_INODELOCK_UPDATE);
2465         if (rc != 0)
2466                 GOTO(put_parent, rc);
2467
2468         if (mdt_object_remote(pobj))
2469                 cos_incompat = true;
2470
2471         rc = mdo_lookup(info->mti_env, mdt_object_child(pobj),
2472                         name, child_fid, &info->mti_spec);
2473         if (rc != 0)
2474                 GOTO(unlock_parent, rc);
2475
2476         if (!lu_fid_eq(child_fid, mdt_object_fid(obj)))
2477                 GOTO(unlock_parent, rc = -EREMCHG);
2478
2479         child_lh = &info->mti_lh[MDT_LH_CHILD];
2480         mdt_lock_reg_init(child_lh, LCK_EX);
2481         rc = mdt_reint_striped_lock(info, obj, child_lh,
2482                                     MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE,
2483                                     einfo, cos_incompat);
2484         if (rc != 0)
2485                 GOTO(unlock_parent, rc);
2486
2487         if (atomic_read(&obj->mot_open_count)) {
2488                 CDEBUG(D_OTHER, "object "DFID" open, skip\n",
2489                        PFID(mdt_object_fid(obj)));
2490                 GOTO(unlock_child, rc = -EBUSY);
2491         }
2492
2493         ma->ma_need = 0;
2494         ma->ma_valid = MA_INODE;
2495         ma->ma_attr.la_valid = LA_CTIME;
2496         ma->ma_attr.la_ctime = ctime;
2497
2498         mutex_lock(&obj->mot_lov_mutex);
2499
2500         rc = mdo_unlink(info->mti_env, mdt_object_child(pobj),
2501                         mdt_object_child(obj), name, ma, 0);
2502
2503         mutex_unlock(&obj->mot_lov_mutex);
2504
2505 unlock_child:
2506         mdt_reint_striped_unlock(info, obj, child_lh, einfo, 1);
2507 unlock_parent:
2508         mdt_object_unlock(info, pobj, parent_lh, 1);
2509 put_parent:
2510         mdt_object_put(info->mti_env, pobj);
2511 out:
2512         RETURN(rc);
2513 }
2514
2515 static int mdt_rmfid_check_permission(struct mdt_thread_info *info,
2516                                         struct mdt_object *obj)
2517 {
2518         struct lu_ucred *uc = lu_ucred(info->mti_env);
2519         struct md_attr *ma = &info->mti_attr;
2520         struct lu_attr *la = &ma->ma_attr;
2521         int rc = 0;
2522         ENTRY;
2523
2524         ma->ma_need = MA_INODE;
2525         rc = mo_attr_get(info->mti_env, mdt_object_child(obj), ma);
2526         if (rc)
2527                 GOTO(out, rc);
2528
2529         if (la->la_flags & LUSTRE_IMMUTABLE_FL)
2530                         rc = -EACCES;
2531
2532         if (cap_raised(uc->uc_cap, CAP_DAC_OVERRIDE))
2533                 RETURN(0);
2534         if (uc->uc_fsuid == la->la_uid) {
2535                 if ((la->la_mode & S_IWUSR) == 0)
2536                         rc = -EACCES;
2537         } else if (uc->uc_fsgid == la->la_gid) {
2538                 if ((la->la_mode & S_IWGRP) == 0)
2539                         rc = -EACCES;
2540         } else if ((la->la_mode & S_IWOTH) == 0) {
2541                         rc = -EACCES;
2542         }
2543
2544 out:
2545         RETURN(rc);
2546 }
2547
2548 static int mdt_rmfid_one(struct mdt_thread_info *info, struct lu_fid *fid,
2549                          s64 ctime)
2550 {
2551         struct mdt_device *mdt = info->mti_mdt;
2552         struct mdt_object *obj = NULL;
2553         struct linkea_data ldata = { NULL };
2554         struct lu_buf *buf = &info->mti_big_buf;
2555         struct lu_name *name = &info->mti_name;
2556         struct lu_fid *pfid = &info->mti_tmp_fid1;
2557         struct link_ea_header *leh;
2558         struct link_ea_entry *lee;
2559         int reclen, count, rc = 0;
2560         ENTRY;
2561
2562         if (!fid_is_sane(fid))
2563                 GOTO(out, rc = -EINVAL);
2564
2565         if (!fid_is_namespace_visible(fid))
2566                 GOTO(out, rc = -EINVAL);
2567
2568         obj = mdt_object_find(info->mti_env, mdt, fid);
2569         if (IS_ERR(obj))
2570                 GOTO(out, rc = PTR_ERR(obj));
2571
2572         if (mdt_object_remote(obj))
2573                 GOTO(out, rc = -EREMOTE);
2574         if (!mdt_object_exists(obj) || lu_object_is_dying(&obj->mot_header))
2575                 GOTO(out, rc = -ENOENT);
2576
2577         rc = mdt_rmfid_check_permission(info, obj);
2578         if (rc)
2579                 GOTO(out, rc);
2580
2581         /* take LinkEA */
2582         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
2583         if (!buf->lb_buf)
2584                 GOTO(out, rc = -ENOMEM);
2585
2586         ldata.ld_buf = buf;
2587         rc = mdt_links_read(info, obj, &ldata);
2588         if (rc)
2589                 GOTO(out, rc);
2590
2591         leh = buf->lb_buf;
2592         lee = (struct link_ea_entry *)(leh + 1);
2593         for (count = 0; count < leh->leh_reccount; count++) {
2594                 /* remove every hardlink */
2595                 linkea_entry_unpack(lee, &reclen, name, pfid);
2596                 lee = (struct link_ea_entry *) ((char *)lee + reclen);
2597                 rc = mdt_rmfid_unlink(info, pfid, name, obj, ctime);
2598                 if (rc)
2599                         break;
2600         }
2601
2602 out:
2603         if (obj && !IS_ERR(obj))
2604                 mdt_object_put(info->mti_env, obj);
2605         if (info->mti_big_buf.lb_buf)
2606                 lu_buf_free(&info->mti_big_buf);
2607
2608         RETURN(rc);
2609 }
2610
2611 static int mdt_rmfid(struct tgt_session_info *tsi)
2612 {
2613         struct mdt_thread_info *mti = tsi2mdt_info(tsi);
2614         struct mdt_body *reqbody;
2615         struct lu_fid *fids, *rfids;
2616         int bufsize, rc;
2617         __u32 *rcs;
2618         int i, nr;
2619         ENTRY;
2620
2621         reqbody = req_capsule_client_get(tsi->tsi_pill, &RMF_MDT_BODY);
2622         if (reqbody == NULL)
2623                 RETURN(-EPROTO);
2624         bufsize = req_capsule_get_size(tsi->tsi_pill, &RMF_FID_ARRAY,
2625                                        RCL_CLIENT);
2626         nr = bufsize / sizeof(struct lu_fid);
2627         if (nr * sizeof(struct lu_fid) != bufsize)
2628                 RETURN(-EINVAL);
2629         req_capsule_set_size(tsi->tsi_pill, &RMF_RCS,
2630                              RCL_SERVER, nr * sizeof(__u32));
2631         req_capsule_set_size(tsi->tsi_pill, &RMF_FID_ARRAY,
2632                              RCL_SERVER, nr * sizeof(struct lu_fid));
2633         rc = req_capsule_server_pack(tsi->tsi_pill);
2634         if (rc)
2635                 GOTO(out, rc = err_serious(rc));
2636         fids = req_capsule_client_get(tsi->tsi_pill, &RMF_FID_ARRAY);
2637         if (fids == NULL)
2638                 RETURN(-EPROTO);
2639         rcs = req_capsule_server_get(tsi->tsi_pill, &RMF_RCS);
2640         LASSERT(rcs);
2641         rfids = req_capsule_server_get(tsi->tsi_pill, &RMF_FID_ARRAY);
2642         LASSERT(rfids);
2643
2644         mdt_init_ucred(mti, reqbody);
2645         for (i = 0; i < nr; i++) {
2646                 rfids[i] = fids[i];
2647                 rcs[i] = mdt_rmfid_one(mti, fids + i, reqbody->mbo_ctime);
2648         }
2649         mdt_exit_ucred(mti);
2650
2651 out:
2652         RETURN(rc);
2653 }
2654
2655 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2656                          void *karg, void __user *uarg);
2657
2658 int mdt_io_set_info(struct tgt_session_info *tsi)
2659 {
2660         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2661         struct ost_body         *body = NULL, *repbody;
2662         void                    *key, *val = NULL;
2663         int                      keylen, vallen, rc = 0;
2664         bool                     is_grant_shrink;
2665
2666         ENTRY;
2667
2668         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
2669         if (key == NULL) {
2670                 DEBUG_REQ(D_HA, req, "no set_info key");
2671                 RETURN(err_serious(-EFAULT));
2672         }
2673         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
2674                                       RCL_CLIENT);
2675
2676         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
2677         if (val == NULL) {
2678                 DEBUG_REQ(D_HA, req, "no set_info val");
2679                 RETURN(err_serious(-EFAULT));
2680         }
2681         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
2682                                       RCL_CLIENT);
2683
2684         is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK);
2685         if (is_grant_shrink)
2686                 /* In this case the value is actually an RMF_OST_BODY, so we
2687                  * transmutate the type of this PTLRPC */
2688                 req_capsule_extend(tsi->tsi_pill, &RQF_OST_SET_GRANT_INFO);
2689
2690         rc = req_capsule_server_pack(tsi->tsi_pill);
2691         if (rc < 0)
2692                 RETURN(rc);
2693
2694         if (is_grant_shrink) {
2695                 body = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_BODY);
2696
2697                 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2698                 *repbody = *body;
2699
2700                 /** handle grant shrink, similar to a read request */
2701                 tgt_grant_prepare_read(tsi->tsi_env, tsi->tsi_exp,
2702                                        &repbody->oa);
2703         } else {
2704                 CERROR("%s: Unsupported key %s\n",
2705                        tgt_name(tsi->tsi_tgt), (char *)key);
2706                 rc = -EOPNOTSUPP;
2707         }
2708
2709         RETURN(rc);
2710 }
2711
2712
2713 static int mdt_set_info(struct tgt_session_info *tsi)
2714 {
2715         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2716         char                    *key;
2717         void                    *val;
2718         int                      keylen, vallen, rc = 0;
2719
2720         ENTRY;
2721
2722         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
2723         if (key == NULL) {
2724                 DEBUG_REQ(D_HA, req, "no set_info key");
2725                 RETURN(err_serious(-EFAULT));
2726         }
2727
2728         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
2729                                       RCL_CLIENT);
2730
2731         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
2732         if (val == NULL) {
2733                 DEBUG_REQ(D_HA, req, "no set_info val");
2734                 RETURN(err_serious(-EFAULT));
2735         }
2736
2737         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
2738                                       RCL_CLIENT);
2739
2740         /* Swab any part of val you need to here */
2741         if (KEY_IS(KEY_READ_ONLY)) {
2742                 spin_lock(&req->rq_export->exp_lock);
2743                 if (*(__u32 *)val)
2744                         *exp_connect_flags_ptr(req->rq_export) |=
2745                                 OBD_CONNECT_RDONLY;
2746                 else
2747                         *exp_connect_flags_ptr(req->rq_export) &=
2748                                 ~OBD_CONNECT_RDONLY;
2749                 spin_unlock(&req->rq_export->exp_lock);
2750         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2751                 struct changelog_setinfo *cs = val;
2752
2753                 if (vallen != sizeof(*cs)) {
2754                         CERROR("%s: bad changelog_clear setinfo size %d\n",
2755                                tgt_name(tsi->tsi_tgt), vallen);
2756                         RETURN(-EINVAL);
2757                 }
2758                 if (req_capsule_req_need_swab(&req->rq_pill)) {
2759                         __swab64s(&cs->cs_recno);
2760                         __swab32s(&cs->cs_id);
2761                 }
2762
2763                 if (!mdt_is_rootadmin(tsi2mdt_info(tsi)))
2764                         RETURN(-EACCES);
2765                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
2766                                    vallen, val, NULL);
2767         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2768                 if (vallen > 0)
2769                         obd_export_evict_by_nid(req->rq_export->exp_obd, val);
2770         } else {
2771                 RETURN(-EINVAL);
2772         }
2773         RETURN(rc);
2774 }
2775
2776 static int mdt_readpage(struct tgt_session_info *tsi)
2777 {
2778         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
2779         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
2780         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
2781         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
2782         struct mdt_body         *repbody;
2783         int                      rc;
2784         int                      i;
2785
2786         ENTRY;
2787
2788         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
2789                 RETURN(err_serious(-ENOMEM));
2790
2791         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
2792         if (repbody == NULL || reqbody == NULL)
2793                 RETURN(err_serious(-EFAULT));
2794
2795         /*
2796          * prepare @rdpg before calling lower layers and transfer itself. Here
2797          * reqbody->size contains offset of where to start to read and
2798          * reqbody->nlink contains number bytes to read.
2799          */
2800         rdpg->rp_hash = reqbody->mbo_size;
2801         if (rdpg->rp_hash != reqbody->mbo_size) {
2802                 CERROR("Invalid hash: %#llx != %#llx\n",
2803                        rdpg->rp_hash, reqbody->mbo_size);
2804                 RETURN(-EFAULT);
2805         }
2806
2807         rdpg->rp_attrs = reqbody->mbo_mode;
2808         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
2809                 rdpg->rp_attrs |= LUDA_64BITHASH;
2810         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
2811                                 exp_max_brw_size(tsi->tsi_exp));
2812         rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >>
2813                           PAGE_SHIFT;
2814         OBD_ALLOC_PTR_ARRAY_LARGE(rdpg->rp_pages, rdpg->rp_npages);
2815         if (rdpg->rp_pages == NULL)
2816                 RETURN(-ENOMEM);
2817
2818         for (i = 0; i < rdpg->rp_npages; ++i) {
2819                 rdpg->rp_pages[i] = alloc_page(GFP_NOFS);
2820                 if (rdpg->rp_pages[i] == NULL)
2821                         GOTO(free_rdpg, rc = -ENOMEM);
2822         }
2823
2824         /* call lower layers to fill allocated pages with directory data */
2825         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
2826         if (rc < 0)
2827                 GOTO(free_rdpg, rc);
2828
2829         /* send pages to client */
2830         rc = tgt_sendpage(tsi, rdpg, rc);
2831
2832         EXIT;
2833 free_rdpg:
2834
2835         for (i = 0; i < rdpg->rp_npages; i++)
2836                 if (rdpg->rp_pages[i] != NULL)
2837                         __free_page(rdpg->rp_pages[i]);
2838         OBD_FREE_PTR_ARRAY_LARGE(rdpg->rp_pages, rdpg->rp_npages);
2839
2840         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
2841                 RETURN(0);
2842
2843         return rc;
2844 }
2845
2846 static int mdt_fix_attr_ucred(struct mdt_thread_info *info, __u32 op)
2847 {
2848         struct lu_ucred *uc = mdt_ucred_check(info);
2849         struct lu_attr *attr = &info->mti_attr.ma_attr;
2850
2851         if (uc == NULL)
2852                 return -EINVAL;
2853
2854         if (op != REINT_SETATTR) {
2855                 if ((attr->la_valid & LA_UID) && (attr->la_uid != -1))
2856                         attr->la_uid = uc->uc_fsuid;
2857                 /* for S_ISGID, inherit gid from his parent, such work will be
2858                  * done in cmm/mdd layer, here set all cases as uc->uc_fsgid. */
2859                 if ((attr->la_valid & LA_GID) && (attr->la_gid != -1))
2860                         attr->la_gid = uc->uc_fsgid;
2861         }
2862
2863         return 0;
2864 }
2865
2866 static inline bool mdt_is_readonly_open(struct mdt_thread_info *info, __u32 op)
2867 {
2868         return op == REINT_OPEN &&
2869              !(info->mti_spec.sp_cr_flags & (MDS_FMODE_WRITE | MDS_OPEN_CREAT));
2870 }
2871
2872 static void mdt_preset_secctx_size(struct mdt_thread_info *info)
2873 {
2874         struct req_capsule *pill = info->mti_pill;
2875
2876         if (req_capsule_has_field(pill, &RMF_FILE_SECCTX,
2877                                   RCL_SERVER) &&
2878             req_capsule_has_field(pill, &RMF_FILE_SECCTX_NAME,
2879                                   RCL_CLIENT)) {
2880                 if (req_capsule_get_size(pill, &RMF_FILE_SECCTX_NAME,
2881                                          RCL_CLIENT) != 0)
2882                         /* pre-set size in server part with max size */
2883                         req_capsule_set_size(pill, &RMF_FILE_SECCTX,
2884                                              RCL_SERVER,
2885                                              OBD_MAX_DEFAULT_EA_SIZE);
2886                 else
2887                         req_capsule_set_size(pill, &RMF_FILE_SECCTX,
2888                                              RCL_SERVER, 0);
2889         }
2890 }
2891
2892 static int mdt_reint_internal(struct mdt_thread_info *info,
2893                               struct mdt_lock_handle *lhc,
2894                               __u32 op)
2895 {
2896         struct req_capsule      *pill = info->mti_pill;
2897         struct mdt_body         *repbody;
2898         int                      rc = 0, rc2;
2899
2900         ENTRY;
2901
2902         rc = mdt_reint_unpack(info, op);
2903         if (rc != 0) {
2904                 CERROR("Can't unpack reint, rc %d\n", rc);
2905                 RETURN(err_serious(rc));
2906         }
2907
2908
2909         /* check if the file system is set to readonly. O_RDONLY open
2910          * is still allowed even the file system is set to readonly mode */
2911         if (mdt_rdonly(info->mti_exp) && !mdt_is_readonly_open(info, op))
2912                 RETURN(err_serious(-EROFS));
2913
2914         /* for replay (no_create) lmm is not needed, client has it already */
2915         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2916                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2917                                      DEF_REP_MD_SIZE);
2918
2919         /* llog cookies are always 0, the field is kept for compatibility */
2920         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2921                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
2922
2923         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
2924          * by default. If the target object has more ACL entries, then
2925          * enlarge the buffer when necessary. */
2926         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
2927                 req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
2928                                      LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
2929
2930         mdt_preset_secctx_size(info);
2931         mdt_preset_encctx_size(info);
2932
2933         rc = req_capsule_server_pack(pill);
2934         if (rc != 0) {
2935                 CERROR("Can't pack response, rc %d\n", rc);
2936                 RETURN(err_serious(rc));
2937         }
2938
2939         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
2940                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
2941                 LASSERT(repbody);
2942                 repbody->mbo_eadatasize = 0;
2943                 repbody->mbo_aclsize = 0;
2944         }
2945
2946         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
2947
2948         /* for replay no cookkie / lmm need, because client have this already */
2949         if (info->mti_spec.no_create)
2950                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2951                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
2952
2953         rc = mdt_init_ucred_reint(info);
2954         if (rc)
2955                 GOTO(out_shrink, rc);
2956
2957         rc = mdt_fix_attr_ucred(info, op);
2958         if (rc != 0)
2959                 GOTO(out_ucred, rc = err_serious(rc));
2960
2961         rc = mdt_check_resent(info, mdt_reconstruct, lhc);
2962         if (rc < 0) {
2963                 GOTO(out_ucred, rc);
2964         } else if (rc == 1) {
2965                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt");
2966                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
2967                 GOTO(out_ucred, rc);
2968         }
2969         rc = mdt_reint_rec(info, lhc);
2970         EXIT;
2971 out_ucred:
2972         mdt_exit_ucred(info);
2973 out_shrink:
2974         mdt_client_compatibility(info);
2975
2976         rc2 = mdt_fix_reply(info);
2977         if (rc == 0)
2978                 rc = rc2;
2979
2980         /*
2981          * Data-on-MDT optimization - read data along with OPEN and return it
2982          * in reply when possible.
2983          */
2984         if (rc == 0 && op == REINT_OPEN && !req_is_replay(pill->rc_req))
2985                 rc = mdt_dom_read_on_open(info, info->mti_mdt,
2986                                           &lhc->mlh_reg_lh);
2987
2988         return rc;
2989 }
2990
2991 static long mdt_reint_opcode(struct ptlrpc_request *req,
2992                              const struct req_format **fmt)
2993 {
2994         struct mdt_device       *mdt;
2995         struct mdt_rec_reint    *rec;
2996         long                     opc;
2997
2998         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
2999         if (rec != NULL) {
3000                 opc = rec->rr_opcode;
3001                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
3002                 if (opc < REINT_MAX && fmt[opc] != NULL)
3003                         req_capsule_extend(&req->rq_pill, fmt[opc]);
3004                 else {
3005                         mdt = mdt_exp2dev(req->rq_export);
3006                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
3007                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
3008                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
3009                         opc = err_serious(-EFAULT);
3010                 }
3011         } else {
3012                 opc = err_serious(-EFAULT);
3013         }
3014         return opc;
3015 }
3016
3017 static int mdt_reint(struct tgt_session_info *tsi)
3018 {
3019         long opc;
3020         int  rc;
3021         static const struct req_format *reint_fmts[REINT_MAX] = {
3022                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
3023                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
3024                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
3025                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
3026                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
3027                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
3028                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
3029                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
3030                 [REINT_MIGRATE]  = &RQF_MDS_REINT_MIGRATE,
3031                 [REINT_RESYNC]   = &RQF_MDS_REINT_RESYNC,
3032         };
3033
3034         ENTRY;
3035
3036         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
3037         if (opc >= 0) {
3038                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
3039                 /*
3040                  * No lock possible here from client to pass it to reint code
3041                  * path.
