lustre/mdt/mdt_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2010, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/mdt/mdt_handler.c
  33  *
  34  * Lustre Metadata Target (mdt) request handler
  35  *
  36  * Author: Peter Braam <braam@clusterfs.com>
  37  * Author: Andreas Dilger <adilger@clusterfs.com>
  38  * Author: Phil Schwan <phil@clusterfs.com>
  39  * Author: Mike Shaver <shaver@clusterfs.com>
  40  * Author: Nikita Danilov <nikita@clusterfs.com>
  41  * Author: Huang Hua <huanghua@clusterfs.com>
  42  * Author: Yury Umanets <umka@clusterfs.com>
  43  */
  44
  45 #define DEBUG_SUBSYSTEM S_MDS
  46
  47 #include <linux/module.h>
  48 #include <linux/pagemap.h>
  49
  50 #include <dt_object.h>
  51 #include <lustre_acl.h>
  52 #include <lustre_export.h>
  53 #include <uapi/linux/lustre/lustre_ioctl.h>
  54 #include <lustre_lfsck.h>
  55 #include <lustre_log.h>
  56 #include <lustre_nodemap.h>
  57 #include <lustre_mds.h>
  58 #include <uapi/linux/lustre/lustre_param.h>
  59 #include <lustre_quota.h>
  60 #include <lustre_swab.h>
  61 #include <obd.h>
  62 #include <obd_support.h>
  63 #include <lustre_barrier.h>
  64 #include <obd_cksum.h>
  65 #include <llog_swab.h>
  66
  67 #include "mdt_internal.h"
  68
  69 static unsigned int max_mod_rpcs_per_client = 8;
  70 module_param(max_mod_rpcs_per_client, uint, 0644);
  71 MODULE_PARM_DESC(max_mod_rpcs_per_client, "maximum number of modify RPCs in flight allowed per client");
  72
  73 mdl_mode_t mdt_mdl_lock_modes[] = {
  74         [LCK_MINMODE] = MDL_MINMODE,
  75         [LCK_EX]      = MDL_EX,
  76         [LCK_PW]      = MDL_PW,
  77         [LCK_PR]      = MDL_PR,
  78         [LCK_CW]      = MDL_CW,
  79         [LCK_CR]      = MDL_CR,
  80         [LCK_NL]      = MDL_NL,
  81         [LCK_GROUP]   = MDL_GROUP
  82 };
  83
  84 enum ldlm_mode mdt_dlm_lock_modes[] = {
  85         [MDL_MINMODE]   = LCK_MINMODE,
  86         [MDL_EX]        = LCK_EX,
  87         [MDL_PW]        = LCK_PW,
  88         [MDL_PR]        = LCK_PR,
  89         [MDL_CW]        = LCK_CW,
  90         [MDL_CR]        = LCK_CR,
  91         [MDL_NL]        = LCK_NL,
  92         [MDL_GROUP]     = LCK_GROUP
  93 };
  94
  95 static struct mdt_device *mdt_dev(struct lu_device *d);
  96
  97 static const struct lu_object_operations mdt_obj_ops;
  98
  99 /* Slab for MDT object allocation */
 100 static struct kmem_cache *mdt_object_kmem;
 101
 102 /* For HSM restore handles */
 103 struct kmem_cache *mdt_hsm_cdt_kmem;
 104
 105 /* For HSM request handles */
 106 struct kmem_cache *mdt_hsm_car_kmem;
 107
 108 static struct lu_kmem_descr mdt_caches[] = {
 109         {
 110                 .ckd_cache = &mdt_object_kmem,
 111                 .ckd_name  = "mdt_obj",
 112                 .ckd_size  = sizeof(struct mdt_object)
 113         },
 114         {
 115                 .ckd_cache      = &mdt_hsm_cdt_kmem,
 116                 .ckd_name       = "mdt_cdt_restore_handle",
 117                 .ckd_size       = sizeof(struct cdt_restore_handle)
 118         },
 119         {
 120                 .ckd_cache      = &mdt_hsm_car_kmem,
 121                 .ckd_name       = "mdt_cdt_agent_req",
 122                 .ckd_size       = sizeof(struct cdt_agent_req)
 123         },
 124         {
 125                 .ckd_cache = NULL
 126         }
 127 };
 128
 129 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag)
 130 {
 131         if (!rep)
 132                 return 0;
 133         return rep->lock_policy_res1 & op_flag;
 134 }
 135
 136 void mdt_clear_disposition(struct mdt_thread_info *info,
 137                            struct ldlm_reply *rep, __u64 op_flag)
 138 {
 139         if (info) {
 140                 info->mti_opdata &= ~op_flag;
 141                 tgt_opdata_clear(info->mti_env, op_flag);
 142         }
 143         if (rep)
 144                 rep->lock_policy_res1 &= ~op_flag;
 145 }
 146
 147 void mdt_set_disposition(struct mdt_thread_info *info,
 148                          struct ldlm_reply *rep, __u64 op_flag)
 149 {
 150         if (info) {
 151                 info->mti_opdata |= op_flag;
 152                 tgt_opdata_set(info->mti_env, op_flag);
 153         }
 154         if (rep)
 155                 rep->lock_policy_res1 |= op_flag;
 156 }
 157
 158 void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm)
 159 {
 160         lh->mlh_pdo_hash = 0;
 161         lh->mlh_reg_mode = lm;
 162         lh->mlh_rreg_mode = lm;
 163         lh->mlh_type = MDT_REG_LOCK;
 164 }
 165
 166 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode,
 167                        const struct lu_name *lname)
 168 {
 169         lh->mlh_reg_mode = lock_mode;
 170         lh->mlh_pdo_mode = LCK_MINMODE;
