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         if (rc == -ENOENT) {
2400                 /* return -ENOKEY instead of -ENOENT to encryption-unaware
2401                  * client if trying to access an encrypted file
2402                  */
2403                 int rc2 = mdt_check_enc(info, parent);
2404
2405                 if (rc2)
2406                         rc = rc2;
2407         }
2408         return rc;
2409 }
2410
2411 /* normal handler: should release the child lock */
2412 static int mdt_getattr_name(struct tgt_session_info *tsi)
2413 {
2414         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
2415         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
2416         struct mdt_body *reqbody;
2417         struct mdt_body *repbody;
2418         int rc, rc2;
2419
2420         ENTRY;
2421
2422         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
2423         LASSERT(reqbody != NULL);
2424         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
2425         LASSERT(repbody != NULL);
2426
2427         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
2428         repbody->mbo_eadatasize = 0;
2429         repbody->mbo_aclsize = 0;
2430
2431         rc = mdt_init_ucred(info, reqbody);
2432         if (unlikely(rc))
2433                 GOTO(out_shrink, rc);
2434
2435         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
2436         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
2437                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
2438                 lhc->mlh_reg_lh.cookie = 0;
2439         }
2440         mdt_exit_ucred(info);
2441         EXIT;
2442 out_shrink:
2443         mdt_client_compatibility(info);
2444         rc2 = mdt_fix_reply(info);
2445         if (rc == 0)
2446                 rc = rc2;
2447         mdt_thread_info_fini(info);
2448         return rc;
2449 }
2450
2451 static int mdt_rmfid_unlink(struct mdt_thread_info *info,
2452                             const struct lu_fid *pfid,
2453                             const struct lu_name *name,
2454                             struct mdt_object *obj, s64 ctime)
2455 {
2456         struct lu_fid *child_fid = &info->mti_tmp_fid1;
2457         struct ldlm_enqueue_info *einfo = &info->mti_einfo[0];
2458         struct mdt_device *mdt = info->mti_mdt;
2459         struct md_attr *ma = &info->mti_attr;
2460         struct mdt_lock_handle *parent_lh;
2461         struct mdt_lock_handle *child_lh;
2462         struct mdt_object *pobj;
2463         bool cos_incompat = false;
2464         int rc;
2465         ENTRY;
2466
2467         pobj = mdt_object_find(info->mti_env, mdt, pfid);
2468         if (IS_ERR(pobj))
2469                 GOTO(out, rc = PTR_ERR(pobj));
2470
2471         parent_lh = &info->mti_lh[MDT_LH_PARENT];
2472         mdt_lock_pdo_init(parent_lh, LCK_PW, name);
2473         rc = mdt_object_lock(info, pobj, parent_lh, MDS_INODELOCK_UPDATE);
2474         if (rc != 0)
2475                 GOTO(put_parent, rc);
2476
2477         if (mdt_object_remote(pobj))
2478                 cos_incompat = true;
2479
2480         rc = mdo_lookup(info->mti_env, mdt_object_child(pobj),
2481                         name, child_fid, &info->mti_spec);
2482         if (rc != 0)
2483                 GOTO(unlock_parent, rc);
2484
2485         if (!lu_fid_eq(child_fid, mdt_object_fid(obj)))
2486                 GOTO(unlock_parent, rc = -EREMCHG);
2487
2488         child_lh = &info->mti_lh[MDT_LH_CHILD];
2489         mdt_lock_reg_init(child_lh, LCK_EX);
2490         rc = mdt_reint_striped_lock(info, obj, child_lh,
2491                                     MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE,
2492                                     einfo, cos_incompat);
2493         if (rc != 0)
2494                 GOTO(unlock_parent, rc);
2495
2496         if (atomic_read(&obj->mot_open_count)) {
2497                 CDEBUG(D_OTHER, "object "DFID" open, skip\n",
2498                        PFID(mdt_object_fid(obj)));
2499                 GOTO(unlock_child, rc = -EBUSY);
2500         }
2501
2502         ma->ma_need = 0;
2503         ma->ma_valid = MA_INODE;
2504         ma->ma_attr.la_valid = LA_CTIME;
2505         ma->ma_attr.la_ctime = ctime;
2506
2507         mutex_lock(&obj->mot_lov_mutex);
2508
2509         rc = mdo_unlink(info->mti_env, mdt_object_child(pobj),
2510                         mdt_object_child(obj), name, ma, 0);
2511
2512         mutex_unlock(&obj->mot_lov_mutex);
2513
2514 unlock_child:
2515         mdt_reint_striped_unlock(info, obj, child_lh, einfo, 1);
2516 unlock_parent:
2517         mdt_object_unlock(info, pobj, parent_lh, 1);
2518 put_parent:
2519         mdt_object_put(info->mti_env, pobj);
2520 out:
2521         RETURN(rc);
2522 }
2523
2524 static int mdt_rmfid_check_permission(struct mdt_thread_info *info,
2525                                         struct mdt_object *obj)
2526 {
2527         struct lu_ucred *uc = lu_ucred(info->mti_env);
2528         struct md_attr *ma = &info->mti_attr;
2529         struct lu_attr *la = &ma->ma_attr;
2530         int rc = 0;
2531         ENTRY;
2532
2533         ma->ma_need = MA_INODE;
2534         rc = mo_attr_get(info->mti_env, mdt_object_child(obj), ma);
2535         if (rc)
2536                 GOTO(out, rc);
2537
2538         if (la->la_flags & LUSTRE_IMMUTABLE_FL)
2539                         rc = -EACCES;
2540
2541         if (cap_raised(uc->uc_cap, CAP_DAC_OVERRIDE))
2542                 RETURN(0);
2543         if (uc->uc_fsuid == la->la_uid) {
2544                 if ((la->la_mode & S_IWUSR) == 0)
2545                         rc = -EACCES;
2546         } else if (uc->uc_fsgid == la->la_gid) {
2547                 if ((la->la_mode & S_IWGRP) == 0)
2548                         rc = -EACCES;
2549         } else if ((la->la_mode & S_IWOTH) == 0) {
2550                         rc = -EACCES;
2551         }
2552
2553 out:
2554         RETURN(rc);
2555 }
2556
2557 static int mdt_rmfid_one(struct mdt_thread_info *info, struct lu_fid *fid,
2558                          s64 ctime)
2559 {
2560         struct mdt_device *mdt = info->mti_mdt;
2561         struct mdt_object *obj = NULL;
2562         struct linkea_data ldata = { NULL };
2563         struct lu_buf *buf = &info->mti_big_buf;
2564         struct lu_name *name = &info->mti_name;
2565         struct lu_fid *pfid = &info->mti_tmp_fid1;
2566         struct link_ea_header *leh;
2567         struct link_ea_entry *lee;
2568         int reclen, count, rc = 0;
2569         ENTRY;
2570
2571         if (!fid_is_sane(fid))
2572                 GOTO(out, rc = -EINVAL);
2573
2574         if (!fid_is_namespace_visible(fid))
2575                 GOTO(out, rc = -EINVAL);
2576
2577         obj = mdt_object_find(info->mti_env, mdt, fid);
2578         if (IS_ERR(obj))
2579                 GOTO(out, rc = PTR_ERR(obj));
2580
2581         if (mdt_object_remote(obj))
2582                 GOTO(out, rc = -EREMOTE);
2583         if (!mdt_object_exists(obj) || lu_object_is_dying(&obj->mot_header))
2584                 GOTO(out, rc = -ENOENT);
2585
2586         rc = mdt_rmfid_check_permission(info, obj);
2587         if (rc)
2588                 GOTO(out, rc);
2589
2590         /* take LinkEA */
2591         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
2592         if (!buf->lb_buf)
2593                 GOTO(out, rc = -ENOMEM);
2594
2595         ldata.ld_buf = buf;
2596         rc = mdt_links_read(info, obj, &ldata);
2597         if (rc)
2598                 GOTO(out, rc);
2599
2600         leh = buf->lb_buf;
2601         lee = (struct link_ea_entry *)(leh + 1);
2602         for (count = 0; count < leh->leh_reccount; count++) {
2603                 /* remove every hardlink */
2604                 linkea_entry_unpack(lee, &reclen, name, pfid);
2605                 lee = (struct link_ea_entry *) ((char *)lee + reclen);
2606                 rc = mdt_rmfid_unlink(info, pfid, name, obj, ctime);
2607                 if (rc)
2608                         break;
2609         }
2610
2611 out:
2612         if (obj && !IS_ERR(obj))
2613                 mdt_object_put(info->mti_env, obj);
2614         if (info->mti_big_buf.lb_buf)
2615                 lu_buf_free(&info->mti_big_buf);
2616
2617         RETURN(rc);
2618 }
2619
2620 static int mdt_rmfid(struct tgt_session_info *tsi)
2621 {
2622         struct mdt_thread_info *mti = tsi2mdt_info(tsi);
2623         struct mdt_body *reqbody;
2624         struct lu_fid *fids, *rfids;
2625         int bufsize, rc;
2626         __u32 *rcs;
2627         int i, nr;
2628         ENTRY;
2629
2630         reqbody = req_capsule_client_get(tsi->tsi_pill, &RMF_MDT_BODY);
2631         if (reqbody == NULL)
2632                 RETURN(-EPROTO);
2633         bufsize = req_capsule_get_size(tsi->tsi_pill, &RMF_FID_ARRAY,
2634                                        RCL_CLIENT);
2635         nr = bufsize / sizeof(struct lu_fid);
2636         if (nr * sizeof(struct lu_fid) != bufsize)
2637                 RETURN(-EINVAL);
2638         req_capsule_set_size(tsi->tsi_pill, &RMF_RCS,
2639                              RCL_SERVER, nr * sizeof(__u32));
2640         req_capsule_set_size(tsi->tsi_pill, &RMF_FID_ARRAY,
2641                              RCL_SERVER, nr * sizeof(struct lu_fid));
2642         rc = req_capsule_server_pack(tsi->tsi_pill);
2643         if (rc)
2644                 GOTO(out, rc = err_serious(rc));
2645         fids = req_capsule_client_get(tsi->tsi_pill, &RMF_FID_ARRAY);
2646         if (fids == NULL)
2647                 RETURN(-EPROTO);
2648         rcs = req_capsule_server_get(tsi->tsi_pill, &RMF_RCS);
2649         LASSERT(rcs);
2650         rfids = req_capsule_server_get(tsi->tsi_pill, &RMF_FID_ARRAY);
2651         LASSERT(rfids);
2652
2653         mdt_init_ucred(mti, reqbody);
2654         for (i = 0; i < nr; i++) {
2655                 rfids[i] = fids[i];
2656                 rcs[i] = mdt_rmfid_one(mti, fids + i, reqbody->mbo_ctime);
2657         }
2658         mdt_exit_ucred(mti);
2659
2660 out:
2661         RETURN(rc);
2662 }
2663
2664 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2665                          void *karg, void __user *uarg);
2666
2667 int mdt_io_set_info(struct tgt_session_info *tsi)
2668 {
2669         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2670         struct ost_body         *body = NULL, *repbody;
2671         void                    *key, *val = NULL;
2672         int                      keylen, vallen, rc = 0;
2673         bool                     is_grant_shrink;
2674
2675         ENTRY;
2676
2677         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
2678         if (key == NULL) {
2679                 DEBUG_REQ(D_HA, req, "no set_info key");
2680                 RETURN(err_serious(-EFAULT));
2681         }
2682         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
2683                                       RCL_CLIENT);
2684
2685         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
2686         if (val == NULL) {
2687                 DEBUG_REQ(D_HA, req, "no set_info val");
2688                 RETURN(err_serious(-EFAULT));
2689         }
2690         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
2691                                       RCL_CLIENT);
2692
2693         is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK);
2694         if (is_grant_shrink)
2695                 /* In this case the value is actually an RMF_OST_BODY, so we
2696                  * transmutate the type of this PTLRPC */
2697                 req_capsule_extend(tsi->tsi_pill, &RQF_OST_SET_GRANT_INFO);
2698
2699         rc = req_capsule_server_pack(tsi->tsi_pill);
2700         if (rc < 0)
2701                 RETURN(rc);
2702
2703         if (is_grant_shrink) {
2704                 body = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_BODY);
2705
2706                 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2707                 *repbody = *body;
2708
2709                 /** handle grant shrink, similar to a read request */
2710                 tgt_grant_prepare_read(tsi->tsi_env, tsi->tsi_exp,
2711                                        &repbody->oa);
2712         } else {
2713                 CERROR("%s: Unsupported key %s\n",
2714                        tgt_name(tsi->tsi_tgt), (char *)key);
2715                 rc = -EOPNOTSUPP;
2716         }
2717
2718         RETURN(rc);
2719 }
2720
2721
2722 static int mdt_set_info(struct tgt_session_info *tsi)
2723 {
2724         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2725         char                    *key;
2726         void                    *val;
2727         int                      keylen, vallen, rc = 0;
2728
2729         ENTRY;
2730
2731         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
2732         if (key == NULL) {
2733                 DEBUG_REQ(D_HA, req, "no set_info key");
2734                 RETURN(err_serious(-EFAULT));
2735         }
2736
2737         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
2738                                       RCL_CLIENT);
2739
2740         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
2741         if (val == NULL) {
2742                 DEBUG_REQ(D_HA, req, "no set_info val");
2743                 RETURN(err_serious(-EFAULT));
2744         }
2745
2746         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
2747                                       RCL_CLIENT);
2748
2749         /* Swab any part of val you need to here */
2750         if (KEY_IS(KEY_READ_ONLY)) {
2751                 spin_lock(&req->rq_export->exp_lock);
2752                 if (*(__u32 *)val)
2753                         *exp_connect_flags_ptr(req->rq_export) |=
2754                                 OBD_CONNECT_RDONLY;
2755                 else
2756                         *exp_connect_flags_ptr(req->rq_export) &=
2757                                 ~OBD_CONNECT_RDONLY;
2758                 spin_unlock(&req->rq_export->exp_lock);
2759         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2760                 struct changelog_setinfo *cs = val;
2761
2762                 if (vallen != sizeof(*cs)) {
2763                         CERROR("%s: bad changelog_clear setinfo size %d\n",
2764                                tgt_name(tsi->tsi_tgt), vallen);
2765                         RETURN(-EINVAL);
2766                 }
2767                 if (req_capsule_req_need_swab(&req->rq_pill)) {
2768                         __swab64s(&cs->cs_recno);
2769                         __swab32s(&cs->cs_id);
2770                 }
2771
2772                 if (!mdt_is_rootadmin(tsi2mdt_info(tsi)))
2773                         RETURN(-EACCES);
2774                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
2775                                    vallen, val, NULL);
2776         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2777                 if (vallen > 0)
2778                         obd_export_evict_by_nid(req->rq_export->exp_obd, val);
2779         } else {
2780                 RETURN(-EINVAL);
2781         }
2782         RETURN(rc);
2783 }
2784
2785 static int mdt_readpage(struct tgt_session_info *tsi)
2786 {
2787         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
2788         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
2789         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
2790         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
2791         struct mdt_body         *repbody;
2792         int                      rc;
2793         int                      i;
2794
2795         ENTRY;
2796
2797         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
2798                 RETURN(err_serious(-ENOMEM));
2799
2800         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
2801         if (repbody == NULL || reqbody == NULL)
2802                 RETURN(err_serious(-EFAULT));
2803
2804         /*
2805          * prepare @rdpg before calling lower layers and transfer itself. Here
2806          * reqbody->size contains offset of where to start to read and
2807          * reqbody->nlink contains number bytes to read.
2808          */
2809         rdpg->rp_hash = reqbody->mbo_size;
2810         if (rdpg->rp_hash != reqbody->mbo_size) {
2811                 CERROR("Invalid hash: %#llx != %#llx\n",
2812                        rdpg->rp_hash, reqbody->mbo_size);
2813                 RETURN(-EFAULT);
2814         }
2815
2816         rdpg->rp_attrs = reqbody->mbo_mode;
2817         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
2818                 rdpg->rp_attrs |= LUDA_64BITHASH;
2819         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
2820                                 exp_max_brw_size(tsi->tsi_exp));
2821         rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >>
2822                           PAGE_SHIFT;
2823         OBD_ALLOC_PTR_ARRAY_LARGE(rdpg->rp_pages, rdpg->rp_npages);
2824         if (rdpg->rp_pages == NULL)
2825                 RETURN(-ENOMEM);
2826
2827         for (i = 0; i < rdpg->rp_npages; ++i) {
2828                 rdpg->rp_pages[i] = alloc_page(GFP_NOFS);
2829                 if (rdpg->rp_pages[i] == NULL)
2830                         GOTO(free_rdpg, rc = -ENOMEM);
2831         }
2832
2833         /* call lower layers to fill allocated pages with directory data */
2834         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
2835         if (rc < 0)
2836                 GOTO(free_rdpg, rc);
2837
2838         /* send pages to client */
2839         rc = tgt_sendpage(tsi, rdpg, rc);
2840
2841         EXIT;
2842 free_rdpg:
2843
2844         for (i = 0; i < rdpg->rp_npages; i++)
2845                 if (rdpg->rp_pages[i] != NULL)
2846                         __free_page(rdpg->rp_pages[i]);
2847         OBD_FREE_PTR_ARRAY_LARGE(rdpg->rp_pages, rdpg->rp_npages);
2848
2849         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
2850                 RETURN(0);
2851
2852         return rc;
2853 }
2854
2855 static int mdt_fix_attr_ucred(struct mdt_thread_info *info, __u32 op)
2856 {
2857         struct lu_ucred *uc = mdt_ucred_check(info);
2858         struct lu_attr *attr = &info->mti_attr.ma_attr;
2859
2860         if (uc == NULL)
2861                 return -EINVAL;
2862
2863         if (op != REINT_SETATTR) {
2864                 if ((attr->la_valid & LA_UID) && (attr->la_uid != -1))
2865                         attr->la_uid = uc->uc_fsuid;
2866                 /* for S_ISGID, inherit gid from his parent, such work will be
2867                  * done in cmm/mdd layer, here set all cases as uc->uc_fsgid. */
2868                 if ((attr->la_valid & LA_GID) && (attr->la_gid != -1))
2869                         attr->la_gid = uc->uc_fsgid;
2870         }
2871
2872         return 0;
2873 }
2874
2875 static inline bool mdt_is_readonly_open(struct mdt_thread_info *info, __u32 op)
2876 {
2877         return op == REINT_OPEN &&
2878              !(info->mti_spec.sp_cr_flags & (MDS_FMODE_WRITE | MDS_OPEN_CREAT));
2879 }
2880
2881 static void mdt_preset_secctx_size(struct mdt_thread_info *info)
2882 {
2883         struct req_capsule *pill = info->mti_pill;
2884
2885         if (req_capsule_has_field(pill, &RMF_FILE_SECCTX,
2886                                   RCL_SERVER) &&
2887             req_capsule_has_field(pill, &RMF_FILE_SECCTX_NAME,
2888                                   RCL_CLIENT)) {
2889                 if (req_capsule_get_size(pill, &RMF_FILE_SECCTX_NAME,
2890                                          RCL_CLIENT) != 0)
2891                         /* pre-set size in server part with max size */
2892                         req_capsule_set_size(pill, &RMF_FILE_SECCTX,
2893                                              RCL_SERVER,
2894                                              OBD_MAX_DEFAULT_EA_SIZE);
2895                 else
2896                         req_capsule_set_size(pill, &RMF_FILE_SECCTX,
2897                                              RCL_SERVER, 0);
2898         }
2899 }
2900
2901 static int mdt_reint_internal(struct mdt_thread_info *info,
2902                               struct mdt_lock_handle *lhc,
2903                               __u32 op)
2904 {
2905         struct req_capsule      *pill = info->mti_pill;
2906         struct mdt_body         *repbody;
2907         int                      rc = 0, rc2;
2908
2909         ENTRY;
2910
2911         rc = mdt_reint_unpack(info, op);
2912         if (rc != 0) {
2913                 CERROR("Can't unpack reint, rc %d\n", rc);
2914                 RETURN(err_serious(rc));
2915         }
2916
2917
2918         /* check if the file system is set to readonly. O_RDONLY open
2919          * is still allowed even the file system is set to readonly mode */
2920         if (mdt_rdonly(info->mti_exp) && !mdt_is_readonly_open(info, op))
2921                 RETURN(err_serious(-EROFS));
2922
2923         /* for replay (no_create) lmm is not needed, client has it already */
2924         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2925                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2926                                      DEF_REP_MD_SIZE);
2927
2928         /* llog cookies are always 0, the field is kept for compatibility */
2929         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2930                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
2931
2932         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
2933          * by default. If the target object has more ACL entries, then
2934          * enlarge the buffer when necessary. */
2935         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
2936                 req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
2937                                      LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
2938
2939         mdt_preset_secctx_size(info);
2940         mdt_preset_encctx_size(info);
2941
2942         rc = req_capsule_server_pack(pill);
2943         if (rc != 0) {
2944                 CERROR("Can't pack response, rc %d\n", rc);
2945                 RETURN(err_serious(rc));
2946         }
2947
2948         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
2949                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
2950                 LASSERT(repbody);
2951                 repbody->mbo_eadatasize = 0;
2952                 repbody->mbo_aclsize = 0;
2953         }
2954
2955         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
2956
2957         /* for replay no cookkie / lmm need, because client have this already */
2958         if (info->mti_spec.no_create)
2959                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2960                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
2961
2962         rc = mdt_init_ucred_reint(info);
2963         if (rc)
2964                 GOTO(out_shrink, rc);
2965
2966         rc = mdt_fix_attr_ucred(info, op);
2967         if (rc != 0)
2968                 GOTO(out_ucred, rc = err_serious(rc));
2969
2970         rc = mdt_check_resent(info, mdt_reconstruct, lhc);
2971         if (rc < 0) {
2972                 GOTO(out_ucred, rc);
2973         } else if (rc == 1) {
2974                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt");
2975                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
2976                 GOTO(out_ucred, rc);
2977         }
2978         rc = mdt_reint_rec(info, lhc);
2979         EXIT;
2980 out_ucred:
2981         mdt_exit_ucred(info);
2982 out_shrink:
2983         mdt_client_compatibility(info);
2984
2985         rc2 = mdt_fix_reply(info);
2986         if (rc == 0)
2987                 rc = rc2;
2988
2989         /*
2990          * Data-on-MDT optimization - read data along with OPEN and return it
2991          * in reply when possible.
2992          */
2993         if (rc == 0 && op == REINT_OPEN && !req_is_replay(pill->rc_req))
2994                 rc = mdt_dom_read_on_open(info, info->mti_mdt,
2995                                           &lhc->mlh_reg_lh);
2996
2997         return rc;
2998 }
2999
3000 static long mdt_reint_opcode(struct ptlrpc_request *req,
3001                              const struct req_format **fmt)
3002 {
3003         struct mdt_device       *mdt;
3004         struct mdt_rec_reint    *rec;
3005         long                     opc;
3006
3007         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
3008         if (rec != NULL) {
3009                 opc = rec->rr_opcode;
3010                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
3011                 if (opc < REINT_MAX && fmt[opc] != NULL)
3012                         req_capsule_extend(&req->rq_pill, fmt[opc]);
3013                 else {
3014                         mdt = mdt_exp2dev(req->rq_export);
3015                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
3016                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
3017                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
3018                         opc = err_serious(-EFAULT);
3019                 }
3020         } else {
3021                 opc = err_serious(-EFAULT);
3022         }
3023         return opc;
3024 }
3025
3026 static int mdt_reint(struct tgt_session_info *tsi)
3027 {
3028         long opc;
3029         int  rc;
3030         static const struct req_format *reint_fmts[REINT_MAX] = {
3031                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
3032                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
3033                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
3034                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
3035                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
3036                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
3037                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
3038                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
3039                 [REINT_MIGRATE]  = &RQF_MDS_REINT_MIGRATE,
3040                 [REINT_RESYNC]   = &RQF_MDS_REINT_RESYNC,
3041         };
3042
3043         ENTRY;
3044
3045         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
3046         if (opc >= 0) {
3047                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
3048                 /*
3049                  * No lock possible here from client to pass it to reint code
3050                  * path.