3042                  */
3043                 rc = mdt_reint_internal(info, NULL, opc);
3044                 mdt_thread_info_fini(info);
3045         } else {
3046                 rc = opc;
3047         }
3048
3049         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
3050         RETURN(rc);
3051 }
3052
3053 /* this should sync the whole device */
3054 int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
3055 {
3056         struct dt_device *dt = mdt->mdt_bottom;
3057         int rc;
3058         ENTRY;
3059
3060         rc = dt->dd_ops->dt_sync(env, dt);
3061         RETURN(rc);
3062 }
3063
3064 /* this should sync this object */
3065 static int mdt_object_sync(const struct lu_env *env, struct obd_export *exp,
3066                            struct mdt_object *mo)
3067 {
3068         int rc = 0;
3069
3070         ENTRY;
3071
3072         if (!mdt_object_exists(mo)) {
3073                 CWARN("%s: non existing object "DFID": rc = %d\n",
3074                       exp->exp_obd->obd_name, PFID(mdt_object_fid(mo)),
3075                       -ESTALE);
3076                 RETURN(-ESTALE);
3077         }
3078
3079         if (S_ISREG(lu_object_attr(&mo->mot_obj))) {
3080                 struct lu_target *tgt = tgt_ses_info(env)->tsi_tgt;
3081                 dt_obj_version_t version;
3082
3083                 version = dt_version_get(env, mdt_obj2dt(mo));
3084                 if (version > tgt->lut_obd->obd_last_committed)
3085                         rc = mo_object_sync(env, mdt_object_child(mo));
3086         } else {
3087                 rc = mo_object_sync(env, mdt_object_child(mo));
3088         }
3089
3090         RETURN(rc);
3091 }
3092
3093 static int mdt_sync(struct tgt_session_info *tsi)
3094 {
3095         struct ptlrpc_request   *req = tgt_ses_req(tsi);
3096         struct req_capsule      *pill = tsi->tsi_pill;
3097         struct mdt_body         *body;
3098         ktime_t                  kstart = ktime_get();
3099         int                      rc;
3100
3101         ENTRY;
3102
3103         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
3104                 RETURN(err_serious(-ENOMEM));
3105
3106         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
3107                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
3108         } else {
3109                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
3110
3111                 if (unlikely(info->mti_object == NULL))
3112                         RETURN(-EPROTO);
3113
3114                 /* sync an object */
3115                 rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp,
3116                                      info->mti_object);
3117                 if (rc == 0) {
3118                         const struct lu_fid *fid;
3119                         struct lu_attr *la = &info->mti_attr.ma_attr;
3120
3121                         info->mti_attr.ma_need = MA_INODE;
3122                         info->mti_attr.ma_valid = 0;
3123                         rc = mdt_attr_get_complex(info, info->mti_object,
3124                                                   &info->mti_attr);
3125                         if (rc == 0) {
3126                                 body = req_capsule_server_get(pill,
3127                                                               &RMF_MDT_BODY);
3128                                 fid = mdt_object_fid(info->mti_object);
3129                                 mdt_pack_attr2body(info, body, la, fid);
3130                         }
3131                 }
3132                 mdt_thread_info_fini(info);
3133         }
3134         if (rc == 0)
3135                 mdt_counter_incr(req, LPROC_MDT_SYNC,
3136                                  ktime_us_delta(ktime_get(), kstart));
3137
3138         RETURN(rc);
3139 }
3140
3141 static int mdt_data_sync(struct tgt_session_info *tsi)
3142 {
3143         struct mdt_thread_info *info;
3144         struct mdt_device *mdt = mdt_exp2dev(tsi->tsi_exp);
3145         struct ost_body *body = tsi->tsi_ost_body;
3146         struct ost_body *repbody;
3147         struct mdt_object *mo = NULL;
3148         struct md_attr *ma;
3149         int rc = 0;
3150
3151         ENTRY;
3152
3153         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
3154
3155         /* if no fid is specified then do nothing,
3156          * device sync is done via MDS_SYNC */
3157         if (fid_is_zero(&tsi->tsi_fid))
3158                 RETURN(0);
3159
3160         mo = mdt_object_find(tsi->tsi_env, mdt, &tsi->tsi_fid);
3161         if (IS_ERR(mo))
3162                 RETURN(PTR_ERR(mo));
3163
3164         rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp, mo);
3165         if (rc)
3166                 GOTO(put, rc);
3167
3168         repbody->oa.o_oi = body->oa.o_oi;
3169         repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
3170
3171         info = tsi2mdt_info(tsi);
3172         ma = &info->mti_attr;
3173         ma->ma_need = MA_INODE;
3174         ma->ma_valid = 0;
3175         rc = mdt_attr_get_complex(info, mo, ma);
3176         if (rc == 0)
3177                 obdo_from_la(&repbody->oa, &ma->ma_attr, VALID_FLAGS);
3178         else
3179                 rc = 0;
3180         mdt_thread_info_fini(info);
3181
3182         EXIT;
3183 put:
3184         if (mo != NULL)
3185                 mdt_object_put(tsi->tsi_env, mo);
3186         return rc;
3187 }
3188
3189 /*
3190  * Handle quota control requests to consult current usage/limit, but also
3191  * to configure quota enforcement
3192  */
3193 static int mdt_quotactl(struct tgt_session_info *tsi)
3194 {
3195         struct obd_export *exp  = tsi->tsi_exp;
3196         struct req_capsule *pill = tsi->tsi_pill;
3197         struct obd_quotactl *oqctl, *repoqc;
3198         int id, rc;
3199         struct mdt_device *mdt = mdt_exp2dev(exp);
3200         struct lu_device *qmt = mdt->mdt_qmt_dev;
3201         struct lu_nodemap *nodemap;
3202         ENTRY;
3203
3204         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
3205         if (!oqctl)
3206                 RETURN(err_serious(-EPROTO));
3207
3208         rc = req_capsule_server_pack(pill);
3209         if (rc)
3210                 RETURN(err_serious(rc));
3211
3212         nodemap = nodemap_get_from_exp(exp);
3213         if (IS_ERR(nodemap))
3214                 RETURN(PTR_ERR(nodemap));
3215
3216         switch (oqctl->qc_cmd) {
3217                 /* master quotactl */
3218         case Q_SETINFO:
3219         case Q_SETQUOTA:
3220         case LUSTRE_Q_SETDEFAULT:
3221         case LUSTRE_Q_SETQUOTAPOOL:
3222         case LUSTRE_Q_SETINFOPOOL:
3223         case LUSTRE_Q_SETDEFAULT_POOL:
3224         case LUSTRE_Q_DELETEQID:
3225                 if (!nodemap_can_setquota(nodemap, oqctl->qc_type,
3226                                           oqctl->qc_id))
3227                         GOTO(out_nodemap, rc = -EPERM);
3228                 fallthrough;
3229         case Q_GETINFO:
3230         case Q_GETQUOTA:
3231         case LUSTRE_Q_GETDEFAULT:
3232         case LUSTRE_Q_GETQUOTAPOOL:
3233         case LUSTRE_Q_GETINFOPOOL:
3234         case LUSTRE_Q_GETDEFAULT_POOL:
3235                 if (qmt == NULL)
3236                         GOTO(out_nodemap, rc = -EOPNOTSUPP);
3237                 /* slave quotactl */
3238                 fallthrough;
3239         case Q_GETOINFO:
3240         case Q_GETOQUOTA:
3241                 break;
3242         default:
3243                 rc = -EFAULT;
3244                 CERROR("%s: unsupported quotactl command %d: rc = %d\n",
3245                        mdt_obd_name(mdt), oqctl->qc_cmd, rc);
3246                 GOTO(out_nodemap, rc);
3247         }
3248
3249         id = oqctl->qc_id;
3250         switch (oqctl->qc_type) {
3251         case USRQUOTA:
3252                 id = nodemap_map_id(nodemap, NODEMAP_UID,
3253                                     NODEMAP_CLIENT_TO_FS, id);
3254                 break;
3255         case GRPQUOTA:
3256                 id = nodemap_map_id(nodemap, NODEMAP_GID,
3257                                     NODEMAP_CLIENT_TO_FS, id);
3258                 break;
3259         case PRJQUOTA:
3260                 id = nodemap_map_id(nodemap, NODEMAP_PROJID,
3261                                     NODEMAP_CLIENT_TO_FS, id);
3262                 break;
3263         default:
3264                 GOTO(out_nodemap, rc = -EOPNOTSUPP);
3265         }
3266         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
3267         if (repoqc == NULL)
3268                 GOTO(out_nodemap, rc = err_serious(-EFAULT));
3269
3270         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA)
3271                 barrier_exit(tsi->tsi_tgt->lut_bottom);
3272
3273         if (oqctl->qc_id != id)
3274                 swap(oqctl->qc_id, id);
3275
3276         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA) {
3277                 if (unlikely(!barrier_entry(tsi->tsi_tgt->lut_bottom)))
3278                         RETURN(-EINPROGRESS);
3279         }
3280
3281         switch (oqctl->qc_cmd) {
3282
3283         case Q_GETINFO:
3284         case Q_SETINFO:
3285         case Q_SETQUOTA:
3286         case Q_GETQUOTA:
3287         case LUSTRE_Q_SETDEFAULT:
3288         case LUSTRE_Q_GETDEFAULT:
3289         case LUSTRE_Q_SETQUOTAPOOL:
3290         case LUSTRE_Q_GETQUOTAPOOL:
3291         case LUSTRE_Q_SETINFOPOOL:
3292         case LUSTRE_Q_GETINFOPOOL:
3293         case LUSTRE_Q_SETDEFAULT_POOL:
3294         case LUSTRE_Q_GETDEFAULT_POOL:
3295         case LUSTRE_Q_DELETEQID:
3296                 /* forward quotactl request to QMT */
3297                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
3298                 break;
3299
3300         case Q_GETOINFO:
3301         case Q_GETOQUOTA:
3302                 /* slave quotactl */
3303                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
3304                                    oqctl);
3305                 break;
3306
3307         default:
3308                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
3309                 GOTO(out_nodemap, rc = -EFAULT);
3310         }
3311
3312         if (oqctl->qc_id != id)
3313                 swap(oqctl->qc_id, id);
3314
3315         QCTL_COPY(repoqc, oqctl);
3316         EXIT;
3317
3318 out_nodemap:
3319         nodemap_putref(nodemap);
3320
3321         return rc;
3322 }
3323
3324 /** clone llog ctxt from child (mdd)
3325  * This allows remote llog (replicator) access.
3326  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
3327  * context was originally set up, or we can handle them directly.
3328  * I choose the latter, but that means I need any llog
3329  * contexts set up by child to be accessable by the mdt.  So we clone the
3330  * context into our context list here.
3331  */
3332 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
3333                                int idx)
3334 {
3335         struct md_device  *next = mdt->mdt_child;
3336         struct llog_ctxt *ctxt;
3337         int rc;
3338
3339         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
3340                 return 0;
3341
3342         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
3343         if (rc || ctxt == NULL) {
3344                 return 0;
3345         }
3346
3347         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
3348         if (rc)
3349                 CERROR("Can't set mdt ctxt %d\n", rc);
3350
3351         return rc;
3352 }
3353
3354 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
3355                                  struct mdt_device *mdt, int idx)
3356 {
3357         struct llog_ctxt *ctxt;
3358
3359         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
3360         if (ctxt == NULL)
3361                 return 0;
3362         /* Put once for the get we just did, and once for the clone */
3363         llog_ctxt_put(ctxt);
3364         llog_ctxt_put(ctxt);
3365         return 0;
3366 }
3367
3368 /*
3369  * sec context handlers
3370  */
3371 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
3372 {
3373         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
3374
3375         return 0;
3376 }
3377
3378 /*
3379  * quota request handlers
3380  */
3381 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
3382 {
3383         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
3384         struct lu_device        *qmt = mdt->mdt_qmt_dev;
3385         int                      rc;
3386         ENTRY;
3387
3388         if (qmt == NULL)
3389                 RETURN(err_serious(-EOPNOTSUPP));
3390
3391         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
3392         RETURN(rc);
3393 }
3394
3395 struct mdt_object *mdt_object_new(const struct lu_env *env,
3396                                   struct mdt_device *d,
3397                                   const struct lu_fid *f)
3398 {
3399         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
3400         struct lu_object *o;
3401         struct mdt_object *m;
3402         ENTRY;
3403
3404         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
3405         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
3406         if (unlikely(IS_ERR(o)))
3407                 m = (struct mdt_object *)o;
3408         else
3409                 m = mdt_obj(o);
3410         RETURN(m);
3411 }
3412
3413 struct mdt_object *mdt_object_find(const struct lu_env *env,
3414                                    struct mdt_device *d,
3415                                    const struct lu_fid *f)
3416 {
3417         struct lu_object *o;
3418         struct mdt_object *m;
3419         ENTRY;
3420
3421         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
3422         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
3423         if (unlikely(IS_ERR(o)))
3424                 m = (struct mdt_object *)o;
3425         else
3426                 m = mdt_obj(o);
3427
3428         RETURN(m);
3429 }
3430
3431 /**
3432  * Asyncronous commit for mdt device.
3433  *
3434  * Pass asynchonous commit call down the MDS stack.
3435  *
3436  * \param env environment
3437  * \param mdt the mdt device
3438  */
3439 static void mdt_device_commit_async(const struct lu_env *env,
3440                                     struct mdt_device *mdt)
3441 {
3442         struct dt_device *dt = mdt->mdt_bottom;
3443         int rc;
3444         ENTRY;
3445
3446         rc = dt->dd_ops->dt_commit_async(env, dt);
3447         if (unlikely(rc != 0))
3448                 CWARN("%s: async commit start failed: rc = %d\n",
3449                       mdt_obd_name(mdt), rc);
3450         atomic_inc(&mdt->mdt_async_commit_count);
3451         EXIT;
3452 }
3453
3454 /**
3455  * Mark the lock as "synchonous".
3456  *
3457  * Mark the lock to deffer transaction commit to the unlock time.
3458  *
3459  * \param lock the lock to mark as "synchonous"
3460  *
3461  * \see mdt_is_lock_sync
3462  * \see mdt_save_lock
3463  */
3464 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
3465 {
3466         lock->l_ast_data = (void*)1;
3467 }
3468
3469 /**
3470  * Check whehter the lock "synchonous" or not.
3471  *
3472  * \param lock the lock to check
3473  * \retval 1 the lock is "synchonous"
3474  * \retval 0 the lock isn't "synchronous"
3475  *
3476  * \see mdt_set_lock_sync
3477  * \see mdt_save_lock
3478  */
3479 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
3480 {
3481         return lock->l_ast_data != NULL;
3482 }
3483
3484 /**
3485  * Blocking AST for mdt locks.
3486  *
3487  * Starts transaction commit if in case of COS lock conflict or
3488  * deffers such a commit to the mdt_save_lock.
3489  *
3490  * \param lock the lock which blocks a request or cancelling lock
3491  * \param desc unused
3492  * \param data unused
3493  * \param flag indicates whether this cancelling or blocking callback
3494  * \retval 0
3495  * \see ldlm_blocking_ast_nocheck
3496  */
3497 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
3498                      void *data, int flag)
3499 {
3500         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
3501         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
3502         struct ldlm_cb_set_arg *arg = data;
3503         bool commit_async = false;
3504         int rc;
3505         ENTRY;
3506
3507         if (flag == LDLM_CB_CANCELING)
3508                 RETURN(0);
3509
3510         lock_res_and_lock(lock);
3511         if (lock->l_blocking_ast != mdt_blocking_ast) {
3512                 unlock_res_and_lock(lock);
3513                 RETURN(0);
3514         }
3515
3516         /* A blocking ast may be sent from ldlm_lock_decref_internal
3517          * when the last reference to a local lock was released and
3518          * during blocking event from ldlm_work_bl_ast_lock().
3519          * The 'data' parameter is l_ast_data in the first case and
3520          * callback arguments in the second one. Distinguish them by that.
3521          */
3522         if (!data || data == lock->l_ast_data || !arg->bl_desc)
3523                 goto skip_cos_checks;
3524
3525         if (lock->l_req_mode & (LCK_PW | LCK_EX)) {
3526                 if (mdt_cos_is_enabled(mdt)) {
3527                         if (!arg->bl_desc->bl_same_client)
3528                                 mdt_set_lock_sync(lock);
3529                 } else if (mdt_slc_is_enabled(mdt) &&
3530                            arg->bl_desc->bl_cos_incompat) {
3531                         mdt_set_lock_sync(lock);
3532                         /*
3533                          * we may do extra commit here, but there is a small
3534                          * window to miss a commit: lock was unlocked (saved),
3535                          * then a conflict lock queued and we come here, but
3536                          * REP-ACK not received, so lock was not converted to
3537                          * COS mode yet.
3538                          * Fortunately this window is quite small, so the
3539                          * extra commit should be rare (not to say distributed
3540                          * operation is rare too).
3541                          */
3542                         commit_async = true;
3543                 }
3544         } else if (lock->l_req_mode == LCK_COS) {
3545                 commit_async = true;
3546         }
3547
3548 skip_cos_checks:
3549         rc = ldlm_blocking_ast_nocheck(lock);
3550
3551         if (commit_async) {
3552                 struct lu_env env;
3553
3554                 rc = lu_env_init(&env, LCT_LOCAL);
3555                 if (unlikely(rc != 0))
3556                         CWARN("%s: lu_env initialization failed, cannot "
3557                               "start asynchronous commit: rc = %d\n",
3558                               obd->obd_name, rc);
3559                 else
3560                         mdt_device_commit_async(&env, mdt);
3561                 lu_env_fini(&env);
3562         }
3563         RETURN(rc);
3564 }
3565
3566 /*
3567  * Blocking AST for cross-MDT lock
3568  *
3569  * Discard lock from uncommitted_slc_locks and cancel it.
3570  *
3571  * \param lock  the lock which blocks a request or cancelling lock
3572  * \param desc  unused
3573  * \param data  unused
3574  * \param flag  indicates whether this cancelling or blocking callback
3575  * \retval      0 on success
3576  * \retval      negative number on error
3577  */
3578 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
3579                             void *data, int flag)
3580 {
3581         int rc = 0;
3582         ENTRY;
3583
3584         switch (flag) {
3585         case LDLM_CB_BLOCKING: {
3586                 struct lustre_handle lockh;
3587
3588                 ldlm_lock2handle(lock, &lockh);
3589                 rc = ldlm_cli_cancel(&lockh,
3590                         ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
3591                 if (rc < 0) {
3592                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
3593                         RETURN(rc);
3594                 }
3595                 break;
3596         }
3597         case LDLM_CB_CANCELING: {
3598                 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
3599                 struct mdt_device *mdt =
3600                                 mdt_dev(obd->obd_lu_dev->ld_site->ls_top_dev);
3601
3602                 LDLM_DEBUG(lock, "Revoke remote lock\n");
3603
3604                 /* discard slc lock here so that it can be cleaned anytime,
3605                  * especially for cleanup_resource() */
3606                 tgt_discard_slc_lock(&mdt->mdt_lut, lock);
3607
3608                 /* once we cache lock, l_ast_data is set to mdt_object */
3609                 if (lock->l_ast_data != NULL) {
3610                         struct mdt_object *mo = lock->l_ast_data;
3611                         struct lu_env env;
3612
3613                         rc = lu_env_init(&env, LCT_MD_THREAD);
3614                         if (unlikely(rc != 0)) {
3615                                 CWARN("%s: lu_env initialization failed, object %p "DFID" is leaked!: rc = %d\n",
3616                                       obd->obd_name, mo,
3617                                       PFID(mdt_object_fid(mo)), rc);
3618                                 RETURN(rc);
3619                         }
3620
3621                         if (lock->l_policy_data.l_inodebits.bits &
3622                             (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)) {
3623                                 rc = mo_invalidate(&env, mdt_object_child(mo));
3624                                 mo->mot_cache_attr = 0;
3625                         }
3626                         mdt_object_put(&env, mo);
3627                         lu_env_fini(&env);
3628                 }
3629                 break;
3630         }
3631         default:
3632                 LBUG();
3633         }
3634
3635         RETURN(rc);
3636 }
3637
3638 int mdt_check_resent_lock(struct mdt_thread_info *info,
3639                           struct mdt_object *mo,
3640                           struct mdt_lock_handle *lhc)
3641 {
3642         /* the lock might already be gotten in ldlm_handle_enqueue() */
3643         if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
3644                 struct ptlrpc_request *req = mdt_info_req(info);
3645                 struct ldlm_lock      *lock;
3646
3647                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
3648                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
3649                 if (lock == NULL) {
3650                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
3651                          * a resend case, however, it could be already destroyed
3652                          * due to client eviction or a raced cancel RPC. */
3653                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx",
3654                                           lhc->mlh_reg_lh.cookie);
3655                         RETURN(-ESTALE);
3656                 }
3657
3658                 if (!fid_res_name_eq(mdt_object_fid(mo),
3659                                      &lock->l_resource->lr_name)) {
3660                         CWARN("%s: Although resent, but still not "
3661                               "get child lock:"DFID"\n",
3662                               info->mti_exp->exp_obd->obd_name,
3663                               PFID(mdt_object_fid(mo)));
3664                         LDLM_LOCK_PUT(lock);
3665                         RETURN(-EPROTO);
3666                 }
3667                 LDLM_LOCK_PUT(lock);
3668                 return 0;
3669         }
3670         return 1;
3671 }
3672
3673 static void mdt_remote_object_lock_created_cb(struct ldlm_lock *lock)
3674 {
3675         mdt_object_get(NULL, lock->l_ast_data);
3676 }
3677
3678 int mdt_remote_object_lock_try(struct mdt_thread_info *mti,
3679                                struct mdt_object *o, const struct lu_fid *fid,
3680                                struct lustre_handle *lh, enum ldlm_mode mode,
3681                                __u64 *ibits, __u64 trybits, bool cache)
3682 {
3683         struct ldlm_enqueue_info *einfo = &mti->mti_remote_einfo;
3684         union ldlm_policy_data *policy = &mti->mti_policy;
3685         struct ldlm_res_id *res_id = &mti->mti_res_id;
3686         int rc = 0;
3687         ENTRY;
3688
3689         LASSERT(mdt_object_remote(o));
3690
3691         fid_build_reg_res_name(fid, res_id);
3692
3693         memset(einfo, 0, sizeof(*einfo));
3694         einfo->ei_type = LDLM_IBITS;
3695         einfo->ei_mode = mode;
3696         einfo->ei_cb_bl = mdt_remote_blocking_ast;
3697         einfo->ei_cb_cp = ldlm_completion_ast;
3698         einfo->ei_enq_slave = 0;
3699         einfo->ei_res_id = res_id;
3700         einfo->ei_req_slot = 1;
3701
3702         if (cache) {
3703                 /*
3704                  * if we cache lock, couple lock with mdt_object, so that object
3705                  * can be easily found in lock ASTs.
3706                  */
3707                 einfo->ei_cbdata = o;
3708                 einfo->ei_cb_created = mdt_remote_object_lock_created_cb;
3709         }
3710
3711         memset(policy, 0, sizeof(*policy));
3712         policy->l_inodebits.bits = *ibits;
3713         policy->l_inodebits.try_bits = trybits;
3714
3715         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
3716                             policy);
3717
3718         /* Return successfully acquired bits to a caller */
3719         if (rc == 0) {
3720                 struct ldlm_lock *lock = ldlm_handle2lock(lh);
3721
3722                 LASSERT(lock);
3723                 *ibits = lock->l_policy_data.l_inodebits.bits;
3724                 LDLM_LOCK_PUT(lock);
3725         }
3726         RETURN(rc);
3727 }
3728
3729 int mdt_remote_object_lock(struct mdt_thread_info *mti, struct mdt_object *o,
3730                            const struct lu_fid *fid, struct lustre_handle *lh,
3731                            enum ldlm_mode mode, __u64 ibits, bool cache)
3732 {
3733         return mdt_remote_object_lock_try(mti, o, fid, lh, mode, &ibits, 0,
3734                                           cache);
3735 }
3736
3737 int mdt_object_local_lock(struct mdt_thread_info *info, struct mdt_object *o,
3738                           struct mdt_lock_handle *lh, __u64 *ibits,
3739                           __u64 trybits, bool cos_incompat)
3740 {
3741         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
3742         union ldlm_policy_data *policy = &info->mti_policy;
3743         struct ldlm_res_id *res_id = &info->mti_res_id;
3744         __u64 dlmflags = 0, *cookie = NULL;
3745         int rc;
3746         ENTRY;
3747
3748         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
3749         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
3750         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
3751         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
3752
3753         if (cos_incompat) {
3754                 LASSERT(lh->mlh_reg_mode == LCK_PW ||
3755                         lh->mlh_reg_mode == LCK_EX);
3756                 dlmflags |= LDLM_FL_COS_INCOMPAT;
3757         } else if (mdt_cos_is_enabled(info->mti_mdt)) {
3758                 dlmflags |= LDLM_FL_COS_ENABLED;
3759         }
3760
3761         /* Only enqueue LOOKUP lock for remote object */
3762         LASSERT(ergo(mdt_object_remote(o), *ibits == MDS_INODELOCK_LOOKUP));
3763
3764         /* Lease lock are granted with LDLM_FL_CANCEL_ON_BLOCK */
3765         if (lh->mlh_type == MDT_REG_LOCK && lh->mlh_reg_mode == LCK_EX &&
3766             *ibits == MDS_INODELOCK_OPEN)
3767                 dlmflags |= LDLM_FL_CANCEL_ON_BLOCK;
3768
3769         if (lh->mlh_type == MDT_PDO_LOCK) {
3770                 /* check for exists after object is locked */
3771                 if (mdt_object_exists(o) == 0) {
3772                         /* Non-existent object shouldn't have PDO lock */
3773                         RETURN(-ESTALE);
3774                 } else {
3775                         /* Non-dir object shouldn't have PDO lock */
3776                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
3777                                 RETURN(-ENOTDIR);
3778                 }
3779         }
3780
3781         fid_build_reg_res_name(mdt_object_fid(o), res_id);
3782         dlmflags |= LDLM_FL_ATOMIC_CB;
3783
3784         if (info->mti_exp)
3785                 cookie = &info->mti_exp->exp_handle.h_cookie;
3786
3787         /*
3788          * Take PDO lock on whole directory and build correct @res_id for lock
3789          * on part of directory.