 171         lh->mlh_rreg_mode = lock_mode;
 172         lh->mlh_type = MDT_PDO_LOCK;
 173
 174         if (lu_name_is_valid(lname)) {
 175                 lh->mlh_pdo_hash = ll_full_name_hash(NULL, lname->ln_name,
 176                                                      lname->ln_namelen);
 177                 /* XXX Workaround for LU-2856
 178                  *
 179                  * Zero is a valid return value of full_name_hash, but
 180                  * several users of mlh_pdo_hash assume a non-zero
 181                  * hash value. We therefore map zero onto an
 182                  * arbitrary, but consistent value (1) to avoid
 183                  * problems further down the road. */
 184                 if (unlikely(lh->mlh_pdo_hash == 0))
 185                         lh->mlh_pdo_hash = 1;
 186         } else {
 187                 lh->mlh_pdo_hash = 0;
 188         }
 189 }
 190
 191 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 192                               struct mdt_lock_handle *lh)
 193 {
 194         mdl_mode_t mode;
 195         ENTRY;
 196
 197         /*
 198          * Any dir access needs couple of locks:
 199          *
 200          * 1) on part of dir we gonna take lookup/modify;
 201          *
 202          * 2) on whole dir to protect it from concurrent splitting and/or to
 203          * flush client's cache for readdir().
 204          *
 205          * so, for a given mode and object this routine decides what lock mode
 206          * to use for lock #2:
 207          *
 208          * 1) if caller's gonna lookup in dir then we need to protect dir from
 209          * being splitted only - LCK_CR
 210          *
 211          * 2) if caller's gonna modify dir then we need to protect dir from
 212          * being splitted and to flush cache - LCK_CW
 213          *
 214          * 3) if caller's gonna modify dir and that dir seems ready for
 215          * splitting then we need to protect it from any type of access
 216          * (lookup/modify/split) - LCK_EX --bzzz
 217          */
 218
 219         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 220         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 221
 222         /*
 223          * Ask underlaying level its opinion about preferable PDO lock mode
 224          * having access type passed as regular lock mode:
 225          *
 226          * - MDL_MINMODE means that lower layer does not want to specify lock
 227          * mode;
 228          *
 229          * - MDL_NL means that no PDO lock should be taken. This is used in some
 230          * cases. Say, for non-splittable directories no need to use PDO locks
 231          * at all.
 232          */
 233         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 234                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 235
 236         if (mode != MDL_MINMODE) {
 237                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 238         } else {
 239                 /*
 240                  * Lower layer does not want to specify locking mode. We do it
 241                  * our selves. No special protection is needed, just flush
 242                  * client's cache on modification and allow concurrent
 243                  * mondification.
 244                  */
 245                 switch (lh->mlh_reg_mode) {
 246                 case LCK_EX:
 247                         lh->mlh_pdo_mode = LCK_EX;
 248                         break;
 249                 case LCK_PR:
 250                         lh->mlh_pdo_mode = LCK_CR;
 251                         break;
 252                 case LCK_PW:
 253                         lh->mlh_pdo_mode = LCK_CW;
 254                         break;
 255                 default:
 256                         CERROR("Not expected lock type (0x%x)\n",
 257                                (int)lh->mlh_reg_mode);
 258                         LBUG();
 259                 }
 260         }
 261
 262         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 263         EXIT;
 264 }
 265
 266 static int mdt_lookup_fileset(struct mdt_thread_info *info, const char *fileset,
 267                               struct lu_fid *fid)
 268 {
 269         struct mdt_device *mdt = info->mti_mdt;
 270         struct lu_name *lname = &info->mti_name;
 271         char *filename = info->mti_filename;
 272         struct mdt_object *parent;
 273         u32 mode;
 274         int rc = 0;
 275
 276         LASSERT(!info->mti_cross_ref);
 277
 278         /*
 279          * We may want to allow this to mount a completely separate
 280          * fileset from the MDT in the future, but keeping it to
 281          * ROOT/ only for now avoid potential security issues.