3051                  */
3052                 rc = mdt_reint_internal(info, NULL, opc);
3053                 mdt_thread_info_fini(info);
3054         } else {
3055                 rc = opc;
3056         }
3057
3058         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
3059         RETURN(rc);
3060 }
3061
3062 /* this should sync the whole device */
3063 int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
3064 {
3065         struct dt_device *dt = mdt->mdt_bottom;
3066         int rc;
3067         ENTRY;
3068
3069         rc = dt->dd_ops->dt_sync(env, dt);
3070         RETURN(rc);
3071 }
3072
3073 /* this should sync this object */
3074 static int mdt_object_sync(const struct lu_env *env, struct obd_export *exp,
3075                            struct mdt_object *mo)
3076 {
3077         int rc = 0;
3078
3079         ENTRY;
3080
3081         if (!mdt_object_exists(mo)) {
3082                 CWARN("%s: non existing object "DFID": rc = %d\n",
3083                       exp->exp_obd->obd_name, PFID(mdt_object_fid(mo)),
3084                       -ESTALE);
3085                 RETURN(-ESTALE);
3086         }
3087
3088         if (S_ISREG(lu_object_attr(&mo->mot_obj))) {
3089                 struct lu_target *tgt = tgt_ses_info(env)->tsi_tgt;
3090                 dt_obj_version_t version;
3091
3092                 version = dt_version_get(env, mdt_obj2dt(mo));
3093                 if (version > tgt->lut_obd->obd_last_committed)
3094                         rc = mo_object_sync(env, mdt_object_child(mo));
3095         } else {
3096                 rc = mo_object_sync(env, mdt_object_child(mo));
3097         }
3098
3099         RETURN(rc);
3100 }
3101
3102 static int mdt_sync(struct tgt_session_info *tsi)
3103 {
3104         struct ptlrpc_request   *req = tgt_ses_req(tsi);
3105         struct req_capsule      *pill = tsi->tsi_pill;
3106         struct mdt_body         *body;
3107         ktime_t                  kstart = ktime_get();
3108         int                      rc;
3109
3110         ENTRY;
3111
3112         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
3113                 RETURN(err_serious(-ENOMEM));
3114
3115         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
3116                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
3117         } else {
3118                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
3119
3120                 if (unlikely(info->mti_object == NULL))
3121                         RETURN(-EPROTO);
3122
3123                 /* sync an object */
3124                 rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp,
3125                                      info->mti_object);
3126                 if (rc == 0) {
3127                         const struct lu_fid *fid;
3128                         struct lu_attr *la = &info->mti_attr.ma_attr;
3129
3130                         info->mti_attr.ma_need = MA_INODE;
3131                         info->mti_attr.ma_valid = 0;
3132                         rc = mdt_attr_get_complex(info, info->mti_object,
3133                                                   &info->mti_attr);
3134                         if (rc == 0) {
3135                                 body = req_capsule_server_get(pill,
3136                                                               &RMF_MDT_BODY);
3137                                 fid = mdt_object_fid(info->mti_object);
3138                                 mdt_pack_attr2body(info, body, la, fid);
3139                         }
3140                 }
3141                 mdt_thread_info_fini(info);
3142         }
3143         if (rc == 0)
3144                 mdt_counter_incr(req, LPROC_MDT_SYNC,
3145                                  ktime_us_delta(ktime_get(), kstart));
3146
3147         RETURN(rc);
3148 }
3149
3150 static int mdt_data_sync(struct tgt_session_info *tsi)
3151 {
3152         struct mdt_thread_info *info;
3153         struct mdt_device *mdt = mdt_exp2dev(tsi->tsi_exp);
3154         struct ost_body *body = tsi->tsi_ost_body;
3155         struct ost_body *repbody;
3156         struct mdt_object *mo = NULL;
3157         struct md_attr *ma;
3158         int rc = 0;
3159
3160         ENTRY;
3161
3162         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
3163
3164         /* if no fid is specified then do nothing,
3165          * device sync is done via MDS_SYNC */
3166         if (fid_is_zero(&tsi->tsi_fid))
3167                 RETURN(0);
3168
3169         mo = mdt_object_find(tsi->tsi_env, mdt, &tsi->tsi_fid);
3170         if (IS_ERR(mo))
3171                 RETURN(PTR_ERR(mo));
3172
3173         rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp, mo);
3174         if (rc)
3175                 GOTO(put, rc);
3176
3177         repbody->oa.o_oi = body->oa.o_oi;
3178         repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
3179
3180         info = tsi2mdt_info(tsi);
3181         ma = &info->mti_attr;
3182         ma->ma_need = MA_INODE;
3183         ma->ma_valid = 0;
3184         rc = mdt_attr_get_complex(info, mo, ma);
3185         if (rc == 0)
3186                 obdo_from_la(&repbody->oa, &ma->ma_attr, VALID_FLAGS);
3187         else
3188                 rc = 0;
3189         mdt_thread_info_fini(info);
3190
3191         EXIT;
3192 put:
3193         if (mo != NULL)
3194                 mdt_object_put(tsi->tsi_env, mo);
3195         return rc;
3196 }
3197
3198 /*
3199  * Handle quota control requests to consult current usage/limit, but also
3200  * to configure quota enforcement
3201  */
3202 static int mdt_quotactl(struct tgt_session_info *tsi)
3203 {
3204         struct obd_export *exp  = tsi->tsi_exp;
3205         struct req_capsule *pill = tsi->tsi_pill;
3206         struct obd_quotactl *oqctl, *repoqc;
3207         int id, rc;
3208         struct mdt_device *mdt = mdt_exp2dev(exp);
3209         struct lu_device *qmt = mdt->mdt_qmt_dev;
3210         struct lu_nodemap *nodemap;
3211         ENTRY;
3212
3213         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
3214         if (!oqctl)
3215                 RETURN(err_serious(-EPROTO));
3216
3217         rc = req_capsule_server_pack(pill);
3218         if (rc)
3219                 RETURN(err_serious(rc));
3220
3221         nodemap = nodemap_get_from_exp(exp);
3222         if (IS_ERR(nodemap))
3223                 RETURN(PTR_ERR(nodemap));
3224
3225         switch (oqctl->qc_cmd) {
3226                 /* master quotactl */
3227         case Q_SETINFO:
3228         case Q_SETQUOTA:
3229         case LUSTRE_Q_SETDEFAULT:
3230         case LUSTRE_Q_SETQUOTAPOOL:
3231         case LUSTRE_Q_SETINFOPOOL:
3232         case LUSTRE_Q_SETDEFAULT_POOL:
3233         case LUSTRE_Q_DELETEQID:
3234                 if (!nodemap_can_setquota(nodemap, oqctl->qc_type,
3235                                           oqctl->qc_id))
3236                         GOTO(out_nodemap, rc = -EPERM);
3237                 fallthrough;
3238         case Q_GETINFO:
3239         case Q_GETQUOTA:
3240         case LUSTRE_Q_GETDEFAULT:
3241         case LUSTRE_Q_GETQUOTAPOOL:
3242         case LUSTRE_Q_GETINFOPOOL:
3243         case LUSTRE_Q_GETDEFAULT_POOL:
3244                 if (qmt == NULL)
3245                         GOTO(out_nodemap, rc = -EOPNOTSUPP);
3246                 /* slave quotactl */
3247                 fallthrough;
3248         case Q_GETOINFO:
3249         case Q_GETOQUOTA:
3250                 break;
3251         default:
3252                 rc = -EFAULT;
3253                 CERROR("%s: unsupported quotactl command %d: rc = %d\n",
3254                        mdt_obd_name(mdt), oqctl->qc_cmd, rc);
3255                 GOTO(out_nodemap, rc);
3256         }
3257
3258         id = oqctl->qc_id;
3259         switch (oqctl->qc_type) {
3260         case USRQUOTA:
3261                 id = nodemap_map_id(nodemap, NODEMAP_UID,
3262                                     NODEMAP_CLIENT_TO_FS, id);
3263                 break;
3264         case GRPQUOTA:
3265                 id = nodemap_map_id(nodemap, NODEMAP_GID,
3266                                     NODEMAP_CLIENT_TO_FS, id);
3267                 break;
3268         case PRJQUOTA:
3269                 id = nodemap_map_id(nodemap, NODEMAP_PROJID,
3270                                     NODEMAP_CLIENT_TO_FS, id);
3271                 break;
3272         default:
3273                 GOTO(out_nodemap, rc = -EOPNOTSUPP);
3274         }
3275         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
3276         if (repoqc == NULL)
3277                 GOTO(out_nodemap, rc = err_serious(-EFAULT));
3278
3279         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA)
3280                 barrier_exit(tsi->tsi_tgt->lut_bottom);
3281
3282         if (oqctl->qc_id != id)
3283                 swap(oqctl->qc_id, id);
3284
3285         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA) {
3286                 if (unlikely(!barrier_entry(tsi->tsi_tgt->lut_bottom)))
3287                         RETURN(-EINPROGRESS);
3288         }
3289
3290         switch (oqctl->qc_cmd) {
3291
3292         case Q_GETINFO:
3293         case Q_SETINFO:
3294         case Q_SETQUOTA:
3295         case Q_GETQUOTA:
3296         case LUSTRE_Q_SETDEFAULT:
3297         case LUSTRE_Q_GETDEFAULT:
3298         case LUSTRE_Q_SETQUOTAPOOL:
3299         case LUSTRE_Q_GETQUOTAPOOL:
3300         case LUSTRE_Q_SETINFOPOOL:
3301         case LUSTRE_Q_GETINFOPOOL:
3302         case LUSTRE_Q_SETDEFAULT_POOL:
3303         case LUSTRE_Q_GETDEFAULT_POOL:
3304         case LUSTRE_Q_DELETEQID:
3305                 /* forward quotactl request to QMT */
3306                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
3307                 break;
3308
3309         case Q_GETOINFO:
3310         case Q_GETOQUOTA:
3311                 /* slave quotactl */
3312                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
3313                                    oqctl);
3314                 break;
3315
3316         default:
3317                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
3318                 GOTO(out_nodemap, rc = -EFAULT);
3319         }
3320
3321         if (oqctl->qc_id != id)
3322                 swap(oqctl->qc_id, id);
3323
3324         QCTL_COPY(repoqc, oqctl);
3325         EXIT;
3326
3327 out_nodemap:
3328         nodemap_putref(nodemap);
3329
3330         return rc;
3331 }
3332
3333 /** clone llog ctxt from child (mdd)
3334  * This allows remote llog (replicator) access.
3335  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
3336  * context was originally set up, or we can handle them directly.
3337  * I choose the latter, but that means I need any llog
3338  * contexts set up by child to be accessable by the mdt.  So we clone the
3339  * context into our context list here.
3340  */
3341 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
3342                                int idx)
3343 {
3344         struct md_device  *next = mdt->mdt_child;
3345         struct llog_ctxt *ctxt;
3346         int rc;
3347
3348         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
3349                 return 0;
3350
3351         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
3352         if (rc || ctxt == NULL) {
3353                 return 0;
3354         }
3355
3356         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
3357         if (rc)
3358                 CERROR("Can't set mdt ctxt %d\n", rc);
3359
3360         return rc;
3361 }
3362
3363 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
3364                                  struct mdt_device *mdt, int idx)
3365 {
3366         struct llog_ctxt *ctxt;
3367
3368         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
3369         if (ctxt == NULL)
3370                 return 0;
3371         /* Put once for the get we just did, and once for the clone */
3372         llog_ctxt_put(ctxt);
3373         llog_ctxt_put(ctxt);
3374         return 0;
3375 }
3376
3377 /*
3378  * sec context handlers
3379  */
3380 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
3381 {
3382         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
3383
3384         return 0;
3385 }
3386
3387 /*
3388  * quota request handlers
3389  */
3390 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
3391 {
3392         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
3393         struct lu_device        *qmt = mdt->mdt_qmt_dev;
3394         int                      rc;
3395         ENTRY;
3396
3397         if (qmt == NULL)
3398                 RETURN(err_serious(-EOPNOTSUPP));
3399
3400         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
3401         RETURN(rc);
3402 }
3403
3404 struct mdt_object *mdt_object_new(const struct lu_env *env,
3405                                   struct mdt_device *d,
3406                                   const struct lu_fid *f)
3407 {
3408         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
3409         struct lu_object *o;
3410         struct mdt_object *m;
3411         ENTRY;
3412
3413         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
3414         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
3415         if (unlikely(IS_ERR(o)))
3416                 m = (struct mdt_object *)o;
3417         else
3418                 m = mdt_obj(o);
3419         RETURN(m);
3420 }
3421
3422 struct mdt_object *mdt_object_find(const struct lu_env *env,
3423                                    struct mdt_device *d,
3424                                    const struct lu_fid *f)
3425 {
3426         struct lu_object *o;
3427         struct mdt_object *m;
3428         ENTRY;
3429
3430         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
3431         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
3432         if (unlikely(IS_ERR(o)))
3433                 m = (struct mdt_object *)o;
3434         else
3435                 m = mdt_obj(o);
3436
3437         RETURN(m);
3438 }
3439
3440 /**
3441  * Asyncronous commit for mdt device.
3442  *
3443  * Pass asynchonous commit call down the MDS stack.
3444  *
3445  * \param env environment
3446  * \param mdt the mdt device
3447  */
3448 static void mdt_device_commit_async(const struct lu_env *env,
3449                                     struct mdt_device *mdt)
3450 {
3451         struct dt_device *dt = mdt->mdt_bottom;
3452         int rc;
3453         ENTRY;
3454
3455         rc = dt->dd_ops->dt_commit_async(env, dt);
3456         if (unlikely(rc != 0))
3457                 CWARN("%s: async commit start failed: rc = %d\n",
3458                       mdt_obd_name(mdt), rc);
3459         atomic_inc(&mdt->mdt_async_commit_count);
3460         EXIT;
3461 }
3462
3463 /**
3464  * Mark the lock as "synchonous".
3465  *
3466  * Mark the lock to deffer transaction commit to the unlock time.
3467  *
3468  * \param lock the lock to mark as "synchonous"
3469  *
3470  * \see mdt_is_lock_sync
3471  * \see mdt_save_lock
3472  */
3473 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
3474 {
3475         lock->l_ast_data = (void*)1;
3476 }
3477
3478 /**
3479  * Check whehter the lock "synchonous" or not.
3480  *
3481  * \param lock the lock to check
3482  * \retval 1 the lock is "synchonous"
3483  * \retval 0 the lock isn't "synchronous"
3484  *
3485  * \see mdt_set_lock_sync
3486  * \see mdt_save_lock
3487  */
3488 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
3489 {
3490         return lock->l_ast_data != NULL;
3491 }
3492
3493 /**
3494  * Blocking AST for mdt locks.
3495  *
3496  * Starts transaction commit if in case of COS lock conflict or
3497  * deffers such a commit to the mdt_save_lock.
3498  *
3499  * \param lock the lock which blocks a request or cancelling lock
3500  * \param desc unused
3501  * \param data unused
3502  * \param flag indicates whether this cancelling or blocking callback
3503  * \retval 0
3504  * \see ldlm_blocking_ast_nocheck
3505  */
3506 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
3507                      void *data, int flag)
3508 {
3509         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
3510         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
3511         struct ldlm_cb_set_arg *arg = data;
3512         bool commit_async = false;
3513         int rc;
3514         ENTRY;
3515
3516         if (flag == LDLM_CB_CANCELING)
3517                 RETURN(0);
3518
3519         lock_res_and_lock(lock);
3520         if (lock->l_blocking_ast != mdt_blocking_ast) {
3521                 unlock_res_and_lock(lock);
3522                 RETURN(0);
3523         }
3524
3525         /* A blocking ast may be sent from ldlm_lock_decref_internal
3526          * when the last reference to a local lock was released and
3527          * during blocking event from ldlm_work_bl_ast_lock().
3528          * The 'data' parameter is l_ast_data in the first case and
3529          * callback arguments in the second one. Distinguish them by that.
3530          */
3531         if (!data || data == lock->l_ast_data || !arg->bl_desc)
3532                 goto skip_cos_checks;
3533
3534         if (lock->l_req_mode & (LCK_PW | LCK_EX)) {
3535                 if (mdt_cos_is_enabled(mdt)) {
3536                         if (!arg->bl_desc->bl_same_client)
3537                                 mdt_set_lock_sync(lock);
3538                 } else if (mdt_slc_is_enabled(mdt) &&
3539                            arg->bl_desc->bl_cos_incompat) {
3540                         mdt_set_lock_sync(lock);
3541                         /*
3542                          * we may do extra commit here, but there is a small
3543                          * window to miss a commit: lock was unlocked (saved),
3544                          * then a conflict lock queued and we come here, but
3545                          * REP-ACK not received, so lock was not converted to
3546                          * COS mode yet.
3547                          * Fortunately this window is quite small, so the
3548                          * extra commit should be rare (not to say distributed
3549                          * operation is rare too).
3550                          */
3551                         commit_async = true;
3552                 }
3553         } else if (lock->l_req_mode == LCK_COS) {
3554                 commit_async = true;
3555         }
3556
3557 skip_cos_checks:
3558         rc = ldlm_blocking_ast_nocheck(lock);
3559
3560         if (commit_async) {
3561                 struct lu_env env;
3562
3563                 rc = lu_env_init(&env, LCT_LOCAL);
3564                 if (unlikely(rc != 0))
3565                         CWARN("%s: lu_env initialization failed, cannot "
3566                               "start asynchronous commit: rc = %d\n",
3567                               obd->obd_name, rc);
3568                 else
3569                         mdt_device_commit_async(&env, mdt);
3570                 lu_env_fini(&env);
3571         }
3572         RETURN(rc);
3573 }
3574
3575 /*
3576  * Blocking AST for cross-MDT lock
3577  *
3578  * Discard lock from uncommitted_slc_locks and cancel it.
3579  *
3580  * \param lock  the lock which blocks a request or cancelling lock
3581  * \param desc  unused
3582  * \param data  unused
3583  * \param flag  indicates whether this cancelling or blocking callback
3584  * \retval      0 on success
3585  * \retval      negative number on error
3586  */
3587 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
3588                             void *data, int flag)
3589 {
3590         int rc = 0;
3591         ENTRY;
3592
3593         switch (flag) {
3594         case LDLM_CB_BLOCKING: {
3595                 struct lustre_handle lockh;
3596
3597                 ldlm_lock2handle(lock, &lockh);
3598                 rc = ldlm_cli_cancel(&lockh,
3599                         ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
3600                 if (rc < 0) {
3601                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
3602                         RETURN(rc);
3603                 }
3604                 break;
3605         }
3606         case LDLM_CB_CANCELING: {
3607                 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
3608                 struct mdt_device *mdt =
3609                                 mdt_dev(obd->obd_lu_dev->ld_site->ls_top_dev);
3610
3611                 LDLM_DEBUG(lock, "Revoke remote lock");
3612
3613                 /* discard slc lock here so that it can be cleaned anytime,
3614                  * especially for cleanup_resource() */
3615                 tgt_discard_slc_lock(&mdt->mdt_lut, lock);
3616
3617                 /* once we cache lock, l_ast_data is set to mdt_object */
3618                 if (lock->l_ast_data != NULL) {
3619                         struct mdt_object *mo = lock->l_ast_data;
3620                         struct lu_env env;
3621
3622                         rc = lu_env_init(&env, LCT_MD_THREAD);
3623                         if (unlikely(rc != 0)) {
3624                                 CWARN("%s: lu_env initialization failed, object %p "DFID" is leaked!: rc = %d\n",
3625                                       obd->obd_name, mo,
3626                                       PFID(mdt_object_fid(mo)), rc);
3627                                 RETURN(rc);
3628                         }
3629
3630                         if (lock->l_policy_data.l_inodebits.bits &
3631                             (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)) {
3632                                 rc = mo_invalidate(&env, mdt_object_child(mo));
3633                                 mo->mot_cache_attr = 0;
3634                         }
3635                         mdt_object_put(&env, mo);
3636                         lu_env_fini(&env);
3637                 }
3638                 break;
3639         }
3640         default:
3641                 LBUG();
3642         }
3643
3644         RETURN(rc);
3645 }
3646
3647 int mdt_check_resent_lock(struct mdt_thread_info *info,
3648                           struct mdt_object *mo,
3649                           struct mdt_lock_handle *lhc)
3650 {
3651         /* the lock might already be gotten in ldlm_handle_enqueue() */
3652         if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
3653                 struct ptlrpc_request *req = mdt_info_req(info);
3654                 struct ldlm_lock      *lock;
3655
3656                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
3657                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
3658                 if (lock == NULL) {
3659                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
3660                          * a resend case, however, it could be already destroyed
3661                          * due to client eviction or a raced cancel RPC. */
3662                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx",
3663                                           lhc->mlh_reg_lh.cookie);
3664                         RETURN(-ESTALE);
3665                 }
3666
3667                 if (!fid_res_name_eq(mdt_object_fid(mo),
3668                                      &lock->l_resource->lr_name)) {
3669                         CWARN("%s: Although resent, but still not "
3670                               "get child lock:"DFID"\n",
3671                               info->mti_exp->exp_obd->obd_name,
3672                               PFID(mdt_object_fid(mo)));
3673                         LDLM_LOCK_PUT(lock);
3674                         RETURN(-EPROTO);
3675                 }
3676                 LDLM_LOCK_PUT(lock);
3677                 return 0;
3678         }
3679         return 1;
3680 }
3681
3682 static void mdt_remote_object_lock_created_cb(struct ldlm_lock *lock)
3683 {
3684         mdt_object_get(NULL, lock->l_ast_data);
3685 }
3686
3687 int mdt_remote_object_lock_try(struct mdt_thread_info *mti,
3688                                struct mdt_object *o, const struct lu_fid *fid,
3689                                struct lustre_handle *lh, enum ldlm_mode mode,
3690                                __u64 *ibits, __u64 trybits, bool cache)
3691 {
3692         struct ldlm_enqueue_info *einfo = &mti->mti_remote_einfo;
3693         union ldlm_policy_data *policy = &mti->mti_policy;
3694         struct ldlm_res_id *res_id = &mti->mti_res_id;
3695         int rc = 0;
3696         ENTRY;
3697
3698         LASSERT(mdt_object_remote(o));
3699
3700         fid_build_reg_res_name(fid, res_id);
3701
3702         memset(einfo, 0, sizeof(*einfo));
3703         einfo->ei_type = LDLM_IBITS;
3704         einfo->ei_mode = mode;
3705         einfo->ei_cb_bl = mdt_remote_blocking_ast;
3706         einfo->ei_cb_cp = ldlm_completion_ast;
3707         einfo->ei_enq_slave = 0;
3708         einfo->ei_res_id = res_id;
3709         einfo->ei_req_slot = 1;
3710
3711         if (cache) {
3712                 /*
3713                  * if we cache lock, couple lock with mdt_object, so that object
3714                  * can be easily found in lock ASTs.