3790          */
3791         if (lh->mlh_pdo_hash != 0) {
3792                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
3793                 mdt_lock_pdo_mode(info, o, lh);
3794                 if (lh->mlh_pdo_mode != LCK_NL) {
3795                         /*
3796                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
3797                          * is never going to be sent to client and we do not
3798                          * want it slowed down due to possible cancels.
3799                          */
3800                         policy->l_inodebits.bits =
3801                                 *ibits & MDS_INODELOCK_UPDATE;
3802                         policy->l_inodebits.try_bits =
3803                                 trybits & MDS_INODELOCK_UPDATE;
3804                         /* at least one of them should be set */
3805                         LASSERT(policy->l_inodebits.bits |
3806                                 policy->l_inodebits.try_bits);
3807                         rc = mdt_fid_lock(info->mti_env, ns, &lh->mlh_pdo_lh,
3808                                           lh->mlh_pdo_mode, policy, res_id,
3809                                           dlmflags, cookie);
3810                         if (unlikely(rc != 0))
3811                                 GOTO(out_unlock, rc);
3812                 }
3813
3814                 /*
3815                  * Finish res_id initializing by name hash marking part of
3816                  * directory which is taking modification.
3817                  */
3818                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
3819         }
3820
3821         policy->l_inodebits.bits = *ibits;
3822         policy->l_inodebits.try_bits = trybits;
3823         policy->l_inodebits.li_gid = lh->mlh_gid;
3824
3825         /*
3826          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
3827          * going to be sent to client. If it is - mdt_intent_policy() path will
3828          * fix it up and turn FL_LOCAL flag off.
3829          */
3830         rc = mdt_fid_lock(info->mti_env, ns, &lh->mlh_reg_lh, lh->mlh_reg_mode,
3831                           policy, res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
3832                           cookie);
3833 out_unlock:
3834         if (rc != 0)
3835                 mdt_object_unlock(info, o, lh, 1);
3836         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
3837                    lh->mlh_pdo_hash != 0 &&
3838                    (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
3839                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
3840
3841         /* Return successfully acquired bits to a caller */
3842         if (rc == 0) {
3843                 struct ldlm_lock *lock = ldlm_handle2lock(&lh->mlh_reg_lh);
3844
3845                 LASSERT(lock);
3846                 *ibits = lock->l_policy_data.l_inodebits.bits;
3847                 LDLM_LOCK_PUT(lock);
3848         }
3849         RETURN(rc);
3850 }
3851
3852 static int
3853 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
3854                          struct mdt_lock_handle *lh, __u64 *ibits,
3855                          __u64 trybits, bool cos_incompat)
3856 {
3857         struct mdt_lock_handle *local_lh = NULL;
3858         int rc;
3859         ENTRY;
3860
3861         if (!mdt_object_remote(o)) {
3862                 rc = mdt_object_local_lock(info, o, lh, ibits, trybits,
3863                                            cos_incompat);
3864                 RETURN(rc);
3865         }
3866
3867         /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
3868         *ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
3869                     MDS_INODELOCK_XATTR);
3870
3871         /* Only enqueue LOOKUP lock for remote object */
3872         if (*ibits & MDS_INODELOCK_LOOKUP) {
3873                 __u64 local = MDS_INODELOCK_LOOKUP;
3874
3875                 rc = mdt_object_local_lock(info, o, lh, &local, 0,
3876                                            cos_incompat);
3877                 if (rc != ELDLM_OK)
3878                         RETURN(rc);
3879
3880                 local_lh = lh;
3881         }
3882
3883         if ((*ibits | trybits) & MDS_INODELOCK_UPDATE) {
3884                 /* Sigh, PDO needs to enqueue 2 locks right now, but
3885                  * enqueue RPC can only request 1 lock, to avoid extra
3886                  * RPC, so it will instead enqueue EX lock for remote
3887                  * object anyway XXX*/
3888                 if (lh->mlh_type == MDT_PDO_LOCK &&
3889                     lh->mlh_pdo_hash != 0) {
3890                         CDEBUG(D_INFO,
3891                                "%s: "DFID" convert PDO lock to EX lock.\n",
3892                                mdt_obd_name(info->mti_mdt),
3893                                PFID(mdt_object_fid(o)));
3894                         lh->mlh_pdo_hash = 0;
3895                         lh->mlh_rreg_mode = LCK_EX;
3896                         lh->mlh_type = MDT_REG_LOCK;
3897                 }
3898
3899                 rc = mdt_remote_object_lock_try(info, o, mdt_object_fid(o),
3900                                                 &lh->mlh_rreg_lh,
3901                                                 lh->mlh_rreg_mode,
3902                                                 ibits, trybits, false);
3903                 if (rc != ELDLM_OK) {
3904                         if (local_lh != NULL)
3905                                 mdt_object_unlock(info, o, local_lh, rc);
3906                         RETURN(rc);
3907                 }
3908         }
3909
3910         /* other components like LFSCK can use lockless access
3911          * and populate cache, so we better invalidate it */
3912         mo_invalidate(info->mti_env, mdt_object_child(o));
3913
3914         RETURN(0);
3915 }
3916
3917 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
3918                     struct mdt_lock_handle *lh, __u64 ibits)
3919 {
3920         return mdt_object_lock_internal(info, o, lh, &ibits, 0, false);
3921 }
3922
3923 int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
3924                           struct mdt_lock_handle *lh, __u64 ibits,
3925                           bool cos_incompat)
3926 {
3927         LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
3928         return mdt_object_lock_internal(info, o, lh, &ibits, 0,
3929                                         cos_incompat);
3930 }
3931
3932 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
3933                         struct mdt_lock_handle *lh, __u64 *ibits,
3934                         __u64 trybits, bool cos_incompat)
3935 {
3936         bool trylock_only = *ibits == 0;
3937         int rc;
3938
3939         LASSERT(!(*ibits & trybits));
3940         rc = mdt_object_lock_internal(info, o, lh, ibits, trybits,
3941                                       cos_incompat);
3942         if (rc && trylock_only) { /* clear error for try ibits lock only */
3943                 LASSERT(*ibits == 0);
3944                 rc = 0;
3945         }
3946         return rc;
3947 }
3948
3949 /**
3950  * Save a lock within request object.
3951  *
3952  * Keep the lock referenced until whether client ACK or transaction
3953  * commit happens or release the lock immediately depending on input
3954  * parameters. If COS is ON, a write lock is converted to COS lock
3955  * before saving.
3956  *
3957  * \param info thead info object
3958  * \param h lock handle
3959  * \param mode lock mode
3960  * \param decref force immediate lock releasing
3961  */
3962 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
3963                    enum ldlm_mode mode, int decref)
3964 {
3965         ENTRY;
3966
3967         if (lustre_handle_is_used(h)) {
3968                 if (decref || !info->mti_has_trans ||
3969                     !(mode & (LCK_PW | LCK_EX))) {
3970                         mdt_fid_unlock(h, mode);
3971                 } else {
3972                         struct mdt_device *mdt = info->mti_mdt;
3973                         struct ldlm_lock *lock = ldlm_handle2lock(h);
3974                         struct ptlrpc_request *req = mdt_info_req(info);
3975                         bool cos = mdt_cos_is_enabled(mdt);
3976                         bool convert_lock = !cos && mdt_slc_is_enabled(mdt);
3977
3978                         LASSERTF(lock != NULL, "no lock for cookie %#llx\n",
3979                                  h->cookie);
3980
3981                         /* there is no request if mdt_object_unlock() is called
3982                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
3983                         if (likely(req != NULL)) {
3984                                 LDLM_DEBUG(lock, "save lock request %p reply "
3985                                         "state %p transno %lld\n", req,
3986                                         req->rq_reply_state, req->rq_transno);
3987                                 if (cos) {
3988                                         ldlm_lock_mode_downgrade(lock, LCK_COS);
3989                                         mode = LCK_COS;
3990                                 }
3991                                 if (req->rq_export->exp_disconnected)
3992                                         mdt_fid_unlock(h, mode);
3993                                 else
3994                                         ptlrpc_save_lock(req, h, mode, cos,
3995                                                          convert_lock);
3996                         } else {
3997                                 mdt_fid_unlock(h, mode);
3998                         }
3999                         if (mdt_is_lock_sync(lock)) {
4000                                 CDEBUG(D_HA, "found sync-lock,"
4001                                        " async commit started\n");
4002                                 mdt_device_commit_async(info->mti_env,
4003                                                         mdt);
4004                         }
4005                         LDLM_LOCK_PUT(lock);
4006                 }
4007                 h->cookie = 0ull;
4008         }
4009
4010         EXIT;
4011 }
4012
4013 /**
4014  * Save cross-MDT lock in uncommitted_slc_locks
4015  *
4016  * Keep the lock referenced until transaction commit happens or release the lock
4017  * immediately depending on input parameters.
4018  *
4019  * \param info thead info object
4020  * \param h lock handle
4021  * \param mode lock mode
4022  * \param decref force immediate lock releasing
4023  */
4024 static void mdt_save_remote_lock(struct mdt_thread_info *info,
4025                                  struct mdt_object *o, struct lustre_handle *h,
4026                                  enum ldlm_mode mode, int decref)
4027 {
4028         ENTRY;
4029
4030         if (lustre_handle_is_used(h)) {
4031                 struct ldlm_lock *lock = ldlm_handle2lock(h);
4032                 struct ptlrpc_request *req = mdt_info_req(info);
4033
4034                 if (o != NULL &&
4035                     (lock->l_policy_data.l_inodebits.bits &
4036                      (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)))
4037                         mo_invalidate(info->mti_env, mdt_object_child(o));
4038
4039                 if (decref || !info->mti_has_trans || !req ||
4040                     !(mode & (LCK_PW | LCK_EX))) {
4041                         ldlm_lock_decref_and_cancel(h, mode);
4042                         LDLM_LOCK_PUT(lock);
4043                 } else {
4044                         tgt_save_slc_lock(&info->mti_mdt->mdt_lut, lock,
4045                                           req->rq_transno);
4046                         ldlm_lock_decref(h, mode);
4047                 }
4048                 h->cookie = 0ull;
4049         }
4050
4051         EXIT;
4052 }
4053
4054 /**
4055  * Unlock mdt object.
4056  *
4057  * Immeditely release the regular lock and the PDO lock or save the
4058  * lock in request and keep them referenced until client ACK or
4059  * transaction commit.
4060  *
4061  * \param info thread info object
4062  * \param o mdt object
4063  * \param lh mdt lock handle referencing regular and PDO locks
4064  * \param decref force immediate lock releasing
4065  *
4066  * XXX o is not used and may be NULL, see hsm_cdt_request_completed().
4067  */
4068 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
4069                        struct mdt_lock_handle *lh, int decref)
4070 {
4071         ENTRY;
4072
4073         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
4074         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
4075         mdt_save_remote_lock(info, o, &lh->mlh_rreg_lh, lh->mlh_rreg_mode,
4076                              decref);
4077
4078         EXIT;
4079 }
4080
4081 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
4082                                         const struct lu_fid *f,
4083                                         struct mdt_lock_handle *lh,
4084                                         __u64 ibits)
4085 {
4086         struct mdt_object *o;
4087
4088         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
4089         if (!IS_ERR(o)) {
4090                 int rc;
4091
4092                 rc = mdt_object_lock(info, o, lh, ibits);
4093                 if (rc != 0) {
4094                         mdt_object_put(info->mti_env, o);
4095                         o = ERR_PTR(rc);
4096                 }
4097         }
4098         return o;
4099 }
4100
4101 void mdt_object_unlock_put(struct mdt_thread_info * info,
4102                            struct mdt_object * o,
4103                            struct mdt_lock_handle *lh,
4104                            int decref)
4105 {
4106         mdt_object_unlock(info, o, lh, decref);
4107         mdt_object_put(info->mti_env, o);
4108 }
4109
4110 /*
4111  * Generic code handling requests that have struct mdt_body passed in:
4112  *
4113  *  - extract mdt_body from request and save it in @info, if present;
4114  *
4115  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
4116  *  @info;
4117  *
4118  *  - if HAS_BODY flag is set for this request type check whether object
4119  *  actually exists on storage (lu_object_exists()).
4120  *
4121  */
4122 static int mdt_body_unpack(struct mdt_thread_info *info,
4123                            enum tgt_handler_flags flags)
4124 {
4125         const struct mdt_body    *body;
4126         struct mdt_object        *obj;
4127         const struct lu_env      *env;
4128         struct req_capsule       *pill;
4129         int                       rc;
4130         ENTRY;
4131
4132         env = info->mti_env;
4133         pill = info->mti_pill;
4134
4135         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
4136         if (body == NULL)
4137                 RETURN(-EFAULT);
4138
4139         if (!(body->mbo_valid & OBD_MD_FLID))
4140                 RETURN(0);
4141
4142         if (!fid_is_sane(&body->mbo_fid1)) {
4143                 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
4144                 RETURN(-EINVAL);
4145         }
4146
4147         obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
4148         if (!IS_ERR(obj)) {
4149                 if ((flags & HAS_BODY) && !mdt_object_exists(obj)) {
4150                         mdt_object_put(env, obj);
4151                         rc = -ENOENT;
4152                 } else {
4153                         info->mti_object = obj;
4154                         rc = 0;
4155                 }
4156         } else
4157                 rc = PTR_ERR(obj);
4158
4159         RETURN(rc);
4160 }
4161
4162 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info,
4163                                    enum tgt_handler_flags flags)
4164 {
4165         struct req_capsule *pill = info->mti_pill;
4166         int rc;
4167
4168         ENTRY;
4169
4170         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
4171                 rc = mdt_body_unpack(info, flags);
4172         else
4173                 rc = 0;
4174
4175         if (rc == 0 && (flags & HAS_REPLY)) {
4176                 /* Pack reply. */
4177                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
4178                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
4179                                              DEF_REP_MD_SIZE);
4180                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
4181                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
4182                                              RCL_SERVER, 0);
4183
4184                 /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
4185                  * by default. If the target object has more ACL entries, then
4186                  * enlarge the buffer when necessary. */
4187                 if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
4188                         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
4189                                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
4190
4191                 mdt_preset_secctx_size(info);
4192                 mdt_preset_encctx_size(info);
4193
4194                 rc = req_capsule_server_pack(pill);
4195                 if (rc)
4196                         CWARN("%s: cannot pack response: rc = %d\n",
4197                                       mdt_obd_name(info->mti_mdt), rc);
4198         }
4199         RETURN(rc);
4200 }
4201
4202 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
4203 {
4204         lh->mlh_type = MDT_NUL_LOCK;
4205         lh->mlh_reg_lh.cookie = 0ull;
4206         lh->mlh_reg_mode = LCK_MINMODE;
4207         lh->mlh_pdo_lh.cookie = 0ull;
4208         lh->mlh_pdo_mode = LCK_MINMODE;
4209         lh->mlh_rreg_lh.cookie = 0ull;
4210         lh->mlh_rreg_mode = LCK_MINMODE;
4211 }
4212
4213 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
4214 {
4215         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
4216         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
4217 }
4218
4219 /*
4220  * Initialize fields of struct mdt_thread_info. Other fields are left in
4221  * uninitialized state, because it's too expensive to zero out whole
4222  * mdt_thread_info (> 1K) on each request arrival.
4223  */
4224 void mdt_thread_info_init(struct ptlrpc_request *req,
4225                           struct mdt_thread_info *info)
4226 {
4227         int i;
4228
4229         info->mti_pill = &req->rq_pill;
4230
4231         /* lock handle */
4232         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
4233                 mdt_lock_handle_init(&info->mti_lh[i]);
4234
4235         /* mdt device: it can be NULL while CONNECT */
4236         if (req->rq_export) {
4237                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
4238                 info->mti_exp = req->rq_export;
4239         } else
4240                 info->mti_mdt = NULL;
4241         info->mti_env = req->rq_svc_thread->t_env;
4242         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
4243
4244         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
4245         info->mti_big_buf = LU_BUF_NULL;
4246         info->mti_body = NULL;
4247         info->mti_object = NULL;
4248         info->mti_dlm_req = NULL;
4249         info->mti_has_trans = 0;
4250         info->mti_cross_ref = 0;
4251         info->mti_opdata = 0;
4252         info->mti_big_lmm_used = 0;
4253         info->mti_big_acl_used = 0;
4254         info->mti_som_strict = 0;
4255
4256         info->mti_spec.no_create = 0;
4257         info->mti_spec.sp_rm_entry = 0;
4258         info->mti_spec.sp_permitted = 0;
4259
4260         info->mti_spec.u.sp_ea.eadata = NULL;
4261         info->mti_spec.u.sp_ea.eadatalen = 0;
4262 }
4263
4264 void mdt_thread_info_fini(struct mdt_thread_info *info)
4265 {
4266         int i;
4267
4268         if (info->mti_object != NULL) {
4269                 mdt_object_put(info->mti_env, info->mti_object);
4270                 info->mti_object = NULL;
4271         }
4272
4273         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
4274                 mdt_lock_handle_fini(&info->mti_lh[i]);
4275         info->mti_env = NULL;
4276         info->mti_pill = NULL;
4277         info->mti_exp = NULL;
4278         info->mti_mdt = NULL;
4279
4280         if (unlikely(info->mti_big_buf.lb_buf != NULL))
4281                 lu_buf_free(&info->mti_big_buf);
4282 }
4283
4284 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
4285 {
4286         struct mdt_thread_info  *mti;
4287
4288         mti = mdt_th_info(tsi->tsi_env);
4289         LASSERT(mti != NULL);
4290
4291         mdt_thread_info_init(tgt_ses_req(tsi), mti);
4292         if (tsi->tsi_corpus != NULL) {
4293                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
4294                 lu_object_get(tsi->tsi_corpus);
4295         }
4296         mti->mti_body = tsi->tsi_mdt_body;
4297         mti->mti_dlm_req = tsi->tsi_dlm_req;
4298
4299         return mti;
4300 }
4301
4302 static int mdt_tgt_connect(struct tgt_session_info *tsi)
4303 {
4304         if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
4305             cfs_fail_val ==
4306             tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id)
4307                 schedule_timeout_uninterruptible(cfs_time_seconds(3));
4308
4309         return tgt_connect(tsi);
4310 }
4311
4312 static int mdt_intent_glimpse(enum ldlm_intent_flags it_opc,
4313                               struct mdt_thread_info *info,
4314                               struct ldlm_lock **lockp, __u64 flags)
4315 {
4316         return mdt_glimpse_enqueue(info, info->mti_mdt->mdt_namespace,
4317                                    lockp, flags);
4318 }
4319 static int mdt_intent_brw(enum ldlm_intent_flags it_opc,
4320                           struct mdt_thread_info *info,
4321                           struct ldlm_lock **lockp, __u64 flags)
4322 {
4323         return mdt_brw_enqueue(info, info->mti_mdt->mdt_namespace,
4324                                lockp, flags);
4325 }
4326
4327 int mdt_intent_lock_replace(struct mdt_thread_info *info,
4328                             struct ldlm_lock **lockp,
4329                             struct mdt_lock_handle *lh,
4330                             __u64 flags, int result)
4331 {
4332         struct ptlrpc_request  *req = mdt_info_req(info);
4333         struct ldlm_lock       *lock = *lockp;
4334         struct ldlm_lock       *new_lock;
4335
4336         /* If possible resent found a lock, @lh is set to its handle */
4337         new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
4338
4339         if (new_lock == NULL) {
4340                 if (flags & LDLM_FL_INTENT_ONLY) {
4341                         result = 0;
4342                 } else if (flags & LDLM_FL_RESENT) {
4343                         /* Lock is pinned by ldlm_handle_enqueue0() as it is a
4344                          * resend case, however, it could be already destroyed
4345                          * due to client eviction or a raced cancel RPC.
4346                          */
4347                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx\n",
4348                                           lh->mlh_reg_lh.cookie);
4349                         result = -ESTALE;
4350                 } else {
4351                         CERROR("%s: Invalid lockh=%#llx flags=%#llx fid1="DFID" fid2="DFID": rc = %d\n",
4352                                mdt_obd_name(info->mti_mdt),
4353                                lh->mlh_reg_lh.cookie, flags,
4354                                PFID(&info->mti_tmp_fid1),
4355                                PFID(&info->mti_tmp_fid2), result);
4356                         result = -ESTALE;
4357                 }
4358                 lh->mlh_reg_lh.cookie = 0;
4359                 RETURN(result);
4360         }
4361
4362         /*
4363          * If we've already given this lock to a client once, then we should
4364          * have no readers or writers.  Otherwise, we should have one reader
4365          * _or_ writer ref (which will be zeroed below) before returning the
4366          * lock to a client.
4367          */
4368         if (new_lock->l_export == req->rq_export) {
4369                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
4370         } else {
4371                 LASSERT(new_lock->l_export == NULL);
4372                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
4373         }
4374
4375         *lockp = new_lock;
4376
4377         if (new_lock->l_export == req->rq_export) {
4378                 /*
4379                  * Already gave this to the client, which means that we
4380                  * reconstructed a reply.
4381                  */
4382                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
4383                         MSG_RESENT);
4384
4385                 LDLM_LOCK_RELEASE(new_lock);
4386                 lh->mlh_reg_lh.cookie = 0;
4387                 RETURN(ELDLM_LOCK_REPLACED);
4388         }
4389
4390         /*
4391          * Fixup the lock to be given to the client.
4392          */
4393         lock_res_and_lock(new_lock);
4394         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
4395          * possible blocking AST. */
4396         while (new_lock->l_readers > 0) {
4397                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
4398                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
4399                 new_lock->l_readers--;
4400         }
4401         while (new_lock->l_writers > 0) {
4402                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
4403                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
4404                 new_lock->l_writers--;
4405         }
4406
4407         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
4408         new_lock->l_blocking_ast = lock->l_blocking_ast;
4409         new_lock->l_completion_ast = lock->l_completion_ast;
4410         if (ldlm_has_dom(new_lock))
4411                 new_lock->l_glimpse_ast = ldlm_server_glimpse_ast;
4412         new_lock->l_remote_handle = lock->l_remote_handle;
4413         new_lock->l_flags &= ~LDLM_FL_LOCAL;
4414
4415         unlock_res_and_lock(new_lock);
4416
4417         cfs_hash_add(new_lock->l_export->exp_lock_hash,
4418                      &new_lock->l_remote_handle,
4419                      &new_lock->l_exp_hash);
4420
4421         LDLM_LOCK_RELEASE(new_lock);
4422         lh->mlh_reg_lh.cookie = 0;
4423
4424         RETURN(ELDLM_LOCK_REPLACED);
4425 }
4426
4427 void mdt_intent_fixup_resent(struct mdt_thread_info *info,
4428                              struct ldlm_lock *new_lock,
4429                              struct mdt_lock_handle *lh, __u64 flags)
4430 {
4431         struct ptlrpc_request  *req = mdt_info_req(info);
4432         struct ldlm_request    *dlmreq;
4433
4434         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
4435                 return;
4436
4437         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
4438
4439         /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
4440          * lock was found by ldlm_handle_enqueue(); if so @lh must be
4441          * initialized. */
4442         if (flags & LDLM_FL_RESENT) {
4443                 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
4444                 lh->mlh_reg_mode = new_lock->l_granted_mode;
4445
4446                 LDLM_DEBUG(new_lock, "Restoring lock cookie");
4447                 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie %#llx",
4448                           lh->mlh_reg_lh.cookie);
4449                 return;
4450         }
4451
4452         /*
4453          * If the xid matches, then we know this is a resent request, and allow
4454          * it. (It's probably an OPEN, for which we don't send a lock.