 282          */
 283         *fid = mdt->mdt_md_root_fid;
 284
 285         while (rc == 0 && fileset != NULL && *fileset != '\0') {
 286                 const char *s1 = fileset;
 287                 const char *s2;
 288
 289                 while (*++s1 == '/')
 290                         ;
 291                 s2 = s1;
 292                 while (*s2 != '/' && *s2 != '\0')
 293                         s2++;
 294
 295                 if (s2 == s1)
 296                         break;
 297
 298                 fileset = s2;
 299
 300                 lname->ln_namelen = s2 - s1;
 301                 if (lname->ln_namelen > NAME_MAX) {
 302                         rc = -EINVAL;
 303                         break;
 304                 }
 305
 306                 /* reject .. as a path component */
 307                 if (lname->ln_namelen == 2 &&
 308                     strncmp(s1, "..", 2) == 0) {
 309                         rc = -EINVAL;
 310                         break;
 311                 }
 312
 313                 strncpy(filename, s1, lname->ln_namelen);
 314                 filename[lname->ln_namelen] = '\0';
 315                 lname->ln_name = filename;
 316
 317                 parent = mdt_object_find(info->mti_env, mdt, fid);
 318                 if (IS_ERR(parent)) {
 319                         rc = PTR_ERR(parent);
 320                         break;
 321                 }
 322                 /* Only got the fid of this obj by name */
 323                 fid_zero(fid);
 324                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
 325                                 fid, &info->mti_spec);
 326                 mdt_object_put(info->mti_env, parent);
 327         }
 328         if (!rc) {
 329                 parent = mdt_object_find(info->mti_env, mdt, fid);
 330                 if (IS_ERR(parent))
 331                         rc = PTR_ERR(parent);
 332                 else {
 333                         mode = lu_object_attr(&parent->mot_obj);
 334                         mdt_object_put(info->mti_env, parent);
 335                         if (!S_ISDIR(mode))
 336                                 rc = -ENOTDIR;
 337                 }
 338         }
 339
 340         return rc;
 341 }
 342
 343 static int mdt_get_root(struct tgt_session_info *tsi)
 344 {
 345         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
 346         struct mdt_device       *mdt = info->mti_mdt;
 347         struct mdt_body         *repbody;
 348         char                    *fileset = NULL, *buffer = NULL;
 349         int                      rc;
 350         struct obd_export       *exp = info->mti_exp;
 351         char                    *nodemap_fileset;
 352
 353         ENTRY;
 354
 355         rc = mdt_check_ucred(info);
 356         if (rc)
 357                 GOTO(out, rc = err_serious(rc));
 358
 359         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GET_ROOT_PACK))
 360                 GOTO(out, rc = err_serious(-ENOMEM));
 361
 362         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 363         if (req_capsule_get_size(info->mti_pill, &RMF_NAME, RCL_CLIENT) > 0) {
 364                 fileset = req_capsule_client_get(info->mti_pill, &RMF_NAME);
 365                 if (fileset == NULL)
 366                         GOTO(out, rc = err_serious(-EFAULT));
 367         }
 368
 369         nodemap_fileset = nodemap_get_fileset(exp->exp_target_data.ted_nodemap);
 370         if (nodemap_fileset && nodemap_fileset[0]) {
 371                 CDEBUG(D_INFO, "nodemap fileset is %s\n", nodemap_fileset);
 372                 if (fileset) {
 373                         /* consider fileset from client as a sub-fileset
 374                          * of the nodemap one */
 375                         OBD_ALLOC(buffer, PATH_MAX + 1);
 376                         if (buffer == NULL)
 377                                 GOTO(out, rc = err_serious(-ENOMEM));
 378                         if (snprintf(buffer, PATH_MAX + 1, "%s/%s",
 379                                      nodemap_fileset, fileset) >= PATH_MAX + 1)
 380                                 GOTO(out, rc = err_serious(-EINVAL));
 381                         fileset = buffer;
 382                 } else {
 383                         /* enforce fileset as specified in the nodemap */
 384                         fileset = nodemap_fileset;
 385                 }
 386         }
 387
 388         if (fileset) {
 389                 CDEBUG(D_INFO, "Getting fileset %s\n", fileset);
 390                 rc = mdt_lookup_fileset(info, fileset, &repbody->mbo_fid1);
 391                 if (rc < 0)
 392                         GOTO(out, rc = err_serious(rc));
 393         } else {
 394                 repbody->mbo_fid1 = mdt->mdt_md_root_fid;
 395         }
 396         repbody->mbo_valid |= OBD_MD_FLID;
 397
 398         EXIT;
 399 out:
 400         mdt_thread_info_fini(info);
 401         if (buffer)
 402                 OBD_FREE(buffer, PATH_MAX+1);
 403         return rc;
 404 }
 405
 406 static int mdt_statfs(struct tgt_session_info *tsi)
 407 {
 408         struct ptlrpc_request *req = tgt_ses_req(tsi);