3715                  */
3716                 einfo->ei_cbdata = o;
3717                 einfo->ei_cb_created = mdt_remote_object_lock_created_cb;
3718         }
3719
3720         memset(policy, 0, sizeof(*policy));
3721         policy->l_inodebits.bits = *ibits;
3722         policy->l_inodebits.try_bits = trybits;
3723
3724         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
3725                             policy);
3726
3727         /* Return successfully acquired bits to a caller */
3728         if (rc == 0) {
3729                 struct ldlm_lock *lock = ldlm_handle2lock(lh);
3730
3731                 LASSERT(lock);
3732                 *ibits = lock->l_policy_data.l_inodebits.bits;
3733                 LDLM_LOCK_PUT(lock);
3734         }
3735         RETURN(rc);
3736 }
3737
3738 int mdt_remote_object_lock(struct mdt_thread_info *mti, struct mdt_object *o,
3739                            const struct lu_fid *fid, struct lustre_handle *lh,
3740                            enum ldlm_mode mode, __u64 ibits, bool cache)
3741 {
3742         return mdt_remote_object_lock_try(mti, o, fid, lh, mode, &ibits, 0,
3743                                           cache);
3744 }
3745
3746 int mdt_object_local_lock(struct mdt_thread_info *info, struct mdt_object *o,
3747                           struct mdt_lock_handle *lh, __u64 *ibits,
3748                           __u64 trybits, bool cos_incompat)
3749 {
3750         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
3751         union ldlm_policy_data *policy = &info->mti_policy;
3752         struct ldlm_res_id *res_id = &info->mti_res_id;
3753         __u64 dlmflags = 0, *cookie = NULL;
3754         int rc;
3755         ENTRY;
3756
3757         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
3758         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
3759         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
3760         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
3761
3762         if (cos_incompat) {
3763                 LASSERT(lh->mlh_reg_mode == LCK_PW ||
3764                         lh->mlh_reg_mode == LCK_EX);
3765                 dlmflags |= LDLM_FL_COS_INCOMPAT;
3766         } else if (mdt_cos_is_enabled(info->mti_mdt)) {
3767                 dlmflags |= LDLM_FL_COS_ENABLED;
3768         }
3769
3770         /* Only enqueue LOOKUP lock for remote object */
3771         LASSERT(ergo(mdt_object_remote(o), *ibits == MDS_INODELOCK_LOOKUP));
3772
3773         /* Lease lock are granted with LDLM_FL_CANCEL_ON_BLOCK */
3774         if (lh->mlh_type == MDT_REG_LOCK && lh->mlh_reg_mode == LCK_EX &&
3775             *ibits == MDS_INODELOCK_OPEN)
3776                 dlmflags |= LDLM_FL_CANCEL_ON_BLOCK;
3777
3778         if (lh->mlh_type == MDT_PDO_LOCK) {
3779                 /* check for exists after object is locked */
3780                 if (mdt_object_exists(o) == 0) {
3781                         /* Non-existent object shouldn't have PDO lock */
3782                         RETURN(-ESTALE);
3783                 } else {
3784                         /* Non-dir object shouldn't have PDO lock */
3785                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
3786                                 RETURN(-ENOTDIR);
3787                 }
3788         }
3789
3790         fid_build_reg_res_name(mdt_object_fid(o), res_id);
3791         dlmflags |= LDLM_FL_ATOMIC_CB;
3792
3793         if (info->mti_exp)
3794                 cookie = &info->mti_exp->exp_handle.h_cookie;
3795
3796         /*
3797          * Take PDO lock on whole directory and build correct @res_id for lock
3798          * on part of directory.
3799          */
3800         if (lh->mlh_pdo_hash != 0) {
3801                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
3802                 mdt_lock_pdo_mode(info, o, lh);
3803                 if (lh->mlh_pdo_mode != LCK_NL) {
3804                         /*
3805                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
3806                          * is never going to be sent to client and we do not
3807                          * want it slowed down due to possible cancels.
3808                          */
3809                         policy->l_inodebits.bits =
3810                                 *ibits & MDS_INODELOCK_UPDATE;
3811                         policy->l_inodebits.try_bits =
3812                                 trybits & MDS_INODELOCK_UPDATE;
3813                         /* at least one of them should be set */
3814                         LASSERT(policy->l_inodebits.bits |
3815                                 policy->l_inodebits.try_bits);
3816                         rc = mdt_fid_lock(info->mti_env, ns, &lh->mlh_pdo_lh,
3817                                           lh->mlh_pdo_mode, policy, res_id,
3818                                           dlmflags, cookie);
3819                         if (unlikely(rc != 0))
3820                                 GOTO(out_unlock, rc);
3821                 }
3822
3823                 /*
3824                  * Finish res_id initializing by name hash marking part of
3825                  * directory which is taking modification.
3826                  */
3827                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
3828         }
3829
3830         policy->l_inodebits.bits = *ibits;
3831         policy->l_inodebits.try_bits = trybits;
3832         policy->l_inodebits.li_gid = lh->mlh_gid;
3833
3834         /*
3835          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
3836          * going to be sent to client. If it is - mdt_intent_policy() path will
3837          * fix it up and turn FL_LOCAL flag off.
3838          */
3839         rc = mdt_fid_lock(info->mti_env, ns, &lh->mlh_reg_lh, lh->mlh_reg_mode,
3840                           policy, res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
3841                           cookie);
3842 out_unlock:
3843         if (rc != 0)
3844                 mdt_object_unlock(info, o, lh, 1);
3845         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
3846                    lh->mlh_pdo_hash != 0 &&
3847                    (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
3848                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
3849
3850         /* Return successfully acquired bits to a caller */
3851         if (rc == 0) {
3852                 struct ldlm_lock *lock = ldlm_handle2lock(&lh->mlh_reg_lh);
3853
3854                 LASSERT(lock);
3855                 *ibits = lock->l_policy_data.l_inodebits.bits;
3856                 LDLM_LOCK_PUT(lock);
3857         }
3858         RETURN(rc);
3859 }
3860
3861 static int
3862 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
3863                          struct mdt_lock_handle *lh, __u64 *ibits,
3864                          __u64 trybits, bool cos_incompat)
3865 {
3866         struct mdt_lock_handle *local_lh = NULL;
3867         int rc;
3868         ENTRY;
3869
3870         if (!mdt_object_remote(o)) {
3871                 rc = mdt_object_local_lock(info, o, lh, ibits, trybits,
3872                                            cos_incompat);
3873                 RETURN(rc);
3874         }
3875
3876         /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
3877         *ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
3878                     MDS_INODELOCK_XATTR);
3879
3880         /* Only enqueue LOOKUP lock for remote object */
3881         if (*ibits & MDS_INODELOCK_LOOKUP) {
3882                 __u64 local = MDS_INODELOCK_LOOKUP;
3883
3884                 rc = mdt_object_local_lock(info, o, lh, &local, 0,
3885                                            cos_incompat);
3886                 if (rc != ELDLM_OK)
3887                         RETURN(rc);
3888
3889                 local_lh = lh;
3890         }
3891
3892         if ((*ibits | trybits) & MDS_INODELOCK_UPDATE) {
3893                 /* Sigh, PDO needs to enqueue 2 locks right now, but
3894                  * enqueue RPC can only request 1 lock, to avoid extra
3895                  * RPC, so it will instead enqueue EX lock for remote
3896                  * object anyway XXX*/
3897                 if (lh->mlh_type == MDT_PDO_LOCK &&
3898                     lh->mlh_pdo_hash != 0) {
3899                         CDEBUG(D_INFO,
3900                                "%s: "DFID" convert PDO lock to EX lock.\n",
3901                                mdt_obd_name(info->mti_mdt),
3902                                PFID(mdt_object_fid(o)));
3903                         lh->mlh_pdo_hash = 0;
3904                         lh->mlh_rreg_mode = LCK_EX;
3905                         lh->mlh_type = MDT_REG_LOCK;
3906                 }
3907
3908                 rc = mdt_remote_object_lock_try(info, o, mdt_object_fid(o),
3909                                                 &lh->mlh_rreg_lh,
3910                                                 lh->mlh_rreg_mode,
3911                                                 ibits, trybits, false);
3912                 if (rc != ELDLM_OK) {
3913                         if (local_lh != NULL)
3914                                 mdt_object_unlock(info, o, local_lh, rc);
3915                         RETURN(rc);
3916                 }
3917         }
3918
3919         /* other components like LFSCK can use lockless access
3920          * and populate cache, so we better invalidate it */
3921         mo_invalidate(info->mti_env, mdt_object_child(o));
3922
3923         RETURN(0);
3924 }
3925
3926 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
3927                     struct mdt_lock_handle *lh, __u64 ibits)
3928 {
3929         return mdt_object_lock_internal(info, o, lh, &ibits, 0, false);
3930 }
3931
3932 int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
3933                           struct mdt_lock_handle *lh, __u64 ibits,
3934                           bool cos_incompat)
3935 {
3936         LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
3937         return mdt_object_lock_internal(info, o, lh, &ibits, 0,
3938                                         cos_incompat);
3939 }
3940
3941 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
3942                         struct mdt_lock_handle *lh, __u64 *ibits,
3943                         __u64 trybits, bool cos_incompat)
3944 {
3945         bool trylock_only = *ibits == 0;
3946         int rc;
3947
3948         LASSERT(!(*ibits & trybits));
3949         rc = mdt_object_lock_internal(info, o, lh, ibits, trybits,
3950                                       cos_incompat);
3951         if (rc && trylock_only) { /* clear error for try ibits lock only */
3952                 LASSERT(*ibits == 0);
3953                 rc = 0;
3954         }
3955         return rc;
3956 }
3957
3958 /**
3959  * Save a lock within request object.
3960  *
3961  * Keep the lock referenced until whether client ACK or transaction
3962  * commit happens or release the lock immediately depending on input
3963  * parameters. If COS is ON, a write lock is converted to COS lock
3964  * before saving.
3965  *
3966  * \param info thead info object
3967  * \param h lock handle
3968  * \param mode lock mode
3969  * \param decref force immediate lock releasing
3970  */
3971 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
3972                    enum ldlm_mode mode, int decref)
3973 {
3974         struct tgt_session_info *tsi = info->mti_env->le_ses ?
3975                                        tgt_ses_info(info->mti_env) : NULL;
3976
3977         ENTRY;
3978
3979         if (lustre_handle_is_used(h)) {
3980                 bool has_trans = tsi && tsi->tsi_has_trans;
3981
3982                 if (decref || !has_trans || !(mode & (LCK_PW | LCK_EX))) {
3983                         mdt_fid_unlock(h, mode);
3984                 } else {
3985                         struct mdt_device *mdt = info->mti_mdt;
3986                         struct ldlm_lock *lock = ldlm_handle2lock(h);
3987                         struct ptlrpc_request *req = mdt_info_req(info);
3988                         bool cos = mdt_cos_is_enabled(mdt);
3989                         bool convert_lock = !cos && mdt_slc_is_enabled(mdt);
3990
3991                         LASSERTF(lock != NULL, "no lock for cookie %#llx\n",
3992                                  h->cookie);
3993
3994                         /* there is no request if mdt_object_unlock() is called
3995                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
3996                         if (likely(req != NULL)) {
3997                                 LDLM_DEBUG(lock, "save lock request %p reply "
3998                                         "state %p transno %lld\n", req,
3999                                         req->rq_reply_state, req->rq_transno);
4000                                 if (cos) {
4001                                         ldlm_lock_mode_downgrade(lock, LCK_COS);
4002                                         mode = LCK_COS;
4003                                 }
4004                                 if (req->rq_export->exp_disconnected)
4005                                         mdt_fid_unlock(h, mode);
4006                                 else
4007                                         ptlrpc_save_lock(req, h, mode, cos,
4008                                                          convert_lock);
4009                         } else {
4010                                 mdt_fid_unlock(h, mode);
4011                         }
4012
4013                         if (mdt_is_lock_sync(lock)) {
4014                                 CDEBUG(D_HA, "sync_lock, do async commit\n");
4015                                 mdt_device_commit_async(info->mti_env, mdt);
4016                         }
4017                         LDLM_LOCK_PUT(lock);
4018                 }
4019                 h->cookie = 0ull;
4020         }
4021
4022         EXIT;
4023 }
4024
4025 /**
4026  * Save cross-MDT lock in uncommitted_slc_locks
4027  *
4028  * Keep the lock referenced until transaction commit happens or release the lock
4029  * immediately depending on input parameters.
4030  *
4031  * \param info thead info object
4032  * \param h lock handle
4033  * \param mode lock mode
4034  * \param decref force immediate lock releasing
4035  */
4036 static void mdt_save_remote_lock(struct mdt_thread_info *info,
4037                                  struct mdt_object *o, struct lustre_handle *h,
4038                                  enum ldlm_mode mode, int decref)
4039 {
4040         ENTRY;
4041
4042         if (lustre_handle_is_used(h)) {
4043                 struct ldlm_lock *lock = ldlm_handle2lock(h);
4044                 struct ptlrpc_request *req = mdt_info_req(info);
4045
4046                 if (o != NULL &&
4047                     (lock->l_policy_data.l_inodebits.bits &
4048                      (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)))
4049                         mo_invalidate(info->mti_env, mdt_object_child(o));
4050
4051                 if (decref || !req || !(mode & (LCK_PW | LCK_EX)) ||
4052                     !tgt_ses_info(info->mti_env)->tsi_has_trans) {
4053                         ldlm_lock_decref_and_cancel(h, mode);
4054                         LDLM_LOCK_PUT(lock);
4055                 } else {
4056                         tgt_save_slc_lock(&info->mti_mdt->mdt_lut, lock,
4057                                           req->rq_transno);
4058                         ldlm_lock_decref(h, mode);
4059                 }
4060                 h->cookie = 0ull;
4061         }
4062
4063         EXIT;
4064 }
4065
4066 /**
4067  * Unlock mdt object.
4068  *
4069  * Immeditely release the regular lock and the PDO lock or save the
4070  * lock in request and keep them referenced until client ACK or
4071  * transaction commit.
4072  *
4073  * \param info thread info object
4074  * \param o mdt object
4075  * \param lh mdt lock handle referencing regular and PDO locks
4076  * \param decref force immediate lock releasing
4077  *
4078  * XXX o is not used and may be NULL, see hsm_cdt_request_completed().
4079  */
4080 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
4081                        struct mdt_lock_handle *lh, int decref)
4082 {
4083         ENTRY;
4084
4085         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
4086         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
4087         mdt_save_remote_lock(info, o, &lh->mlh_rreg_lh, lh->mlh_rreg_mode,
4088                              decref);
4089
4090         EXIT;
4091 }
4092
4093 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
4094                                         const struct lu_fid *f,
4095                                         struct mdt_lock_handle *lh,
4096                                         __u64 ibits)
4097 {
4098         struct mdt_object *o;
4099
4100         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
4101         if (!IS_ERR(o)) {
4102                 int rc;
4103
4104                 rc = mdt_object_lock(info, o, lh, ibits);
4105                 if (rc != 0) {
4106                         mdt_object_put(info->mti_env, o);
4107                         o = ERR_PTR(rc);
4108                 }
4109         }
4110         return o;
4111 }
4112
4113 void mdt_object_unlock_put(struct mdt_thread_info * info,
4114                            struct mdt_object * o,
4115                            struct mdt_lock_handle *lh,
4116                            int decref)
4117 {
4118         mdt_object_unlock(info, o, lh, decref);
4119         mdt_object_put(info->mti_env, o);
4120 }
4121
4122 /*
4123  * Generic code handling requests that have struct mdt_body passed in:
4124  *
4125  *  - extract mdt_body from request and save it in @info, if present;
4126  *
4127  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
4128  *  @info;
4129  *
4130  *  - if HAS_BODY flag is set for this request type check whether object
4131  *  actually exists on storage (lu_object_exists()).
4132  *
4133  */
4134 static int mdt_body_unpack(struct mdt_thread_info *info,
4135                            enum tgt_handler_flags flags)
4136 {
4137         const struct mdt_body    *body;
4138         struct mdt_object        *obj;
4139         const struct lu_env      *env;
4140         struct req_capsule       *pill;
4141         int                       rc;
4142         ENTRY;
4143
4144         env = info->mti_env;
4145         pill = info->mti_pill;
4146
4147         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
4148         if (body == NULL)
4149                 RETURN(-EFAULT);
4150
4151         if (!(body->mbo_valid & OBD_MD_FLID))
4152                 RETURN(0);
4153
4154         if (!fid_is_sane(&body->mbo_fid1)) {
4155                 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
4156                 RETURN(-EINVAL);
4157         }
4158
4159         obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
4160         if (!IS_ERR(obj)) {
4161                 if ((flags & HAS_BODY) && !mdt_object_exists(obj)) {
4162                         mdt_object_put(env, obj);
4163                         rc = -ENOENT;
4164                 } else {
4165                         info->mti_object = obj;
4166                         rc = 0;
4167                 }
4168         } else
4169                 rc = PTR_ERR(obj);
4170
4171         RETURN(rc);
4172 }
4173
4174 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info,
4175                                    enum tgt_handler_flags flags)
4176 {
4177         struct req_capsule *pill = info->mti_pill;
4178         int rc;
4179
4180         ENTRY;
4181
4182         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
4183                 rc = mdt_body_unpack(info, flags);
4184         else
4185                 rc = 0;
4186
4187         if (rc == 0 && (flags & HAS_REPLY)) {
4188                 /* Pack reply. */
4189                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
4190                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
4191                                              DEF_REP_MD_SIZE);
4192                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
4193                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
4194                                              RCL_SERVER, 0);
4195
4196                 /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
4197                  * by default. If the target object has more ACL entries, then
4198                  * enlarge the buffer when necessary. */
4199                 if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
4200                         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
4201                                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
4202
4203                 mdt_preset_secctx_size(info);
4204                 mdt_preset_encctx_size(info);
4205
4206                 rc = req_capsule_server_pack(pill);
4207                 if (rc)
4208                         CWARN("%s: cannot pack response: rc = %d\n",
4209                                       mdt_obd_name(info->mti_mdt), rc);
4210         }
4211         RETURN(rc);
4212 }
4213
4214 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
4215 {
4216         lh->mlh_type = MDT_NUL_LOCK;
4217         lh->mlh_reg_lh.cookie = 0ull;
4218         lh->mlh_reg_mode = LCK_MINMODE;
4219         lh->mlh_pdo_lh.cookie = 0ull;
4220         lh->mlh_pdo_mode = LCK_MINMODE;
4221         lh->mlh_rreg_lh.cookie = 0ull;
4222         lh->mlh_rreg_mode = LCK_MINMODE;
4223 }
4224
4225 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
4226 {
4227         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
4228         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
4229 }
4230
4231 /*
4232  * Initialize fields of struct mdt_thread_info. Other fields are left in
4233  * uninitialized state, because it's too expensive to zero out whole
4234  * mdt_thread_info (> 1K) on each request arrival.
4235  */
4236 void mdt_thread_info_init(struct ptlrpc_request *req,
4237                           struct mdt_thread_info *info)
4238 {
4239         int i;
4240
4241         info->mti_pill = &req->rq_pill;
4242
4243         /* lock handle */
4244         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
4245                 mdt_lock_handle_init(&info->mti_lh[i]);
4246
4247         /* mdt device: it can be NULL while CONNECT */
4248         if (req->rq_export) {
4249                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
4250                 info->mti_exp = req->rq_export;
4251         } else
4252                 info->mti_mdt = NULL;
4253         info->mti_env = req->rq_svc_thread->t_env;
4254         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
4255
4256         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
4257         info->mti_big_buf = LU_BUF_NULL;
4258         info->mti_body = NULL;
4259         info->mti_object = NULL;
4260         info->mti_dlm_req = NULL;
4261         info->mti_cross_ref = 0;
4262         info->mti_opdata = 0;
4263         info->mti_big_lmm_used = 0;
4264         info->mti_big_acl_used = 0;
4265         info->mti_som_strict = 0;
4266
4267         info->mti_spec.no_create = 0;
4268         info->mti_spec.sp_rm_entry = 0;
4269         info->mti_spec.sp_permitted = 0;
4270
4271         info->mti_spec.u.sp_ea.eadata = NULL;
4272         info->mti_spec.u.sp_ea.eadatalen = 0;
4273 }
4274
4275 void mdt_thread_info_fini(struct mdt_thread_info *info)
4276 {
4277         int i;
4278
4279         if (info->mti_object != NULL) {
4280                 mdt_object_put(info->mti_env, info->mti_object);
4281                 info->mti_object = NULL;
4282         }
4283
4284         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
4285                 mdt_lock_handle_fini(&info->mti_lh[i]);
4286         info->mti_env = NULL;
4287         info->mti_pill = NULL;
4288         info->mti_exp = NULL;
4289         info->mti_mdt = NULL;
4290
4291         if (unlikely(info->mti_big_buf.lb_buf != NULL))
4292                 lu_buf_free(&info->mti_big_buf);
4293 }
4294
4295 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
4296 {
4297         struct mdt_thread_info  *mti;
4298
4299         mti = mdt_th_info(tsi->tsi_env);
4300         LASSERT(mti != NULL);
4301
4302         mdt_thread_info_init(tgt_ses_req(tsi), mti);
4303         if (tsi->tsi_corpus != NULL) {
4304                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
4305                 lu_object_get(tsi->tsi_corpus);
4306         }
4307         mti->mti_body = tsi->tsi_mdt_body;
4308         mti->mti_dlm_req = tsi->tsi_dlm_req;
4309
4310         return mti;
4311 }
4312
4313 static int mdt_tgt_connect(struct tgt_session_info *tsi)
4314 {
4315         if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
4316             cfs_fail_val ==
4317             tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id)
4318                 schedule_timeout_uninterruptible(cfs_time_seconds(3));
4319
4320         return tgt_connect(tsi);
4321 }
4322
4323 static int mdt_intent_glimpse(enum ldlm_intent_flags it_opc,
4324                               struct mdt_thread_info *info,
4325                               struct ldlm_lock **lockp, __u64 flags)
4326 {
4327         return mdt_glimpse_enqueue(info, info->mti_mdt->mdt_namespace,
4328                                    lockp, flags);
4329 }
4330 static int mdt_intent_brw(enum ldlm_intent_flags it_opc,
4331                           struct mdt_thread_info *info,
4332                           struct ldlm_lock **lockp, __u64 flags)
4333 {
4334         return mdt_brw_enqueue(info, info->mti_mdt->mdt_namespace,
4335                                lockp, flags);
4336 }
4337
4338 int mdt_intent_lock_replace(struct mdt_thread_info *info,
4339                             struct ldlm_lock **lockp,
4340                             struct mdt_lock_handle *lh,
4341                             __u64 flags, int result)
4342 {
4343         struct ptlrpc_request  *req = mdt_info_req(info);
4344         struct ldlm_lock       *lock = *lockp;
4345         struct ldlm_lock       *new_lock;
4346
4347         /* If possible resent found a lock, @lh is set to its handle */
4348         new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
4349
4350         if (new_lock == NULL) {
4351                 if (flags & LDLM_FL_INTENT_ONLY) {
4352                         result = 0;
4353                 } else if (flags & LDLM_FL_RESENT) {
4354                         /* Lock is pinned by ldlm_handle_enqueue0() as it is a
4355                          * resend case, however, it could be already destroyed
4356                          * due to client eviction or a raced cancel RPC.
4357                          */
4358                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx\n",
4359                                           lh->mlh_reg_lh.cookie);
4360                         result = -ESTALE;
4361                 } else {
4362                         CERROR("%s: Invalid lockh=%#llx flags=%#llx fid1="DFID" fid2="DFID": rc = %d\n",
4363                                mdt_obd_name(info->mti_mdt),
4364                                lh->mlh_reg_lh.cookie, flags,
4365                                PFID(&info->mti_tmp_fid1),
4366                                PFID(&info->mti_tmp_fid2), result);
4367                         result = -ESTALE;
4368                 }
4369                 lh->mlh_reg_lh.cookie = 0;
4370                 RETURN(result);
4371         }
4372
4373         /*
4374          * If we've already given this lock to a client once, then we should
4375          * have no readers or writers.  Otherwise, we should have one reader
4376          * _or_ writer ref (which will be zeroed below) before returning the
4377          * lock to a client.
4378          */
4379         if (new_lock->l_export == req->rq_export) {
4380                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
4381         } else {
4382                 LASSERT(new_lock->l_export == NULL);
4383                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
4384         }
4385
4386         *lockp = new_lock;
4387
4388         if (new_lock->l_export == req->rq_export) {
4389                 /*
4390                  * Already gave this to the client, which means that we
4391                  * reconstructed a reply.
4392                  */
4393                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
4394                         MSG_RESENT);
4395
4396                 LDLM_LOCK_RELEASE(new_lock);
4397                 lh->mlh_reg_lh.cookie = 0;
4398                 RETURN(ELDLM_LOCK_REPLACED);
4399         }
4400
4401         /*
4402          * Fixup the lock to be given to the client.