4455          */
4456         if (req_can_reconstruct(req, NULL) != 0)
4457                 return;
4458
4459         /*
4460          * This remote handle isn't enqueued, so we never received or processed
4461          * this request.  Clear MSG_RESENT, because it can be handled like any
4462          * normal request now.
4463          */
4464         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
4465
4466         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle %#llx",
4467                   dlmreq->lock_handle[0].cookie);
4468 }
4469
4470 static int mdt_intent_getxattr(enum ldlm_intent_flags it_opc,
4471                                struct mdt_thread_info *info,
4472                                struct ldlm_lock **lockp,
4473                                __u64 flags)
4474 {
4475         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4476         struct ldlm_reply      *ldlm_rep = NULL;
4477         int rc;
4478         ENTRY;
4479
4480         /*
4481          * Initialize lhc->mlh_reg_lh either from a previously granted lock
4482          * (for the resend case) or a new lock. Below we will use it to
4483          * replace the original lock.
4484          */
4485         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4486         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
4487                 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
4488                 rc = mdt_object_lock(info, info->mti_object, lhc,
4489                                      MDS_INODELOCK_XATTR);
4490                 if (rc)
4491                         return rc;
4492         }
4493
4494         rc = mdt_getxattr(info);
4495
4496         if (mdt_info_req(info)->rq_repmsg != NULL)
4497                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4498
4499         if (ldlm_rep == NULL ||
4500             OBD_FAIL_CHECK(OBD_FAIL_MDS_XATTR_REP)) {
4501                 mdt_object_unlock(info,  info->mti_object, lhc, 1);
4502                 if (is_serious(rc))
4503                         RETURN(rc);
4504                 else
4505                         RETURN(err_serious(-EFAULT));
4506         }
4507
4508         ldlm_rep->lock_policy_res2 = clear_serious(rc);
4509
4510         /* This is left for interop instead of adding a new interop flag.
4511          * LU-7433 */
4512 #if LUSTRE_VERSION_CODE > OBD_OCD_VERSION(3, 0, 0, 0)
4513         if (ldlm_rep->lock_policy_res2) {
4514                 mdt_object_unlock(info, info->mti_object, lhc, 1);
4515                 RETURN(ELDLM_LOCK_ABORTED);
4516         }
4517 #endif
4518
4519         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4520         RETURN(rc);
4521 }
4522
4523 static int mdt_intent_getattr(enum ldlm_intent_flags it_opc,
4524                               struct mdt_thread_info *info,
4525                               struct ldlm_lock **lockp,
4526                               __u64 flags)
4527 {
4528         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4529         __u64                   child_bits;
4530         struct ldlm_reply      *ldlm_rep;
4531         struct mdt_body        *reqbody;
4532         struct mdt_body        *repbody;
4533         int                     rc, rc2;
4534         ENTRY;
4535
4536         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
4537         LASSERT(reqbody);
4538
4539         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
4540         LASSERT(repbody);
4541
4542         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
4543         repbody->mbo_eadatasize = 0;
4544         repbody->mbo_aclsize = 0;
4545
4546         switch (it_opc) {
4547         case IT_LOOKUP:
4548                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
4549                 break;
4550         case IT_GETATTR:
4551                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
4552                              MDS_INODELOCK_PERM;
4553                 break;
4554         default:
4555                 CERROR("%s: unsupported intent %#x\n",
4556                        mdt_obd_name(info->mti_mdt), (unsigned int)it_opc);
4557                 GOTO(out_shrink, rc = -EINVAL);
4558         }
4559
4560         rc = mdt_init_ucred(info, reqbody);
4561         if (rc)
4562                 GOTO(out_shrink, rc);
4563
4564         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4565         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
4566
4567         /* Get lock from request for possible resent case. */
4568         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4569
4570         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
4571         ldlm_rep->lock_policy_res2 = clear_serious(rc);
4572
4573         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
4574                 ldlm_rep->lock_policy_res2 = 0;
4575         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
4576             ldlm_rep->lock_policy_res2) {
4577                 lhc->mlh_reg_lh.cookie = 0ull;
4578                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
4579         }
4580
4581         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4582         EXIT;
4583 out_ucred:
4584         mdt_exit_ucred(info);
4585 out_shrink:
4586         mdt_client_compatibility(info);
4587         rc2 = mdt_fix_reply(info);
4588         if (rc == 0)
4589                 rc = rc2;
4590         return rc;
4591 }
4592
4593 static int mdt_intent_layout(enum ldlm_intent_flags it_opc,
4594                              struct mdt_thread_info *info,
4595                              struct ldlm_lock **lockp,
4596                              __u64 flags)
4597 {
4598         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4599         struct md_layout_change layout = { .mlc_opc = MD_LAYOUT_NOP };
4600         struct layout_intent *intent;
4601         struct ldlm_reply *ldlm_rep;
4602         struct lu_fid *fid = &info->mti_tmp_fid2;
4603         struct mdt_object *obj = NULL;
4604         int layout_size = 0;
4605         struct lu_buf *buf = &layout.mlc_buf;
4606         int rc = 0;
4607
4608         ENTRY;
4609
4610         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
4611
4612         intent = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
4613         if (intent == NULL)
4614                 RETURN(-EPROTO);
4615
4616         CDEBUG(D_INFO, DFID "got layout change request from client: "
4617                "opc:%u flags:%#x extent "DEXT"\n",
4618                PFID(fid), intent->li_opc, intent->li_flags,
4619                PEXT(&intent->li_extent));
4620
4621         switch (intent->li_opc) {
4622         case LAYOUT_INTENT_TRUNC:
4623         case LAYOUT_INTENT_WRITE:
4624                 layout.mlc_opc = MD_LAYOUT_WRITE;
4625                 layout.mlc_intent = intent;
4626                 break;
4627         case LAYOUT_INTENT_ACCESS:
4628                 break;
4629         case LAYOUT_INTENT_READ:
4630         case LAYOUT_INTENT_GLIMPSE:
4631         case LAYOUT_INTENT_RELEASE:
4632         case LAYOUT_INTENT_RESTORE:
4633                 CERROR("%s: Unsupported layout intent opc %d\n",
4634                        mdt_obd_name(info->mti_mdt), intent->li_opc);
4635                 RETURN(-ENOTSUPP);
4636         default:
4637                 CERROR("%s: Unknown layout intent opc %d\n",
4638                        mdt_obd_name(info->mti_mdt), intent->li_opc);
4639                 RETURN(-EINVAL);
4640         }
4641
4642         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
4643         if (IS_ERR(obj))
4644                 RETURN(PTR_ERR(obj));
4645
4646         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
4647                 /* if layout is going to be changed don't use the current EA
4648                  * size but the maximum one. That buffer will be shrinked
4649                  * to the actual size in req_capsule_shrink() before reply.
4650                  */
4651                 if (layout.mlc_opc == MD_LAYOUT_WRITE) {
4652                         layout_size = info->mti_mdt->mdt_max_mdsize;
4653                 } else {
4654                         layout_size = mdt_attr_get_eabuf_size(info, obj);
4655                         if (layout_size < 0)
4656                                 GOTO(out, rc = layout_size);
4657
4658                         if (layout_size > info->mti_mdt->mdt_max_mdsize)
4659                                 info->mti_mdt->mdt_max_mdsize = layout_size;
4660                 }
4661                 CDEBUG(D_INFO, "%s: layout_size %d\n",
4662                        mdt_obd_name(info->mti_mdt), layout_size);
4663         }
4664
4665         /*
4666          * set reply buffer size, so that ldlm_handle_enqueue0()->
4667          * ldlm_lvbo_fill() will fill the reply buffer with lovea.
4668          */
4669         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
4670                              layout_size);
4671         rc = req_capsule_server_pack(info->mti_pill);
4672         if (rc)
4673                 GOTO(out, rc);
4674
4675         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4676         if (!ldlm_rep)
4677                 GOTO(out, rc = -EPROTO);
4678
4679         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
4680
4681         /* take lock in ldlm_lock_enqueue() for LAYOUT_INTENT_ACCESS */
4682         if (layout.mlc_opc == MD_LAYOUT_NOP)
4683                 GOTO(out, rc = 0);
4684
4685         rc = mdt_check_resent(info, mdt_reconstruct_generic, lhc);
4686         if (rc < 0)
4687                 GOTO(out, rc);
4688         if (rc == 1) {
4689                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
4690                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
4691                 GOTO(out, rc);
4692         }
4693
4694         buf->lb_buf = NULL;
4695         buf->lb_len = 0;
4696         if (unlikely(req_is_replay(mdt_info_req(info)))) {
4697                 buf->lb_buf = req_capsule_client_get(info->mti_pill,
4698                                                      &RMF_EADATA);
4699                 buf->lb_len = req_capsule_get_size(info->mti_pill,
4700                                                      &RMF_EADATA, RCL_CLIENT);
4701                 /*
4702                  * If it's a replay of layout write intent RPC, the client has
4703                  * saved the extended lovea when it get reply then.
4704                  */
4705                 if (buf->lb_len > 0)
4706                         mdt_fix_lov_magic(info, buf->lb_buf);
4707         }
4708
4709         /* Get lock from request for possible resent case. */
4710         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4711         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
4712
4713         /*
4714          * Instantiate some layout components, if @buf contains lovea, then it's
4715          * a replay of the layout intent write RPC.
4716          */
4717         rc = mdt_layout_change(info, obj, lhc, &layout);
4718         ldlm_rep->lock_policy_res2 = clear_serious(rc);
4719
4720         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
4721                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4722                 if (rc == ELDLM_LOCK_REPLACED &&
4723                     (*lockp)->l_granted_mode == LCK_EX)
4724                         ldlm_lock_mode_downgrade(*lockp, LCK_CR);
4725         }
4726
4727         EXIT;
4728 out:
4729         mdt_object_put(info->mti_env, obj);
4730         return rc;
4731 }
4732
4733 static int mdt_intent_open(enum ldlm_intent_flags it_opc,
4734                            struct mdt_thread_info *info,
4735                            struct ldlm_lock **lockp,
4736                            __u64 flags)
4737 {
4738         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4739         struct ldlm_reply      *rep = NULL;
4740         long                    opc;
4741         int                     rc;
4742         struct ptlrpc_request  *req = mdt_info_req(info);
4743
4744         static const struct req_format *intent_fmts[REINT_MAX] = {
4745                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
4746                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
4747         };
4748
4749         ENTRY;
4750
4751         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
4752         if (opc < 0)
4753                 RETURN(opc);
4754
4755         /* Get lock from request for possible resent case. */
4756         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4757
4758         rc = mdt_reint_internal(info, lhc, opc);
4759
4760         if (rc < 0 && lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
4761                 DEBUG_REQ(D_ERROR, req, "Replay open failed with %d", rc);
4762
4763         /* Check whether the reply has been packed successfully. */
4764         if (mdt_info_req(info)->rq_repmsg != NULL)
4765                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4766         if (rep == NULL) {
4767                 if (is_serious(rc))
4768                         RETURN(rc);
4769                 else
4770                         RETURN(err_serious(-EFAULT));
4771         }
4772
4773         /* MDC expects this in any case */
4774         if (rc != 0)
4775                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
4776
4777         /* the open lock or the lock for cross-ref object should be
4778          * returned to the client */
4779         if (lustre_handle_is_used(&lhc->mlh_reg_lh) &&
4780             (rc == 0 || rc == -MDT_EREMOTE_OPEN)) {
4781                 rep->lock_policy_res2 = 0;
4782                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4783                 RETURN(rc);
4784         }
4785
4786         rep->lock_policy_res2 = clear_serious(rc);
4787
4788         if (rep->lock_policy_res2 == -ENOENT &&
4789             mdt_get_disposition(rep, DISP_LOOKUP_NEG) &&
4790             !mdt_get_disposition(rep, DISP_OPEN_CREATE))
4791                 rep->lock_policy_res2 = 0;
4792
4793         lhc->mlh_reg_lh.cookie = 0ull;
4794         if (rc == -ENOTCONN || rc == -ENODEV ||
4795             rc == -EOVERFLOW) { /**< if VBR failure then return error */
4796                 /*
4797                  * If it is the disconnect error (ENODEV & ENOCONN), the error
4798                  * will be returned by rq_status, and client at ptlrpc layer
4799                  * will detect this, then disconnect, reconnect the import
4800                  * immediately, instead of impacting the following the rpc.
4801                  */
4802                 RETURN(rc);
4803         }
4804         /*
4805          * For other cases, the error will be returned by intent, and client
4806          * will retrieve the result from intent.
4807          */
4808         RETURN(ELDLM_LOCK_ABORTED);
4809 }
4810
4811 static int mdt_intent_opc(enum ldlm_intent_flags it_opc,
4812                           struct mdt_thread_info *info,
4813                           struct ldlm_lock **lockp,
4814                           u64 flags /* LDLM_FL_* */)
4815 {
4816         struct req_capsule *pill = info->mti_pill;
4817         struct ptlrpc_request *req = mdt_info_req(info);
4818         const struct req_format *it_format;
4819         int (*it_handler)(enum ldlm_intent_flags,
4820                           struct mdt_thread_info *,
4821                           struct ldlm_lock **,
4822                           u64);
4823         enum tgt_handler_flags it_handler_flags = 0;
4824         struct ldlm_reply *rep;
4825         bool check_mdt_object = false;
4826         int rc;
4827         ENTRY;
4828
4829         switch (it_opc) {
4830         case IT_OPEN:
4831         case IT_OPEN|IT_CREAT:
4832                 /*
4833                  * OCREAT is not a IS_MUTABLE request since the file may
4834                  * already exist. We do the extra check of
4835                  * OBD_CONNECT_RDONLY in mdt_reint_open() when we
4836                  * really need to create the object.
4837                  */
4838                 it_format = &RQF_LDLM_INTENT;
4839                 it_handler = &mdt_intent_open;
4840                 break;
4841         case IT_GETATTR:
4842                 check_mdt_object = true;
4843                 fallthrough;
4844         case IT_LOOKUP:
4845                 it_format = &RQF_LDLM_INTENT_GETATTR;
4846                 it_handler = &mdt_intent_getattr;
4847                 it_handler_flags = HAS_REPLY;
4848                 break;
4849         case IT_GETXATTR:
4850                 check_mdt_object = true;
4851                 it_format = &RQF_LDLM_INTENT_GETXATTR;
4852                 it_handler = &mdt_intent_getxattr;
4853                 it_handler_flags = HAS_BODY;
4854                 break;
4855         case IT_LAYOUT:
4856                 it_format = &RQF_LDLM_INTENT_LAYOUT;
4857                 it_handler = &mdt_intent_layout;
4858                 break;
4859         case IT_GLIMPSE:
4860                 it_format = &RQF_LDLM_INTENT;
4861                 it_handler = &mdt_intent_glimpse;
4862                 break;
4863         case IT_BRW:
4864                 it_format = &RQF_LDLM_INTENT;
4865                 it_handler = &mdt_intent_brw;
4866                 break;
4867         case IT_QUOTA_DQACQ:
4868         case IT_QUOTA_CONN: {
4869                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
4870
4871                 if (qmt == NULL)
4872                         RETURN(-EOPNOTSUPP);
4873
4874                 if (mdt_rdonly(req->rq_export))
4875                         RETURN(-EROFS);
4876
4877                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
4878                 /* pass the request to quota master */
4879                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
4880                                                  mdt_info_req(info), lockp,
4881                                                  flags);
4882                 RETURN(rc);
4883         }
4884         default:
4885                 CERROR("%s: unknown intent code %#x\n",
4886                        mdt_obd_name(info->mti_mdt), it_opc);
4887                 RETURN(-EPROTO);
4888         }
4889
4890         req_capsule_extend(pill, it_format);
4891
4892         rc = mdt_unpack_req_pack_rep(info, it_handler_flags);
4893         if (rc < 0)
4894                 RETURN(rc);
4895
4896         if (unlikely(info->mti_object == NULL && check_mdt_object))
4897                 RETURN(-EPROTO);
4898
4899         if (it_handler_flags & IS_MUTABLE && mdt_rdonly(req->rq_export))
4900                 RETURN(-EROFS);
4901
4902         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_INTENT_DELAY, 10);
4903
4904         /* execute policy */
4905         rc = (*it_handler)(it_opc, info, lockp, flags);
4906
4907         /* Check whether the reply has been packed successfully. */
4908         if (req->rq_repmsg != NULL) {
4909                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4910                 rep->lock_policy_res2 =
4911                         ptlrpc_status_hton(rep->lock_policy_res2);
4912         }
4913
4914         RETURN(rc);
4915 }
4916
4917 static void mdt_ptlrpc_stats_update(struct ptlrpc_request *req,
4918                                     enum ldlm_intent_flags it_opc)
4919 {
4920         struct lprocfs_stats *srv_stats = ptlrpc_req2svc(req)->srv_stats;
4921
4922         /* update stats when IT code is known */
4923         if (srv_stats != NULL)
4924                 lprocfs_counter_incr(srv_stats,
4925                                 PTLRPC_LAST_CNTR + (it_opc == IT_GLIMPSE ?
4926                                 LDLM_GLIMPSE_ENQUEUE : LDLM_IBITS_ENQUEUE));
4927 }
4928
4929 static int mdt_intent_policy(const struct lu_env *env,
4930                              struct ldlm_namespace *ns,
4931                              struct ldlm_lock **lockp,
4932                              void *req_cookie,
4933                              enum ldlm_mode mode,
4934                              __u64 flags, void *data)
4935 {
4936         struct tgt_session_info *tsi;
4937         struct mdt_thread_info  *info;
4938         struct ptlrpc_request   *req  =  req_cookie;
4939         struct ldlm_intent      *it;
4940         struct req_capsule      *pill;
4941         const struct ldlm_lock_desc *ldesc;
4942         int rc;
4943
4944         ENTRY;
4945
4946         LASSERT(req != NULL);
4947
4948         tsi = tgt_ses_info(env);
4949
4950         info = tsi2mdt_info(tsi);
4951         LASSERT(info != NULL);
4952         pill = info->mti_pill;
4953         LASSERT(pill->rc_req == req);
4954         ldesc = &info->mti_dlm_req->lock_desc;
4955
4956         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
4957                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
4958                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
4959                 if (it != NULL) {
4960                         mdt_ptlrpc_stats_update(req, it->opc);
4961                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
4962                         if (rc == 0)
4963                                 rc = ELDLM_OK;
4964
4965                         /* Lock without inodebits makes no sense and will oops
4966                          * later in ldlm. Let's check it now to see if we have
4967                          * ibits corrupted somewhere in mdt_intent_opc().
4968                          * The case for client miss to set ibits has been
4969                          * processed by others. */
4970                         LASSERT(ergo(ldesc->l_resource.lr_type == LDLM_IBITS,
4971                                 ldesc->l_policy_data.l_inodebits.bits != 0));
4972                 } else {
4973                         rc = err_serious(-EFAULT);
4974                 }
4975         } else if (ldesc->l_resource.lr_type == LDLM_IBITS &&
4976                    ldesc->l_policy_data.l_inodebits.bits == MDS_INODELOCK_DOM) {
4977                 struct ldlm_reply *rep;
4978
4979                 /* No intent was provided but INTENT flag is set along with
4980                  * DOM bit, this is considered as GLIMPSE request.
4981                  * This logic is common for MDT and OST glimpse
4982                  */
4983                 mdt_ptlrpc_stats_update(req, IT_GLIMPSE);
4984                 rc = mdt_glimpse_enqueue(info, ns, lockp, flags);
4985                 /* Check whether the reply has been packed successfully. */
4986                 if (req->rq_repmsg != NULL) {
4987                         rep = req_capsule_server_get(info->mti_pill,
4988                                                      &RMF_DLM_REP);
4989                         rep->lock_policy_res2 =
4990                                 ptlrpc_status_hton(rep->lock_policy_res2);
4991                 }
4992         } else {
4993                 /* No intent was provided */
4994                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
4995                 rc = req_capsule_server_pack(pill);
4996                 if (rc)
4997                         rc = err_serious(rc);
4998         }
4999         mdt_thread_info_fini(info);
5000         RETURN(rc);
5001 }
5002
5003 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
5004 {
5005         struct seq_server_site  *ss = mdt_seq_site(mdt);
5006
5007         if (ss->ss_node_id == 0)
5008                 return;
5009
5010         if (ss->ss_client_seq != NULL) {
5011                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
5012                 ss->ss_client_seq->lcs_exp = NULL;
5013         }
5014
5015         if (ss->ss_server_fld != NULL) {
5016                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
5017                 ss->ss_server_fld->lsf_control_exp = NULL;
5018         }
5019 }
5020
5021 static void mdt_seq_fini_cli(struct mdt_device *mdt)
5022 {
5023         struct seq_server_site *ss = mdt_seq_site(mdt);
5024
5025         if (ss == NULL)
5026                 return;
5027
5028         if (ss->ss_server_seq != NULL)
5029                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
5030 }
5031
5032 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
5033 {
5034         mdt_seq_fini_cli(mdt);
5035         mdt_deregister_seq_exp(mdt);
5036
5037         return seq_site_fini(env, mdt_seq_site(mdt));
5038 }
5039
5040 /**
5041  * It will retrieve its FLDB entries from MDT0, and it only happens
5042  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
5043  * and it needs to refresh FLDB from the MDT0.
5044  **/
5045 static int mdt_register_lwp_callback(void *data)
5046 {
5047         struct lu_env           env;
5048         struct mdt_device       *mdt = data;
5049         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
5050         int                     rc;
5051         ENTRY;
5052
5053         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
5054
5055         rc = lu_env_init(&env, LCT_MD_THREAD);
5056         if (rc < 0) {
5057                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
5058                 RETURN(rc);
5059         }
5060
5061         /* Allocate new sequence now to avoid creating local transaction
5062          * in the normal transaction process */
5063         rc = seq_server_check_and_alloc_super(&env,
5064                                               mdt_seq_site(mdt)->ss_server_seq);
5065         if (rc < 0)
5066                 GOTO(out, rc);
5067
5068         if (fld->lsf_new) {
5069                 rc = fld_update_from_controller(&env, fld);
5070                 if (rc != 0) {
5071                         CERROR("%s: cannot update controller: rc = %d\n",
5072                                mdt_obd_name(mdt), rc);
5073                         GOTO(out, rc);
5074                 }
5075         }
5076 out:
5077         lu_env_fini(&env);
5078         RETURN(rc);
5079 }
5080
5081 static int mdt_register_seq_exp(struct mdt_device *mdt)
5082 {
5083         struct seq_server_site  *ss = mdt_seq_site(mdt);
5084         char                    *lwp_name = NULL;
5085         int                     rc;
5086
5087         if (ss->ss_node_id == 0)
5088                 return 0;
5089
5090         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
5091         if (lwp_name == NULL)
5092                 GOTO(out_free, rc = -ENOMEM);
5093
5094         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
5095         if (rc != 0)
5096                 GOTO(out_free, rc);
5097
5098         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
5099                                       NULL, NULL);
5100         if (rc != 0)
5101                 GOTO(out_free, rc);
5102
5103         rc = lustre_register_lwp_item(lwp_name,
5104                                       &ss->ss_server_fld->lsf_control_exp,
5105                                       mdt_register_lwp_callback, mdt);
5106         if (rc != 0) {
5107                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
5108                 ss->ss_client_seq->lcs_exp = NULL;
5109                 GOTO(out_free, rc);
5110         }
5111 out_free:
5112         if (lwp_name != NULL)
5113                 OBD_FREE(lwp_name, MAX_OBD_NAME);
5114
5115         return rc;
5116 }
5117
5118 /*
5119  * Init client sequence manager which is used by local MDS to talk to sequence
5120  * controller on remote node.