 409         struct mdt_thread_info *info = tsi2mdt_info(tsi);
 410         struct mdt_device *mdt = info->mti_mdt;
 411         struct tg_grants_data *tgd = &mdt->mdt_lut.lut_tgd;
 412         struct md_device *next = mdt->mdt_child;
 413         struct ptlrpc_service_part *svcpt;
 414         struct obd_statfs *osfs;
 415         struct mdt_body *reqbody = NULL;
 416         struct mdt_statfs_cache *msf;
 417         int rc;
 418
 419         ENTRY;
 420
 421         svcpt = req->rq_rqbd->rqbd_svcpt;
 422
 423         /* This will trigger a watchdog timeout */
 424         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 425                          (MDT_SERVICE_WATCHDOG_FACTOR *
 426                           at_get(&svcpt->scp_at_estimate)) + 1);
 427
 428         rc = mdt_check_ucred(info);
 429         if (rc)
 430                 GOTO(out, rc = err_serious(rc));
 431
 432         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
 433                 GOTO(out, rc = err_serious(-ENOMEM));
 434
 435         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 436         if (!osfs)
 437                 GOTO(out, rc = -EPROTO);
 438
 439         if (mdt_is_sum_statfs_client(req->rq_export) &&
 440                 lustre_packed_msg_size(req->rq_reqmsg) ==
 441                 req_capsule_fmt_size(req->rq_reqmsg->lm_magic,
 442                                      &RQF_MDS_STATFS_NEW, RCL_CLIENT)) {
 443                 req_capsule_extend(info->mti_pill, &RQF_MDS_STATFS_NEW);
 444                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
 445         }
 446
 447         if (reqbody && reqbody->mbo_valid & OBD_MD_FLAGSTATFS)
 448                 msf = &mdt->mdt_sum_osfs;
 449         else
 450                 msf = &mdt->mdt_osfs;
 451
 452         if (msf->msf_age + OBD_STATFS_CACHE_SECONDS <= ktime_get_seconds()) {
 453                         /** statfs data is too old, get up-to-date one */
 454                         if (reqbody && reqbody->mbo_valid & OBD_MD_FLAGSTATFS)
 455                                 rc = next->md_ops->mdo_statfs(info->mti_env,
 456                                                               next, osfs);
 457                         else
 458                                 rc = dt_statfs(info->mti_env, mdt->mdt_bottom,
 459                                                osfs);
 460                         if (rc)
 461                                 GOTO(out, rc);
 462                         spin_lock(&mdt->mdt_lock);
 463                         msf->msf_osfs = *osfs;
 464                         msf->msf_age = ktime_get_seconds();
 465                         spin_unlock(&mdt->mdt_lock);
 466         } else {
 467                         /** use cached statfs data */
 468                         spin_lock(&mdt->mdt_lock);
 469                         *osfs = msf->msf_osfs;
 470                         spin_unlock(&mdt->mdt_lock);
 471         }
 472
 473         /* at least try to account for cached pages.  its still racy and
 474          * might be under-reporting if clients haven't announced their
 475          * caches with brw recently */
 476         CDEBUG(D_SUPER | D_CACHE, "blocks cached %llu granted %llu"
 477                " pending %llu free %llu avail %llu\n",
 478                tgd->tgd_tot_dirty, tgd->tgd_tot_granted,
 479                tgd->tgd_tot_pending,
 480                osfs->os_bfree << tgd->tgd_blockbits,
 481                osfs->os_bavail << tgd->tgd_blockbits);
 482
 483         osfs->os_bavail -= min_t(u64, osfs->os_bavail,
 484                                  ((tgd->tgd_tot_dirty + tgd->tgd_tot_pending +
 485                                    osfs->os_bsize - 1) >> tgd->tgd_blockbits));
 486
 487         tgt_grant_sanity_check(mdt->mdt_lu_dev.ld_obd, __func__);
 488         CDEBUG(D_CACHE, "%llu blocks: %llu free, %llu avail; "
 489                "%llu objects: %llu free; state %x\n",
 490                osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
 491                osfs->os_files, osfs->os_ffree, osfs->os_state);
 492
 493         if (!exp_grant_param_supp(tsi->tsi_exp) &&
 494             tgd->tgd_blockbits > COMPAT_BSIZE_SHIFT) {
 495                 /* clients which don't support OBD_CONNECT_GRANT_PARAM
 496                  * should not see a block size > page size, otherwise
 497                  * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
 498                  * block size which is the biggest block size known to work
 499                  * with all client's page size. */
 500                 osfs->os_blocks <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
 501                 osfs->os_bfree  <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
 502                 osfs->os_bavail <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
 503                 osfs->os_bsize = 1 << COMPAT_BSIZE_SHIFT;
 504         }
 505         if (rc == 0)
 506                 mdt_counter_incr(req, LPROC_MDT_STATFS);
 507 out:
 508         mdt_thread_info_fini(info);
 509         RETURN(rc);
 510 }
 511
 512 /**
 513  * Pack size attributes into the reply.