4403          */
4404         lock_res_and_lock(new_lock);
4405         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
4406          * possible blocking AST. */
4407         while (new_lock->l_readers > 0) {
4408                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
4409                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
4410                 new_lock->l_readers--;
4411         }
4412         while (new_lock->l_writers > 0) {
4413                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
4414                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
4415                 new_lock->l_writers--;
4416         }
4417
4418         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
4419         new_lock->l_blocking_ast = lock->l_blocking_ast;
4420         new_lock->l_completion_ast = lock->l_completion_ast;
4421         if (ldlm_has_dom(new_lock))
4422                 new_lock->l_glimpse_ast = ldlm_server_glimpse_ast;
4423         new_lock->l_remote_handle = lock->l_remote_handle;
4424         new_lock->l_flags &= ~LDLM_FL_LOCAL;
4425
4426         unlock_res_and_lock(new_lock);
4427
4428         cfs_hash_add(new_lock->l_export->exp_lock_hash,
4429                      &new_lock->l_remote_handle,
4430                      &new_lock->l_exp_hash);
4431
4432         LDLM_LOCK_RELEASE(new_lock);
4433         lh->mlh_reg_lh.cookie = 0;
4434
4435         RETURN(ELDLM_LOCK_REPLACED);
4436 }
4437
4438 void mdt_intent_fixup_resent(struct mdt_thread_info *info,
4439                              struct ldlm_lock *new_lock,
4440                              struct mdt_lock_handle *lh, __u64 flags)
4441 {
4442         struct ptlrpc_request  *req = mdt_info_req(info);
4443         struct ldlm_request    *dlmreq;
4444
4445         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
4446                 return;
4447
4448         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
4449
4450         /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
4451          * lock was found by ldlm_handle_enqueue(); if so @lh must be
4452          * initialized. */
4453         if (flags & LDLM_FL_RESENT) {
4454                 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
4455                 lh->mlh_reg_mode = new_lock->l_granted_mode;
4456
4457                 LDLM_DEBUG(new_lock, "Restoring lock cookie");
4458                 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie %#llx",
4459                           lh->mlh_reg_lh.cookie);
4460                 return;
4461         }
4462
4463         /*
4464          * If the xid matches, then we know this is a resent request, and allow
4465          * it. (It's probably an OPEN, for which we don't send a lock.
4466          */
4467         if (req_can_reconstruct(req, NULL) != 0)
4468                 return;
4469
4470         /*
4471          * This remote handle isn't enqueued, so we never received or processed
4472          * this request.  Clear MSG_RESENT, because it can be handled like any
4473          * normal request now.
4474          */
4475         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
4476
4477         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle %#llx",
4478                   dlmreq->lock_handle[0].cookie);
4479 }
4480
4481 static int mdt_intent_getxattr(enum ldlm_intent_flags it_opc,
4482                                struct mdt_thread_info *info,
4483                                struct ldlm_lock **lockp,
4484                                __u64 flags)
4485 {
4486         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4487         struct ldlm_reply      *ldlm_rep = NULL;
4488         int rc;
4489         ENTRY;
4490
4491         /*
4492          * Initialize lhc->mlh_reg_lh either from a previously granted lock
4493          * (for the resend case) or a new lock. Below we will use it to
4494          * replace the original lock.
4495          */
4496         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4497         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
4498                 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
4499                 rc = mdt_object_lock(info, info->mti_object, lhc,
4500                                      MDS_INODELOCK_XATTR);
4501                 if (rc)
4502                         return rc;
4503         }
4504
4505         rc = mdt_getxattr(info);
4506
4507         if (mdt_info_req(info)->rq_repmsg != NULL)
4508                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4509
4510         if (ldlm_rep == NULL ||
4511             OBD_FAIL_CHECK(OBD_FAIL_MDS_XATTR_REP)) {
4512                 mdt_object_unlock(info,  info->mti_object, lhc, 1);
4513                 if (is_serious(rc))
4514                         RETURN(rc);
4515                 else
4516                         RETURN(err_serious(-EFAULT));
4517         }
4518
4519         ldlm_rep->lock_policy_res2 = clear_serious(rc);
4520
4521         /* This is left for interop instead of adding a new interop flag.
4522          * LU-7433 */
4523 #if LUSTRE_VERSION_CODE > OBD_OCD_VERSION(3, 0, 0, 0)
4524         if (ldlm_rep->lock_policy_res2) {
4525                 mdt_object_unlock(info, info->mti_object, lhc, 1);
4526                 RETURN(ELDLM_LOCK_ABORTED);
4527         }
4528 #endif
4529
4530         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4531         RETURN(rc);
4532 }
4533
4534 static int mdt_intent_getattr(enum ldlm_intent_flags it_opc,
4535                               struct mdt_thread_info *info,
4536                               struct ldlm_lock **lockp,
4537                               __u64 flags)
4538 {
4539         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4540         __u64                   child_bits;
4541         struct ldlm_reply      *ldlm_rep;
4542         struct mdt_body        *reqbody;
4543         struct mdt_body        *repbody;
4544         int                     rc, rc2;
4545         ENTRY;
4546
4547         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
4548         LASSERT(reqbody);
4549
4550         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
4551         LASSERT(repbody);
4552
4553         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
4554         repbody->mbo_eadatasize = 0;
4555         repbody->mbo_aclsize = 0;
4556
4557         switch (it_opc) {
4558         case IT_LOOKUP:
4559                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
4560                 break;
4561         case IT_GETATTR:
4562                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
4563                              MDS_INODELOCK_PERM;
4564                 break;
4565         default:
4566                 CERROR("%s: unsupported intent %#x\n",
4567                        mdt_obd_name(info->mti_mdt), (unsigned int)it_opc);
4568                 GOTO(out_shrink, rc = -EINVAL);
4569         }
4570
4571         rc = mdt_init_ucred(info, reqbody);
4572         if (rc)
4573                 GOTO(out_shrink, rc);
4574
4575         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4576         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
4577
4578         /* Get lock from request for possible resent case. */
4579         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4580
4581         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
4582         ldlm_rep->lock_policy_res2 = clear_serious(rc);
4583
4584         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG) &&
4585             ldlm_rep->lock_policy_res2 != -ENOKEY)
4586                 ldlm_rep->lock_policy_res2 = 0;
4587         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
4588             ldlm_rep->lock_policy_res2) {
4589                 lhc->mlh_reg_lh.cookie = 0ull;
4590                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
4591         }
4592
4593         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4594         EXIT;
4595 out_ucred:
4596         mdt_exit_ucred(info);
4597 out_shrink:
4598         mdt_client_compatibility(info);
4599         rc2 = mdt_fix_reply(info);
4600         if (rc == 0)
4601                 rc = rc2;
4602         return rc;
4603 }
4604
4605 static int mdt_intent_layout(enum ldlm_intent_flags it_opc,
4606                              struct mdt_thread_info *info,
4607                              struct ldlm_lock **lockp,
4608                              __u64 flags)
4609 {
4610         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4611         struct md_layout_change layout = { .mlc_opc = MD_LAYOUT_NOP };
4612         struct layout_intent *intent;
4613         struct ldlm_reply *ldlm_rep;
4614         struct lu_fid *fid = &info->mti_tmp_fid2;
4615         struct mdt_object *obj = NULL;
4616         int layout_size = 0;
4617         struct lu_buf *buf = &layout.mlc_buf;
4618         int rc = 0;
4619
4620         ENTRY;
4621
4622         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
4623
4624         intent = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
4625         if (intent == NULL)
4626                 RETURN(-EPROTO);
4627
4628         CDEBUG(D_INFO, DFID "got layout change request from client: "
4629                "opc:%u flags:%#x extent "DEXT"\n",
4630                PFID(fid), intent->li_opc, intent->li_flags,
4631                PEXT(&intent->li_extent));
4632
4633         switch (intent->li_opc) {
4634         case LAYOUT_INTENT_TRUNC:
4635         case LAYOUT_INTENT_WRITE:
4636                 layout.mlc_opc = MD_LAYOUT_WRITE;
4637                 layout.mlc_intent = intent;
4638                 break;
4639         case LAYOUT_INTENT_ACCESS:
4640                 break;
4641         case LAYOUT_INTENT_READ:
4642         case LAYOUT_INTENT_GLIMPSE:
4643         case LAYOUT_INTENT_RELEASE:
4644         case LAYOUT_INTENT_RESTORE:
4645                 CERROR("%s: Unsupported layout intent opc %d\n",
4646                        mdt_obd_name(info->mti_mdt), intent->li_opc);
4647                 RETURN(-ENOTSUPP);
4648         default:
4649                 CERROR("%s: Unknown layout intent opc %d\n",
4650                        mdt_obd_name(info->mti_mdt), intent->li_opc);
4651                 RETURN(-EINVAL);
4652         }
4653
4654         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
4655         if (IS_ERR(obj))
4656                 RETURN(PTR_ERR(obj));
4657
4658         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
4659                 /* if layout is going to be changed don't use the current EA
4660                  * size but the maximum one. That buffer will be shrinked
4661                  * to the actual size in req_capsule_shrink() before reply.
4662                  */
4663                 if (layout.mlc_opc == MD_LAYOUT_WRITE) {
4664                         layout_size = info->mti_mdt->mdt_max_mdsize;
4665                 } else {
4666                         layout_size = mdt_attr_get_eabuf_size(info, obj);
4667                         if (layout_size < 0)
4668                                 GOTO(out, rc = layout_size);
4669
4670                         if (layout_size > info->mti_mdt->mdt_max_mdsize)
4671                                 info->mti_mdt->mdt_max_mdsize = layout_size;
4672                 }
4673                 CDEBUG(D_INFO, "%s: layout_size %d\n",
4674                        mdt_obd_name(info->mti_mdt), layout_size);
4675         }
4676
4677         /*
4678          * set reply buffer size, so that ldlm_handle_enqueue0()->
4679          * ldlm_lvbo_fill() will fill the reply buffer with lovea.
4680          */
4681         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
4682                              layout_size);
4683         rc = req_capsule_server_pack(info->mti_pill);
4684         if (rc)
4685                 GOTO(out, rc);
4686
4687         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4688         if (!ldlm_rep)
4689                 GOTO(out, rc = -EPROTO);
4690
4691         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
4692
4693         /* take lock in ldlm_lock_enqueue() for LAYOUT_INTENT_ACCESS */
4694         if (layout.mlc_opc == MD_LAYOUT_NOP)
4695                 GOTO(out, rc = 0);
4696
4697         rc = mdt_check_resent(info, mdt_reconstruct_generic, lhc);
4698         if (rc < 0)
4699                 GOTO(out, rc);
4700         if (rc == 1) {
4701                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
4702                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
4703                 GOTO(out, rc);
4704         }
4705
4706         buf->lb_buf = NULL;
4707         buf->lb_len = 0;
4708         if (unlikely(req_is_replay(mdt_info_req(info)))) {
4709                 buf->lb_buf = req_capsule_client_get(info->mti_pill,
4710                                                      &RMF_EADATA);
4711                 buf->lb_len = req_capsule_get_size(info->mti_pill,
4712                                                      &RMF_EADATA, RCL_CLIENT);
4713                 /*
4714                  * If it's a replay of layout write intent RPC, the client has
4715                  * saved the extended lovea when it get reply then.
4716                  */
4717                 if (buf->lb_len > 0)
4718                         mdt_fix_lov_magic(info, buf->lb_buf);
4719         }
4720
4721         /* Get lock from request for possible resent case. */
4722         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4723         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
4724
4725         /*
4726          * Instantiate some layout components, if @buf contains lovea, then it's
4727          * a replay of the layout intent write RPC.
4728          */
4729         rc = mdt_layout_change(info, obj, lhc, &layout);
4730         ldlm_rep->lock_policy_res2 = clear_serious(rc);
4731
4732         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
4733                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4734                 if (rc == ELDLM_LOCK_REPLACED &&
4735                     (*lockp)->l_granted_mode == LCK_EX)
4736                         ldlm_lock_mode_downgrade(*lockp, LCK_CR);
4737         }
4738
4739         EXIT;
4740 out:
4741         mdt_object_put(info->mti_env, obj);
4742         return rc;
4743 }
4744
4745 static int mdt_intent_open(enum ldlm_intent_flags it_opc,
4746                            struct mdt_thread_info *info,
4747                            struct ldlm_lock **lockp,
4748                            __u64 flags)
4749 {
4750         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4751         struct ldlm_reply      *rep = NULL;
4752         long                    opc;
4753         int                     rc;
4754         struct ptlrpc_request  *req = mdt_info_req(info);
4755
4756         static const struct req_format *intent_fmts[REINT_MAX] = {
4757                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
4758                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
4759         };
4760
4761         ENTRY;
4762
4763         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
4764         if (opc < 0)
4765                 RETURN(opc);
4766
4767         /* Get lock from request for possible resent case. */
4768         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4769
4770         rc = mdt_reint_internal(info, lhc, opc);
4771
4772         if (rc < 0 && lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
4773                 DEBUG_REQ(D_ERROR, req, "Replay open failed with %d", rc);
4774
4775         /* Check whether the reply has been packed successfully. */
4776         if (mdt_info_req(info)->rq_repmsg != NULL)
4777                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4778         if (rep == NULL) {
4779                 if (is_serious(rc))
4780                         RETURN(rc);
4781                 else
4782                         RETURN(err_serious(-EFAULT));
4783         }
4784
4785         /* MDC expects this in any case */
4786         if (rc != 0)
4787                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
4788
4789         /* the open lock or the lock for cross-ref object should be
4790          * returned to the client */
4791         if (lustre_handle_is_used(&lhc->mlh_reg_lh) &&
4792             (rc == 0 || rc == -MDT_EREMOTE_OPEN)) {
4793                 rep->lock_policy_res2 = 0;
4794                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4795                 RETURN(rc);
4796         }
4797
4798         rep->lock_policy_res2 = clear_serious(rc);
4799
4800         if (rep->lock_policy_res2 == -ENOENT &&
4801             mdt_get_disposition(rep, DISP_LOOKUP_NEG) &&
4802             !mdt_get_disposition(rep, DISP_OPEN_CREATE))
4803                 rep->lock_policy_res2 = 0;
4804
4805         lhc->mlh_reg_lh.cookie = 0ull;
4806         if (rc == -ENOTCONN || rc == -ENODEV ||
4807             rc == -EOVERFLOW) { /**< if VBR failure then return error */
4808                 /*
4809                  * If it is the disconnect error (ENODEV & ENOCONN), the error
4810                  * will be returned by rq_status, and client at ptlrpc layer
4811                  * will detect this, then disconnect, reconnect the import
4812                  * immediately, instead of impacting the following the rpc.
4813                  */
4814                 RETURN(rc);
4815         }
4816         /*
4817          * For other cases, the error will be returned by intent, and client
4818          * will retrieve the result from intent.
4819          */
4820         RETURN(ELDLM_LOCK_ABORTED);
4821 }
4822
4823 static int mdt_intent_opc(enum ldlm_intent_flags it_opc,
4824                           struct mdt_thread_info *info,
4825                           struct ldlm_lock **lockp,
4826                           u64 flags /* LDLM_FL_* */)
4827 {
4828         struct req_capsule *pill = info->mti_pill;
4829         struct ptlrpc_request *req = mdt_info_req(info);
4830         const struct req_format *it_format;
4831         int (*it_handler)(enum ldlm_intent_flags,
4832                           struct mdt_thread_info *,
4833                           struct ldlm_lock **,
4834                           u64);
4835         enum tgt_handler_flags it_handler_flags = 0;
4836         struct ldlm_reply *rep;
4837         bool check_mdt_object = false;
4838         int rc;
4839         ENTRY;
4840
4841         switch (it_opc) {
4842         case IT_OPEN:
4843         case IT_OPEN|IT_CREAT:
4844                 /*
4845                  * OCREAT is not a IS_MUTABLE request since the file may
4846                  * already exist. We do the extra check of
4847                  * OBD_CONNECT_RDONLY in mdt_reint_open() when we
4848                  * really need to create the object.
4849                  */
4850                 it_format = &RQF_LDLM_INTENT;
4851                 it_handler = &mdt_intent_open;
4852                 break;
4853         case IT_GETATTR:
4854                 check_mdt_object = true;
4855                 fallthrough;
4856         case IT_LOOKUP:
4857                 it_format = &RQF_LDLM_INTENT_GETATTR;
4858                 it_handler = &mdt_intent_getattr;
4859                 it_handler_flags = HAS_REPLY;
4860                 break;
4861         case IT_GETXATTR:
4862                 check_mdt_object = true;
4863                 it_format = &RQF_LDLM_INTENT_GETXATTR;
4864                 it_handler = &mdt_intent_getxattr;
4865                 it_handler_flags = HAS_BODY;
4866                 break;
4867         case IT_LAYOUT:
4868                 it_format = &RQF_LDLM_INTENT_LAYOUT;
4869                 it_handler = &mdt_intent_layout;
4870                 break;
4871         case IT_GLIMPSE:
4872                 it_format = &RQF_LDLM_INTENT;
4873                 it_handler = &mdt_intent_glimpse;
4874                 break;
4875         case IT_BRW:
4876                 it_format = &RQF_LDLM_INTENT;
4877                 it_handler = &mdt_intent_brw;
4878                 break;
4879         case IT_QUOTA_DQACQ:
4880         case IT_QUOTA_CONN: {
4881                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
4882
4883                 if (qmt == NULL)
4884                         RETURN(-EOPNOTSUPP);
4885
4886                 if (mdt_rdonly(req->rq_export))
4887                         RETURN(-EROFS);
4888
4889                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
4890                 /* pass the request to quota master */
4891                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
4892                                                  mdt_info_req(info), lockp,
4893                                                  flags);
4894                 RETURN(rc);
4895         }
4896         default:
4897                 CERROR("%s: unknown intent code %#x\n",
4898                        mdt_obd_name(info->mti_mdt), it_opc);
4899                 RETURN(-EPROTO);
4900         }
4901
4902         req_capsule_extend(pill, it_format);
4903
4904         rc = mdt_unpack_req_pack_rep(info, it_handler_flags);
4905         if (rc < 0)
4906                 RETURN(rc);
4907
4908         if (unlikely(info->mti_object == NULL && check_mdt_object))
4909                 RETURN(-EPROTO);
4910
4911         if (it_handler_flags & IS_MUTABLE && mdt_rdonly(req->rq_export))
4912                 RETURN(-EROFS);
4913
4914         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_INTENT_DELAY, 10);
4915
4916         /* execute policy */
4917         rc = (*it_handler)(it_opc, info, lockp, flags);
4918
4919         /* Check whether the reply has been packed successfully. */
4920         if (req->rq_repmsg != NULL) {
4921                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4922                 rep->lock_policy_res2 =
4923                         ptlrpc_status_hton(rep->lock_policy_res2);
4924         }
4925
4926         RETURN(rc);
4927 }
4928
4929 static void mdt_ptlrpc_stats_update(struct ptlrpc_request *req,
4930                                     enum ldlm_intent_flags it_opc)
4931 {
4932         struct lprocfs_stats *srv_stats = ptlrpc_req2svc(req)->srv_stats;
4933
4934         /* update stats when IT code is known */
4935         if (srv_stats != NULL)
4936                 lprocfs_counter_incr(srv_stats,
4937                                 PTLRPC_LAST_CNTR + (it_opc == IT_GLIMPSE ?
4938                                 LDLM_GLIMPSE_ENQUEUE : LDLM_IBITS_ENQUEUE));
4939 }
4940
4941 static int mdt_intent_policy(const struct lu_env *env,
4942                              struct ldlm_namespace *ns,
4943                              struct ldlm_lock **lockp,
4944                              void *req_cookie,
4945                              enum ldlm_mode mode,
4946                              __u64 flags, void *data)
4947 {
4948         struct tgt_session_info *tsi;
4949         struct mdt_thread_info  *info;
4950         struct ptlrpc_request   *req  =  req_cookie;
4951         struct ldlm_intent      *it;
4952         struct req_capsule      *pill;
4953         const struct ldlm_lock_desc *ldesc;
4954         int rc;
4955
4956         ENTRY;
4957
4958         LASSERT(req != NULL);
4959
4960         tsi = tgt_ses_info(env);
4961
4962         info = tsi2mdt_info(tsi);
4963         LASSERT(info != NULL);
4964         pill = info->mti_pill;
4965         LASSERT(pill->rc_req == req);
4966         ldesc = &info->mti_dlm_req->lock_desc;
4967
4968         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
4969                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
4970                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
4971                 if (it != NULL) {
4972                         mdt_ptlrpc_stats_update(req, it->opc);
4973                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
4974                         if (rc == 0)
4975                                 rc = ELDLM_OK;
4976
4977                         /* Lock without inodebits makes no sense and will oops
4978                          * later in ldlm. Let's check it now to see if we have
4979                          * ibits corrupted somewhere in mdt_intent_opc().
4980                          * The case for client miss to set ibits has been
4981                          * processed by others. */
4982                         LASSERT(ergo(ldesc->l_resource.lr_type == LDLM_IBITS,
4983                                 ldesc->l_policy_data.l_inodebits.bits != 0));
4984                 } else {
4985                         rc = err_serious(-EFAULT);
4986                 }
4987         } else if (ldesc->l_resource.lr_type == LDLM_IBITS &&
4988                    ldesc->l_policy_data.l_inodebits.bits == MDS_INODELOCK_DOM) {
4989                 struct ldlm_reply *rep;
4990
4991                 /* No intent was provided but INTENT flag is set along with
4992                  * DOM bit, this is considered as GLIMPSE request.
4993                  * This logic is common for MDT and OST glimpse
4994                  */
4995                 mdt_ptlrpc_stats_update(req, IT_GLIMPSE);
4996                 rc = mdt_glimpse_enqueue(info, ns, lockp, flags);
4997                 /* Check whether the reply has been packed successfully. */
4998                 if (req->rq_repmsg != NULL) {
4999                         rep = req_capsule_server_get(info->mti_pill,
5000                                                      &RMF_DLM_REP);
5001                         rep->lock_policy_res2 =
5002                                 ptlrpc_status_hton(rep->lock_policy_res2);
5003                 }
5004         } else {
5005                 /* No intent was provided */
5006                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
5007                 rc = req_capsule_server_pack(pill);
5008                 if (rc)
5009                         rc = err_serious(rc);
5010         }
5011         mdt_thread_info_fini(info);
5012         RETURN(rc);
5013 }
5014
5015 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
5016 {
5017         struct seq_server_site  *ss = mdt_seq_site(mdt);
5018
5019         if (ss->ss_node_id == 0)
5020                 return;
5021
5022         if (ss->ss_client_seq != NULL) {
5023                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
5024                 ss->ss_client_seq->lcs_exp = NULL;
5025         }
5026
5027         if (ss->ss_server_fld != NULL) {
5028                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
5029                 ss->ss_server_fld->lsf_control_exp = NULL;
5030         }
5031 }
5032
5033 static void mdt_seq_fini_cli(struct mdt_device *mdt)
5034 {
5035         struct seq_server_site *ss = mdt_seq_site(mdt);
5036
5037         if (ss == NULL)
5038                 return;
5039
5040         if (ss->ss_server_seq != NULL)
5041                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
5042 }
5043
5044 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
5045 {
5046         mdt_seq_fini_cli(mdt);
5047         mdt_deregister_seq_exp(mdt);
5048
5049         return seq_site_fini(env, mdt_seq_site(mdt));
5050 }
5051
5052 /**
5053  * It will retrieve its FLDB entries from MDT0, and it only happens
5054  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
5055  * and it needs to refresh FLDB from the MDT0.