5121  */
5122 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
5123 {
5124         struct seq_server_site *ss = mdt_seq_site(mdt);
5125         char *prefix;
5126         ENTRY;
5127
5128         /* check if this is adding the first MDC and controller is not yet
5129          * initialized. */
5130         OBD_ALLOC_PTR(ss->ss_client_seq);
5131         if (ss->ss_client_seq == NULL)
5132                 RETURN(-ENOMEM);
5133
5134         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
5135         if (prefix == NULL) {
5136                 OBD_FREE_PTR(ss->ss_client_seq);
5137                 ss->ss_client_seq = NULL;
5138                 RETURN(-ENOMEM);
5139         }
5140
5141         /* Note: seq_client_fini will be called in seq_site_fini */
5142         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
5143         seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
5144                         prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
5145                                                             NULL);
5146         OBD_FREE(prefix, MAX_OBD_NAME + 5);
5147
5148         RETURN(seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq));
5149 }
5150
5151 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
5152 {
5153         struct seq_server_site  *ss;
5154         int                     rc;
5155         ENTRY;
5156
5157         ss = mdt_seq_site(mdt);
5158         /* init sequence controller server(MDT0) */
5159         if (ss->ss_node_id == 0) {
5160                 OBD_ALLOC_PTR(ss->ss_control_seq);
5161                 if (ss->ss_control_seq == NULL)
5162                         RETURN(-ENOMEM);
5163
5164                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
5165                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
5166                                      ss);
5167                 if (rc)
5168                         GOTO(out_seq_fini, rc);
5169         }
5170
5171         /* Init normal sequence server */
5172         OBD_ALLOC_PTR(ss->ss_server_seq);
5173         if (ss->ss_server_seq == NULL)
5174                 GOTO(out_seq_fini, rc = -ENOMEM);
5175
5176         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
5177                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
5178         if (rc)
5179                 GOTO(out_seq_fini, rc);
5180
5181         /* init seq client for seq server to talk to seq controller(MDT0) */
5182         rc = mdt_seq_init_cli(env, mdt);
5183         if (rc != 0)
5184                 GOTO(out_seq_fini, rc);
5185
5186         if (ss->ss_node_id != 0)
5187                 /* register controller export through lwp */
5188                 rc = mdt_register_seq_exp(mdt);
5189
5190         EXIT;
5191 out_seq_fini:
5192         if (rc)
5193                 mdt_seq_fini(env, mdt);
5194
5195         return rc;
5196 }
5197
5198 /*
5199  * FLD wrappers
5200  */
5201 static int mdt_fld_fini(const struct lu_env *env,
5202                         struct mdt_device *m)
5203 {
5204         struct seq_server_site *ss = mdt_seq_site(m);
5205         ENTRY;
5206
5207         if (ss && ss->ss_server_fld) {
5208                 fld_server_fini(env, ss->ss_server_fld);
5209                 OBD_FREE_PTR(ss->ss_server_fld);
5210                 ss->ss_server_fld = NULL;
5211         }
5212
5213         RETURN(0);
5214 }
5215
5216 static int mdt_fld_init(const struct lu_env *env,
5217                         const char *uuid,
5218                         struct mdt_device *m)
5219 {
5220         struct seq_server_site *ss;
5221         int rc;
5222         ENTRY;
5223
5224         ss = mdt_seq_site(m);
5225
5226         OBD_ALLOC_PTR(ss->ss_server_fld);
5227         if (ss->ss_server_fld == NULL)
5228                 RETURN(rc = -ENOMEM);
5229
5230         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
5231                              LU_SEQ_RANGE_MDT);
5232         if (rc) {
5233                 OBD_FREE_PTR(ss->ss_server_fld);
5234                 ss->ss_server_fld = NULL;
5235                 RETURN(rc);
5236         }
5237
5238         RETURN(0);
5239 }
5240
5241 static void mdt_stack_pre_fini(const struct lu_env *env,
5242                            struct mdt_device *m, struct lu_device *top)
5243 {
5244         struct lustre_cfg_bufs  *bufs;
5245         struct lustre_cfg       *lcfg;
5246         struct mdt_thread_info  *info;
5247         ENTRY;
5248
5249         LASSERT(top);
5250
5251         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5252         LASSERT(info != NULL);
5253
5254         bufs = &info->mti_u.bufs;
5255
5256         LASSERT(m->mdt_child_exp);
5257         LASSERT(m->mdt_child_exp->exp_obd);
5258
5259         /* process cleanup, pass mdt obd name to get obd umount flags */
5260         /* XXX: this is needed because all layers are referenced by
5261          * objects (some of them are pinned by osd, for example *
5262          * the proper solution should be a model where object used
5263          * by osd only doesn't have mdt/mdd slices -bzzz */
5264         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
5265         lustre_cfg_bufs_set_string(bufs, 1, NULL);
5266         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5267         if (!lcfg)
5268                 RETURN_EXIT;
5269         lustre_cfg_init(lcfg, LCFG_PRE_CLEANUP, bufs);
5270
5271         top->ld_ops->ldo_process_config(env, top, lcfg);
5272         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5273         EXIT;
5274 }
5275
5276 static void mdt_stack_fini(const struct lu_env *env,
5277                            struct mdt_device *m, struct lu_device *top)
5278 {
5279         struct obd_device       *obd = mdt2obd_dev(m);
5280         struct lustre_cfg_bufs  *bufs;
5281         struct lustre_cfg       *lcfg;
5282         struct mdt_thread_info  *info;
5283         char                     flags[3] = "";
5284         ENTRY;
5285
5286         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5287         LASSERT(info != NULL);
5288
5289         lu_dev_del_linkage(top->ld_site, top);
5290
5291         lu_site_purge(env, top->ld_site, -1);
5292
5293         bufs = &info->mti_u.bufs;
5294         /* process cleanup, pass mdt obd name to get obd umount flags */
5295         /* another purpose is to let all layers to release their objects */
5296         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
5297         if (obd->obd_force)
5298                 strcat(flags, "F");
5299         if (obd->obd_fail)
5300                 strcat(flags, "A");
5301         lustre_cfg_bufs_set_string(bufs, 1, flags);
5302         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5303         if (!lcfg)
5304                 RETURN_EXIT;
5305         lustre_cfg_init(lcfg, LCFG_CLEANUP, bufs);
5306
5307         LASSERT(top);
5308         top->ld_ops->ldo_process_config(env, top, lcfg);
5309         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5310
5311         lu_site_purge(env, top->ld_site, -1);
5312
5313         m->mdt_child = NULL;
5314         m->mdt_bottom = NULL;
5315
5316         obd_disconnect(m->mdt_child_exp);
5317         m->mdt_child_exp = NULL;
5318
5319         obd_disconnect(m->mdt_bottom_exp);
5320         m->mdt_child_exp = NULL;
5321 }
5322
5323 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
5324                                const char *next, struct obd_export **exp)
5325 {
5326         struct obd_connect_data *data = NULL;
5327         struct obd_device       *obd;
5328         int                      rc;
5329         ENTRY;
5330
5331         OBD_ALLOC_PTR(data);
5332         if (data == NULL)
5333                 GOTO(out, rc = -ENOMEM);
5334
5335         obd = class_name2obd(next);
5336         if (obd == NULL) {
5337                 CERROR("%s: can't locate next device: %s\n",
5338                        mdt_obd_name(m), next);
5339                 GOTO(out, rc = -ENOTCONN);
5340         }
5341
5342         data->ocd_connect_flags = OBD_CONNECT_VERSION;
5343         data->ocd_version = LUSTRE_VERSION_CODE;
5344
5345         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
5346         if (rc) {
5347                 CERROR("%s: cannot connect to next dev %s (%d)\n",
5348                        mdt_obd_name(m), next, rc);
5349                 GOTO(out, rc);
5350         }
5351
5352 out:
5353         if (data)
5354                 OBD_FREE_PTR(data);
5355         RETURN(rc);
5356 }
5357
5358 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
5359                           struct lustre_cfg *cfg)
5360 {
5361         char                   *dev = lustre_cfg_string(cfg, 0);
5362         int                     rc, name_size, uuid_size;
5363         char                   *name, *uuid, *p;
5364         struct lustre_cfg_bufs *bufs;
5365         struct lustre_cfg      *lcfg;
5366         struct obd_device      *obd;
5367         struct lustre_profile  *lprof;
5368         struct lu_site         *site;
5369         ENTRY;
5370
5371         /* in 1.8 we had the only device in the stack - MDS.
5372          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
5373          * in 2.3 OSD is instantiated by obd_mount.c, so we need
5374          * to generate names and setup MDT, MDD. MDT will be using
5375          * generated name to connect to MDD. for MDD the next device
5376          * will be LOD with name taken from so called "profile" which
5377          * is generated by mount_option line
5378          *
5379          * 1.8 MGS generates config. commands like this:
5380          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
5381          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
5382          * 2.0 MGS generates config. commands like this:
5383          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
5384          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
5385          *                    3:lustre-MDT0000-mdtlov  4:f
5386          *
5387          * we generate MDD name from MDT one, just replacing T with D
5388          *
5389          * after all the preparations, the logical equivalent will be
5390          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
5391          *                    3:lustre-MDT0000-mdtlov  4:f
5392          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
5393          *                    3:lustre-MDD0000  4:f
5394          *
5395          *  notice we build the stack from down to top: MDD first, then MDT */
5396
5397         name_size = MAX_OBD_NAME;
5398         uuid_size = MAX_OBD_NAME;
5399
5400         OBD_ALLOC(name, name_size);
5401         OBD_ALLOC(uuid, uuid_size);
5402         if (name == NULL || uuid == NULL)
5403                 GOTO(cleanup_mem, rc = -ENOMEM);
5404
5405         OBD_ALLOC_PTR(bufs);
5406         if (!bufs)
5407                 GOTO(cleanup_mem, rc = -ENOMEM);
5408
5409         strcpy(name, dev);
5410         p = strstr(name, "-MDT");
5411         if (p == NULL)
5412                 GOTO(free_bufs, rc = -ENOMEM);
5413         p[3] = 'D';
5414
5415         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
5416
5417         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
5418         if (lprof == NULL || lprof->lp_dt == NULL) {
5419                 CERROR("can't find the profile: %s\n",
5420                        lustre_cfg_string(cfg, 0));
5421                 GOTO(free_bufs, rc = -EINVAL);
5422         }
5423
5424         lustre_cfg_bufs_reset(bufs, name);
5425         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
5426         lustre_cfg_bufs_set_string(bufs, 2, uuid);
5427         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
5428
5429         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5430         if (!lcfg)
5431                 GOTO(put_profile, rc = -ENOMEM);
5432         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
5433
5434         rc = class_attach(lcfg);
5435         if (rc)
5436                 GOTO(lcfg_cleanup, rc);
5437
5438         obd = class_name2obd(name);
5439         if (!obd) {
5440                 CERROR("Can not find obd %s (%s in config)\n",
5441                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
5442                 GOTO(lcfg_cleanup, rc = -EINVAL);
5443         }
5444
5445         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5446
5447         lustre_cfg_bufs_reset(bufs, name);
5448         lustre_cfg_bufs_set_string(bufs, 1, uuid);
5449         lustre_cfg_bufs_set_string(bufs, 2, dev);
5450         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
5451
5452         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5453         if (!lcfg)
5454                 GOTO(class_detach, rc = -ENOMEM);
5455         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
5456
5457         rc = class_setup(obd, lcfg);
5458         if (rc)
5459                 GOTO(class_detach, rc);
5460
5461         /* connect to MDD we just setup */
5462         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
5463         if (rc)
5464                 GOTO(class_detach, rc);
5465
5466         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
5467         LASSERT(site);
5468         LASSERT(mdt_lu_site(mdt) == NULL);
5469         mdt->mdt_lu_dev.ld_site = site;
5470         site->ls_top_dev = &mdt->mdt_lu_dev;
5471         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
5472
5473         /* now connect to bottom OSD */
5474         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
5475         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
5476         if (rc)
5477                 GOTO(class_detach, rc);
5478         mdt->mdt_bottom =
5479                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
5480
5481         rc = lu_env_refill((struct lu_env *)env);
5482         if (rc != 0)
5483                 CERROR("Failure to refill session: '%d'\n", rc);
5484
5485         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
5486
5487         EXIT;
5488 class_detach:
5489         if (rc)
5490                 class_detach(obd, lcfg);
5491 lcfg_cleanup:
5492         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5493 put_profile:
5494         class_put_profile(lprof);
5495 free_bufs:
5496         OBD_FREE_PTR(bufs);
5497 cleanup_mem:
5498         if (name)
5499                 OBD_FREE(name, name_size);
5500         if (uuid)
5501                 OBD_FREE(uuid, uuid_size);
5502         RETURN(rc);
5503 }
5504
5505 /* setup quota master target on MDT0 */
5506 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
5507                           struct lustre_cfg *cfg)
5508 {
5509         struct obd_device       *obd;
5510         char                    *dev = lustre_cfg_string(cfg, 0);
5511         char                    *qmtname, *uuid, *p;
5512         struct lustre_cfg_bufs  *bufs;
5513         struct lustre_cfg       *lcfg;
5514         struct lustre_profile   *lprof;
5515         struct obd_connect_data *data;
5516         int                      rc;
5517         ENTRY;
5518
5519         LASSERT(mdt->mdt_qmt_exp == NULL);
5520         LASSERT(mdt->mdt_qmt_dev == NULL);
5521
5522         /* quota master is on MDT0 only for now */
5523         if (mdt->mdt_seq_site.ss_node_id != 0)
5524                 RETURN(0);
5525
5526         /* MGS generates config commands which look as follows:
5527          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
5528          *                    3:lustre-MDT0000-mdtlov  4:f
5529          *
5530          * We generate the QMT name from the MDT one, just replacing MD with QM
5531          * after all the preparations, the logical equivalent will be:
5532          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
5533          *                    3:lustre-MDT0000-osd  4:f */
5534         OBD_ALLOC(qmtname, MAX_OBD_NAME);
5535         OBD_ALLOC(uuid, UUID_MAX);
5536         OBD_ALLOC_PTR(bufs);
5537         OBD_ALLOC_PTR(data);
5538         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
5539                 GOTO(cleanup_mem, rc = -ENOMEM);
5540
5541         strcpy(qmtname, dev);
5542         p = strstr(qmtname, "-MDT");
5543         if (p == NULL)
5544                 GOTO(cleanup_mem, rc = -ENOMEM);
5545         /* replace MD with QM */
5546         p[1] = 'Q';
5547         p[2] = 'M';
5548
5549         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
5550
5551         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
5552         if (lprof == NULL || lprof->lp_dt == NULL) {
5553                 CERROR("can't find profile for %s\n",
5554                        lustre_cfg_string(cfg, 0));
5555                 GOTO(cleanup_mem, rc = -EINVAL);
5556         }
5557
5558         lustre_cfg_bufs_reset(bufs, qmtname);
5559         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
5560         lustre_cfg_bufs_set_string(bufs, 2, uuid);
5561         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
5562
5563         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5564         if (!lcfg)
5565                 GOTO(put_profile, rc = -ENOMEM);
5566         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
5567
5568         rc = class_attach(lcfg);
5569         if (rc)
5570                 GOTO(lcfg_cleanup, rc);
5571
5572         obd = class_name2obd(qmtname);
5573         if (!obd) {
5574                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
5575                        lustre_cfg_string(cfg, 0));
5576                 GOTO(lcfg_cleanup, rc = -EINVAL);
5577         }
5578
5579         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5580
5581         lustre_cfg_bufs_reset(bufs, qmtname);
5582         lustre_cfg_bufs_set_string(bufs, 1, uuid);
5583         lustre_cfg_bufs_set_string(bufs, 2, dev);
5584
5585         /* for quota, the next device should be the OSD device */
5586         lustre_cfg_bufs_set_string(bufs, 3,
5587                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
5588
5589         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5590         if (!lcfg)
5591                 GOTO(class_detach, rc = -ENOMEM);
5592         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
5593
5594         rc = class_setup(obd, lcfg);
5595         if (rc)
5596                 GOTO(class_detach, rc);
5597
5598         mdt->mdt_qmt_dev = obd->obd_lu_dev;
5599
5600         /* configure local quota objects */
5601         if (OBD_FAIL_CHECK(OBD_FAIL_QUOTA_INIT))
5602                 rc = -EBADF;
5603         else
5604                 rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
5605                                                            &mdt->mdt_lu_dev,
5606                                                            mdt->mdt_qmt_dev);
5607         if (rc)
5608                 GOTO(class_cleanup, rc);
5609
5610         /* connect to quota master target */
5611         data->ocd_connect_flags = OBD_CONNECT_VERSION;
5612         data->ocd_version = LUSTRE_VERSION_CODE;
5613         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
5614                          data, NULL);
5615         if (rc) {
5616                 CERROR("cannot connect to quota master device %s (%d)\n",
5617                        qmtname, rc);
5618                 GOTO(class_cleanup, rc);
5619         }
5620
5621         EXIT;
5622 class_cleanup:
5623         if (rc) {
5624                 class_manual_cleanup(obd);
5625                 mdt->mdt_qmt_dev = NULL;
5626                 GOTO(lcfg_cleanup, rc);
5627         }
5628 class_detach:
5629         if (rc)
5630                 class_detach(obd, lcfg);
5631 lcfg_cleanup:
5632         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5633 put_profile:
5634         class_put_profile(lprof);
5635 cleanup_mem:
5636         if (bufs)
5637                 OBD_FREE_PTR(bufs);
5638         if (qmtname)
5639                 OBD_FREE(qmtname, MAX_OBD_NAME);
5640         if (uuid)
5641                 OBD_FREE(uuid, UUID_MAX);
5642         if (data)
5643                 OBD_FREE_PTR(data);
5644         return rc;
5645 }
5646
5647 /* Shutdown quota master target associated with mdt */
5648 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
5649 {
5650         ENTRY;
5651
5652         if (mdt->mdt_qmt_exp == NULL)
5653                 RETURN_EXIT;
5654         LASSERT(mdt->mdt_qmt_dev != NULL);
5655
5656         /* the qmt automatically shuts down when the mdt disconnects */
5657         obd_disconnect(mdt->mdt_qmt_exp);
5658         mdt->mdt_qmt_exp = NULL;
5659         mdt->mdt_qmt_dev = NULL;
5660         EXIT;
5661 }
5662
5663 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
5664  * for now. This will be removed along with converting rest of MDT code
5665  * to use tgt_session_info */
5666 static int mdt_tgt_getxattr(struct tgt_session_info *tsi)
5667 {
5668         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
5669         int                      rc;
5670
5671         if (unlikely(info->mti_object == NULL))
5672                 return -EPROTO;
5673
5674         rc = mdt_getxattr(info);
5675
5676         mdt_thread_info_fini(info);
5677         return rc;
5678 }
5679
5680 static int mdt_llog_open(struct tgt_session_info *tsi)
5681 {
5682         ENTRY;
5683
5684         if (!mdt_is_rootadmin(tsi2mdt_info(tsi)))
5685                 RETURN(err_serious(-EACCES));
5686
5687         RETURN(tgt_llog_open(tsi));
5688 }
5689
5690 #define OBD_FAIL_OST_READ_NET   OBD_FAIL_OST_BRW_NET
5691 #define OBD_FAIL_OST_WRITE_NET  OBD_FAIL_OST_BRW_NET
5692 #define OST_BRW_READ    OST_READ
5693 #define OST_BRW_WRITE   OST_WRITE
5694
5695 static struct tgt_handler mdt_tgt_handlers[] = {
5696 TGT_RPC_HANDLER(MDS_FIRST_OPC,
5697                 0,                      MDS_CONNECT,    mdt_tgt_connect,
5698                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
5699 TGT_RPC_HANDLER(MDS_FIRST_OPC,
5700                 0,                      MDS_DISCONNECT, tgt_disconnect,
5701                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
5702 TGT_RPC_HANDLER(MDS_FIRST_OPC,
5703                 HAS_REPLY,              MDS_SET_INFO,   mdt_set_info,
5704                 &RQF_MDT_SET_INFO, LUSTRE_MDS_VERSION),
5705 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
5706 TGT_MDT_HDL(HAS_REPLY,          MDS_GET_ROOT,   mdt_get_root),
5707 TGT_MDT_HDL(HAS_BODY,           MDS_GETATTR,    mdt_getattr),
5708 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_GETATTR_NAME,
5709                                                         mdt_getattr_name),
5710 TGT_MDT_HDL(HAS_BODY,           MDS_GETXATTR,   mdt_tgt_getxattr),
5711 TGT_MDT_HDL(HAS_REPLY,          MDS_STATFS,     mdt_statfs),
5712 TGT_MDT_HDL(IS_MUTABLE,         MDS_REINT,      mdt_reint),
5713 TGT_MDT_HDL(HAS_BODY,           MDS_CLOSE,      mdt_close),
5714 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_READPAGE,   mdt_readpage),
5715 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_SYNC,       mdt_sync),
5716 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
5717 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_PROGRESS,
5718                                                         mdt_hsm_progress),
5719 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_CT_REGISTER,
5720                                                         mdt_hsm_ct_register),
5721 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_CT_UNREGISTER,
5722                                                         mdt_hsm_ct_unregister),
5723 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_STATE_GET,
5724                                                         mdt_hsm_state_get),
5725 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_STATE_SET,
5726                                                         mdt_hsm_state_set),
5727 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_ACTION,       mdt_hsm_action),
5728 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_REQUEST,
5729                                                         mdt_hsm_request),
5730 TGT_MDT_HDL(HAS_KEY | HAS_BODY | HAS_REPLY | IS_MUTABLE,
5731             MDS_SWAP_LAYOUTS,
5732             mdt_swap_layouts),
5733 TGT_MDT_HDL(IS_MUTABLE,         MDS_RMFID,      mdt_rmfid),
5734 };
5735
5736 static struct tgt_handler mdt_io_ops[] = {
5737 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY, OST_BRW_READ, tgt_brw_read,
5738                                                         mdt_hp_brw),
5739 TGT_OST_HDL_HP(HAS_BODY | IS_MUTABLE,    OST_BRW_WRITE, tgt_brw_write,
5740                                                         mdt_hp_brw),
5741 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY | IS_MUTABLE,
5742                                          OST_PUNCH,     mdt_punch_hdl,
5743                                                         mdt_hp_punch),
5744 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SYNC,     mdt_data_sync),
5745 TGT_OST_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, OST_FALLOCATE,
5746                                                         mdt_fallocate_hdl),
5747 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SEEK, tgt_lseek),
5748 TGT_RPC_HANDLER(OST_FIRST_OPC,
5749                 0,                      OST_SET_INFO,   mdt_io_set_info,
5750                 &RQF_OBD_SET_INFO, LUSTRE_OST_VERSION),
5751 };
5752
5753 static struct tgt_handler mdt_sec_ctx_ops[] = {
5754 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
5755 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
5756 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
5757 };
5758
5759 static struct tgt_handler mdt_quota_ops[] = {
5760 TGT_QUOTA_HDL(HAS_REPLY,                QUOTA_DQACQ,      mdt_quota_dqacq),
5761 };
5762
5763 static struct tgt_handler mdt_llog_handlers[] = {
5764         TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_CREATE,      mdt_llog_open),
5765         TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_NEXT_BLOCK,  tgt_llog_next_block),
5766         TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
5767         TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_PREV_BLOCK,  tgt_llog_prev_block),
5768 };
5769
5770 static struct tgt_opc_slice mdt_common_slice[] = {
5771         {
5772                 .tos_opc_start  = MDS_FIRST_OPC,
5773                 .tos_opc_end    = MDS_LAST_OPC,
5774                 .tos_hs         = mdt_tgt_handlers
5775         },
5776         {
5777                 .tos_opc_start  = OBD_FIRST_OPC,
5778                 .tos_opc_end    = OBD_LAST_OPC,
5779                 .tos_hs         = tgt_obd_handlers
5780         },
5781         {
5782                 .tos_opc_start  = LDLM_FIRST_OPC,
5783                 .tos_opc_end    = LDLM_LAST_OPC,
5784                 .tos_hs         = tgt_dlm_handlers
5785         },
5786         {
5787                 .tos_opc_start  = SEC_FIRST_OPC,
5788                 .tos_opc_end    = SEC_LAST_OPC,
5789                 .tos_hs         = mdt_sec_ctx_ops
5790         },
5791         {
5792                 .tos_opc_start  = OUT_UPDATE_FIRST_OPC,
5793                 .tos_opc_end    = OUT_UPDATE_LAST_OPC,
5794                 .tos_hs         = tgt_out_handlers
5795         },
5796         {
5797                 .tos_opc_start  = FLD_FIRST_OPC,
5798                 .tos_opc_end    = FLD_LAST_OPC,
5799                 .tos_hs         = fld_handlers
5800         },
5801         {
5802                 .tos_opc_start  = SEQ_FIRST_OPC,
5803                 .tos_opc_end    = SEQ_LAST_OPC,
5804                 .tos_hs         = seq_handlers
5805         },
5806         {
5807                 .tos_opc_start  = QUOTA_DQACQ,
5808                 .tos_opc_end    = QUOTA_LAST_OPC,
5809                 .tos_hs         = mdt_quota_ops
5810         },
5811         {
5812                 .tos_opc_start  = LLOG_FIRST_OPC,
5813                 .tos_opc_end    = LLOG_LAST_OPC,
5814                 .tos_hs         = mdt_llog_handlers
5815         },
5816         {
5817                 .tos_opc_start  = LFSCK_FIRST_OPC,
5818                 .tos_opc_end    = LFSCK_LAST_OPC,
5819                 .tos_hs         = tgt_lfsck_handlers
5820         },
5821         {
5822                 .tos_opc_start  = OST_FIRST_OPC,
5823                 .tos_opc_end    = OST_LAST_OPC,
5824                 .tos_hs         = mdt_io_ops
5825         },
5826         {
5827                 .tos_hs         = NULL
5828         }
5829 };
5830
5831 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
5832 {
5833         struct md_device *next = m->mdt_child;
5834         struct lu_device *d = &m->mdt_lu_dev;
5835         struct obd_device *obd = mdt2obd_dev(m);
5836         struct lfsck_stop stop;
5837
5838         ENTRY;
5839         stop.ls_status = LS_PAUSED;
5840         stop.ls_flags = 0;
5841         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
5842
5843         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
5844
5845         mdt_restriper_stop(m);
5846         ping_evictor_stop();
5847
5848         /* Remove the HSM /proc entry so the coordinator cannot be
5849          * restarted by a user while it's shutting down.