 514  */
 515 int mdt_pack_size2body(struct mdt_thread_info *info,
 516                         const struct lu_fid *fid, struct lustre_handle *lh)
 517 {
 518         struct mdt_body *b;
 519         struct md_attr *ma = &info->mti_attr;
 520         int dom_stripe;
 521         bool dom_lock = false;
 522
 523         ENTRY;
 524
 525         LASSERT(ma->ma_attr.la_valid & LA_MODE);
 526
 527         if (!S_ISREG(ma->ma_attr.la_mode) ||
 528             !(ma->ma_valid & MA_LOV && ma->ma_lmm != NULL))
 529                 RETURN(-ENODATA);
 530
 531         dom_stripe = mdt_lmm_dom_entry(ma->ma_lmm);
 532         /* no DoM stripe, no size in reply */
 533         if (dom_stripe == LMM_NO_DOM)
 534                 RETURN(-ENOENT);
 535
 536         if (lustre_handle_is_used(lh)) {
 537                 struct ldlm_lock *lock;
 538
 539                 lock = ldlm_handle2lock(lh);
 540                 if (lock != NULL) {
 541                         dom_lock = ldlm_has_dom(lock);
 542                         LDLM_LOCK_PUT(lock);
 543                 }
 544         }
 545
 546         /* no DoM lock, no size in reply */
 547         if (!dom_lock)
 548                 RETURN(0);
 549
 550         /* Either DoM lock exists or LMM has only DoM stripe then
 551          * return size on body. */
 552         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 553
 554         mdt_dom_object_size(info->mti_env, info->mti_mdt, fid, b, dom_lock);
 555         RETURN(0);
 556 }
 557
 558 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
 559 /*
 560  * Pack ACL data into the reply. UIDs/GIDs are mapped and filtered by nodemap.
 561  *
 562  * \param       info    thread info object
 563  * \param       repbody reply to pack ACLs into
 564  * \param       o       mdt object of file to examine
 565  * \param       nodemap nodemap of client to reply to
 566  * \retval      0       success
 567  * \retval      -errno  error getting or parsing ACL from disk
 568  */
 569 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
 570                       struct mdt_object *o, struct lu_nodemap *nodemap)
 571 {
 572         const struct lu_env     *env = info->mti_env;
 573         struct md_object        *next = mdt_object_child(o);
 574         struct lu_buf           *buf = &info->mti_buf;
 575         struct mdt_device       *mdt = info->mti_mdt;
 576         struct req_capsule *pill = info->mti_pill;
 577         int rc;
 578
 579         ENTRY;
 580
 581         buf->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
 582         buf->lb_len = req_capsule_get_size(pill, &RMF_ACL, RCL_SERVER);
 583         if (buf->lb_len == 0)
 584                 RETURN(0);
 585
 586 again:
 587         rc = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_ACCESS);
 588         if (rc < 0) {
 589                 if (rc == -ENODATA) {
 590                         repbody->mbo_aclsize = 0;
 591                         repbody->mbo_valid |= OBD_MD_FLACL;
 592                         rc = 0;
 593                 } else if (rc == -EOPNOTSUPP) {
 594                         rc = 0;
 595                 } else {
 596                         if (rc == -ERANGE &&
 597                             exp_connect_large_acl(info->mti_exp) &&
 598                             buf->lb_buf != info->mti_big_acl) {
 599                                 if (info->mti_big_acl == NULL) {
 600                                         info->mti_big_aclsize =
 601                                                         min_t(unsigned int,
 602                                                               mdt->mdt_max_ea_size,
 603                                                               XATTR_SIZE_MAX);
 604                                         OBD_ALLOC_LARGE(info->mti_big_acl,
 605                                                         info->mti_big_aclsize);
 606                                         if (info->mti_big_acl == NULL) {
 607                                                 info->mti_big_aclsize = 0;
 608                                                 CERROR("%s: unable to grow "
 609                                                        DFID" ACL buffer\n",
 610                                                        mdt_obd_name(mdt),
 611                                                        PFID(mdt_object_fid(o)));