5056  **/
5057 static int mdt_register_lwp_callback(void *data)
5058 {
5059         struct lu_env           env;
5060         struct mdt_device       *mdt = data;
5061         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
5062         int                     rc;
5063         ENTRY;
5064
5065         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
5066
5067         rc = lu_env_init(&env, LCT_MD_THREAD);
5068         if (rc < 0) {
5069                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
5070                 RETURN(rc);
5071         }
5072
5073         /* Allocate new sequence now to avoid creating local transaction
5074          * in the normal transaction process */
5075         rc = seq_server_check_and_alloc_super(&env,
5076                                               mdt_seq_site(mdt)->ss_server_seq);
5077         if (rc < 0)
5078                 GOTO(out, rc);
5079
5080         if (fld->lsf_new) {
5081                 rc = fld_update_from_controller(&env, fld);
5082                 if (rc != 0) {
5083                         CERROR("%s: cannot update controller: rc = %d\n",
5084                                mdt_obd_name(mdt), rc);
5085                         GOTO(out, rc);
5086                 }
5087         }
5088 out:
5089         lu_env_fini(&env);
5090         RETURN(rc);
5091 }
5092
5093 static int mdt_register_seq_exp(struct mdt_device *mdt)
5094 {
5095         struct seq_server_site  *ss = mdt_seq_site(mdt);
5096         char                    *lwp_name = NULL;
5097         int                     rc;
5098
5099         if (ss->ss_node_id == 0)
5100                 return 0;
5101
5102         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
5103         if (lwp_name == NULL)
5104                 GOTO(out_free, rc = -ENOMEM);
5105
5106         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
5107         if (rc != 0)
5108                 GOTO(out_free, rc);
5109
5110         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
5111                                       NULL, NULL);
5112         if (rc != 0)
5113                 GOTO(out_free, rc);
5114
5115         rc = lustre_register_lwp_item(lwp_name,
5116                                       &ss->ss_server_fld->lsf_control_exp,
5117                                       mdt_register_lwp_callback, mdt);
5118         if (rc != 0) {
5119                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
5120                 ss->ss_client_seq->lcs_exp = NULL;
5121                 GOTO(out_free, rc);
5122         }
5123 out_free:
5124         if (lwp_name != NULL)
5125                 OBD_FREE(lwp_name, MAX_OBD_NAME);
5126
5127         return rc;
5128 }
5129
5130 /*
5131  * Init client sequence manager which is used by local MDS to talk to sequence
5132  * controller on remote node.
5133  */
5134 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
5135 {
5136         struct seq_server_site *ss = mdt_seq_site(mdt);
5137         char *prefix;
5138         ENTRY;
5139
5140         /* check if this is adding the first MDC and controller is not yet
5141          * initialized. */
5142         OBD_ALLOC_PTR(ss->ss_client_seq);
5143         if (ss->ss_client_seq == NULL)
5144                 RETURN(-ENOMEM);
5145
5146         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
5147         if (prefix == NULL) {
5148                 OBD_FREE_PTR(ss->ss_client_seq);
5149                 ss->ss_client_seq = NULL;
5150                 RETURN(-ENOMEM);
5151         }
5152
5153         /* Note: seq_client_fini will be called in seq_site_fini */
5154         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
5155         seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
5156                         prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
5157                                                             NULL);
5158         OBD_FREE(prefix, MAX_OBD_NAME + 5);
5159
5160         RETURN(seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq));
5161 }
5162
5163 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
5164 {
5165         struct seq_server_site  *ss;
5166         int                     rc;
5167         ENTRY;
5168
5169         ss = mdt_seq_site(mdt);
5170         /* init sequence controller server(MDT0) */
5171         if (ss->ss_node_id == 0) {
5172                 OBD_ALLOC_PTR(ss->ss_control_seq);
5173                 if (ss->ss_control_seq == NULL)
5174                         RETURN(-ENOMEM);
5175
5176                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
5177                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
5178                                      ss);
5179                 if (rc)
5180                         GOTO(out_seq_fini, rc);
5181         }
5182
5183         /* Init normal sequence server */
5184         OBD_ALLOC_PTR(ss->ss_server_seq);
5185         if (ss->ss_server_seq == NULL)
5186                 GOTO(out_seq_fini, rc = -ENOMEM);
5187
5188         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
5189                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
5190         if (rc)
5191                 GOTO(out_seq_fini, rc);
5192
5193         /* init seq client for seq server to talk to seq controller(MDT0) */
5194         rc = mdt_seq_init_cli(env, mdt);
5195         if (rc != 0)
5196                 GOTO(out_seq_fini, rc);
5197
5198         if (ss->ss_node_id != 0)
5199                 /* register controller export through lwp */
5200                 rc = mdt_register_seq_exp(mdt);
5201
5202         EXIT;
5203 out_seq_fini:
5204         if (rc)
5205                 mdt_seq_fini(env, mdt);
5206
5207         return rc;
5208 }
5209
5210 /*
5211  * FLD wrappers
5212  */
5213 static int mdt_fld_fini(const struct lu_env *env,
5214                         struct mdt_device *m)
5215 {
5216         struct seq_server_site *ss = mdt_seq_site(m);
5217         ENTRY;
5218
5219         if (ss && ss->ss_server_fld) {
5220                 fld_server_fini(env, ss->ss_server_fld);
5221                 OBD_FREE_PTR(ss->ss_server_fld);
5222                 ss->ss_server_fld = NULL;
5223         }
5224
5225         RETURN(0);
5226 }
5227
5228 static int mdt_fld_init(const struct lu_env *env,
5229                         const char *uuid,
5230                         struct mdt_device *m)
5231 {
5232         struct seq_server_site *ss;
5233         int rc;
5234         ENTRY;
5235
5236         ss = mdt_seq_site(m);
5237
5238         OBD_ALLOC_PTR(ss->ss_server_fld);
5239         if (ss->ss_server_fld == NULL)
5240                 RETURN(rc = -ENOMEM);
5241
5242         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
5243                              LU_SEQ_RANGE_MDT);
5244         if (rc) {
5245                 OBD_FREE_PTR(ss->ss_server_fld);
5246                 ss->ss_server_fld = NULL;
5247                 RETURN(rc);
5248         }
5249
5250         RETURN(0);
5251 }
5252
5253 static void mdt_stack_pre_fini(const struct lu_env *env,
5254                            struct mdt_device *m, struct lu_device *top)
5255 {
5256         struct lustre_cfg_bufs  *bufs;
5257         struct lustre_cfg       *lcfg;
5258         struct mdt_thread_info  *info;
5259         ENTRY;
5260
5261         LASSERT(top);
5262
5263         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5264         LASSERT(info != NULL);
5265
5266         bufs = &info->mti_u.bufs;
5267
5268         LASSERT(m->mdt_child_exp);
5269         LASSERT(m->mdt_child_exp->exp_obd);
5270
5271         /* process cleanup, pass mdt obd name to get obd umount flags */
5272         /* XXX: this is needed because all layers are referenced by
5273          * objects (some of them are pinned by osd, for example *
5274          * the proper solution should be a model where object used
5275          * by osd only doesn't have mdt/mdd slices -bzzz */
5276         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
5277         lustre_cfg_bufs_set_string(bufs, 1, NULL);
5278         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5279         if (!lcfg)
5280                 RETURN_EXIT;
5281         lustre_cfg_init(lcfg, LCFG_PRE_CLEANUP, bufs);
5282
5283         top->ld_ops->ldo_process_config(env, top, lcfg);
5284         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5285         EXIT;
5286 }
5287
5288 static void mdt_stack_fini(const struct lu_env *env,
5289                            struct mdt_device *m, struct lu_device *top)
5290 {
5291         struct obd_device       *obd = mdt2obd_dev(m);
5292         struct lustre_cfg_bufs  *bufs;
5293         struct lustre_cfg       *lcfg;
5294         struct mdt_thread_info  *info;
5295         char                     flags[3] = "";
5296         ENTRY;
5297
5298         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5299         LASSERT(info != NULL);
5300
5301         lu_dev_del_linkage(top->ld_site, top);
5302
5303         lu_site_purge(env, top->ld_site, -1);
5304
5305         bufs = &info->mti_u.bufs;
5306         /* process cleanup, pass mdt obd name to get obd umount flags */
5307         /* another purpose is to let all layers to release their objects */
5308         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
5309         if (obd->obd_force)
5310                 strcat(flags, "F");
5311         if (obd->obd_fail)
5312                 strcat(flags, "A");
5313         lustre_cfg_bufs_set_string(bufs, 1, flags);
5314         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5315         if (!lcfg)
5316                 RETURN_EXIT;
5317         lustre_cfg_init(lcfg, LCFG_CLEANUP, bufs);
5318
5319         LASSERT(top);
5320         top->ld_ops->ldo_process_config(env, top, lcfg);
5321         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5322
5323         lu_site_purge(env, top->ld_site, -1);
5324
5325         m->mdt_child = NULL;
5326         m->mdt_bottom = NULL;
5327
5328         obd_disconnect(m->mdt_child_exp);
5329         m->mdt_child_exp = NULL;
5330
5331         obd_disconnect(m->mdt_bottom_exp);
5332         m->mdt_child_exp = NULL;
5333 }
5334
5335 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
5336                                const char *next, struct obd_export **exp)
5337 {
5338         struct obd_connect_data *data = NULL;
5339         struct obd_device       *obd;
5340         int                      rc;
5341         ENTRY;
5342
5343         OBD_ALLOC_PTR(data);
5344         if (data == NULL)
5345                 GOTO(out, rc = -ENOMEM);
5346
5347         obd = class_name2obd(next);
5348         if (obd == NULL) {
5349                 CERROR("%s: can't locate next device: %s\n",
5350                        mdt_obd_name(m), next);
5351                 GOTO(out, rc = -ENOTCONN);
5352         }
5353
5354         data->ocd_connect_flags = OBD_CONNECT_VERSION;
5355         data->ocd_version = LUSTRE_VERSION_CODE;
5356
5357         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
5358         if (rc) {
5359                 CERROR("%s: cannot connect to next dev %s (%d)\n",
5360                        mdt_obd_name(m), next, rc);
5361                 GOTO(out, rc);
5362         }
5363
5364 out:
5365         if (data)
5366                 OBD_FREE_PTR(data);
5367         RETURN(rc);
5368 }
5369
5370 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
5371                           struct lustre_cfg *cfg)
5372 {
5373         char                   *dev = lustre_cfg_string(cfg, 0);
5374         int                     rc, name_size, uuid_size;
5375         char                   *name, *uuid, *p;
5376         struct lustre_cfg_bufs *bufs;
5377         struct lustre_cfg      *lcfg;
5378         struct obd_device      *obd;
5379         struct lustre_profile  *lprof;
5380         struct lu_site         *site;
5381         ENTRY;
5382
5383         /* in 1.8 we had the only device in the stack - MDS.
5384          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
5385          * in 2.3 OSD is instantiated by obd_mount.c, so we need
5386          * to generate names and setup MDT, MDD. MDT will be using
5387          * generated name to connect to MDD. for MDD the next device
5388          * will be LOD with name taken from so called "profile" which
5389          * is generated by mount_option line
5390          *
5391          * 1.8 MGS generates config. commands like this:
5392          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
5393          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
5394          * 2.0 MGS generates config. commands like this:
5395          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
5396          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
5397          *                    3:lustre-MDT0000-mdtlov  4:f
5398          *
5399          * we generate MDD name from MDT one, just replacing T with D
5400          *
5401          * after all the preparations, the logical equivalent will be
5402          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
5403          *                    3:lustre-MDT0000-mdtlov  4:f
5404          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
5405          *                    3:lustre-MDD0000  4:f
5406          *
5407          *  notice we build the stack from down to top: MDD first, then MDT */
5408
5409         name_size = MAX_OBD_NAME;
5410         uuid_size = MAX_OBD_NAME;
5411
5412         OBD_ALLOC(name, name_size);
5413         OBD_ALLOC(uuid, uuid_size);
5414         if (name == NULL || uuid == NULL)
5415                 GOTO(cleanup_mem, rc = -ENOMEM);
5416
5417         OBD_ALLOC_PTR(bufs);
5418         if (!bufs)
5419                 GOTO(cleanup_mem, rc = -ENOMEM);
5420
5421         strcpy(name, dev);
5422         p = strstr(name, "-MDT");
5423         if (p == NULL)
5424                 GOTO(free_bufs, rc = -ENOMEM);
5425         p[3] = 'D';
5426
5427         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
5428
5429         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
5430         if (lprof == NULL || lprof->lp_dt == NULL) {
5431                 CERROR("can't find the profile: %s\n",
5432                        lustre_cfg_string(cfg, 0));
5433                 GOTO(free_bufs, rc = -EINVAL);
5434         }
5435
5436         lustre_cfg_bufs_reset(bufs, name);
5437         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
5438         lustre_cfg_bufs_set_string(bufs, 2, uuid);
5439         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
5440
5441         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5442         if (!lcfg)
5443                 GOTO(put_profile, rc = -ENOMEM);
5444         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
5445
5446         rc = class_attach(lcfg);
5447         if (rc)
5448                 GOTO(lcfg_cleanup, rc);
5449
5450         obd = class_name2obd(name);
5451         if (!obd) {
5452                 CERROR("Can not find obd %s (%s in config)\n",
5453                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
5454                 GOTO(lcfg_cleanup, rc = -EINVAL);
5455         }
5456
5457         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5458
5459         lustre_cfg_bufs_reset(bufs, name);
5460         lustre_cfg_bufs_set_string(bufs, 1, uuid);
5461         lustre_cfg_bufs_set_string(bufs, 2, dev);
5462         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
5463
5464         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5465         if (!lcfg)
5466                 GOTO(class_detach, rc = -ENOMEM);
5467         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
5468
5469         rc = class_setup(obd, lcfg);
5470         if (rc)
5471                 GOTO(class_detach, rc);
5472
5473         /* connect to MDD we just setup */
5474         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
5475         if (rc)
5476                 GOTO(class_detach, rc);
5477
5478         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
5479         LASSERT(site);
5480         LASSERT(mdt_lu_site(mdt) == NULL);
5481         mdt->mdt_lu_dev.ld_site = site;
5482         site->ls_top_dev = &mdt->mdt_lu_dev;
5483         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
5484
5485         /* now connect to bottom OSD */
5486         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
5487         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
5488         if (rc)
5489                 GOTO(class_detach, rc);
5490         mdt->mdt_bottom =
5491                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
5492
5493         rc = lu_env_refill((struct lu_env *)env);
5494         if (rc != 0)
5495                 CERROR("Failure to refill session: '%d'\n", rc);
5496
5497         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
5498
5499         EXIT;
5500 class_detach:
5501         if (rc)
5502                 class_detach(obd, lcfg);
5503 lcfg_cleanup:
5504         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5505 put_profile:
5506         class_put_profile(lprof);
5507 free_bufs:
5508         OBD_FREE_PTR(bufs);
5509 cleanup_mem:
5510         if (name)
5511                 OBD_FREE(name, name_size);
5512         if (uuid)
5513                 OBD_FREE(uuid, uuid_size);
5514         RETURN(rc);
5515 }
5516
5517 /* setup quota master target on MDT0 */
5518 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
5519                           struct lustre_cfg *cfg)
5520 {
5521         struct obd_device       *obd;
5522         char                    *dev = lustre_cfg_string(cfg, 0);
5523         char                    *qmtname, *uuid, *p;
5524         struct lustre_cfg_bufs  *bufs;
5525         struct lustre_cfg       *lcfg;
5526         struct lustre_profile   *lprof;
5527         struct obd_connect_data *data;
5528         int                      rc;
5529         ENTRY;
5530
5531         LASSERT(mdt->mdt_qmt_exp == NULL);
5532         LASSERT(mdt->mdt_qmt_dev == NULL);
5533
5534         /* quota master is on MDT0 only for now */
5535         if (mdt->mdt_seq_site.ss_node_id != 0)
5536                 RETURN(0);
5537
5538         /* MGS generates config commands which look as follows:
5539          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
5540          *                    3:lustre-MDT0000-mdtlov  4:f
5541          *
5542          * We generate the QMT name from the MDT one, just replacing MD with QM
5543          * after all the preparations, the logical equivalent will be:
5544          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
5545          *                    3:lustre-MDT0000-osd  4:f */
5546         OBD_ALLOC(qmtname, MAX_OBD_NAME);
5547         OBD_ALLOC(uuid, UUID_MAX);
5548         OBD_ALLOC_PTR(bufs);
5549         OBD_ALLOC_PTR(data);
5550         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
5551                 GOTO(cleanup_mem, rc = -ENOMEM);
5552
5553         strcpy(qmtname, dev);
5554         p = strstr(qmtname, "-MDT");
5555         if (p == NULL)
5556                 GOTO(cleanup_mem, rc = -ENOMEM);
5557         /* replace MD with QM */
5558         p[1] = 'Q';
5559         p[2] = 'M';
5560
5561         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
5562
5563         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
5564         if (lprof == NULL || lprof->lp_dt == NULL) {
5565                 CERROR("can't find profile for %s\n",
5566                        lustre_cfg_string(cfg, 0));
5567                 GOTO(cleanup_mem, rc = -EINVAL);
5568         }
5569
5570         lustre_cfg_bufs_reset(bufs, qmtname);
5571         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
5572         lustre_cfg_bufs_set_string(bufs, 2, uuid);
5573         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
5574
5575         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5576         if (!lcfg)
5577                 GOTO(put_profile, rc = -ENOMEM);
5578         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
5579
5580         rc = class_attach(lcfg);
5581         if (rc)
5582                 GOTO(lcfg_cleanup, rc);
5583
5584         obd = class_name2obd(qmtname);
5585         if (!obd) {
5586                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
5587                        lustre_cfg_string(cfg, 0));
5588                 GOTO(lcfg_cleanup, rc = -EINVAL);
5589         }
5590
5591         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5592
5593         lustre_cfg_bufs_reset(bufs, qmtname);
5594         lustre_cfg_bufs_set_string(bufs, 1, uuid);
5595         lustre_cfg_bufs_set_string(bufs, 2, dev);
5596
5597         /* for quota, the next device should be the OSD device */
5598         lustre_cfg_bufs_set_string(bufs, 3,
5599                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
5600
5601         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5602         if (!lcfg)
5603                 GOTO(class_detach, rc = -ENOMEM);
5604         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
5605
5606         rc = class_setup(obd, lcfg);
5607         if (rc)
5608                 GOTO(class_detach, rc);
5609
5610         mdt->mdt_qmt_dev = obd->obd_lu_dev;
5611
5612         /* configure local quota objects */
5613         if (OBD_FAIL_CHECK(OBD_FAIL_QUOTA_INIT))
5614                 rc = -EBADF;
5615         else
5616                 rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
5617                                                            &mdt->mdt_lu_dev,
5618                                                            mdt->mdt_qmt_dev);
5619         if (rc)
5620                 GOTO(class_cleanup, rc);
5621
5622         /* connect to quota master target */
5623         data->ocd_connect_flags = OBD_CONNECT_VERSION;
5624         data->ocd_version = LUSTRE_VERSION_CODE;
5625         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
5626                          data, NULL);
5627         if (rc) {
5628                 CERROR("cannot connect to quota master device %s (%d)\n",
5629                        qmtname, rc);
5630                 GOTO(class_cleanup, rc);
5631         }
5632
5633         EXIT;
5634 class_cleanup:
5635         if (rc) {
5636                 class_manual_cleanup(obd);
5637                 mdt->mdt_qmt_dev = NULL;
5638                 GOTO(lcfg_cleanup, rc);
5639         }
5640 class_detach:
5641         if (rc)
5642                 class_detach(obd, lcfg);
5643 lcfg_cleanup:
5644         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5645 put_profile:
5646         class_put_profile(lprof);
5647 cleanup_mem:
5648         if (bufs)
5649                 OBD_FREE_PTR(bufs);
5650         if (qmtname)
5651                 OBD_FREE(qmtname, MAX_OBD_NAME);
5652         if (uuid)
5653                 OBD_FREE(uuid, UUID_MAX);
5654         if (data)
5655                 OBD_FREE_PTR(data);
5656         return rc;
5657 }
5658
5659 /* Shutdown quota master target associated with mdt */
5660 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
5661 {
5662         ENTRY;
5663
5664         if (mdt->mdt_qmt_exp == NULL)
5665                 RETURN_EXIT;
5666         LASSERT(mdt->mdt_qmt_dev != NULL);
5667
5668         /* the qmt automatically shuts down when the mdt disconnects */
5669         obd_disconnect(mdt->mdt_qmt_exp);
5670         mdt->mdt_qmt_exp = NULL;
5671         mdt->mdt_qmt_dev = NULL;
5672         EXIT;
5673 }
5674
5675 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
5676  * for now. This will be removed along with converting rest of MDT code
5677  * to use tgt_session_info */
5678 static int mdt_tgt_getxattr(struct tgt_session_info *tsi)
5679 {
5680         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
5681         int                      rc;
5682
5683         if (unlikely(info->mti_object == NULL))
5684                 return -EPROTO;
5685
5686         rc = mdt_getxattr(info);
5687
5688         mdt_thread_info_fini(info);
5689         return rc;
5690 }
5691
5692 static int mdt_llog_open(struct tgt_session_info *tsi)
5693 {
5694         ENTRY;
5695
5696         if (!mdt_is_rootadmin(tsi2mdt_info(tsi)))
5697                 RETURN(err_serious(-EACCES));
5698
5699         RETURN(tgt_llog_open(tsi));
5700 }
5701
5702 #define OBD_FAIL_OST_READ_NET   OBD_FAIL_OST_BRW_NET
5703 #define OBD_FAIL_OST_WRITE_NET  OBD_FAIL_OST_BRW_NET
5704 #define OST_BRW_READ    OST_READ
5705 #define OST_BRW_WRITE   OST_WRITE
5706
5707 static struct tgt_handler mdt_tgt_handlers[] = {
5708 TGT_RPC_HANDLER(MDS_FIRST_OPC,
5709                 0,                      MDS_CONNECT,    mdt_tgt_connect,
5710                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
5711 TGT_RPC_HANDLER(MDS_FIRST_OPC,
5712                 0,                      MDS_DISCONNECT, tgt_disconnect,
5713                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
5714 TGT_RPC_HANDLER(MDS_FIRST_OPC,
5715                 HAS_REPLY,              MDS_SET_INFO,   mdt_set_info,
5716                 &RQF_MDT_SET_INFO, LUSTRE_MDS_VERSION),
5717 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
5718 TGT_MDT_HDL(HAS_REPLY,          MDS_GET_ROOT,   mdt_get_root),
5719 TGT_MDT_HDL(HAS_BODY,           MDS_GETATTR,    mdt_getattr),
5720 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_GETATTR_NAME,
5721                                                         mdt_getattr_name),
5722 TGT_MDT_HDL(HAS_BODY,           MDS_GETXATTR,   mdt_tgt_getxattr),
5723 TGT_MDT_HDL(HAS_REPLY,          MDS_STATFS,     mdt_statfs),
5724 TGT_MDT_HDL(IS_MUTABLE,         MDS_REINT,      mdt_reint),
5725 TGT_MDT_HDL(HAS_BODY,           MDS_CLOSE,      mdt_close),
5726 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_READPAGE,   mdt_readpage),
5727 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_SYNC,       mdt_sync),
5728 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
5729 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_PROGRESS,
5730                                                         mdt_hsm_progress),
5731 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_CT_REGISTER,
5732                                                         mdt_hsm_ct_register),
5733 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_CT_UNREGISTER,
5734                                                         mdt_hsm_ct_unregister),
5735 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_STATE_GET,
5736                                                         mdt_hsm_state_get),
5737 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_STATE_SET,
5738                                                         mdt_hsm_state_set),
5739 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_ACTION,       mdt_hsm_action),
5740 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_REQUEST,
5741                                                         mdt_hsm_request),
5742 TGT_MDT_HDL(HAS_KEY | HAS_BODY | HAS_REPLY | IS_MUTABLE,
5743             MDS_SWAP_LAYOUTS,
5744             mdt_swap_layouts),
5745 TGT_MDT_HDL(IS_MUTABLE,         MDS_RMFID,      mdt_rmfid),
5746 };
5747
5748 static struct tgt_handler mdt_io_ops[] = {
5749 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY, OST_BRW_READ, tgt_brw_read,
5750                                                         mdt_hp_brw),
5751 TGT_OST_HDL_HP(HAS_BODY | IS_MUTABLE,    OST_BRW_WRITE, tgt_brw_write,
5752                                                         mdt_hp_brw),
5753 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY | IS_MUTABLE,
5754                                          OST_PUNCH,     mdt_punch_hdl,
5755                                                         mdt_hp_punch),
5756 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SYNC,     mdt_data_sync),
5757 TGT_OST_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, OST_FALLOCATE,
5758                                                         mdt_fallocate_hdl),
5759 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SEEK, tgt_lseek),
5760 TGT_RPC_HANDLER(OST_FIRST_OPC,
5761                 0,                      OST_SET_INFO,   mdt_io_set_info,
5762                 &RQF_OBD_SET_INFO, LUSTRE_OST_VERSION),
5763 };
5764
5765 static struct tgt_handler mdt_sec_ctx_ops[] = {
5766 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
5767 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
5768 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
5769 };
5770
5771 static struct tgt_handler mdt_quota_ops[] = {
5772 TGT_QUOTA_HDL(HAS_REPLY,                QUOTA_DQACQ,      mdt_quota_dqacq),
5773 };
5774
5775 static struct tgt_handler mdt_llog_handlers[] = {
5776         TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_CREATE,      mdt_llog_open),
5777         TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_NEXT_BLOCK,  tgt_llog_next_block),
5778         TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
5779         TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_PREV_BLOCK,  tgt_llog_prev_block),
5780 };
5781
5782 static struct tgt_opc_slice mdt_common_slice[] = {
5783         {
5784                 .tos_opc_start  = MDS_FIRST_OPC,
5785                 .tos_opc_end    = MDS_LAST_OPC,
5786                 .tos_hs         = mdt_tgt_handlers
5787         },
5788         {
5789                 .tos_opc_start  = OBD_FIRST_OPC,
5790                 .tos_opc_end    = OBD_LAST_OPC,
5791                 .tos_hs         = tgt_obd_handlers
5792         },
5793         {
5794                 .tos_opc_start  = LDLM_FIRST_OPC,
5795                 .tos_opc_end    = LDLM_LAST_OPC,
5796                 .tos_hs         = tgt_dlm_handlers
5797         },
5798         {
5799                 .tos_opc_start  = SEC_FIRST_OPC,
5800                 .tos_opc_end    = SEC_LAST_OPC,
5801                 .tos_hs         = mdt_sec_ctx_ops
5802         },
5803         {
5804                 .tos_opc_start  = OUT_UPDATE_FIRST_OPC,
5805                 .tos_opc_end    = OUT_UPDATE_LAST_OPC,
5806                 .tos_hs         = tgt_out_handlers
5807         },
5808         {
5809                 .tos_opc_start  = FLD_FIRST_OPC,
5810                 .tos_opc_end    = FLD_LAST_OPC,
5811                 .tos_hs         = fld_handlers
5812         },
5813         {
5814                 .tos_opc_start  = SEQ_FIRST_OPC,
5815                 .tos_opc_end    = SEQ_LAST_OPC,
5816                 .tos_hs         = seq_handlers
5817         },
5818         {
5819                 .tos_opc_start  = QUOTA_DQACQ,
5820                 .tos_opc_end    = QUOTA_LAST_OPC,
5821                 .tos_hs         = mdt_quota_ops
5822         },
5823         {
5824                 .tos_opc_start  = LLOG_FIRST_OPC,
5825                 .tos_opc_end    = LLOG_LAST_OPC,
5826                 .tos_hs         = mdt_llog_handlers
5827         },
5828         {
5829                 .tos_opc_start  = LFSCK_FIRST_OPC,
5830                 .tos_opc_end    = LFSCK_LAST_OPC,
5831                 .tos_hs         = tgt_lfsck_handlers
5832         },
5833         {
5834                 .tos_opc_start  = OST_FIRST_OPC,
5835                 .tos_opc_end    = OST_LAST_OPC,
5836                 .tos_hs         = mdt_io_ops
5837         },
5838         {
5839                 .tos_hs         = NULL
5840         }
5841 };
5842
5843 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
5844 {
5845         struct md_device *next = m->mdt_child;
5846         struct lu_device *d = &m->mdt_lu_dev;
5847         struct obd_device *obd = mdt2obd_dev(m);
5848         struct lfsck_stop stop;
5849
5850         ENTRY;
5851         stop.ls_status = LS_PAUSED;
5852         stop.ls_flags = 0;
5853         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
5854
5855         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
5856
5857         mdt_restriper_stop(m);
5858         ping_evictor_stop();
5859
5860         /* Remove the HSM /proc entry so the coordinator cannot be
5861          * restarted by a user while it's shutting down.