5850          */
5851         mdt_hsm_cdt_stop(m);
5852
5853         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
5854         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
5855
5856         if (m->mdt_namespace != NULL)
5857                 ldlm_namespace_free_prior(m->mdt_namespace, NULL,
5858                                           d->ld_obd->obd_force);
5859
5860         obd_exports_barrier(obd);
5861         obd_zombie_barrier();
5862
5863         mdt_quota_fini(env, m);
5864
5865         cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
5866
5867         /* Calling the cleanup functions in the same order as in the mdt_init0
5868          * error path
5869          */
5870         mdt_tunables_fini(m);
5871
5872         target_recovery_fini(obd);
5873         upcall_cache_cleanup(m->mdt_identity_cache);
5874         m->mdt_identity_cache = NULL;
5875
5876         mdt_fs_cleanup(env, m);
5877
5878         tgt_fini(env, &m->mdt_lut);
5879
5880         mdt_hsm_cdt_fini(m);
5881
5882         if (m->mdt_los != NULL) {
5883                 local_oid_storage_fini(env, m->mdt_los);
5884                 m->mdt_los = NULL;
5885         }
5886
5887         if (m->mdt_namespace != NULL) {
5888                 ldlm_namespace_free_post(m->mdt_namespace);
5889                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
5890         }
5891
5892         if (m->mdt_md_root != NULL) {
5893                 mdt_object_put(env, m->mdt_md_root);
5894                 m->mdt_md_root = NULL;
5895         }
5896
5897         mdt_seq_fini(env, m);
5898
5899         mdt_fld_fini(env, m);
5900
5901         /*
5902          * Finish the stack
5903          */
5904         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
5905
5906         LASSERT(atomic_read(&d->ld_ref) == 0);
5907
5908         server_put_mount(mdt_obd_name(m), true);
5909
5910         EXIT;
5911 }
5912
5913 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
5914
5915 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
5916                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
5917 {
5918         const struct dt_device_param *dt_conf;
5919         struct mdt_thread_info *info;
5920         struct obd_device *obd;
5921         const char *dev = lustre_cfg_string(cfg, 0);
5922         const char *num = lustre_cfg_string(cfg, 2);
5923         struct tg_grants_data *tgd = &m->mdt_lut.lut_tgd;
5924         struct lustre_mount_info *lmi = NULL;
5925         struct lustre_sb_info *lsi;
5926         struct lu_site *s;
5927         struct seq_server_site *ss_site;
5928         const char *identity_upcall = "NONE";
5929         struct md_device *next;
5930         struct lu_fid fid;
5931         int rc;
5932         long node_id;
5933         mntopt_t mntopts;
5934         ENTRY;
5935
5936         lu_device_init(&m->mdt_lu_dev, ldt);
5937         /*
5938          * Environment (env) might be missing mdt_thread_key values at that
5939          * point, if device is allocated when mdt_thread_key is in QUIESCENT
5940          * mode.
5941          *
5942          * Usually device allocation path doesn't use module key values, but
5943          * mdt has to do a lot of work here, so allocate key value.
5944          */
5945         rc = lu_env_refill((struct lu_env *)env);
5946         if (rc != 0)
5947                 RETURN(rc);
5948
5949         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5950         LASSERT(info != NULL);
5951
5952         obd = class_name2obd(dev);
5953         LASSERT(obd != NULL);
5954
5955         m->mdt_max_mdsize = MAX_MD_SIZE_OLD;
5956         m->mdt_opts.mo_evict_tgt_nids = 1;
5957         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
5958
5959         lmi = server_get_mount(dev);
5960         if (lmi == NULL) {
5961                 CERROR("Cannot get mount info for %s!\n", dev);
5962                 RETURN(-EFAULT);
5963         } else {
5964                 lsi = s2lsi(lmi->lmi_sb);
5965                 LASSERT(lsi->lsi_lmd);
5966                 /* CMD is supported only in IAM mode */
5967                 LASSERT(num);
5968                 rc = kstrtol(num, 10, &node_id);
5969                 if (rc)
5970                         RETURN(rc);
5971
5972                 obd->u.obt.obt_magic = OBT_MAGIC;
5973                 if (lsi->lsi_lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
5974                         m->mdt_skip_lfsck = 1;
5975         }
5976
5977         /* Just try to get a DoM lock by default. Otherwise, having a group
5978          * lock granted, it may get blocked for a long time. */
5979         m->mdt_opts.mo_dom_lock = TRYLOCK_DOM_ON_OPEN;
5980         /* DoM files are read at open and data is packed in the reply */
5981         m->mdt_opts.mo_dom_read_open = 1;
5982
5983         m->mdt_squash.rsi_uid = 0;
5984         m->mdt_squash.rsi_gid = 0;
5985         INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
5986         spin_lock_init(&m->mdt_squash.rsi_lock);
5987         spin_lock_init(&m->mdt_lock);
5988         m->mdt_enable_remote_dir = 1;
5989         m->mdt_enable_striped_dir = 1;
5990         m->mdt_enable_dir_migration = 1;
5991         m->mdt_enable_dir_restripe = 0;
5992         m->mdt_enable_dir_auto_split = 0;
5993         m->mdt_enable_remote_dir_gid = 0;
5994         m->mdt_enable_chprojid_gid = 0;
5995         m->mdt_enable_remote_rename = 1;
5996         m->mdt_dir_restripe_nsonly = 1;
5997         m->mdt_enable_remote_subdir_mount = 1;
5998
5999         atomic_set(&m->mdt_mds_mds_conns, 0);
6000         atomic_set(&m->mdt_async_commit_count, 0);
6001
6002         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
6003         m->mdt_lu_dev.ld_obd = obd;
6004         /* Set this lu_device to obd for error handling purposes. */
6005         obd->obd_lu_dev = &m->mdt_lu_dev;
6006
6007         /* init the stack */
6008         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
6009         if (rc) {
6010                 CERROR("%s: Can't init device stack, rc %d\n",
6011                        mdt_obd_name(m), rc);
6012                 GOTO(err_lmi, rc);
6013         }
6014
6015         s = mdt_lu_site(m);
6016         ss_site = mdt_seq_site(m);
6017         s->ld_seq_site = ss_site;
6018         ss_site->ss_lu = s;
6019
6020         /* set server index */
6021         ss_site->ss_node_id = node_id;
6022
6023         /* failover is the default
6024          * FIXME: we do not failout mds0/mgs, which may cause some problems.
6025          * assumed whose ss_node_id == 0 XXX
6026          * */
6027         obd->obd_replayable = 1;
6028         /* No connection accepted until configurations will finish */
6029         obd->obd_no_conn = 1;
6030
6031         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
6032                 char *str = lustre_cfg_string(cfg, 4);
6033                 if (strchr(str, 'n')) {
6034                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
6035                         obd->obd_replayable = 0;
6036                 }
6037         }
6038
6039         rc = mdt_fld_init(env, mdt_obd_name(m), m);
6040         if (rc)
6041                 GOTO(err_fini_stack, rc);
6042
6043         rc = mdt_seq_init(env, m);
6044         if (rc)
6045                 GOTO(err_fini_fld, rc);
6046
6047         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
6048                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
6049         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
6050                                               LDLM_NAMESPACE_SERVER,
6051                                               LDLM_NAMESPACE_GREEDY,
6052                                               LDLM_NS_TYPE_MDT);
6053         if (IS_ERR(m->mdt_namespace)) {
6054                 rc = PTR_ERR(m->mdt_namespace);
6055                 CERROR("%s: unable to create server namespace: rc = %d\n",
6056                        obd->obd_name, rc);
6057                 m->mdt_namespace = NULL;
6058                 GOTO(err_fini_seq, rc);
6059         }
6060
6061         m->mdt_namespace->ns_lvbp = m;
6062         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
6063
6064         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
6065         /* set obd_namespace for compatibility with old code */
6066         obd->obd_namespace = m->mdt_namespace;
6067
6068         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
6069                       OBD_FAIL_MDS_ALL_REQUEST_NET,
6070                       OBD_FAIL_MDS_ALL_REPLY_NET);
6071         if (rc)
6072                 GOTO(err_free_ns, rc);
6073
6074         /* Amount of available space excluded from granting and reserved
6075          * for metadata. It is a percentage of the total MDT size. */
6076         tgd->tgd_reserved_pcnt = 10;
6077
6078         if (ONE_MB_BRW_SIZE < (1U << tgd->tgd_blockbits))
6079                 m->mdt_brw_size = 1U << tgd->tgd_blockbits;
6080         else
6081                 m->mdt_brw_size = ONE_MB_BRW_SIZE;
6082
6083         rc = mdt_fs_setup(env, m, obd, lsi);
6084         if (rc)
6085                 GOTO(err_tgt, rc);
6086
6087         fid.f_seq = FID_SEQ_LOCAL_NAME;
6088         fid.f_oid = 1;
6089         fid.f_ver = 0;
6090         rc = local_oid_storage_init(env, m->mdt_bottom, &fid, &m->mdt_los);
6091         if (rc != 0)
6092                 GOTO(err_fs_cleanup, rc);
6093
6094         rc = mdt_hsm_cdt_init(m);
6095         if (rc != 0) {
6096                 CERROR("%s: error initializing coordinator, rc %d\n",
6097                        mdt_obd_name(m), rc);
6098                 GOTO(err_los_fini, rc);
6099         }
6100
6101         tgt_adapt_sptlrpc_conf(&m->mdt_lut);
6102
6103         next = m->mdt_child;
6104         dt_conf = next->md_ops->mdo_dtconf_get(env, next);
6105
6106         mntopts = dt_conf->ddp_mntopts;
6107
6108         if (mntopts & MNTOPT_USERXATTR)
6109                 m->mdt_opts.mo_user_xattr = 1;
6110         else
6111                 m->mdt_opts.mo_user_xattr = 0;
6112
6113         m->mdt_max_ea_size = dt_conf->ddp_max_ea_size;
6114
6115         if (mntopts & MNTOPT_ACL)
6116                 m->mdt_opts.mo_acl = 1;
6117         else
6118                 m->mdt_opts.mo_acl = 0;
6119
6120         m->mdt_opts.mo_enable_strict_som = 1;
6121
6122         /* XXX: to support suppgid for ACL, we enable identity_upcall
6123          * by default, otherwise, maybe got unexpected -EACCESS. */
6124         if (m->mdt_opts.mo_acl)
6125                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
6126
6127         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
6128                                                 identity_upcall,
6129                                                 &mdt_identity_upcall_cache_ops);
6130         if (IS_ERR(m->mdt_identity_cache)) {
6131                 rc = PTR_ERR(m->mdt_identity_cache);
6132                 m->mdt_identity_cache = NULL;
6133                 GOTO(err_free_hsm, rc);
6134         }
6135
6136         rc = mdt_tunables_init(m, dev);
6137         if (rc) {
6138                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
6139                 GOTO(err_recovery, rc);
6140         }
6141
6142         rc = mdt_quota_init(env, m, cfg);
6143         if (rc)
6144                 GOTO(err_procfs, rc);
6145
6146         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
6147         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
6148                            "mdt_ldlm_client", m->mdt_ldlm_client);
6149
6150         ping_evictor_start();
6151
6152         /* recovery will be started upon mdt_prepare()
6153          * when the whole stack is complete and ready
6154          * to serve the requests */
6155
6156         /* Reduce the initial timeout on an MDS because it doesn't need such
6157          * a long timeout as an OST does. Adaptive timeouts will adjust this
6158          * value appropriately. */
6159         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
6160                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
6161
6162         if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_LOCAL_RECOV))
6163                 m->mdt_lut.lut_local_recovery = 1;
6164
6165         rc = mdt_restriper_start(m);
6166         if (rc)
6167                 GOTO(err_ping_evictor, rc);
6168
6169         RETURN(0);
6170
6171 err_ping_evictor:
6172         ping_evictor_stop();
6173 err_procfs:
6174         mdt_tunables_fini(m);
6175 err_recovery:
6176         upcall_cache_cleanup(m->mdt_identity_cache);
6177         m->mdt_identity_cache = NULL;
6178 err_free_hsm:
6179         mdt_hsm_cdt_fini(m);
6180 err_los_fini:
6181         local_oid_storage_fini(env, m->mdt_los);
6182         m->mdt_los = NULL;
6183 err_fs_cleanup:
6184         mdt_fs_cleanup(env, m);
6185 err_tgt:
6186         /* keep recoverable clients */
6187         obd->obd_fail = 1;
6188         target_recovery_fini(obd);
6189         obd_exports_barrier(obd);
6190         obd_zombie_barrier();
6191         tgt_fini(env, &m->mdt_lut);
6192 err_free_ns:
6193         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
6194         obd->obd_namespace = m->mdt_namespace = NULL;
6195 err_fini_seq:
6196         mdt_seq_fini(env, m);
6197 err_fini_fld:
6198         mdt_fld_fini(env, m);
6199 err_fini_stack:
6200         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
6201 err_lmi:
6202         if (lmi)
6203                 server_put_mount(dev, true);
6204         return(rc);
6205 }
6206
6207 /* For interoperability, the left element is old parameter, the right one
6208  * is the new version of the parameter, if some parameter is deprecated,
6209  * the new version should be set as NULL. */
6210 static struct cfg_interop_param mdt_interop_param[] = {
6211         { "mdt.group_upcall",   NULL },
6212         { "mdt.quota_type",     NULL },
6213         { "mdd.quota_type",     NULL },
6214         { "mdt.som",            NULL },
6215         { "mdt.rootsquash",     "mdt.root_squash" },
6216         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
6217         { NULL }
6218 };
6219
6220 /* used by MGS to process specific configurations */
6221 static int mdt_process_config(const struct lu_env *env,
6222                               struct lu_device *d, struct lustre_cfg *cfg)
6223 {
6224         struct mdt_device *m = mdt_dev(d);
6225         struct md_device *md_next = m->mdt_child;
6226         struct lu_device *next = md2lu_dev(md_next);
6227         int rc;
6228         ENTRY;
6229
6230         switch (cfg->lcfg_command) {
6231         case LCFG_PARAM: {
6232                 struct obd_device *obd = d->ld_obd;
6233                 /* For interoperability */
6234                 struct cfg_interop_param *ptr = NULL;
6235                 struct lustre_cfg *old_cfg = NULL;
6236                 char *param = NULL;
6237                 ssize_t count;
6238
6239                 param = lustre_cfg_string(cfg, 1);
6240                 if (param == NULL) {
6241                         CERROR("param is empty\n");
6242                         rc = -EINVAL;
6243                         break;
6244                 }
6245
6246                 ptr = class_find_old_param(param, mdt_interop_param);
6247                 if (ptr != NULL) {
6248                         if (ptr->new_param == NULL) {
6249                                 rc = 0;
6250                                 CWARN("For interoperability, skip this %s."
6251                                       " It is obsolete.\n", ptr->old_param);
6252                                 break;
6253                         }
6254
6255                         CWARN("Found old param %s, changed it to %s.\n",
6256                               ptr->old_param, ptr->new_param);
6257
6258                         old_cfg = cfg;
6259                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
6260                         if (IS_ERR(cfg)) {
6261                                 rc = PTR_ERR(cfg);
6262                                 break;
6263                         }
6264                 }
6265
6266                 count = class_modify_config(cfg, PARAM_MDT,
6267                                             &obd->obd_kset.kobj);
6268                 if (count < 0) {
6269                         struct coordinator *cdt = &m->mdt_coordinator;
6270
6271                         /* is it an HSM var ? */
6272                         count = class_modify_config(cfg, PARAM_HSM,
6273                                                     &cdt->cdt_hsm_kobj);
6274                         if (count < 0)
6275                                 /* we don't understand; pass it on */
6276                                 rc = next->ld_ops->ldo_process_config(env, next,
6277                                                                       cfg);
6278                         else
6279                                 rc = count > 0 ? 0 : count;
6280                 } else {
6281                         rc = count > 0 ? 0 : count;
6282                 }
6283
6284                 if (old_cfg)
6285                         OBD_FREE(cfg, lustre_cfg_len(cfg->lcfg_bufcount,
6286                                                      cfg->lcfg_buflens));
6287                 break;
6288         }
6289         default:
6290                 /* others are passed further */
6291                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
6292                 break;
6293         }
6294         RETURN(rc);
6295 }
6296
6297 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
6298                                           const struct lu_object_header *hdr,
6299                                           struct lu_device *d)
6300 {
6301         struct mdt_object *mo;
6302
6303         ENTRY;
6304
6305         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
6306         if (mo != NULL) {
6307                 struct lu_object *o;
6308                 struct lu_object_header *h;
6309
6310                 o = &mo->mot_obj;
6311                 h = &mo->mot_header;
6312                 lu_object_header_init(h);
6313                 lu_object_init(o, h, d);
6314                 lu_object_add_top(h, o);
6315                 o->lo_ops = &mdt_obj_ops;
6316                 spin_lock_init(&mo->mot_write_lock);
6317                 mutex_init(&mo->mot_som_mutex);
6318                 mutex_init(&mo->mot_lov_mutex);
6319                 init_rwsem(&mo->mot_dom_sem);
6320                 init_rwsem(&mo->mot_open_sem);
6321                 atomic_set(&mo->mot_open_count, 0);
6322                 mo->mot_restripe_offset = 0;
6323                 INIT_LIST_HEAD(&mo->mot_restripe_linkage);
6324                 mo->mot_lsom_size = 0;
6325                 mo->mot_lsom_blocks = 0;
6326                 mo->mot_lsom_inited = false;
6327                 RETURN(o);
6328         }
6329         RETURN(NULL);
6330 }
6331
6332 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
6333                            const struct lu_object_conf *unused)
6334 {
6335         struct mdt_device *d = mdt_dev(o->lo_dev);
6336         struct lu_device  *under;
6337         struct lu_object  *below;
6338         int                rc = 0;
6339         ENTRY;
6340
6341         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
6342                PFID(lu_object_fid(o)));
6343
6344         under = &d->mdt_child->md_lu_dev;
6345         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
6346         if (below != NULL) {
6347                 lu_object_add(o, below);
6348         } else
6349                 rc = -ENOMEM;
6350
6351         RETURN(rc);
6352 }
6353
6354 static void mdt_object_free_rcu(struct rcu_head *head)
6355 {
6356         struct mdt_object *mo = container_of(head, struct mdt_object,
6357                                              mot_header.loh_rcu);
6358
6359         kmem_cache_free(mdt_object_kmem, mo);
6360 }
6361
6362 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
6363 {
6364         struct mdt_object *mo = mdt_obj(o);
6365         struct lu_object_header *h;
6366         ENTRY;
6367
6368         h = o->lo_header;
6369         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
6370                PFID(lu_object_fid(o)));
6371
6372         LASSERT(atomic_read(&mo->mot_open_count) == 0);
6373         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
6374
6375         lu_object_fini(o);
6376         lu_object_header_fini(h);
6377         OBD_FREE_PRE(mo, sizeof(*mo), "slab-freed");
6378         call_rcu(&mo->mot_header.loh_rcu, mdt_object_free_rcu);
6379
6380         EXIT;
6381 }
6382
6383 static int mdt_object_print(const struct lu_env *env, void *cookie,
6384                             lu_printer_t p, const struct lu_object *o)
6385 {
6386         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
6387
6388         return (*p)(env, cookie,
6389                     LUSTRE_MDT_NAME"-object@%p(%s %s, writecount=%d)",
6390                     mdto, mdto->mot_lov_created ? "lov_created" : "",
6391                     mdto->mot_cache_attr ? "cache_attr" : "",
6392                     mdto->mot_write_count);
6393 }
6394
6395 static int mdt_prepare(const struct lu_env *env,
6396                 struct lu_device *pdev,
6397                 struct lu_device *cdev)
6398 {
6399         struct mdt_device *mdt = mdt_dev(cdev);
6400         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
6401         struct obd_device *obd = cdev->ld_obd;
6402         int rc;
6403
6404         ENTRY;
6405
6406         LASSERT(obd);
6407
6408         rc = next->ld_ops->ldo_prepare(env, cdev, next);
6409         if (rc)
6410                 RETURN(rc);
6411
6412         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
6413         if (rc)
6414                 RETURN(rc);
6415
6416         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
6417         if (rc)
6418                 RETURN(rc);
6419
6420         rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
6421         /* The LFSCK instance is registered just now, so it must be there when
6422          * register the namespace to such instance. */
6423         LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
6424
6425         if (mdt->mdt_seq_site.ss_node_id == 0) {
6426                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
6427                                                          &mdt->mdt_md_root_fid);
6428                 if (rc)
6429                         RETURN(rc);
6430         }
6431
6432         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
6433
6434         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
6435         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
6436         LASSERT(obd->obd_no_conn);
6437         spin_lock(&obd->obd_dev_lock);
6438         obd->obd_no_conn = 0;
6439         spin_unlock(&obd->obd_dev_lock);
6440
6441         if (obd->obd_recovering == 0)
6442                 mdt_postrecov(env, mdt);
6443
6444         RETURN(rc);
6445 }
6446
6447 const struct lu_device_operations mdt_lu_ops = {
6448         .ldo_object_alloc   = mdt_object_alloc,
6449         .ldo_process_config = mdt_process_config,
6450         .ldo_prepare        = mdt_prepare,
6451 };
6452
6453 static const struct lu_object_operations mdt_obj_ops = {
6454         .loo_object_init    = mdt_object_init,
6455         .loo_object_free    = mdt_object_free,
6456         .loo_object_print   = mdt_object_print
6457 };
6458
6459 static int mdt_obd_set_info_async(const struct lu_env *env,
6460                                   struct obd_export *exp,
6461                                   __u32 keylen, void *key,
6462                                   __u32 vallen, void *val,
6463                                   struct ptlrpc_request_set *set)
6464 {
6465         int rc;
6466
6467         ENTRY;
6468
6469         if (KEY_IS(KEY_SPTLRPC_CONF)) {
6470                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp));
6471                 RETURN(rc);
6472         }
6473
6474         RETURN(0);
6475 }
6476
6477 static inline void mdt_enable_slc(struct mdt_device *mdt)
6478 {
6479         if (mdt->mdt_lut.lut_sync_lock_cancel == SYNC_LOCK_CANCEL_NEVER)
6480                 mdt->mdt_lut.lut_sync_lock_cancel = SYNC_LOCK_CANCEL_BLOCKING;
6481 }
6482
6483 static inline void mdt_disable_slc(struct mdt_device *mdt)
6484 {
6485         if (mdt->mdt_lut.lut_sync_lock_cancel == SYNC_LOCK_CANCEL_BLOCKING)
6486                 mdt->mdt_lut.lut_sync_lock_cancel = SYNC_LOCK_CANCEL_NEVER;
6487 }
6488
6489 /**
6490  * Match client and server connection feature flags.