 612                                                 RETURN(-ENOMEM);
 613                                         }
 614                                 }
 615
 616                                 CDEBUG(D_INODE, "%s: grow the "DFID
 617                                        " ACL buffer to size %d\n",
 618                                        mdt_obd_name(mdt),
 619                                        PFID(mdt_object_fid(o)),
 620                                        info->mti_big_aclsize);
 621
 622                                 buf->lb_buf = info->mti_big_acl;
 623                                 buf->lb_len = info->mti_big_aclsize;
 624
 625                                 goto again;
 626                         }
 627
 628                         CERROR("%s: unable to read "DFID" ACL: rc = %d\n",
 629                                mdt_obd_name(mdt), PFID(mdt_object_fid(o)), rc);
 630                 }
 631         } else {
 632                 int client;
 633                 int server;
 634                 int acl_buflen;
 635                 int lmm_buflen = 0;
 636                 int lmmsize = 0;
 637
 638                 acl_buflen = req_capsule_get_size(pill, &RMF_ACL, RCL_SERVER);
 639                 if (acl_buflen >= rc)
 640                         goto map;
 641
 642                 /* If LOV/LMA EA is small, we can reuse part of their buffer */
 643                 client = ptlrpc_req_get_repsize(pill->rc_req);
 644                 server = lustre_packed_msg_size(pill->rc_req->rq_repmsg);
 645                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
 646                         lmm_buflen = req_capsule_get_size(pill, &RMF_MDT_MD,
 647                                                           RCL_SERVER);
 648                         lmmsize = repbody->mbo_eadatasize;
 649                 }
 650
 651                 if (client < server - acl_buflen - lmm_buflen + rc + lmmsize) {
 652                         CDEBUG(D_INODE, "%s: client prepared buffer size %d "
 653                                "is not big enough with the ACL size %d (%d)\n",
 654                                mdt_obd_name(mdt), client, rc,
 655                                server - acl_buflen - lmm_buflen + rc + lmmsize);
 656                         repbody->mbo_aclsize = 0;
 657                         repbody->mbo_valid &= ~OBD_MD_FLACL;
 658                         RETURN(-ERANGE);
 659                 }
 660
 661 map:
 662                 if (buf->lb_buf == info->mti_big_acl)
 663                         info->mti_big_acl_used = 1;
 664
 665                 rc = nodemap_map_acl(nodemap, buf->lb_buf,
 666                                      rc, NODEMAP_FS_TO_CLIENT);
 667                 /* if all ACLs mapped out, rc is still >= 0 */
 668                 if (rc < 0) {
 669                         CERROR("%s: nodemap_map_acl unable to parse "DFID
 670                                " ACL: rc = %d\n", mdt_obd_name(mdt),
 671                                PFID(mdt_object_fid(o)), rc);
 672                         repbody->mbo_aclsize = 0;
 673                         repbody->mbo_valid &= ~OBD_MD_FLACL;
 674                 } else {
 675                         repbody->mbo_aclsize = rc;
 676                         repbody->mbo_valid |= OBD_MD_FLACL;
 677                         rc = 0;
 678                 }
 679         }
 680
 681         RETURN(rc);
 682 }
 683 #endif
 684
 685 /* XXX Look into layout in MDT layer. */
 686 static inline bool mdt_hsm_is_released(struct lov_mds_md *lmm)
 687 {
 688         struct lov_comp_md_v1   *comp_v1;
 689         struct lov_mds_md       *v1;
 690         int                      i;
 691
 692         if (lmm->lmm_magic == LOV_MAGIC_COMP_V1) {
 693                 comp_v1 = (struct lov_comp_md_v1 *)lmm;
 694
 695                 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
 696                         v1 = (struct lov_mds_md *)((char *)comp_v1 +
 697                                 comp_v1->lcm_entries[i].lcme_offset);
 698                         /* We don't support partial release for now */
 699                         if (!(v1->lmm_pattern & LOV_PATTERN_F_RELEASED))
 700                                 return false;
 701                 }
 702                 return true;
 703         } else {
 704                 return (lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) ?