5862          */
5863         mdt_hsm_cdt_stop(m);
5864
5865         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
5866         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
5867
5868         if (m->mdt_namespace != NULL)
5869                 ldlm_namespace_free_prior(m->mdt_namespace, NULL,
5870                                           d->ld_obd->obd_force);
5871
5872         obd_exports_barrier(obd);
5873         obd_zombie_barrier();
5874
5875         mdt_quota_fini(env, m);
5876
5877         cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
5878
5879         /* Calling the cleanup functions in the same order as in the mdt_init0
5880          * error path
5881          */
5882         mdt_tunables_fini(m);
5883
5884         target_recovery_fini(obd);
5885         upcall_cache_cleanup(m->mdt_identity_cache);
5886         m->mdt_identity_cache = NULL;
5887
5888         tgt_fini(env, &m->mdt_lut);
5889
5890         mdt_hsm_cdt_fini(m);
5891
5892         if (m->mdt_los != NULL) {
5893                 local_oid_storage_fini(env, m->mdt_los);
5894                 m->mdt_los = NULL;
5895         }
5896
5897         if (m->mdt_namespace != NULL) {
5898                 ldlm_namespace_free_post(m->mdt_namespace);
5899                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
5900         }
5901
5902         if (m->mdt_md_root != NULL) {
5903                 mdt_object_put(env, m->mdt_md_root);
5904                 m->mdt_md_root = NULL;
5905         }
5906
5907         mdt_seq_fini(env, m);
5908
5909         mdt_fld_fini(env, m);
5910
5911         /*
5912          * Finish the stack
5913          */
5914         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
5915
5916         LASSERT(atomic_read(&d->ld_ref) == 0);
5917
5918         server_put_mount(mdt_obd_name(m), true);
5919
5920         EXIT;
5921 }
5922
5923 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
5924
5925 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
5926                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
5927 {
5928         const struct dt_device_param *dt_conf;
5929         struct mdt_thread_info *info;
5930         struct obd_device *obd;
5931         const char *dev = lustre_cfg_string(cfg, 0);
5932         const char *num = lustre_cfg_string(cfg, 2);
5933         struct tg_grants_data *tgd = &m->mdt_lut.lut_tgd;
5934         struct lustre_mount_info *lmi = NULL;
5935         struct lustre_sb_info *lsi;
5936         struct lu_site *s;
5937         struct seq_server_site *ss_site;
5938         const char *identity_upcall = "NONE";
5939         struct md_device *next;
5940         struct lu_fid fid;
5941         int rc;
5942         long node_id;
5943         mntopt_t mntopts;
5944         ENTRY;
5945
5946         lu_device_init(&m->mdt_lu_dev, ldt);
5947         /*
5948          * Environment (env) might be missing mdt_thread_key values at that
5949          * point, if device is allocated when mdt_thread_key is in QUIESCENT
5950          * mode.
5951          *
5952          * Usually device allocation path doesn't use module key values, but
5953          * mdt has to do a lot of work here, so allocate key value.
5954          */
5955         rc = lu_env_refill((struct lu_env *)env);
5956         if (rc != 0)
5957                 RETURN(rc);
5958
5959         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5960         LASSERT(info != NULL);
5961
5962         obd = class_name2obd(dev);
5963         LASSERT(obd != NULL);
5964
5965         m->mdt_max_mdsize = MAX_MD_SIZE_OLD;
5966         m->mdt_opts.mo_evict_tgt_nids = 1;
5967         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
5968
5969         lmi = server_get_mount(dev);
5970         if (lmi == NULL) {
5971                 CERROR("Cannot get mount info for %s!\n", dev);
5972                 RETURN(-EFAULT);
5973         } else {
5974                 lsi = s2lsi(lmi->lmi_sb);
5975                 LASSERT(lsi->lsi_lmd);
5976                 /* CMD is supported only in IAM mode */
5977                 LASSERT(num);
5978                 rc = kstrtol(num, 10, &node_id);
5979                 if (rc)
5980                         RETURN(rc);
5981
5982                 obd_obt_init(obd);
5983                 if (lsi->lsi_lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
5984                         m->mdt_skip_lfsck = 1;
5985         }
5986
5987         /* Just try to get a DoM lock by default. Otherwise, having a group
5988          * lock granted, it may get blocked for a long time. */
5989         m->mdt_opts.mo_dom_lock = TRYLOCK_DOM_ON_OPEN;
5990         /* DoM files are read at open and data is packed in the reply */
5991         m->mdt_opts.mo_dom_read_open = 1;
5992
5993         m->mdt_squash.rsi_uid = 0;
5994         m->mdt_squash.rsi_gid = 0;
5995         INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
5996         spin_lock_init(&m->mdt_squash.rsi_lock);
5997         spin_lock_init(&m->mdt_lock);
5998         m->mdt_enable_chprojid_gid = 0;
5999         m->mdt_enable_dir_migration = 1;
6000         m->mdt_enable_dir_restripe = 0;
6001         m->mdt_enable_dir_auto_split = 0;
6002         m->mdt_enable_parallel_rename_dir = 1;
6003         m->mdt_enable_parallel_rename_file = 1;
6004         m->mdt_enable_remote_dir = 1;
6005         m->mdt_enable_remote_dir_gid = 0;
6006         m->mdt_enable_remote_rename = 1;
6007         m->mdt_enable_remote_subdir_mount = 1;
6008         m->mdt_enable_striped_dir = 1;
6009         m->mdt_dir_restripe_nsonly = 1;
6010
6011         atomic_set(&m->mdt_mds_mds_conns, 0);
6012         atomic_set(&m->mdt_async_commit_count, 0);
6013
6014         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
6015         m->mdt_lu_dev.ld_obd = obd;
6016         /* Set this lu_device to obd for error handling purposes. */
6017         obd->obd_lu_dev = &m->mdt_lu_dev;
6018
6019         /* init the stack */
6020         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
6021         if (rc) {
6022                 CERROR("%s: Can't init device stack, rc %d\n",
6023                        mdt_obd_name(m), rc);
6024                 GOTO(err_lmi, rc);
6025         }
6026
6027         s = mdt_lu_site(m);
6028         ss_site = mdt_seq_site(m);
6029         s->ld_seq_site = ss_site;
6030         ss_site->ss_lu = s;
6031
6032         /* set server index */
6033         ss_site->ss_node_id = node_id;
6034
6035         /* failover is the default
6036          * FIXME: we do not failout mds0/mgs, which may cause some problems.
6037          * assumed whose ss_node_id == 0 XXX
6038          * */
6039         obd->obd_replayable = 1;
6040         /* No connection accepted until configurations will finish */
6041         obd->obd_no_conn = 1;
6042
6043         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
6044                 char *str = lustre_cfg_string(cfg, 4);
6045                 if (strchr(str, 'n')) {
6046                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
6047                         obd->obd_replayable = 0;
6048                 }
6049         }
6050
6051         rc = mdt_fld_init(env, mdt_obd_name(m), m);
6052         if (rc)
6053                 GOTO(err_fini_stack, rc);
6054
6055         rc = mdt_seq_init(env, m);
6056         if (rc)
6057                 GOTO(err_fini_fld, rc);
6058
6059         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
6060                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
6061         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
6062                                               LDLM_NAMESPACE_SERVER,
6063                                               LDLM_NAMESPACE_GREEDY,
6064                                               LDLM_NS_TYPE_MDT);
6065         if (IS_ERR(m->mdt_namespace)) {
6066                 rc = PTR_ERR(m->mdt_namespace);
6067                 CERROR("%s: unable to create server namespace: rc = %d\n",
6068                        obd->obd_name, rc);
6069                 m->mdt_namespace = NULL;
6070                 GOTO(err_fini_seq, rc);
6071         }
6072
6073         m->mdt_namespace->ns_lvbp = m;
6074         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
6075
6076         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
6077         /* set obd_namespace for compatibility with old code */
6078         obd->obd_namespace = m->mdt_namespace;
6079
6080         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
6081                       OBD_FAIL_MDS_ALL_REQUEST_NET,
6082                       OBD_FAIL_MDS_ALL_REPLY_NET);
6083         if (rc)
6084                 GOTO(err_free_ns, rc);
6085
6086         /* Amount of available space excluded from granting and reserved
6087          * for metadata. It is a percentage of the total MDT size. */
6088         tgd->tgd_reserved_pcnt = 10;
6089
6090         if (ONE_MB_BRW_SIZE < (1U << tgd->tgd_blockbits))
6091                 m->mdt_brw_size = 1U << tgd->tgd_blockbits;
6092         else
6093                 m->mdt_brw_size = ONE_MB_BRW_SIZE;
6094
6095         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_FS_SETUP))
6096                 GOTO(err_tgt, rc = -ENOENT);
6097
6098         fid.f_seq = FID_SEQ_LOCAL_NAME;
6099         fid.f_oid = 1;
6100         fid.f_ver = 0;
6101         rc = local_oid_storage_init(env, m->mdt_bottom, &fid, &m->mdt_los);
6102         if (rc != 0)
6103                 GOTO(err_tgt, rc);
6104
6105         rc = mdt_hsm_cdt_init(m);
6106         if (rc != 0) {
6107                 CERROR("%s: error initializing coordinator, rc %d\n",
6108                        mdt_obd_name(m), rc);
6109                 GOTO(err_los_fini, rc);
6110         }
6111
6112         tgt_adapt_sptlrpc_conf(&m->mdt_lut);
6113
6114         next = m->mdt_child;
6115         dt_conf = next->md_ops->mdo_dtconf_get(env, next);
6116
6117         mntopts = dt_conf->ddp_mntopts;
6118
6119         if (mntopts & MNTOPT_USERXATTR)
6120                 m->mdt_opts.mo_user_xattr = 1;
6121         else
6122                 m->mdt_opts.mo_user_xattr = 0;
6123
6124         m->mdt_max_ea_size = dt_conf->ddp_max_ea_size;
6125
6126         if (mntopts & MNTOPT_ACL)
6127                 m->mdt_opts.mo_acl = 1;
6128         else
6129                 m->mdt_opts.mo_acl = 0;
6130
6131         m->mdt_opts.mo_enable_strict_som = 1;
6132
6133         /* XXX: to support suppgid for ACL, we enable identity_upcall
6134          * by default, otherwise, maybe got unexpected -EACCESS. */
6135         if (m->mdt_opts.mo_acl)
6136                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
6137
6138         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
6139                                                 identity_upcall,
6140                                                 &mdt_identity_upcall_cache_ops);
6141         if (IS_ERR(m->mdt_identity_cache)) {
6142                 rc = PTR_ERR(m->mdt_identity_cache);
6143                 m->mdt_identity_cache = NULL;
6144                 GOTO(err_free_hsm, rc);
6145         }
6146
6147         rc = mdt_tunables_init(m, dev);
6148         if (rc) {
6149                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
6150                 GOTO(err_recovery, rc);
6151         }
6152
6153         rc = mdt_quota_init(env, m, cfg);
6154         if (rc)
6155                 GOTO(err_procfs, rc);
6156
6157         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
6158         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
6159                            "mdt_ldlm_client", m->mdt_ldlm_client);
6160
6161         ping_evictor_start();
6162
6163         /* recovery will be started upon mdt_prepare()
6164          * when the whole stack is complete and ready
6165          * to serve the requests */
6166
6167         /* Reduce the initial timeout on an MDS because it doesn't need such
6168          * a long timeout as an OST does. Adaptive timeouts will adjust this
6169          * value appropriately. */
6170         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
6171                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
6172
6173         if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_LOCAL_RECOV))
6174                 m->mdt_lut.lut_local_recovery = 1;
6175
6176         rc = mdt_restriper_start(m);
6177         if (rc)
6178                 GOTO(err_ping_evictor, rc);
6179
6180         RETURN(0);
6181
6182 err_ping_evictor:
6183         ping_evictor_stop();
6184 err_procfs:
6185         mdt_tunables_fini(m);
6186 err_recovery:
6187         upcall_cache_cleanup(m->mdt_identity_cache);
6188         m->mdt_identity_cache = NULL;
6189 err_free_hsm:
6190         mdt_hsm_cdt_fini(m);
6191 err_los_fini:
6192         local_oid_storage_fini(env, m->mdt_los);
6193         m->mdt_los = NULL;
6194 err_tgt:
6195         /* keep recoverable clients */
6196         obd->obd_fail = 1;
6197         target_recovery_fini(obd);
6198         obd_exports_barrier(obd);
6199         obd_zombie_barrier();
6200         tgt_fini(env, &m->mdt_lut);
6201 err_free_ns:
6202         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
6203         obd->obd_namespace = m->mdt_namespace = NULL;
6204 err_fini_seq:
6205         mdt_seq_fini(env, m);
6206 err_fini_fld:
6207         mdt_fld_fini(env, m);
6208 err_fini_stack:
6209         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
6210 err_lmi:
6211         if (lmi)
6212                 server_put_mount(dev, true);
6213         return(rc);
6214 }
6215
6216 /* For interoperability, the left element is old parameter, the right one
6217  * is the new version of the parameter, if some parameter is deprecated,
6218  * the new version should be set as NULL. */
6219 static struct cfg_interop_param mdt_interop_param[] = {
6220         { "mdt.group_upcall",   NULL },
6221         { "mdt.quota_type",     NULL },
6222         { "mdd.quota_type",     NULL },
6223         { "mdt.som",            NULL },
6224         { "mdt.rootsquash",     "mdt.root_squash" },
6225         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
6226         { NULL }
6227 };
6228
6229 /* used by MGS to process specific configurations */
6230 static int mdt_process_config(const struct lu_env *env,
6231                               struct lu_device *d, struct lustre_cfg *cfg)
6232 {
6233         struct mdt_device *m = mdt_dev(d);
6234         struct md_device *md_next = m->mdt_child;
6235         struct lu_device *next = md2lu_dev(md_next);
6236         int rc;
6237         ENTRY;
6238
6239         switch (cfg->lcfg_command) {
6240         case LCFG_PARAM: {
6241                 struct obd_device *obd = d->ld_obd;
6242                 /* For interoperability */
6243                 struct cfg_interop_param *ptr = NULL;
6244                 struct lustre_cfg *old_cfg = NULL;
6245                 char *param = NULL;
6246                 ssize_t count;
6247
6248                 param = lustre_cfg_string(cfg, 1);
6249                 if (param == NULL) {
6250                         CERROR("param is empty\n");
6251                         rc = -EINVAL;
6252                         break;
6253                 }
6254
6255                 ptr = class_find_old_param(param, mdt_interop_param);
6256                 if (ptr != NULL) {
6257                         if (ptr->new_param == NULL) {
6258                                 rc = 0;
6259                                 CWARN("For interoperability, skip this %s."
6260                                       " It is obsolete.\n", ptr->old_param);
6261                                 break;
6262                         }
6263
6264                         CWARN("Found old param %s, changed it to %s.\n",
6265                               ptr->old_param, ptr->new_param);
6266
6267                         old_cfg = cfg;
6268                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
6269                         if (IS_ERR(cfg)) {
6270                                 rc = PTR_ERR(cfg);
6271                                 break;
6272                         }
6273                 }
6274
6275                 count = class_modify_config(cfg, PARAM_MDT,
6276                                             &obd->obd_kset.kobj);
6277                 if (count < 0) {
6278                         struct coordinator *cdt = &m->mdt_coordinator;
6279
6280                         /* is it an HSM var ? */
6281                         count = class_modify_config(cfg, PARAM_HSM,
6282                                                     &cdt->cdt_hsm_kobj);
6283                         if (count < 0)
6284                                 /* we don't understand; pass it on */
6285                                 rc = next->ld_ops->ldo_process_config(env, next,
6286                                                                       cfg);
6287                         else
6288                                 rc = count > 0 ? 0 : count;
6289                 } else {
6290                         rc = count > 0 ? 0 : count;
6291                 }
6292
6293                 if (old_cfg)
6294                         OBD_FREE(cfg, lustre_cfg_len(cfg->lcfg_bufcount,
6295                                                      cfg->lcfg_buflens));
6296                 break;
6297         }
6298         default:
6299                 /* others are passed further */
6300                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
6301                 break;
6302         }
6303         RETURN(rc);
6304 }
6305
6306 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
6307                                           const struct lu_object_header *hdr,
6308                                           struct lu_device *d)
6309 {
6310         struct mdt_object *mo;
6311
6312         ENTRY;
6313
6314         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
6315         if (mo != NULL) {
6316                 struct lu_object *o;
6317                 struct lu_object_header *h;
6318
6319                 o = &mo->mot_obj;
6320                 h = &mo->mot_header;
6321                 lu_object_header_init(h);
6322                 lu_object_init(o, h, d);
6323                 lu_object_add_top(h, o);
6324                 o->lo_ops = &mdt_obj_ops;
6325                 spin_lock_init(&mo->mot_write_lock);
6326                 mutex_init(&mo->mot_som_mutex);
6327                 mutex_init(&mo->mot_lov_mutex);
6328                 init_rwsem(&mo->mot_dom_sem);
6329                 init_rwsem(&mo->mot_open_sem);
6330                 atomic_set(&mo->mot_open_count, 0);
6331                 mo->mot_restripe_offset = 0;
6332                 INIT_LIST_HEAD(&mo->mot_restripe_linkage);
6333                 mo->mot_lsom_size = 0;
6334                 mo->mot_lsom_blocks = 0;
6335                 mo->mot_lsom_inited = false;
6336                 RETURN(o);
6337         }
6338         RETURN(NULL);
6339 }
6340
6341 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
6342                            const struct lu_object_conf *unused)
6343 {
6344         struct mdt_device *d = mdt_dev(o->lo_dev);
6345         struct lu_device  *under;
6346         struct lu_object  *below;
6347         int                rc = 0;
6348         ENTRY;
6349
6350         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
6351                PFID(lu_object_fid(o)));
6352
6353         under = &d->mdt_child->md_lu_dev;
6354         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
6355         if (below != NULL) {
6356                 lu_object_add(o, below);
6357         } else
6358                 rc = -ENOMEM;
6359
6360         RETURN(rc);
6361 }
6362
6363 static void mdt_object_free_rcu(struct rcu_head *head)
6364 {
6365         struct mdt_object *mo = container_of(head, struct mdt_object,
6366                                              mot_header.loh_rcu);
6367
6368         kmem_cache_free(mdt_object_kmem, mo);
6369 }
6370
6371 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
6372 {
6373         struct mdt_object *mo = mdt_obj(o);
6374         struct lu_object_header *h;
6375         ENTRY;
6376
6377         h = o->lo_header;
6378         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
6379                PFID(lu_object_fid(o)));
6380
6381         LASSERT(atomic_read(&mo->mot_open_count) == 0);
6382         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
6383
6384         lu_object_fini(o);
6385         lu_object_header_fini(h);
6386         OBD_FREE_PRE(mo, sizeof(*mo), "slab-freed");
6387         call_rcu(&mo->mot_header.loh_rcu, mdt_object_free_rcu);
6388
6389         EXIT;
6390 }
6391
6392 static int mdt_object_print(const struct lu_env *env, void *cookie,
6393                             lu_printer_t p, const struct lu_object *o)
6394 {
6395         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
6396
6397         return (*p)(env, cookie,
6398                     LUSTRE_MDT_NAME"-object@%p(%s %s, writecount=%d)",
6399                     mdto, mdto->mot_lov_created ? "lov_created" : "",
6400                     mdto->mot_cache_attr ? "cache_attr" : "",
6401                     mdto->mot_write_count);
6402 }
6403
6404 static int mdt_prepare(const struct lu_env *env,
6405                 struct lu_device *pdev,
6406                 struct lu_device *cdev)
6407 {
6408         struct mdt_device *mdt = mdt_dev(cdev);
6409         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
6410         struct obd_device *obd = cdev->ld_obd;
6411         int rc;
6412
6413         ENTRY;
6414
6415         LASSERT(obd);
6416
6417         rc = next->ld_ops->ldo_prepare(env, cdev, next);
6418         if (rc)
6419                 RETURN(rc);
6420
6421         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
6422         if (rc)
6423                 RETURN(rc);
6424
6425         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
6426         if (rc)
6427                 RETURN(rc);
6428
6429         rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
6430         /* The LFSCK instance is registered just now, so it must be there when
6431          * register the namespace to such instance. */
6432         LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
6433
6434         if (mdt->mdt_seq_site.ss_node_id == 0) {
6435                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
6436                                                          &mdt->mdt_md_root_fid);
6437                 if (rc)
6438                         RETURN(rc);
6439         }
6440
6441         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
6442
6443         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
6444         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
6445         LASSERT(obd->obd_no_conn);
6446         spin_lock(&obd->obd_dev_lock);
6447         obd->obd_no_conn = 0;
6448         spin_unlock(&obd->obd_dev_lock);
6449
6450         if (obd->obd_recovering == 0)
6451                 mdt_postrecov(env, mdt);
6452
6453         RETURN(rc);
6454 }
6455
6456 const struct lu_device_operations mdt_lu_ops = {
6457         .ldo_object_alloc   = mdt_object_alloc,
6458         .ldo_process_config = mdt_process_config,
6459         .ldo_prepare        = mdt_prepare,
6460 };
6461
6462 static const struct lu_object_operations mdt_obj_ops = {
6463         .loo_object_init    = mdt_object_init,
6464         .loo_object_free    = mdt_object_free,
6465         .loo_object_print   = mdt_object_print
6466 };
6467
6468 static int mdt_obd_set_info_async(const struct lu_env *env,
6469                                   struct obd_export *exp,
6470                                   __u32 keylen, void *key,
6471                                   __u32 vallen, void *val,
6472                                   struct ptlrpc_request_set *set)
6473 {
6474         int rc;
6475
6476         ENTRY;
6477
6478         if (KEY_IS(KEY_SPTLRPC_CONF)) {
6479                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp));
6480                 RETURN(rc);
6481         }
6482
6483         RETURN(0);
6484 }
6485
6486 static inline void mdt_enable_slc(struct mdt_device *mdt)
6487 {
6488         if (mdt->mdt_lut.lut_sync_lock_cancel == SYNC_LOCK_CANCEL_NEVER)
6489                 mdt->mdt_lut.lut_sync_lock_cancel = SYNC_LOCK_CANCEL_BLOCKING;
6490 }
6491
6492 static inline void mdt_disable_slc(struct mdt_device *mdt)
6493 {
6494         if (mdt->mdt_lut.lut_sync_lock_cancel == SYNC_LOCK_CANCEL_BLOCKING)
6495                 mdt->mdt_lut.lut_sync_lock_cancel = SYNC_LOCK_CANCEL_NEVER;
6496 }
6497
6498 /**
6499  * Match client and server connection feature flags.