6491  *
6492  * Compute the compatibility flags for a connection request based on
6493  * features mutually supported by client and server.
6494  *
6495  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
6496  * must not be updated here, otherwise a partially initialized value may
6497  * be exposed. After the connection request is successfully processed,
6498  * the top-level MDT connect request handler atomically updates the export
6499  * connect flags from the obd_connect_data::ocd_connect_flags field of the
6500  * reply. \see mdt_connect().
6501  *
6502  * Before 2.7.50 clients will send a struct obd_connect_data_v1 rather than a
6503  * full struct obd_connect_data. So care must be taken when accessing fields
6504  * that are not present in struct obd_connect_data_v1. See LU-16.
6505  *
6506  * \param exp   the obd_export associated with this client/target pair
6507  * \param mdt   the target device for the connection
6508  * \param data  stores data for this connect request
6509  *
6510  * \retval 0       success
6511  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
6512  * \retval -EBADE  client and server feature requirements are incompatible
6513  */
6514 static int mdt_connect_internal(const struct lu_env *env,
6515                                 struct obd_export *exp,
6516                                 struct mdt_device *mdt,
6517                                 struct obd_connect_data *data, bool reconnect)
6518 {
6519         const char *obd_name = mdt_obd_name(mdt);
6520         LASSERT(data != NULL);
6521
6522         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
6523
6524         if (mdt->mdt_bottom->dd_rdonly &&
6525             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
6526             !(data->ocd_connect_flags & OBD_CONNECT_RDONLY))
6527                 RETURN(-EACCES);
6528
6529         if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
6530                 data->ocd_connect_flags2 &= MDT_CONNECT_SUPPORTED2;
6531
6532         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
6533
6534         if (!mdt->mdt_opts.mo_acl)
6535                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
6536
6537         if (!mdt->mdt_opts.mo_user_xattr)
6538                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
6539
6540         if (OCD_HAS_FLAG(data, BRW_SIZE)) {
6541                 data->ocd_brw_size = min(data->ocd_brw_size,
6542                                          mdt->mdt_brw_size);
6543                 if (data->ocd_brw_size == 0) {
6544                         CERROR("%s: cli %s/%p ocd_connect_flags: %#llx "
6545                                "ocd_version: %x ocd_grant: %d ocd_index: %u "
6546                                "ocd_brw_size unexpectedly zero, network data "
6547                                "corruption? Refusing to connect this client\n",
6548                                obd_name, exp->exp_client_uuid.uuid,
6549                                exp, data->ocd_connect_flags, data->ocd_version,
6550                                data->ocd_grant, data->ocd_index);
6551                         return -EPROTO;
6552                 }
6553         }
6554
6555         if (OCD_HAS_FLAG(data, GRANT_PARAM)) {
6556                 struct dt_device_param *ddp = &mdt->mdt_lut.lut_dt_conf;
6557
6558                 /* client is reporting its page size, for future use */
6559                 exp->exp_target_data.ted_pagebits = data->ocd_grant_blkbits;
6560                 data->ocd_grant_blkbits  = mdt->mdt_lut.lut_tgd.tgd_blockbits;
6561                 /* ddp_inodespace may not be power-of-two value, eg. for ldiskfs
6562                  * it's LDISKFS_DIR_REC_LEN(20) = 28. */
6563                 data->ocd_grant_inobits = fls(ddp->ddp_inodespace - 1);
6564                 /* ocd_grant_tax_kb is in 1K byte blocks */
6565                 data->ocd_grant_tax_kb = ddp->ddp_extent_tax >> 10;
6566                 data->ocd_grant_max_blks = ddp->ddp_max_extent_blks;
6567         }
6568
6569         /* Save connect_data we have so far because tgt_grant_connect()
6570          * uses it to calculate grant, and we want to save the client
6571          * version before it is overwritten by LUSTRE_VERSION_CODE. */
6572         exp->exp_connect_data = *data;
6573         if (OCD_HAS_FLAG(data, GRANT))
6574                 tgt_grant_connect(env, exp, data, !reconnect);
6575
6576         if (OCD_HAS_FLAG(data, MAXBYTES))
6577                 data->ocd_maxbytes = mdt->mdt_lut.lut_dt_conf.ddp_maxbytes;
6578
6579         /* NB: Disregard the rule against updating
6580          * exp_connect_data.ocd_connect_flags in this case, since
6581          * tgt_client_new() needs to know if this is a lightweight
6582          * connection, and it is safe to expose this flag before
6583          * connection processing completes. */
6584         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
6585                 spin_lock(&exp->exp_lock);
6586                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
6587                 spin_unlock(&exp->exp_lock);
6588         }
6589
6590         data->ocd_version = LUSTRE_VERSION_CODE;
6591
6592         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
6593                 CWARN("%s: MDS requires FID support, but client not\n",
6594                       obd_name);
6595                 return -EBADE;
6596         }
6597
6598         if (OCD_HAS_FLAG(data, PINGLESS)) {
6599                 if (ptlrpc_pinger_suppress_pings()) {
6600                         spin_lock(&exp->exp_obd->obd_dev_lock);
6601                         list_del_init(&exp->exp_obd_chain_timed);
6602                         spin_unlock(&exp->exp_obd->obd_dev_lock);
6603                 } else {
6604                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
6605                 }
6606         }
6607
6608         data->ocd_max_easize = mdt->mdt_max_ea_size;
6609
6610         /* NB: Disregard the rule against updating
6611          * exp_connect_data.ocd_connect_flags in this case, since
6612          * tgt_client_new() needs to know if this is client supports
6613          * multiple modify RPCs, and it is safe to expose this flag before
6614          * connection processing completes. */
6615         if (data->ocd_connect_flags & OBD_CONNECT_MULTIMODRPCS) {
6616                 data->ocd_maxmodrpcs = max_mod_rpcs_per_client;
6617                 spin_lock(&exp->exp_lock);
6618                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_MULTIMODRPCS;
6619                 spin_unlock(&exp->exp_lock);
6620         }
6621
6622         if (OCD_HAS_FLAG(data, CKSUM)) {
6623                 __u32 cksum_types = data->ocd_cksum_types;
6624
6625                 tgt_mask_cksum_types(&mdt->mdt_lut, &data->ocd_cksum_types);
6626
6627                 if (unlikely(data->ocd_cksum_types == 0)) {
6628                         CERROR("%s: Connect with checksum support but no "
6629                                "ocd_cksum_types is set\n",
6630                                exp->exp_obd->obd_name);
6631                         RETURN(-EPROTO);
6632                 }
6633
6634                 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
6635                        "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
6636                        cksum_types, data->ocd_cksum_types);
6637         } else {
6638                 /* This client does not support OBD_CONNECT_CKSUM
6639                  * fall back to CRC32 */
6640                 CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
6641                        "OBD_CONNECT_CKSUM, CRC32 will be used\n",
6642                        exp->exp_obd->obd_name, obd_export_nid2str(exp));
6643         }
6644
6645         if ((data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
6646             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
6647                 atomic_inc(&mdt->mdt_mds_mds_conns);
6648                 mdt_enable_slc(mdt);
6649         }
6650
6651         if (!mdt->mdt_lut.lut_dt_conf.ddp_has_lseek_data_hole)
6652                 data->ocd_connect_flags2 &= ~OBD_CONNECT2_LSEEK;
6653
6654         return 0;
6655 }
6656
6657 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
6658                                    struct mdt_thread_info *info,
6659                                    struct mdt_file_data *mfd)
6660 {
6661         struct lu_context ses;
6662         int rc;
6663         ENTRY;
6664
6665         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
6666         if (rc)
6667                 RETURN(rc);
6668
6669         env->le_ses = &ses;
6670         lu_context_enter(&ses);
6671
6672         mdt_ucred(info)->uc_valid = UCRED_OLD;
6673         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
6674
6675         lu_context_exit(&ses);
6676         lu_context_fini(&ses);
6677         env->le_ses = NULL;
6678
6679         RETURN(rc);
6680 }
6681
6682 static int mdt_export_cleanup(struct obd_export *exp)
6683 {
6684         LIST_HEAD(closing_list);
6685         struct mdt_export_data  *med = &exp->exp_mdt_data;
6686         struct obd_device       *obd = exp->exp_obd;
6687         struct mdt_device       *mdt;
6688         struct mdt_thread_info  *info;
6689         struct lu_env            env;
6690         struct mdt_file_data    *mfd, *n;
6691         int rc = 0;
6692         ENTRY;
6693
6694         spin_lock(&med->med_open_lock);
6695         while (!list_empty(&med->med_open_head)) {
6696                 struct list_head *tmp = med->med_open_head.next;
6697                 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
6698
6699                 /* Remove mfd handle so it can't be found again.
6700                  * We are consuming the mfd_list reference here. */
6701                 class_handle_unhash(&mfd->mfd_open_handle);
6702                 list_move_tail(&mfd->mfd_list, &closing_list);
6703         }
6704         spin_unlock(&med->med_open_lock);
6705         mdt = mdt_dev(obd->obd_lu_dev);
6706         LASSERT(mdt != NULL);
6707
6708         rc = lu_env_init(&env, LCT_MD_THREAD);
6709         if (rc)
6710                 RETURN(rc);
6711
6712         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6713         LASSERT(info != NULL);
6714         memset(info, 0, sizeof *info);
6715         info->mti_env = &env;
6716         info->mti_mdt = mdt;
6717         info->mti_exp = exp;
6718
6719         if (!list_empty(&closing_list)) {
6720                 struct md_attr *ma = &info->mti_attr;
6721
6722                 /* Close any open files (which may also cause orphan
6723                  * unlinking). */
6724                 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
6725                         list_del_init(&mfd->mfd_list);
6726                         ma->ma_need = ma->ma_valid = 0;
6727
6728                         /* This file is being closed due to an eviction, it
6729                          * could have been modified and now dirty regarding to
6730                          * HSM archive, check this!
6731                          * The logic here is to mark a file dirty if there's a
6732                          * chance it was dirtied before the client was evicted,
6733                          * so that we don't have to wait for a release attempt
6734                          * before finding out the file was actually dirty and
6735                          * fail the release. Aggressively marking it dirty here
6736                          * will cause the policy engine to attempt to
6737                          * re-archive it; when rearchiving, we can compare the
6738                          * current version to the HSM data_version and make the
6739                          * archive request into a noop if it's not actually
6740                          * dirty.
6741                          */
6742                         if (mfd->mfd_open_flags & MDS_FMODE_WRITE)
6743                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
6744
6745                         /* Don't unlink orphan on failover umount, LU-184 */
6746                         if (exp->exp_flags & OBD_OPT_FAILOVER ||
6747                             exp->exp_obd->obd_stopping) {
6748                                 ma->ma_valid = MA_FLAGS;
6749                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
6750                         }
6751                         ma->ma_valid |= MA_FORCE_LOG;
6752                         mdt_mfd_close(info, mfd);
6753                 }
6754         }
6755         info->mti_mdt = NULL;
6756         /* cleanup client slot early */
6757         /* Do not erase record for recoverable client. */
6758         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
6759                 tgt_client_del(&env, exp);
6760         lu_env_fini(&env);
6761
6762         RETURN(rc);
6763 }
6764
6765 static int mdt_obd_disconnect(struct obd_export *exp)
6766 {
6767         int rc;
6768
6769         ENTRY;
6770
6771         LASSERT(exp);
6772         class_export_get(exp);
6773
6774         if (!(exp->exp_flags & OBD_OPT_FORCE))
6775                 tgt_grant_sanity_check(exp->exp_obd, __func__);
6776
6777         if ((exp_connect_flags(exp) & OBD_CONNECT_MDS_MDS) &&
6778             !(exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT)) {
6779                 struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
6780
6781                 if (atomic_dec_and_test(&mdt->mdt_mds_mds_conns))
6782                         mdt_disable_slc(mdt);
6783         }
6784
6785         rc = server_disconnect_export(exp);
6786         if (rc != 0)
6787                 CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
6788
6789         tgt_grant_discard(exp);
6790
6791         rc = mdt_export_cleanup(exp);
6792         nodemap_del_member(exp);
6793         class_export_put(exp);
6794         RETURN(rc);
6795 }
6796
6797 /* mds_connect copy */
6798 static int mdt_obd_connect(const struct lu_env *env,
6799                            struct obd_export **exp, struct obd_device *obd,
6800                            struct obd_uuid *cluuid,
6801                            struct obd_connect_data *data,
6802                            void *localdata)
6803 {
6804         struct obd_export       *lexp;
6805         struct lustre_handle    conn = { 0 };
6806         struct mdt_device       *mdt;
6807         int                      rc;
6808         lnet_nid_t              *client_nid = localdata;
6809         ENTRY;
6810
6811         LASSERT(env != NULL);
6812         LASSERT(data != NULL);
6813
6814         if (!exp || !obd || !cluuid)
6815                 RETURN(-EINVAL);
6816
6817         mdt = mdt_dev(obd->obd_lu_dev);
6818
6819         /*
6820          * first, check whether the stack is ready to handle requests
6821          * XXX: probably not very appropriate method is used now
6822          *      at some point we should find a better one
6823          */
6824         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) &&
6825             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) &&
6826             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
6827                 rc = obd_get_info(env, mdt->mdt_child_exp,
6828                                   sizeof(KEY_OSP_CONNECTED),
6829                                   KEY_OSP_CONNECTED, NULL, NULL);
6830                 if (rc)
6831                         RETURN(-EAGAIN);
6832                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
6833         }
6834
6835         rc = class_connect(&conn, obd, cluuid);
6836         if (rc)
6837                 RETURN(rc);
6838
6839         lexp = class_conn2export(&conn);
6840         LASSERT(lexp != NULL);
6841
6842         rc = nodemap_add_member(*client_nid, lexp);
6843         if (rc != 0 && rc != -EEXIST)
6844                 GOTO(out, rc);
6845
6846         rc = mdt_connect_internal(env, lexp, mdt, data, false);
6847         if (rc == 0) {
6848                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
6849
6850                 LASSERT(lcd);
6851                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
6852                 rc = tgt_client_new(env, lexp);
6853                 if (rc == 0)
6854                         mdt_export_stats_init(obd, lexp, localdata);
6855         }
6856 out:
6857         if (rc != 0) {
6858                 class_disconnect(lexp);
6859                 nodemap_del_member(lexp);
6860                 *exp = NULL;
6861         } else {
6862                 *exp = lexp;
6863                 /* Because we do not want this export to be evicted by pinger,
6864                  * let's not add this export to the timed chain list. */
6865                 if (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) {
6866                         spin_lock(&lexp->exp_obd->obd_dev_lock);
6867                         list_del_init(&lexp->exp_obd_chain_timed);
6868                         spin_unlock(&lexp->exp_obd->obd_dev_lock);
6869                 }
6870         }
6871
6872         RETURN(rc);
6873 }
6874
6875 static int mdt_obd_reconnect(const struct lu_env *env,
6876                              struct obd_export *exp, struct obd_device *obd,
6877                              struct obd_uuid *cluuid,
6878                              struct obd_connect_data *data,
6879                              void *localdata)
6880 {
6881         lnet_nid_t             *client_nid = localdata;
6882         int                     rc;
6883         ENTRY;
6884
6885         if (exp == NULL || obd == NULL || cluuid == NULL)
6886                 RETURN(-EINVAL);
6887
6888         rc = nodemap_add_member(*client_nid, exp);
6889         if (rc != 0 && rc != -EEXIST)
6890                 RETURN(rc);
6891
6892         rc = mdt_connect_internal(env, exp, mdt_dev(obd->obd_lu_dev), data,
6893                                   true);
6894         if (rc == 0)
6895                 mdt_export_stats_init(obd, exp, localdata);
6896         else
6897                 nodemap_del_member(exp);
6898
6899         RETURN(rc);
6900 }
6901
6902 /* FIXME: Can we avoid using these two interfaces? */
6903 static int mdt_init_export(struct obd_export *exp)
6904 {
6905         struct mdt_export_data *med = &exp->exp_mdt_data;
6906         int                     rc;
6907         ENTRY;
6908
6909         INIT_LIST_HEAD(&med->med_open_head);
6910         spin_lock_init(&med->med_open_lock);
6911         spin_lock(&exp->exp_lock);
6912         exp->exp_connecting = 1;
6913         spin_unlock(&exp->exp_lock);
6914
6915         OBD_ALLOC(exp->exp_used_slots,
6916                   BITS_TO_LONGS(OBD_MAX_RIF_MAX) * sizeof(long));
6917         if (exp->exp_used_slots == NULL)
6918                 RETURN(-ENOMEM);
6919
6920         /* self-export doesn't need client data and ldlm initialization */
6921         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
6922                                      &exp->exp_client_uuid)))
6923                 RETURN(0);
6924
6925         rc = tgt_client_alloc(exp);
6926         if (rc)
6927                 GOTO(err, rc);
6928
6929         rc = ldlm_init_export(exp);
6930         if (rc)
6931                 GOTO(err_free, rc);
6932
6933         RETURN(rc);
6934
6935 err_free:
6936         tgt_client_free(exp);
6937 err:
6938         OBD_FREE(exp->exp_used_slots,
6939                  BITS_TO_LONGS(OBD_MAX_RIF_MAX) * sizeof(long));
6940         exp->exp_used_slots = NULL;
6941
6942         CERROR("%s: Failed to initialize export: rc = %d\n",
6943                exp->exp_obd->obd_name, rc);
6944         return rc;
6945 }
6946
6947 static int mdt_destroy_export(struct obd_export *exp)
6948 {
6949         ENTRY;
6950
6951         target_destroy_export(exp);
6952         if (exp->exp_used_slots)
6953                 OBD_FREE(exp->exp_used_slots,
6954                          BITS_TO_LONGS(OBD_MAX_RIF_MAX) * sizeof(long));
6955
6956         /* destroy can be called from failed obd_setup, so
6957          * checking uuid is safer than obd_self_export */
6958         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
6959                                      &exp->exp_client_uuid)))
6960                 RETURN(0);
6961
6962         ldlm_destroy_export(exp);
6963         tgt_client_free(exp);
6964
6965         LASSERT(list_empty(&exp->exp_outstanding_replies));
6966         LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
6967
6968         /*
6969          * discard grants once we're sure no more
6970          * interaction with the client is possible
6971          */
6972         tgt_grant_discard(exp);
6973         if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
6974                 exp->exp_obd->u.obt.obt_lut->lut_tgd.tgd_tot_granted_clients--;
6975
6976         if (!(exp->exp_flags & OBD_OPT_FORCE))
6977                 tgt_grant_sanity_check(exp->exp_obd, __func__);
6978
6979         RETURN(0);
6980 }
6981
6982 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
6983                    struct linkea_data *ldata)
6984 {
6985         int rc;
6986
6987         LASSERT(ldata->ld_buf->lb_buf != NULL);
6988
6989         if (!mdt_object_exists(mdt_obj))
6990                 return -ENODATA;
6991
6992         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6993                           ldata->ld_buf, XATTR_NAME_LINK);
6994         if (rc == -ERANGE) {
6995                 /* Buf was too small, figure out what we need. */
6996                 lu_buf_free(ldata->ld_buf);
6997                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6998                                   ldata->ld_buf, XATTR_NAME_LINK);
6999                 if (rc < 0)
7000                         return rc;
7001                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
7002                 if (ldata->ld_buf->lb_buf == NULL)
7003                         return -ENOMEM;
7004                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
7005                                   ldata->ld_buf, XATTR_NAME_LINK);
7006         }
7007         if (rc < 0)
7008                 return rc;
7009
7010         return linkea_init_with_rec(ldata);
7011 }
7012
7013 /**
7014  * Given an MDT object, try to look up the full path to the object.
7015  * Part of the MDT layer implementation of lfs fid2path.
7016  *
7017  * \param[in]     info  Per-thread common data shared by MDT level handlers.