 705                         true : false;
 706         }
 707 }
 708
 709 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 710                         const struct lu_attr *attr, const struct lu_fid *fid)
 711 {
 712         struct md_attr *ma = &info->mti_attr;
 713         struct obd_export *exp = info->mti_exp;
 714         struct lu_nodemap *nodemap = NULL;
 715
 716         LASSERT(ma->ma_valid & MA_INODE);
 717
 718         if (attr->la_valid & LA_ATIME) {
 719                 b->mbo_atime = attr->la_atime;
 720                 b->mbo_valid |= OBD_MD_FLATIME;
 721         }
 722         if (attr->la_valid & LA_MTIME) {
 723                 b->mbo_mtime = attr->la_mtime;
 724                 b->mbo_valid |= OBD_MD_FLMTIME;
 725         }
 726         if (attr->la_valid & LA_CTIME) {
 727                 b->mbo_ctime = attr->la_ctime;
 728                 b->mbo_valid |= OBD_MD_FLCTIME;
 729         }
 730         if (attr->la_valid & LA_FLAGS) {
 731                 b->mbo_flags = attr->la_flags;
 732                 b->mbo_valid |= OBD_MD_FLFLAGS;
 733         }
 734         if (attr->la_valid & LA_NLINK) {
 735                 b->mbo_nlink = attr->la_nlink;
 736                 b->mbo_valid |= OBD_MD_FLNLINK;
 737         }
 738         if (attr->la_valid & (LA_UID|LA_GID)) {
 739                 nodemap = nodemap_get_from_exp(exp);
 740                 if (IS_ERR(nodemap))
 741                         goto out;
 742         }
 743         if (attr->la_valid & LA_UID) {
 744                 b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 745                                             NODEMAP_FS_TO_CLIENT,
 746                                             attr->la_uid);
 747                 b->mbo_valid |= OBD_MD_FLUID;
 748         }
 749         if (attr->la_valid & LA_GID) {
 750                 b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 751                                             NODEMAP_FS_TO_CLIENT,
 752                                             attr->la_gid);
 753                 b->mbo_valid |= OBD_MD_FLGID;
 754         }
 755
 756         if (attr->la_valid & LA_PROJID) {
 757                 /* TODO, nodemap for project id */
 758                 b->mbo_projid = attr->la_projid;
 759                 b->mbo_valid |= OBD_MD_FLPROJID;
 760         }
 761
 762         b->mbo_mode = attr->la_mode;
 763         if (attr->la_valid & LA_MODE)
 764                 b->mbo_valid |= OBD_MD_FLMODE;
 765         if (attr->la_valid & LA_TYPE)
 766                 b->mbo_valid |= OBD_MD_FLTYPE;
 767
 768         if (fid != NULL) {
 769                 b->mbo_fid1 = *fid;
 770                 b->mbo_valid |= OBD_MD_FLID;
 771                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid=%#llx\n",
 772                        PFID(fid), b->mbo_nlink, b->mbo_mode, b->mbo_valid);
 773         }
 774
 775         if (!(attr->la_valid & LA_TYPE))
 776                 return;
 777
 778         b->mbo_rdev   = attr->la_rdev;
 779         b->mbo_size   = attr->la_size;
 780         b->mbo_blocks = attr->la_blocks;
 781
 782         if (!S_ISREG(attr->la_mode)) {
 783                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 784         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
 785                 /* means no objects are allocated on osts. */
 786                 LASSERT(!(ma->ma_valid & MA_LOV));
 787                 /* just ignore blocks occupied by extend attributes on MDS */
 788                 b->mbo_blocks = 0;
 789                 /* if no object is allocated on osts, the size on mds is valid.