6500  *
6501  * Compute the compatibility flags for a connection request based on
6502  * features mutually supported by client and server.
6503  *
6504  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
6505  * must not be updated here, otherwise a partially initialized value may
6506  * be exposed. After the connection request is successfully processed,
6507  * the top-level MDT connect request handler atomically updates the export
6508  * connect flags from the obd_connect_data::ocd_connect_flags field of the
6509  * reply. \see mdt_connect().
6510  *
6511  * Before 2.7.50 clients will send a struct obd_connect_data_v1 rather than a
6512  * full struct obd_connect_data. So care must be taken when accessing fields
6513  * that are not present in struct obd_connect_data_v1. See LU-16.
6514  *
6515  * \param exp   the obd_export associated with this client/target pair
6516  * \param mdt   the target device for the connection
6517  * \param data  stores data for this connect request
6518  *
6519  * \retval 0       success
6520  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
6521  * \retval -EBADE  client and server feature requirements are incompatible
6522  */
6523 static int mdt_connect_internal(const struct lu_env *env,
6524                                 struct obd_export *exp,
6525                                 struct mdt_device *mdt,
6526                                 struct obd_connect_data *data, bool reconnect)
6527 {
6528         const char *obd_name = mdt_obd_name(mdt);
6529         LASSERT(data != NULL);
6530
6531         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
6532
6533         if (mdt->mdt_bottom->dd_rdonly &&
6534             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
6535             !(data->ocd_connect_flags & OBD_CONNECT_RDONLY))
6536                 RETURN(-EACCES);
6537
6538         if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
6539                 data->ocd_connect_flags2 &= MDT_CONNECT_SUPPORTED2;
6540
6541         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
6542
6543         if (!mdt->mdt_opts.mo_acl)
6544                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
6545
6546         if (!mdt->mdt_opts.mo_user_xattr)
6547                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
6548
6549         if (OCD_HAS_FLAG(data, BRW_SIZE)) {
6550                 data->ocd_brw_size = min(data->ocd_brw_size,
6551                                          mdt->mdt_brw_size);
6552                 if (data->ocd_brw_size == 0) {
6553                         CERROR("%s: cli %s/%p ocd_connect_flags: %#llx "
6554                                "ocd_version: %x ocd_grant: %d ocd_index: %u "
6555                                "ocd_brw_size unexpectedly zero, network data "
6556                                "corruption? Refusing to connect this client\n",
6557                                obd_name, exp->exp_client_uuid.uuid,
6558                                exp, data->ocd_connect_flags, data->ocd_version,
6559                                data->ocd_grant, data->ocd_index);
6560                         return -EPROTO;
6561                 }
6562         }
6563
6564         if (OCD_HAS_FLAG(data, GRANT_PARAM)) {
6565                 struct dt_device_param *ddp = &mdt->mdt_lut.lut_dt_conf;
6566
6567                 /* client is reporting its page size, for future use */
6568                 exp->exp_target_data.ted_pagebits = data->ocd_grant_blkbits;
6569                 data->ocd_grant_blkbits  = mdt->mdt_lut.lut_tgd.tgd_blockbits;
6570                 /* ddp_inodespace may not be power-of-two value, eg. for ldiskfs
6571                  * it's LDISKFS_DIR_REC_LEN(20) = 28. */
6572                 data->ocd_grant_inobits = fls(ddp->ddp_inodespace - 1);
6573                 /* ocd_grant_tax_kb is in 1K byte blocks */
6574                 data->ocd_grant_tax_kb = ddp->ddp_extent_tax >> 10;
6575                 data->ocd_grant_max_blks = ddp->ddp_max_extent_blks;
6576         }
6577
6578         /* Save connect_data we have so far because tgt_grant_connect()
6579          * uses it to calculate grant, and we want to save the client
6580          * version before it is overwritten by LUSTRE_VERSION_CODE. */
6581         exp->exp_connect_data = *data;
6582         if (OCD_HAS_FLAG(data, GRANT))
6583                 tgt_grant_connect(env, exp, data, !reconnect);
6584
6585         if (OCD_HAS_FLAG(data, MAXBYTES))
6586                 data->ocd_maxbytes = mdt->mdt_lut.lut_dt_conf.ddp_maxbytes;
6587
6588         /* NB: Disregard the rule against updating
6589          * exp_connect_data.ocd_connect_flags in this case, since
6590          * tgt_client_new() needs to know if this is a lightweight
6591          * connection, and it is safe to expose this flag before
6592          * connection processing completes. */
6593         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
6594                 spin_lock(&exp->exp_lock);
6595                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
6596                 spin_unlock(&exp->exp_lock);
6597         }
6598
6599         data->ocd_version = LUSTRE_VERSION_CODE;
6600
6601         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
6602                 CWARN("%s: MDS requires FID support, but client not\n",
6603                       obd_name);
6604                 return -EBADE;
6605         }
6606
6607         if (OCD_HAS_FLAG(data, PINGLESS)) {
6608                 if (ptlrpc_pinger_suppress_pings()) {
6609                         spin_lock(&exp->exp_obd->obd_dev_lock);
6610                         list_del_init(&exp->exp_obd_chain_timed);
6611                         spin_unlock(&exp->exp_obd->obd_dev_lock);
6612                 } else {
6613                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
6614                 }
6615         }
6616
6617         data->ocd_max_easize = mdt->mdt_max_ea_size;
6618
6619         /* NB: Disregard the rule against updating
6620          * exp_connect_data.ocd_connect_flags in this case, since
6621          * tgt_client_new() needs to know if this is client supports
6622          * multiple modify RPCs, and it is safe to expose this flag before
6623          * connection processing completes. */
6624         if (data->ocd_connect_flags & OBD_CONNECT_MULTIMODRPCS) {
6625                 data->ocd_maxmodrpcs = max_mod_rpcs_per_client;
6626                 spin_lock(&exp->exp_lock);
6627                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_MULTIMODRPCS;
6628                 spin_unlock(&exp->exp_lock);
6629         }
6630
6631         if (OCD_HAS_FLAG(data, CKSUM)) {
6632                 __u32 cksum_types = data->ocd_cksum_types;
6633
6634                 tgt_mask_cksum_types(&mdt->mdt_lut, &data->ocd_cksum_types);
6635
6636                 if (unlikely(data->ocd_cksum_types == 0)) {
6637                         CERROR("%s: Connect with checksum support but no "
6638                                "ocd_cksum_types is set\n",
6639                                exp->exp_obd->obd_name);
6640                         RETURN(-EPROTO);
6641                 }
6642
6643                 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
6644                        "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
6645                        cksum_types, data->ocd_cksum_types);
6646         } else {
6647                 /* This client does not support OBD_CONNECT_CKSUM
6648                  * fall back to CRC32 */
6649                 CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
6650                        "OBD_CONNECT_CKSUM, CRC32 will be used\n",
6651                        exp->exp_obd->obd_name, obd_export_nid2str(exp));
6652         }
6653
6654         if ((data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
6655             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
6656                 atomic_inc(&mdt->mdt_mds_mds_conns);
6657                 mdt_enable_slc(mdt);
6658         }
6659
6660         if (!mdt->mdt_lut.lut_dt_conf.ddp_has_lseek_data_hole)
6661                 data->ocd_connect_flags2 &= ~OBD_CONNECT2_LSEEK;
6662
6663         return 0;
6664 }
6665
6666 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
6667                                    struct mdt_thread_info *info,
6668                                    struct mdt_file_data *mfd)
6669 {
6670         struct lu_context ses;
6671         int rc;
6672         ENTRY;
6673
6674         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
6675         if (rc)
6676                 RETURN(rc);
6677
6678         env->le_ses = &ses;
6679         lu_context_enter(&ses);
6680
6681         mdt_ucred(info)->uc_valid = UCRED_OLD;
6682         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
6683
6684         lu_context_exit(&ses);
6685         lu_context_fini(&ses);
6686         env->le_ses = NULL;
6687
6688         RETURN(rc);
6689 }
6690
6691 static int mdt_export_cleanup(struct obd_export *exp)
6692 {
6693         LIST_HEAD(closing_list);
6694         struct mdt_export_data  *med = &exp->exp_mdt_data;
6695         struct obd_device       *obd = exp->exp_obd;
6696         struct mdt_device       *mdt;
6697         struct mdt_thread_info  *info;
6698         struct lu_env            env;
6699         struct mdt_file_data    *mfd, *n;
6700         int rc = 0;
6701         ENTRY;
6702
6703         spin_lock(&med->med_open_lock);
6704         while (!list_empty(&med->med_open_head)) {
6705                 struct list_head *tmp = med->med_open_head.next;
6706                 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
6707
6708                 /* Remove mfd handle so it can't be found again.
6709                  * We are consuming the mfd_list reference here. */
6710                 class_handle_unhash(&mfd->mfd_open_handle);
6711                 list_move_tail(&mfd->mfd_list, &closing_list);
6712         }
6713         spin_unlock(&med->med_open_lock);
6714         mdt = mdt_dev(obd->obd_lu_dev);
6715         LASSERT(mdt != NULL);
6716
6717         rc = lu_env_init(&env, LCT_MD_THREAD);
6718         if (rc)
6719                 RETURN(rc);
6720
6721         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6722         LASSERT(info != NULL);
6723         memset(info, 0, sizeof *info);
6724         info->mti_env = &env;
6725         info->mti_mdt = mdt;
6726         info->mti_exp = exp;
6727
6728         if (!list_empty(&closing_list)) {
6729                 struct md_attr *ma = &info->mti_attr;
6730
6731                 /* Close any open files (which may also cause orphan
6732                  * unlinking). */
6733                 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
6734                         list_del_init(&mfd->mfd_list);
6735                         ma->ma_need = ma->ma_valid = 0;
6736
6737                         /* This file is being closed due to an eviction, it
6738                          * could have been modified and now dirty regarding to
6739                          * HSM archive, check this!
6740                          * The logic here is to mark a file dirty if there's a
6741                          * chance it was dirtied before the client was evicted,
6742                          * so that we don't have to wait for a release attempt
6743                          * before finding out the file was actually dirty and
6744                          * fail the release. Aggressively marking it dirty here
6745                          * will cause the policy engine to attempt to
6746                          * re-archive it; when rearchiving, we can compare the
6747                          * current version to the HSM data_version and make the
6748                          * archive request into a noop if it's not actually
6749                          * dirty.
6750                          */
6751                         if (mfd->mfd_open_flags & MDS_FMODE_WRITE)
6752                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
6753
6754                         /* Don't unlink orphan on failover umount, LU-184 */
6755                         if (exp->exp_flags & OBD_OPT_FAILOVER ||
6756                             exp->exp_obd->obd_stopping) {
6757                                 ma->ma_valid = MA_FLAGS;
6758                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
6759                         }
6760                         ma->ma_valid |= MA_FORCE_LOG;
6761                         mdt_mfd_close(info, mfd);
6762                 }
6763         }
6764         info->mti_mdt = NULL;
6765         /* cleanup client slot early */
6766         /* Do not erase record for recoverable client. */
6767         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
6768                 tgt_client_del(&env, exp);
6769         lu_env_fini(&env);
6770
6771         RETURN(rc);
6772 }
6773
6774 static int mdt_obd_disconnect(struct obd_export *exp)
6775 {
6776         int rc;
6777
6778         ENTRY;
6779
6780         LASSERT(exp);
6781         class_export_get(exp);
6782
6783         if (!(exp->exp_flags & OBD_OPT_FORCE))
6784                 tgt_grant_sanity_check(exp->exp_obd, __func__);
6785
6786         if ((exp_connect_flags(exp) & OBD_CONNECT_MDS_MDS) &&
6787             !(exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT)) {
6788                 struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
6789
6790                 if (atomic_dec_and_test(&mdt->mdt_mds_mds_conns))
6791                         mdt_disable_slc(mdt);
6792         }
6793
6794         rc = server_disconnect_export(exp);
6795         if (rc != 0)
6796                 CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
6797
6798         tgt_grant_discard(exp);
6799
6800         rc = mdt_export_cleanup(exp);
6801         nodemap_del_member(exp);
6802         class_export_put(exp);
6803         RETURN(rc);
6804 }
6805
6806 /* mds_connect copy */
6807 static int mdt_obd_connect(const struct lu_env *env,
6808                            struct obd_export **exp, struct obd_device *obd,
6809                            struct obd_uuid *cluuid,
6810                            struct obd_connect_data *data,
6811                            void *localdata)
6812 {
6813         struct obd_export       *lexp;
6814         struct lustre_handle    conn = { 0 };
6815         struct mdt_device       *mdt;
6816         int                      rc;
6817         lnet_nid_t              *client_nid = localdata;
6818         ENTRY;
6819
6820         LASSERT(env != NULL);
6821         LASSERT(data != NULL);
6822
6823         if (!exp || !obd || !cluuid)
6824                 RETURN(-EINVAL);
6825
6826         mdt = mdt_dev(obd->obd_lu_dev);
6827
6828         /*
6829          * first, check whether the stack is ready to handle requests
6830          * XXX: probably not very appropriate method is used now
6831          *      at some point we should find a better one
6832          */
6833         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) &&
6834             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) &&
6835             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
6836                 rc = obd_get_info(env, mdt->mdt_child_exp,
6837                                   sizeof(KEY_OSP_CONNECTED),
6838                                   KEY_OSP_CONNECTED, NULL, NULL);
6839                 if (rc)
6840                         RETURN(-EAGAIN);
6841                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
6842         }
6843
6844         rc = class_connect(&conn, obd, cluuid);
6845         if (rc)
6846                 RETURN(rc);
6847
6848         lexp = class_conn2export(&conn);
6849         LASSERT(lexp != NULL);
6850
6851         rc = nodemap_add_member(*client_nid, lexp);
6852         if (rc != 0 && rc != -EEXIST)
6853                 GOTO(out, rc);
6854
6855         rc = mdt_connect_internal(env, lexp, mdt, data, false);
6856         if (rc == 0) {
6857                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
6858
6859                 LASSERT(lcd);
6860                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
6861                 rc = tgt_client_new(env, lexp);
6862                 if (rc == 0)
6863                         mdt_export_stats_init(obd, lexp, localdata);
6864         }
6865 out:
6866         if (rc != 0) {
6867                 class_disconnect(lexp);
6868                 nodemap_del_member(lexp);
6869                 *exp = NULL;
6870         } else {
6871                 *exp = lexp;
6872                 /* Because we do not want this export to be evicted by pinger,
6873                  * let's not add this export to the timed chain list. */
6874                 if (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) {
6875                         spin_lock(&lexp->exp_obd->obd_dev_lock);
6876                         list_del_init(&lexp->exp_obd_chain_timed);
6877                         spin_unlock(&lexp->exp_obd->obd_dev_lock);
6878                 }
6879         }
6880
6881         RETURN(rc);
6882 }
6883
6884 static int mdt_obd_reconnect(const struct lu_env *env,
6885                              struct obd_export *exp, struct obd_device *obd,
6886                              struct obd_uuid *cluuid,
6887                              struct obd_connect_data *data,
6888                              void *localdata)
6889 {
6890         lnet_nid_t             *client_nid = localdata;
6891         int                     rc;
6892         ENTRY;
6893
6894         if (exp == NULL || obd == NULL || cluuid == NULL)
6895                 RETURN(-EINVAL);
6896
6897         rc = nodemap_add_member(*client_nid, exp);
6898         if (rc != 0 && rc != -EEXIST)
6899                 RETURN(rc);
6900
6901         rc = mdt_connect_internal(env, exp, mdt_dev(obd->obd_lu_dev), data,
6902                                   true);
6903         if (rc == 0)
6904                 mdt_export_stats_init(obd, exp, localdata);
6905         else
6906                 nodemap_del_member(exp);
6907
6908         RETURN(rc);
6909 }
6910
6911 /* FIXME: Can we avoid using these two interfaces? */
6912 static int mdt_init_export(struct obd_export *exp)
6913 {
6914         struct mdt_export_data *med = &exp->exp_mdt_data;
6915         int                     rc;
6916         ENTRY;
6917
6918         INIT_LIST_HEAD(&med->med_open_head);
6919         spin_lock_init(&med->med_open_lock);
6920         spin_lock(&exp->exp_lock);
6921         exp->exp_connecting = 1;
6922         spin_unlock(&exp->exp_lock);
6923
6924         OBD_ALLOC(exp->exp_used_slots,
6925                   BITS_TO_LONGS(OBD_MAX_RIF_MAX) * sizeof(long));
6926         if (exp->exp_used_slots == NULL)
6927                 RETURN(-ENOMEM);
6928
6929         /* self-export doesn't need client data and ldlm initialization */
6930         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
6931                                      &exp->exp_client_uuid)))
6932                 RETURN(0);
6933
6934         rc = tgt_client_alloc(exp);
6935         if (rc)
6936                 GOTO(err, rc);
6937
6938         rc = ldlm_init_export(exp);
6939         if (rc)
6940                 GOTO(err_free, rc);
6941
6942         RETURN(rc);
6943
6944 err_free:
6945         tgt_client_free(exp);
6946 err:
6947         OBD_FREE(exp->exp_used_slots,
6948                  BITS_TO_LONGS(OBD_MAX_RIF_MAX) * sizeof(long));
6949         exp->exp_used_slots = NULL;
6950
6951         CERROR("%s: Failed to initialize export: rc = %d\n",
6952                exp->exp_obd->obd_name, rc);
6953         return rc;
6954 }
6955
6956 static int mdt_destroy_export(struct obd_export *exp)
6957 {
6958         ENTRY;
6959
6960         target_destroy_export(exp);
6961         if (exp->exp_used_slots)
6962                 OBD_FREE(exp->exp_used_slots,
6963                          BITS_TO_LONGS(OBD_MAX_RIF_MAX) * sizeof(long));
6964
6965         /* destroy can be called from failed obd_setup, so
6966          * checking uuid is safer than obd_self_export */
6967         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
6968                                      &exp->exp_client_uuid)))
6969                 RETURN(0);
6970
6971         ldlm_destroy_export(exp);
6972         tgt_client_free(exp);
6973
6974         LASSERT(list_empty(&exp->exp_outstanding_replies));
6975         LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
6976
6977         /*
6978          * discard grants once we're sure no more
6979          * interaction with the client is possible
6980          */
6981         tgt_grant_discard(exp);
6982         if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
6983                 obd2obt(exp->exp_obd)->obt_lut->lut_tgd.tgd_tot_granted_clients--;
6984
6985         if (!(exp->exp_flags & OBD_OPT_FORCE))
6986                 tgt_grant_sanity_check(exp->exp_obd, __func__);
6987
6988         RETURN(0);
6989 }
6990
6991 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
6992                    struct linkea_data *ldata)
6993 {
6994         int rc;
6995
6996         LASSERT(ldata->ld_buf->lb_buf != NULL);
6997
6998         if (!mdt_object_exists(mdt_obj))
6999                 return -ENODATA;
7000
7001         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
7002                           ldata->ld_buf, XATTR_NAME_LINK);
7003         if (rc == -ERANGE) {
7004                 /* Buf was too small, figure out what we need. */
7005                 lu_buf_free(ldata->ld_buf);
7006                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
7007                                   ldata->ld_buf, XATTR_NAME_LINK);
7008                 if (rc < 0)
7009                         return rc;
7010                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
7011                 if (ldata->ld_buf->lb_buf == NULL)
7012                         return -ENOMEM;
7013                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
7014                                   ldata->ld_buf, XATTR_NAME_LINK);
7015         }
7016         if (rc < 0)
7017                 return rc;
7018
7019         return linkea_init_with_rec(ldata);
7020 }
7021
7022 /**
7023  * Given an MDT object, try to look up the full path to the object.
7024  * Part of the MDT layer implementation of lfs fid2path.
7025  *
7026  * \param[in]     info  Per-thread common data shared by MDT level handlers.