7018  * \param[in]     obj   Object to do path lookup of
7019  * \param[in,out] fp    User-provided struct to store path information
7020  * \param[in]     root_fid Root FID of current path should reach
7021  *
7022  * \retval 0 Lookup successful, path information stored in fp
7023  * \retval -EAGAIN Lookup failed, usually because object is being moved
7024  * \retval negative errno if there was a problem
7025  */
7026 static int mdt_path_current(struct mdt_thread_info *info,
7027                             struct mdt_object *obj,
7028                             struct getinfo_fid2path *fp,
7029                             struct lu_fid *root_fid)
7030 {
7031         struct mdt_device *mdt = info->mti_mdt;
7032         struct lu_name *tmpname = &info->mti_name;
7033         struct lu_fid *tmpfid = &info->mti_tmp_fid1;
7034         struct lu_buf *buf = &info->mti_big_buf;
7035         struct linkea_data ldata = { NULL };
7036         bool first = true;
7037         struct mdt_object *mdt_obj;
7038         struct link_ea_header *leh;
7039         struct link_ea_entry *lee;
7040         char *ptr;
7041         int reclen;
7042         int rc = 0;
7043
7044         ENTRY;
7045
7046         /* temp buffer for path element, the buffer will be finally freed
7047          * in mdt_thread_info_fini */
7048         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
7049         if (buf->lb_buf == NULL)
7050                 RETURN(-ENOMEM);
7051
7052         ldata.ld_buf = buf;
7053         ptr = fp->gf_u.gf_path + fp->gf_pathlen - 1;
7054         *ptr = 0;
7055         --ptr;
7056         *tmpfid = fp->gf_fid = *mdt_object_fid(obj);
7057
7058         while (!lu_fid_eq(root_fid, &fp->gf_fid)) {
7059                 if (!lu_fid_eq(root_fid, &mdt->mdt_md_root_fid) &&
7060                     lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid))
7061                         GOTO(out, rc = -ENOENT);
7062
7063                 if (lu_fid_eq(mdt_object_fid(obj), tmpfid)) {
7064                         mdt_obj = obj;
7065                         mdt_object_get(info->mti_env, mdt_obj);
7066                 } else {
7067                         mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
7068                         if (IS_ERR(mdt_obj))
7069                                 GOTO(out, rc = PTR_ERR(mdt_obj));
7070                 }
7071
7072                 if (!mdt_object_exists(mdt_obj)) {
7073                         mdt_object_put(info->mti_env, mdt_obj);
7074                         GOTO(out, rc = -ENOENT);
7075                 }
7076
7077                 if (mdt_object_remote(mdt_obj)) {
7078                         mdt_object_put(info->mti_env, mdt_obj);
7079                         GOTO(remote_out, rc = -EREMOTE);
7080                 }
7081
7082                 rc = mdt_links_read(info, mdt_obj, &ldata);
7083                 if (rc != 0) {
7084                         mdt_object_put(info->mti_env, mdt_obj);
7085                         GOTO(out, rc);
7086                 }
7087
7088                 leh = buf->lb_buf;
7089                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
7090                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
7091                 /* If set, use link #linkno for path lookup, otherwise use
7092                    link #0.  Only do this for the final path element. */
7093                 if (first && fp->gf_linkno < leh->leh_reccount) {
7094                         int count;
7095                         for (count = 0; count < fp->gf_linkno; count++) {
7096                                 lee = (struct link_ea_entry *)
7097                                      ((char *)lee + reclen);
7098                                 linkea_entry_unpack(lee, &reclen, tmpname,
7099                                                     tmpfid);
7100                         }
7101                         if (fp->gf_linkno < leh->leh_reccount - 1)
7102                                 /* indicate to user there are more links */
7103                                 fp->gf_linkno++;
7104                 }
7105
7106                 /* Check if it is slave stripes */
7107                 rc = mdt_is_dir_stripe(info, mdt_obj);
7108                 mdt_object_put(info->mti_env, mdt_obj);
7109                 if (rc < 0)
7110                         GOTO(out, rc);
7111                 if (rc == 1) {
7112                         fp->gf_fid = *tmpfid;
7113                         continue;
7114                 }
7115
7116                 /* Pack the name in the end of the buffer */
7117                 ptr -= tmpname->ln_namelen;
7118                 if (ptr - 1 <= fp->gf_u.gf_path)
7119                         GOTO(out, rc = -EOVERFLOW);
7120                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
7121                 *(--ptr) = '/';
7122
7123                 /* keep the last resolved fid to the client, so the
7124                  * client will build the left path on another MDT for
7125                  * remote object */
7126                 fp->gf_fid = *tmpfid;
7127
7128                 first = false;
7129         }
7130
7131         /* non-zero will be treated as an error */
7132         rc = 0;
7133
7134 remote_out:
7135         ptr++; /* skip leading / */
7136         memmove(fp->gf_u.gf_path, ptr,
7137                 fp->gf_u.gf_path + fp->gf_pathlen - ptr);
7138
7139 out:
7140         RETURN(rc);
7141 }
7142
7143 /**
7144  * Given an MDT object, use mdt_path_current to get the path.
7145  * Essentially a wrapper to retry mdt_path_current a set number of times
7146  * if -EAGAIN is returned (usually because an object is being moved).
7147  *
7148  * Part of the MDT layer implementation of lfs fid2path.
7149  *
7150  * \param[in]     info  Per-thread common data shared by mdt level handlers.
7151  * \param[in]     obj   Object to do path lookup of
7152  * \param[in,out] fp    User-provided struct for arguments and to store path
7153  *                      information
7154  *
7155  * \retval 0 Lookup successful, path information stored in fp
7156  * \retval negative errno if there was a problem
7157  */
7158 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
7159                     struct getinfo_fid2path *fp, struct lu_fid *root_fid)
7160 {
7161         struct mdt_device       *mdt = info->mti_mdt;
7162         int                     tries = 3;
7163         int                     rc = -EAGAIN;
7164         ENTRY;
7165
7166         if (fp->gf_pathlen < 3)
7167                 RETURN(-EOVERFLOW);
7168
7169         if (root_fid == NULL)
7170                 root_fid = &mdt->mdt_md_root_fid;
7171
7172         if (lu_fid_eq(root_fid, mdt_object_fid(obj))) {
7173                 fp->gf_u.gf_path[0] = '\0';
7174                 RETURN(0);
7175         }
7176
7177         /* Retry multiple times in case file is being moved */
7178         while (tries-- && rc == -EAGAIN)
7179                 rc = mdt_path_current(info, obj, fp, root_fid);
7180
7181         RETURN(rc);
7182 }
7183
7184 /**
7185  * Get the full path of the provided FID, as of changelog record recno.
7186  *
7187  * This checks sanity and looks up object for user provided FID
7188  * before calling the actual path lookup code.
7189  *
7190  * Part of the MDT layer implementation of lfs fid2path.
7191  *
7192  * \param[in]     info  Per-thread common data shared by mdt level handlers.
7193  * \param[in,out] fp    User-provided struct for arguments and to store path
7194  *                      information
7195  *
7196  * \retval 0 Lookup successful, path information and recno stored in fp
7197  * \retval -ENOENT, object does not exist
7198  * \retval negative errno if there was a problem
7199  */
7200 static int mdt_fid2path(struct mdt_thread_info *info,
7201                         struct lu_fid *root_fid,
7202                         struct getinfo_fid2path *fp)
7203 {
7204         struct mdt_device *mdt = info->mti_mdt;
7205         struct mdt_object *obj;
7206         int    rc;
7207         ENTRY;
7208
7209         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
7210                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
7211
7212         if (!fid_is_sane(&fp->gf_fid))
7213                 RETURN(-EINVAL);
7214
7215         if (!fid_is_namespace_visible(&fp->gf_fid)) {
7216                 CDEBUG(D_INFO, "%s: "DFID" is invalid, f_seq should be >= %#llx"
7217                        ", or f_oid != 0, or f_ver == 0\n", mdt_obd_name(mdt),
7218                        PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
7219                 RETURN(-EINVAL);
7220         }
7221
7222         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
7223         if (IS_ERR(obj)) {
7224                 rc = PTR_ERR(obj);
7225                 CDEBUG(D_IOCTL, "cannot find "DFID": rc = %d\n",
7226                        PFID(&fp->gf_fid), rc);
7227                 RETURN(rc);
7228         }
7229
7230         if (mdt_object_remote(obj)) {
7231                 rc = -EREMOTE;
7232         } else if (!mdt_object_exists(obj)) {
7233                 rc = -ENOENT;
7234         } else {
7235                 struct lu_attr la = { 0 };
7236                 struct dt_object *dt = mdt_obj2dt(obj);
7237
7238                 if (dt && dt->do_ops && dt->do_ops->do_attr_get)
7239                         dt_attr_get(info->mti_env, mdt_obj2dt(obj), &la);
7240                 if (la.la_valid & LA_FLAGS && la.la_flags & LUSTRE_ENCRYPT_FL)
7241                         /* path resolution cannot be carried out on server
7242                          * side for encrypted files
7243                          */
7244                         rc = -ENODATA;
7245                 else
7246                         rc = 0;
7247         }
7248
7249         if (rc < 0) {
7250                 mdt_object_put(info->mti_env, obj);
7251                 CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n",
7252                        PFID(&fp->gf_fid), rc);
7253                 RETURN(rc);
7254         }
7255
7256         rc = mdt_path(info, obj, fp, root_fid);
7257
7258         CDEBUG(D_INFO, "fid "DFID", path %s recno %#llx linkno %u\n",
7259                PFID(&fp->gf_fid), fp->gf_u.gf_path,
7260                fp->gf_recno, fp->gf_linkno);
7261
7262         mdt_object_put(info->mti_env, obj);
7263
7264         RETURN(rc);
7265 }
7266
7267 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, int keylen,
7268                             void *val, int vallen)
7269 {
7270         struct getinfo_fid2path *fpout, *fpin;
7271         struct lu_fid *root_fid = NULL;
7272         int rc = 0;
7273
7274         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
7275         fpout = val;
7276
7277         if (req_capsule_req_need_swab(info->mti_pill))
7278                 lustre_swab_fid2path(fpin);
7279
7280         memcpy(fpout, fpin, sizeof(*fpin));
7281         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
7282                 RETURN(-EINVAL);
7283
7284         if (keylen >= cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*fpin) +
7285                       sizeof(struct lu_fid)) {
7286                 /* client sent its root FID, which is normally fileset FID */
7287                 root_fid = fpin->gf_u.gf_root_fid;
7288                 if (req_capsule_req_need_swab(info->mti_pill))
7289                         lustre_swab_lu_fid(root_fid);
7290
7291                 if (root_fid != NULL && !fid_is_sane(root_fid))
7292                         RETURN(-EINVAL);
7293         }
7294
7295         rc = mdt_fid2path(info, root_fid, fpout);
7296         RETURN(rc);
7297 }
7298
7299 int mdt_get_info(struct tgt_session_info *tsi)
7300 {
7301         char    *key;
7302         int      keylen;
7303         __u32   *vallen;
7304         void    *valout;
7305         int      rc;
7306
7307         ENTRY;
7308
7309         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
7310         if (key == NULL) {
7311                 CDEBUG(D_IOCTL, "No GETINFO key\n");
7312                 RETURN(err_serious(-EFAULT));
7313         }
7314         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
7315                                       RCL_CLIENT);
7316
7317         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
7318         if (vallen == NULL) {
7319                 CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n",
7320                                 tgt_name(tsi->tsi_tgt));
7321                 RETURN(err_serious(-EFAULT));
7322         }
7323
7324         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
7325                              *vallen);
7326         rc = req_capsule_server_pack(tsi->tsi_pill);
7327         if (rc)
7328                 RETURN(err_serious(rc));
7329
7330         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
7331         if (valout == NULL) {
7332                 CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n",
7333                                 tgt_name(tsi->tsi_tgt));
7334                 RETURN(err_serious(-EFAULT));
7335         }
7336
7337         if (KEY_IS(KEY_FID2PATH)) {
7338                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
7339
7340                 rc = mdt_rpc_fid2path(info, key, keylen, valout, *vallen);
7341                 mdt_thread_info_fini(info);
7342         } else {
7343                 rc = -EINVAL;
7344         }
7345         RETURN(rc);
7346 }
7347
7348 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
7349 {
7350         struct obd_ioctl_data *data = karg;
7351         struct lu_fid *fid;
7352         __u64 version;
7353         struct mdt_object *obj;
7354         struct mdt_lock_handle  *lh;
7355         int rc;
7356         ENTRY;
7357
7358         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
7359             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
7360                 RETURN(-EINVAL);
7361
7362         fid = (struct lu_fid *)data->ioc_inlbuf1;
7363
7364         if (!fid_is_sane(fid))
7365                 RETURN(-EINVAL);
7366
7367         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
7368
7369         lh = &mti->mti_lh[MDT_LH_PARENT];
7370         mdt_lock_reg_init(lh, LCK_CR);
7371
7372         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
7373         if (IS_ERR(obj))
7374                 RETURN(PTR_ERR(obj));
7375
7376         if (mdt_object_remote(obj)) {
7377                 rc = -EREMOTE;
7378                 /**
7379                  * before calling version get the correct MDS should be
7380                  * fid, this is error to find remote object here
7381                  */
7382                 CERROR("nonlocal object "DFID"\n", PFID(fid));
7383         } else if (!mdt_object_exists(obj)) {
7384                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
7385                 rc = -ENOENT;
7386         } else {
7387                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
7388                *(__u64 *)data->ioc_inlbuf2 = version;
7389                 rc = 0;
7390         }
7391         mdt_object_unlock_put(mti, obj, lh, 1);
7392         RETURN(rc);
7393 }
7394
7395 /* ioctls on obd dev */
7396 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
7397                          void *karg, void __user *uarg)
7398 {
7399         struct lu_env      env;
7400         struct obd_device *obd = exp->exp_obd;
7401         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
7402         struct dt_device  *dt = mdt->mdt_bottom;
7403         int rc;
7404
7405         ENTRY;
7406         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
7407         rc = lu_env_init(&env, LCT_MD_THREAD);
7408         if (rc)
7409                 RETURN(rc);
7410
7411         switch (cmd) {
7412         case OBD_IOC_SYNC:
7413                 rc = mdt_device_sync(&env, mdt);
7414                 break;
7415         case OBD_IOC_SET_READONLY:
7416                 rc = dt_sync(&env, dt);
7417                 if (rc == 0)
7418                         rc = dt_ro(&env, dt);
7419                 break;
7420         case OBD_IOC_ABORT_RECOVERY: {
7421                 struct obd_ioctl_data *data = karg;
7422
7423                 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
7424                 if (data->ioc_type & OBD_FLG_ABORT_RECOV_MDT) {
7425                         obd->obd_abort_recov_mdt = 1;
7426                         wake_up(&obd->obd_next_transno_waitq);
7427                 } else { /* if (data->ioc_type & OBD_FLG_ABORT_RECOV_OST) */
7428                         /* lctl didn't set OBD_FLG_ABORT_RECOV_OST < 2.13.57 */
7429                         obd->obd_abort_recovery = 1;
7430                         target_stop_recovery_thread(obd);
7431                 }
7432                 rc = 0;
7433                 break;
7434         }
7435         case OBD_IOC_CHANGELOG_REG:
7436         case OBD_IOC_CHANGELOG_DEREG:
7437         case OBD_IOC_CHANGELOG_CLEAR:
7438                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(&env,
7439                                                            mdt->mdt_child,
7440                                                            cmd, len, karg);
7441                 break;
7442         case OBD_IOC_START_LFSCK: {
7443                 struct md_device *next = mdt->mdt_child;
7444                 struct obd_ioctl_data *data = karg;
7445                 struct lfsck_start_param lsp;
7446
7447                 if (unlikely(data == NULL)) {
7448                         rc = -EINVAL;
7449                         break;
7450                 }
7451
7452                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
7453                 lsp.lsp_index_valid = 0;
7454                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
7455                 break;
7456         }
7457         case OBD_IOC_STOP_LFSCK: {
7458                 struct md_device        *next = mdt->mdt_child;
7459                 struct obd_ioctl_data   *data = karg;
7460                 struct lfsck_stop        stop;
7461
7462                 stop.ls_status = LS_STOPPED;
7463                 /* Old lfsck utils may pass NULL @stop. */
7464                 if (data->ioc_inlbuf1 == NULL)
7465                         stop.ls_flags = 0;
7466                 else
7467                         stop.ls_flags =
7468                         ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
7469
7470                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
7471                 break;
7472         }
7473         case OBD_IOC_QUERY_LFSCK: {
7474                 struct md_device        *next = mdt->mdt_child;
7475                 struct obd_ioctl_data   *data = karg;
7476
7477                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0,
7478                                                  data->ioc_inlbuf1);
7479                 break;
7480         }
7481         case OBD_IOC_GET_OBJ_VERSION: {
7482                 struct mdt_thread_info *mti;
7483                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
7484                 memset(mti, 0, sizeof *mti);
7485                 mti->mti_env = &env;
7486                 mti->mti_mdt = mdt;
7487                 mti->mti_exp = exp;
7488
7489                 rc = mdt_ioc_version_get(mti, karg);
7490                 break;
7491         }
7492         case OBD_IOC_CATLOGLIST: {
7493                 struct mdt_thread_info *mti;
7494
7495                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
7496                 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
7497                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
7498                                        &mti->mti_tmp_fid1);
7499                 break;
7500          }
7501         default:
7502                 rc = -EOPNOTSUPP;
7503                 CERROR("%s: Not supported cmd = %d, rc = %d\n",
7504                         mdt_obd_name(mdt), cmd, rc);
7505         }
7506
7507         lu_env_fini(&env);
7508         RETURN(rc);
7509 }
7510
7511 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
7512 {
7513         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
7514         int rc;
7515         ENTRY;
7516
7517         if (!mdt->mdt_skip_lfsck && !mdt->mdt_bottom->dd_rdonly) {
7518                 struct lfsck_start_param lsp;
7519
7520                 lsp.lsp_start = NULL;
7521                 lsp.lsp_index_valid = 0;
7522                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
7523                                                            OBD_IOC_START_LFSCK,
7524                                                            0, &lsp);
7525                 if (rc != 0 && rc != -EALREADY)
7526                         CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
7527                               mdt_obd_name(mdt), rc);
7528         }
7529
7530         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
7531         RETURN(rc);
7532 }
7533
7534 static int mdt_obd_postrecov(struct obd_device *obd)
7535 {
7536         struct lu_env env;
7537         int rc;
7538
7539         rc = lu_env_init(&env, LCT_MD_THREAD);
7540         if (rc)
7541                 RETURN(rc);
7542         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
7543         lu_env_fini(&env);
7544         return rc;
7545 }
7546
7547 static const struct obd_ops mdt_obd_device_ops = {
7548         .o_owner          = THIS_MODULE,
7549         .o_set_info_async = mdt_obd_set_info_async,
7550         .o_connect        = mdt_obd_connect,
7551         .o_reconnect      = mdt_obd_reconnect,
7552         .o_disconnect     = mdt_obd_disconnect,
7553         .o_init_export    = mdt_init_export,
7554         .o_destroy_export = mdt_destroy_export,
7555         .o_iocontrol      = mdt_iocontrol,
7556         .o_postrecov      = mdt_obd_postrecov,
7557         /* Data-on-MDT IO methods */
7558         .o_preprw         = mdt_obd_preprw,
7559         .o_commitrw       = mdt_obd_commitrw,
7560 };
7561
7562 static struct lu_device* mdt_device_fini(const struct lu_env *env,
7563                                          struct lu_device *d)
7564 {
7565         struct mdt_device *m = mdt_dev(d);
7566         ENTRY;
7567
7568         mdt_fini(env, m);
7569         RETURN(NULL);
7570 }
7571
7572 static struct lu_device *mdt_device_free(const struct lu_env *env,
7573                                          struct lu_device *d)
7574 {
7575         struct mdt_device *m = mdt_dev(d);
7576         ENTRY;
7577
7578         lu_device_fini(&m->mdt_lu_dev);
7579         OBD_FREE_PTR(m);
7580
7581         RETURN(NULL);
7582 }
7583
7584 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
7585                                           struct lu_device_type *t,
7586                                           struct lustre_cfg *cfg)
7587 {
7588         struct lu_device  *l;
7589         struct mdt_device *m;
7590
7591         OBD_ALLOC_PTR(m);
7592         if (m != NULL) {
7593                 int rc;
7594
7595                 l = &m->mdt_lu_dev;
7596                 rc = mdt_init0(env, m, t, cfg);
7597                 if (rc != 0) {
7598                         mdt_device_free(env, l);
7599                         l = ERR_PTR(rc);
7600                         return l;
7601                 }
7602         } else
7603                 l = ERR_PTR(-ENOMEM);
7604         return l;
7605 }
7606
7607 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
7608 LU_KEY_INIT(mdt, struct mdt_thread_info);
7609
7610 static void mdt_key_fini(const struct lu_context *ctx,
7611                          struct lu_context_key *key, void* data)
7612 {
7613         struct mdt_thread_info *info = data;
7614
7615         if (info->mti_big_lmm) {
7616                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
7617                 info->mti_big_lmm = NULL;
7618                 info->mti_big_lmmsize = 0;
7619         }
7620
7621         if (info->mti_big_acl) {
7622                 OBD_FREE_LARGE(info->mti_big_acl, info->mti_big_aclsize);
7623                 info->mti_big_acl = NULL;
7624                 info->mti_big_aclsize = 0;
7625         }
7626
7627         OBD_FREE_PTR(info);
7628 }
7629
7630 /* context key: mdt_thread_key */
7631 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
7632
7633 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
7634 {
7635         return lu_ucred(info->mti_env);
7636 }
7637
7638 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
7639 {
7640         return lu_ucred_check(info->mti_env);
7641 }
7642
7643 /**
7644  * Enable/disable COS (Commit On Sharing).
7645  *
7646  * Set/Clear the COS flag in mdt options.
7647  *
7648  * \param mdt mdt device
7649  * \param val 0 disables COS, other values enable COS
7650  */
7651 void mdt_enable_cos(struct mdt_device *mdt, bool val)
7652 {
7653         struct lu_env env;
7654         int rc;
7655
7656         mdt->mdt_opts.mo_cos = val;
7657         rc = lu_env_init(&env, LCT_LOCAL);
7658         if (unlikely(rc != 0)) {
7659                 CWARN("%s: lu_env initialization failed, cannot "
7660                       "sync: rc = %d\n", mdt_obd_name(mdt), rc);
7661                 return;
7662         }
7663         mdt_device_sync(&env, mdt);
7664         lu_env_fini(&env);
7665 }
7666
7667 /**
7668  * Check COS (Commit On Sharing) status.
7669  *
7670  * Return COS flag status.
7671  *
7672  * \param mdt mdt device
7673  */
7674 int mdt_cos_is_enabled(struct mdt_device *mdt)
7675 {
7676         return mdt->mdt_opts.mo_cos != 0;
7677 }
7678
7679 static const struct lu_device_type_operations mdt_device_type_ops = {
7680         .ldto_device_alloc = mdt_device_alloc,
7681         .ldto_device_free  = mdt_device_free,
7682         .ldto_device_fini  = mdt_device_fini
7683 };
7684
7685 static struct lu_device_type mdt_device_type = {
7686         .ldt_tags     = LU_DEVICE_MD,
7687         .ldt_name     = LUSTRE_MDT_NAME,
7688         .ldt_ops      = &mdt_device_type_ops,
7689         .ldt_ctx_tags = LCT_MD_THREAD
7690 };
7691
7692 static int __init mdt_init(void)
7693 {
7694         int rc;
7695
7696         BUILD_BUG_ON(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") !=
7697                      FID_NOBRACE_LEN + 1);
7698         BUILD_BUG_ON(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") !=
7699                      FID_LEN + 1);
7700         rc = lu_kmem_init(mdt_caches);
7701         if (rc)
7702                 return rc;
7703
7704         rc = mds_mod_init();
7705         if (rc)
7706                 GOTO(lu_fini, rc);
7707
7708         rc = class_register_type(&mdt_obd_device_ops, NULL, true,
7709                                  LUSTRE_MDT_NAME, &mdt_device_type);
7710         if (rc)
7711                 GOTO(mds_fini, rc);
7712 lu_fini:
7713         if (rc)
7714                 lu_kmem_fini(mdt_caches);
7715 mds_fini:
7716         if (rc)
7717                 mds_mod_exit();
7718         return rc;
7719 }
7720
7721 static void __exit mdt_exit(void)
7722 {
7723         class_unregister_type(LUSTRE_MDT_NAME);
7724         mds_mod_exit();
7725         lu_kmem_fini(mdt_caches);
7726 }
7727
7728 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
7729 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
7730 MODULE_VERSION(LUSTRE_VERSION_STRING);
7731 MODULE_LICENSE("GPL");
7732
7733 module_init(mdt_init);
7734 module_exit(mdt_exit);