7027  * \param[in]     obj   Object to do path lookup of
7028  * \param[in,out] fp    User-provided struct to store path information
7029  * \param[in]     root_fid Root FID of current path should reach
7030  *
7031  * \retval 0 Lookup successful, path information stored in fp
7032  * \retval -EAGAIN Lookup failed, usually because object is being moved
7033  * \retval negative errno if there was a problem
7034  */
7035 static int mdt_path_current(struct mdt_thread_info *info,
7036                             struct mdt_object *obj,
7037                             struct getinfo_fid2path *fp,
7038                             struct lu_fid *root_fid)
7039 {
7040         struct mdt_device *mdt = info->mti_mdt;
7041         struct lu_name *tmpname = &info->mti_name;
7042         struct lu_fid *tmpfid = &info->mti_tmp_fid1;
7043         struct lu_buf *buf = &info->mti_big_buf;
7044         struct linkea_data ldata = { NULL };
7045         bool first = true;
7046         struct mdt_object *mdt_obj;
7047         struct link_ea_header *leh;
7048         struct link_ea_entry *lee;
7049         char *ptr;
7050         int reclen;
7051         int rc = 0;
7052
7053         ENTRY;
7054
7055         /* temp buffer for path element, the buffer will be finally freed
7056          * in mdt_thread_info_fini */
7057         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
7058         if (buf->lb_buf == NULL)
7059                 RETURN(-ENOMEM);
7060
7061         ldata.ld_buf = buf;
7062         ptr = fp->gf_u.gf_path + fp->gf_pathlen - 1;
7063         *ptr = 0;
7064         --ptr;
7065         *tmpfid = fp->gf_fid = *mdt_object_fid(obj);
7066
7067         while (!lu_fid_eq(root_fid, &fp->gf_fid)) {
7068                 if (!lu_fid_eq(root_fid, &mdt->mdt_md_root_fid) &&
7069                     lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid))
7070                         GOTO(out, rc = -ENOENT);
7071
7072                 if (lu_fid_eq(mdt_object_fid(obj), tmpfid)) {
7073                         mdt_obj = obj;
7074                         mdt_object_get(info->mti_env, mdt_obj);
7075                 } else {
7076                         mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
7077                         if (IS_ERR(mdt_obj))
7078                                 GOTO(out, rc = PTR_ERR(mdt_obj));
7079                 }
7080
7081                 if (!mdt_object_exists(mdt_obj)) {
7082                         mdt_object_put(info->mti_env, mdt_obj);
7083                         GOTO(out, rc = -ENOENT);
7084                 }
7085
7086                 if (mdt_object_remote(mdt_obj)) {
7087                         mdt_object_put(info->mti_env, mdt_obj);
7088                         GOTO(remote_out, rc = -EREMOTE);
7089                 }
7090
7091                 rc = mdt_links_read(info, mdt_obj, &ldata);
7092                 if (rc != 0) {
7093                         mdt_object_put(info->mti_env, mdt_obj);
7094                         GOTO(out, rc);
7095                 }
7096
7097                 leh = buf->lb_buf;
7098                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
7099                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
7100                 /* If set, use link #linkno for path lookup, otherwise use
7101                    link #0.  Only do this for the final path element. */
7102                 if (first && fp->gf_linkno < leh->leh_reccount) {
7103                         int count;
7104                         for (count = 0; count < fp->gf_linkno; count++) {
7105                                 lee = (struct link_ea_entry *)
7106                                      ((char *)lee + reclen);
7107                                 linkea_entry_unpack(lee, &reclen, tmpname,
7108                                                     tmpfid);
7109                         }
7110                         if (fp->gf_linkno < leh->leh_reccount - 1)
7111                                 /* indicate to user there are more links */
7112                                 fp->gf_linkno++;
7113                 }
7114
7115                 /* Check if it is slave stripes */
7116                 rc = mdt_is_dir_stripe(info, mdt_obj);
7117                 mdt_object_put(info->mti_env, mdt_obj);
7118                 if (rc < 0)
7119                         GOTO(out, rc);
7120                 if (rc == 1) {
7121                         fp->gf_fid = *tmpfid;
7122                         continue;
7123                 }
7124
7125                 /* Pack the name in the end of the buffer */
7126                 ptr -= tmpname->ln_namelen;
7127                 if (ptr - 1 <= fp->gf_u.gf_path)
7128                         GOTO(out, rc = -EOVERFLOW);
7129                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
7130                 *(--ptr) = '/';
7131
7132                 /* keep the last resolved fid to the client, so the
7133                  * client will build the left path on another MDT for
7134                  * remote object */
7135                 fp->gf_fid = *tmpfid;
7136
7137                 first = false;
7138         }
7139
7140         /* non-zero will be treated as an error */
7141         rc = 0;
7142
7143 remote_out:
7144         ptr++; /* skip leading / */
7145         memmove(fp->gf_u.gf_path, ptr,
7146                 fp->gf_u.gf_path + fp->gf_pathlen - ptr);
7147
7148 out:
7149         RETURN(rc);
7150 }
7151
7152 /**
7153  * Given an MDT object, use mdt_path_current to get the path.
7154  * Essentially a wrapper to retry mdt_path_current a set number of times
7155  * if -EAGAIN is returned (usually because an object is being moved).
7156  *
7157  * Part of the MDT layer implementation of lfs fid2path.
7158  *
7159  * \param[in]     info  Per-thread common data shared by mdt level handlers.
7160  * \param[in]     obj   Object to do path lookup of
7161  * \param[in,out] fp    User-provided struct for arguments and to store path
7162  *                      information
7163  *
7164  * \retval 0 Lookup successful, path information stored in fp
7165  * \retval negative errno if there was a problem
7166  */
7167 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
7168                     struct getinfo_fid2path *fp, struct lu_fid *root_fid)
7169 {
7170         struct mdt_device       *mdt = info->mti_mdt;
7171         int                     tries = 3;
7172         int                     rc = -EAGAIN;
7173         ENTRY;
7174
7175         if (fp->gf_pathlen < 3)
7176                 RETURN(-EOVERFLOW);
7177
7178         if (root_fid == NULL)
7179                 root_fid = &mdt->mdt_md_root_fid;
7180
7181         if (lu_fid_eq(root_fid, mdt_object_fid(obj))) {
7182                 fp->gf_u.gf_path[0] = '\0';
7183                 RETURN(0);
7184         }
7185
7186         /* Retry multiple times in case file is being moved */
7187         while (tries-- && rc == -EAGAIN)
7188                 rc = mdt_path_current(info, obj, fp, root_fid);
7189
7190         RETURN(rc);
7191 }
7192
7193 /**
7194  * Get the full path of the provided FID, as of changelog record recno.
7195  *
7196  * This checks sanity and looks up object for user provided FID
7197  * before calling the actual path lookup code.
7198  *
7199  * Part of the MDT layer implementation of lfs fid2path.
7200  *
7201  * \param[in]     info  Per-thread common data shared by mdt level handlers.
7202  * \param[in,out] fp    User-provided struct for arguments and to store path
7203  *                      information
7204  *
7205  * \retval 0 Lookup successful, path information and recno stored in fp
7206  * \retval -ENOENT, object does not exist
7207  * \retval negative errno if there was a problem
7208  */
7209 static int mdt_fid2path(struct mdt_thread_info *info,
7210                         struct lu_fid *root_fid,
7211                         struct getinfo_fid2path *fp)
7212 {
7213         struct mdt_device *mdt = info->mti_mdt;
7214         struct mdt_object *obj;
7215         int    rc;
7216         ENTRY;
7217
7218         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
7219                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
7220
7221         if (!fid_is_sane(&fp->gf_fid))
7222                 RETURN(-EINVAL);
7223
7224         if (!fid_is_namespace_visible(&fp->gf_fid)) {
7225                 CDEBUG(D_INFO, "%s: "DFID" is invalid, f_seq should be >= %#llx"
7226                        ", or f_oid != 0, or f_ver == 0\n", mdt_obd_name(mdt),
7227                        PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
7228                 RETURN(-EINVAL);
7229         }
7230
7231         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
7232         if (IS_ERR(obj)) {
7233                 rc = PTR_ERR(obj);
7234                 CDEBUG(D_IOCTL, "cannot find "DFID": rc = %d\n",
7235                        PFID(&fp->gf_fid), rc);
7236                 RETURN(rc);
7237         }
7238
7239         if (mdt_object_remote(obj)) {
7240                 rc = -EREMOTE;
7241         } else if (!mdt_object_exists(obj)) {
7242                 rc = -ENOENT;
7243         } else {
7244                 struct lu_attr la = { 0 };
7245                 struct dt_object *dt = mdt_obj2dt(obj);
7246
7247                 if (dt && dt->do_ops && dt->do_ops->do_attr_get)
7248                         dt_attr_get(info->mti_env, mdt_obj2dt(obj), &la);
7249                 if (la.la_valid & LA_FLAGS && la.la_flags & LUSTRE_ENCRYPT_FL)
7250                         /* path resolution cannot be carried out on server
7251                          * side for encrypted files
7252                          */
7253                         rc = -ENODATA;
7254                 else
7255                         rc = 0;
7256         }
7257
7258         if (rc < 0) {
7259                 mdt_object_put(info->mti_env, obj);
7260                 CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n",
7261                        PFID(&fp->gf_fid), rc);
7262                 RETURN(rc);
7263         }
7264
7265         rc = mdt_path(info, obj, fp, root_fid);
7266
7267         CDEBUG(D_INFO, "fid "DFID", path %s recno %#llx linkno %u\n",
7268                PFID(&fp->gf_fid), fp->gf_u.gf_path,
7269                fp->gf_recno, fp->gf_linkno);
7270
7271         mdt_object_put(info->mti_env, obj);
7272
7273         RETURN(rc);
7274 }
7275
7276 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, int keylen,
7277                             void *val, int vallen)
7278 {
7279         struct getinfo_fid2path *fpout, *fpin;
7280         struct lu_fid *root_fid = NULL;
7281         int rc = 0;
7282
7283         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
7284         fpout = val;
7285
7286         if (req_capsule_req_need_swab(info->mti_pill))
7287                 lustre_swab_fid2path(fpin);
7288
7289         memcpy(fpout, fpin, sizeof(*fpin));
7290         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
7291                 RETURN(-EINVAL);
7292
7293         if (keylen >= cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*fpin) +
7294                       sizeof(struct lu_fid)) {
7295                 /* client sent its root FID, which is normally fileset FID */
7296                 root_fid = fpin->gf_u.gf_root_fid;
7297                 if (req_capsule_req_need_swab(info->mti_pill))
7298                         lustre_swab_lu_fid(root_fid);
7299
7300                 if (root_fid != NULL && !fid_is_sane(root_fid))
7301                         RETURN(-EINVAL);
7302         }
7303
7304         rc = mdt_fid2path(info, root_fid, fpout);
7305         RETURN(rc);
7306 }
7307
7308 int mdt_get_info(struct tgt_session_info *tsi)
7309 {
7310         char    *key;
7311         int      keylen;
7312         __u32   *vallen;
7313         void    *valout;
7314         int      rc;
7315
7316         ENTRY;
7317
7318         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
7319         if (key == NULL) {
7320                 CDEBUG(D_IOCTL, "No GETINFO key\n");
7321                 RETURN(err_serious(-EFAULT));
7322         }
7323         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
7324                                       RCL_CLIENT);
7325
7326         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
7327         if (vallen == NULL) {
7328                 CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n",
7329                                 tgt_name(tsi->tsi_tgt));
7330                 RETURN(err_serious(-EFAULT));
7331         }
7332
7333         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
7334                              *vallen);
7335         rc = req_capsule_server_pack(tsi->tsi_pill);
7336         if (rc)
7337                 RETURN(err_serious(rc));
7338
7339         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
7340         if (valout == NULL) {
7341                 CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n",
7342                                 tgt_name(tsi->tsi_tgt));
7343                 RETURN(err_serious(-EFAULT));
7344         }
7345
7346         if (KEY_IS(KEY_FID2PATH)) {
7347                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
7348
7349                 rc = mdt_rpc_fid2path(info, key, keylen, valout, *vallen);
7350                 mdt_thread_info_fini(info);
7351         } else {
7352                 rc = -EINVAL;
7353         }
7354         RETURN(rc);
7355 }
7356
7357 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
7358 {
7359         struct obd_ioctl_data *data = karg;
7360         struct lu_fid *fid;
7361         __u64 version;
7362         struct mdt_object *obj;
7363         struct mdt_lock_handle  *lh;
7364         int rc;
7365         ENTRY;
7366
7367         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
7368             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
7369                 RETURN(-EINVAL);
7370
7371         fid = (struct lu_fid *)data->ioc_inlbuf1;
7372
7373         if (!fid_is_sane(fid))
7374                 RETURN(-EINVAL);
7375
7376         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
7377
7378         lh = &mti->mti_lh[MDT_LH_PARENT];
7379         mdt_lock_reg_init(lh, LCK_CR);
7380
7381         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
7382         if (IS_ERR(obj))
7383                 RETURN(PTR_ERR(obj));
7384
7385         if (mdt_object_remote(obj)) {
7386                 rc = -EREMOTE;
7387                 /**
7388                  * before calling version get the correct MDS should be
7389                  * fid, this is error to find remote object here
7390                  */
7391                 CERROR("nonlocal object "DFID"\n", PFID(fid));
7392         } else if (!mdt_object_exists(obj)) {
7393                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
7394                 rc = -ENOENT;
7395         } else {
7396                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
7397                *(__u64 *)data->ioc_inlbuf2 = version;
7398                 rc = 0;
7399         }
7400         mdt_object_unlock_put(mti, obj, lh, 1);
7401         RETURN(rc);
7402 }
7403
7404 /* ioctls on obd dev */
7405 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
7406                          void *karg, void __user *uarg)
7407 {
7408         struct lu_env      env;
7409         struct obd_device *obd = exp->exp_obd;
7410         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
7411         struct dt_device  *dt = mdt->mdt_bottom;
7412         int rc;
7413
7414         ENTRY;
7415         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
7416         rc = lu_env_init(&env, LCT_MD_THREAD);
7417         if (rc)
7418                 RETURN(rc);
7419
7420         switch (cmd) {
7421         case OBD_IOC_SYNC:
7422                 rc = mdt_device_sync(&env, mdt);
7423                 break;
7424         case OBD_IOC_SET_READONLY:
7425                 rc = dt_sync(&env, dt);
7426                 if (rc == 0)
7427                         rc = dt_ro(&env, dt);
7428                 break;
7429         case OBD_IOC_ABORT_RECOVERY: {
7430                 struct obd_ioctl_data *data = karg;
7431
7432                 if (data->ioc_type & OBD_FLG_ABORT_RECOV_MDT) {
7433                         CERROR("%s: Aborting MDT recovery\n",
7434                                mdt_obd_name(mdt));
7435                         obd->obd_abort_mdt_recovery = 1;
7436                         wake_up(&obd->obd_next_transno_waitq);
7437                 } else { /* if (data->ioc_type & OBD_FLG_ABORT_RECOV_OST) */
7438                         /* lctl didn't set OBD_FLG_ABORT_RECOV_OST < 2.13.57 */
7439                         CERROR("%s: Aborting client recovery\n",
7440                                mdt_obd_name(mdt));
7441                         obd->obd_abort_recovery = 1;
7442                         target_stop_recovery_thread(obd);
7443                 }
7444                 rc = 0;
7445                 break;
7446         }
7447         case OBD_IOC_CHANGELOG_REG:
7448         case OBD_IOC_CHANGELOG_DEREG:
7449         case OBD_IOC_CHANGELOG_CLEAR:
7450         case OBD_IOC_LLOG_PRINT:
7451         case OBD_IOC_LLOG_CANCEL:
7452                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(&env,
7453                                                            mdt->mdt_child,
7454                                                            cmd, len, karg);
7455                 break;
7456         case OBD_IOC_START_LFSCK: {
7457                 struct md_device *next = mdt->mdt_child;
7458                 struct obd_ioctl_data *data = karg;
7459                 struct lfsck_start_param lsp;
7460
7461                 if (unlikely(data == NULL)) {
7462                         rc = -EINVAL;
7463                         break;
7464                 }
7465
7466                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
7467                 lsp.lsp_index_valid = 0;
7468                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
7469                 break;
7470         }
7471         case OBD_IOC_STOP_LFSCK: {
7472                 struct md_device        *next = mdt->mdt_child;
7473                 struct obd_ioctl_data   *data = karg;
7474                 struct lfsck_stop        stop;
7475
7476                 stop.ls_status = LS_STOPPED;
7477                 /* Old lfsck utils may pass NULL @stop. */
7478                 if (data->ioc_inlbuf1 == NULL)
7479                         stop.ls_flags = 0;
7480                 else
7481                         stop.ls_flags =
7482                         ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
7483
7484                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
7485                 break;
7486         }
7487         case OBD_IOC_QUERY_LFSCK: {
7488                 struct md_device        *next = mdt->mdt_child;
7489                 struct obd_ioctl_data   *data = karg;
7490
7491                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0,
7492                                                  data->ioc_inlbuf1);
7493                 break;
7494         }
7495         case OBD_IOC_GET_OBJ_VERSION: {
7496                 struct mdt_thread_info *mti;
7497
7498                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
7499                 memset(mti, 0, sizeof(*mti));
7500                 mti->mti_env = &env;
7501                 mti->mti_mdt = mdt;
7502                 mti->mti_exp = exp;
7503
7504                 rc = mdt_ioc_version_get(mti, karg);
7505                 break;
7506         }
7507         case OBD_IOC_CATLOGLIST: {
7508                 struct mdt_thread_info *mti;
7509
7510                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
7511                 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
7512                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
7513                                        &mti->mti_tmp_fid1);
7514                 break;
7515          }
7516         default:
7517                 rc = -EOPNOTSUPP;
7518                 CERROR("%s: Not supported cmd = %d, rc = %d\n",
7519                         mdt_obd_name(mdt), cmd, rc);
7520         }
7521
7522         lu_env_fini(&env);
7523         RETURN(rc);
7524 }
7525
7526 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
7527 {
7528         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
7529         int rc;
7530         ENTRY;
7531
7532         if (!mdt->mdt_skip_lfsck && !mdt->mdt_bottom->dd_rdonly) {
7533                 struct lfsck_start_param lsp;
7534
7535                 lsp.lsp_start = NULL;
7536                 lsp.lsp_index_valid = 0;
7537                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
7538                                                            OBD_IOC_START_LFSCK,
7539                                                            0, &lsp);
7540                 if (rc != 0 && rc != -EALREADY)
7541                         CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
7542                               mdt_obd_name(mdt), rc);
7543         }
7544
7545         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
7546         RETURN(rc);
7547 }
7548
7549 static int mdt_obd_postrecov(struct obd_device *obd)
7550 {
7551         struct lu_env env;
7552         int rc;
7553
7554         rc = lu_env_init(&env, LCT_MD_THREAD);
7555         if (rc)
7556                 RETURN(rc);
7557         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
7558         lu_env_fini(&env);
7559         return rc;
7560 }
7561
7562 static const struct obd_ops mdt_obd_device_ops = {
7563         .o_owner          = THIS_MODULE,
7564         .o_set_info_async = mdt_obd_set_info_async,
7565         .o_connect        = mdt_obd_connect,
7566         .o_reconnect      = mdt_obd_reconnect,
7567         .o_disconnect     = mdt_obd_disconnect,
7568         .o_init_export    = mdt_init_export,
7569         .o_destroy_export = mdt_destroy_export,
7570         .o_iocontrol      = mdt_iocontrol,
7571         .o_postrecov      = mdt_obd_postrecov,
7572         /* Data-on-MDT IO methods */
7573         .o_preprw         = mdt_obd_preprw,
7574         .o_commitrw       = mdt_obd_commitrw,
7575 };
7576
7577 static struct lu_device* mdt_device_fini(const struct lu_env *env,
7578                                          struct lu_device *d)
7579 {
7580         struct mdt_device *m = mdt_dev(d);
7581         ENTRY;
7582
7583         mdt_fini(env, m);
7584         RETURN(NULL);
7585 }
7586
7587 static struct lu_device *mdt_device_free(const struct lu_env *env,
7588                                          struct lu_device *d)
7589 {
7590         struct mdt_device *m = mdt_dev(d);
7591         ENTRY;
7592
7593         lu_device_fini(&m->mdt_lu_dev);
7594         OBD_FREE_PTR(m);
7595
7596         RETURN(NULL);
7597 }
7598
7599 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
7600                                           struct lu_device_type *t,
7601                                           struct lustre_cfg *cfg)
7602 {
7603         struct lu_device  *l;
7604         struct mdt_device *m;
7605
7606         OBD_ALLOC_PTR(m);
7607         if (m != NULL) {
7608                 int rc;
7609
7610                 l = &m->mdt_lu_dev;
7611                 rc = mdt_init0(env, m, t, cfg);
7612                 if (rc != 0) {
7613                         mdt_device_free(env, l);
7614                         l = ERR_PTR(rc);
7615                         return l;
7616                 }
7617         } else
7618                 l = ERR_PTR(-ENOMEM);
7619         return l;
7620 }
7621
7622 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
7623 LU_KEY_INIT(mdt, struct mdt_thread_info);
7624
7625 static void mdt_key_fini(const struct lu_context *ctx,
7626                          struct lu_context_key *key, void* data)
7627 {
7628         struct mdt_thread_info *info = data;
7629
7630         if (info->mti_big_lmm) {
7631                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
7632                 info->mti_big_lmm = NULL;
7633                 info->mti_big_lmmsize = 0;
7634         }
7635
7636         if (info->mti_big_acl) {
7637                 OBD_FREE_LARGE(info->mti_big_acl, info->mti_big_aclsize);
7638                 info->mti_big_acl = NULL;
7639                 info->mti_big_aclsize = 0;
7640         }
7641
7642         OBD_FREE_PTR(info);
7643 }
7644
7645 /* context key: mdt_thread_key */
7646 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
7647
7648 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
7649 {
7650         return lu_ucred(info->mti_env);
7651 }
7652
7653 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
7654 {
7655         return lu_ucred_check(info->mti_env);
7656 }
7657
7658 /**
7659  * Enable/disable COS (Commit On Sharing).
7660  *
7661  * Set/Clear the COS flag in mdt options.
7662  *
7663  * \param mdt mdt device
7664  * \param val 0 disables COS, other values enable COS
7665  */
7666 void mdt_enable_cos(struct mdt_device *mdt, bool val)
7667 {
7668         struct lu_env env;
7669         int rc;
7670
7671         mdt->mdt_opts.mo_cos = val;
7672         rc = lu_env_init(&env, LCT_LOCAL);
7673         if (unlikely(rc != 0)) {
7674                 CWARN("%s: lu_env initialization failed, cannot "
7675                       "sync: rc = %d\n", mdt_obd_name(mdt), rc);
7676                 return;
7677         }
7678         mdt_device_sync(&env, mdt);
7679         lu_env_fini(&env);
7680 }
7681
7682 /**
7683  * Check COS (Commit On Sharing) status.
7684  *
7685  * Return COS flag status.
7686  *
7687  * \param mdt mdt device
7688  */
7689 int mdt_cos_is_enabled(struct mdt_device *mdt)
7690 {
7691         return mdt->mdt_opts.mo_cos != 0;
7692 }
7693
7694 static const struct lu_device_type_operations mdt_device_type_ops = {
7695         .ldto_device_alloc = mdt_device_alloc,
7696         .ldto_device_free  = mdt_device_free,
7697         .ldto_device_fini  = mdt_device_fini
7698 };
7699
7700 static struct lu_device_type mdt_device_type = {
7701         .ldt_tags     = LU_DEVICE_MD,
7702         .ldt_name     = LUSTRE_MDT_NAME,
7703         .ldt_ops      = &mdt_device_type_ops,
7704         .ldt_ctx_tags = LCT_MD_THREAD
7705 };
7706
7707 static int __init mdt_init(void)
7708 {
7709         int rc;
7710
7711         BUILD_BUG_ON(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") !=
7712                      FID_NOBRACE_LEN + 1);
7713         BUILD_BUG_ON(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") !=
7714                      FID_LEN + 1);
7715         rc = lu_kmem_init(mdt_caches);
7716         if (rc)
7717                 return rc;
7718
7719         rc = mds_mod_init();
7720         if (rc)
7721                 GOTO(lu_fini, rc);
7722
7723         rc = class_register_type(&mdt_obd_device_ops, NULL, true,
7724                                  LUSTRE_MDT_NAME, &mdt_device_type);
7725         if (rc)
7726                 GOTO(mds_fini, rc);
7727 lu_fini:
7728         if (rc)
7729                 lu_kmem_fini(mdt_caches);
7730 mds_fini:
7731         if (rc)
7732                 mds_mod_exit();
7733         return rc;
7734 }
7735
7736 static void __exit mdt_exit(void)
7737 {
7738         class_unregister_type(LUSTRE_MDT_NAME);
7739         mds_mod_exit();
7740         lu_kmem_fini(mdt_caches);
7741 }
7742
7743 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
7744 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
7745 MODULE_VERSION(LUSTRE_VERSION_STRING);
7746 MODULE_LICENSE("GPL");
7747
7748 module_init(mdt_init);
7749 module_exit(mdt_exit);