 790                  * b=22272 */
 791                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 792         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL) {
 793                 if (mdt_hsm_is_released(ma->ma_lmm)) {
 794                         /* A released file stores its size on MDS. */
 795                         /* But return 1 block for released file, unless tools
 796                          * like tar will consider it fully sparse. (LU-3864)
 797                          */
 798                         if (unlikely(b->mbo_size == 0))
 799                                 b->mbo_blocks = 0;
 800                         else
 801                                 b->mbo_blocks = 1;
 802                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 803                 } else if (info->mti_som_valid) { /* som is valid */
 804                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 805                 } else if (ma->ma_valid & MA_SOM) { /* lsom is valid */
 806                         b->mbo_valid |= OBD_MD_FLLAZYSIZE | OBD_MD_FLLAZYBLOCKS;
 807                         b->mbo_size = ma->ma_som.ms_size;
 808                         b->mbo_blocks = ma->ma_som.ms_blocks;
 809                 }
 810         }
 811
 812         if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE ||
 813                             b->mbo_valid & OBD_MD_FLLAZYSIZE))
 814                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 815                        PFID(fid), (unsigned long long)b->mbo_size);
 816
 817 out:
 818         if (!IS_ERR_OR_NULL(nodemap))
 819                 nodemap_putref(nodemap);
 820 }
 821
 822 static inline int mdt_body_has_lov(const struct lu_attr *la,
 823                                    const struct mdt_body *body)
 824 {
 825         return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
 826                (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
 827 }
 828
 829 void mdt_client_compatibility(struct mdt_thread_info *info)
 830 {
 831         struct mdt_body       *body;
 832         struct ptlrpc_request *req = mdt_info_req(info);
 833         struct obd_export     *exp = req->rq_export;
 834         struct md_attr        *ma = &info->mti_attr;
 835         struct lu_attr        *la = &ma->ma_attr;
 836         ENTRY;
 837
 838         if (exp_connect_layout(exp))
 839                 /* the client can deal with 16-bit lmm_stripe_count */
 840                 RETURN_EXIT;
 841
 842         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 843
 844         if (!mdt_body_has_lov(la, body))
 845                 RETURN_EXIT;
 846
 847         /* now we have a reply with a lov for a client not compatible with the
 848          * layout lock so we have to clean the layout generation number */
 849         if (S_ISREG(la->la_mode))
 850                 ma->ma_lmm->lmm_layout_gen = 0;
 851         EXIT;
 852 }
 853
 854 static int mdt_attr_get_eabuf_size(struct mdt_thread_info *info,
 855                                    struct mdt_object *o)
 856 {
 857         const struct lu_env *env = info->mti_env;
 858         int rc, rc2;
 859
 860         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 861                           XATTR_NAME_LOV);
 862
 863         if (rc == -ENODATA)
 864                 rc = 0;
 865
 866         if (rc < 0)
 867                 goto out;
 868
 869         /* Is it a directory? Let's check for the LMV as well */
 870         if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
 871                 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 872                                    XATTR_NAME_LMV);
 873
 874                 if (rc2 == -ENODATA)
 875                         rc2 = mo_xattr_get(env, mdt_object_child(o),
 876                                            &LU_BUF_NULL,
 877                                            XATTR_NAME_DEFAULT_LMV);
 878
 879                 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
 880                         rc = rc2;
 881         }
 882
 883 out:
 884         return rc;
 885 }
 886
 887 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
 888                       const char *name)
 889 {
 890         const struct lu_env *env = info->mti_env;
 891         int rc;
 892         ENTRY;
 893
 894         LASSERT(info->mti_big_lmm_used == 0);
 895         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
 896         if (rc < 0)
 897                 RETURN(rc);
 898
 899         /* big_lmm may need to be grown */
 900         if (info->mti_big_lmmsize < rc) {
 901                 int size = size_roundup_power2(rc);
 902
 903                 if (info->mti_big_lmmsize > 0) {
 904                         /* free old buffer */
 905                         LASSERT(info->mti_big_lmm);
 906                         OBD_FREE_LARGE(info->mti_big_lmm,
 907                                        info->mti_big_lmmsize);
 908                         info->mti_big_lmm = NULL;
 909                         info->mti_big_lmmsize = 0;
 910                 }
 911
 912                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
 913                 if (info->mti_big_lmm == NULL)
 914                         RETURN(-ENOMEM);
 915                 info->mti_big_lmmsize = size;
 916         }
 917         LASSERT(info->mti_big_lmmsize >= rc);
 918
 919         info->mti_buf.lb_buf = info->mti_big_lmm;
 920         info->mti_buf.lb_len = info->mti_big_lmmsize;
 921         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
 922
 923         RETURN(rc);
 924 }
 925
 926 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
 927                    struct md_attr *ma, const char *name)
 928 {
 929         struct md_object *next = mdt_object_child(o);
 930         struct lu_buf    *buf = &info->mti_buf;
 931         int rc;
 932
 933         if (strcmp(name, XATTR_NAME_LOV) == 0) {
 934                 buf->lb_buf = ma->ma_lmm;
 935                 buf->lb_len = ma->ma_lmm_size;
 936                 LASSERT(!(ma->ma_valid & MA_LOV));
 937         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 938                 buf->lb_buf = ma->ma_lmv;
 939                 buf->lb_len = ma->ma_lmv_size;
 940                 LASSERT(!(ma->ma_valid & MA_LMV));
 941         } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 942                 buf->lb_buf = ma->ma_default_lmv;
 943                 buf->lb_len = ma->ma_default_lmv_size;
 944                 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
 945         } else {
 946                 return -EINVAL;
 947         }
 948
 949         LASSERT(buf->lb_buf);
 950
 951         rc = mo_xattr_get(info->mti_env, next, buf, name);
 952         if (rc > 0) {
 953
 954 got:
 955                 if (strcmp(name, XATTR_NAME_LOV) == 0) {
 956                         if (info->mti_big_lmm_used)
 957                                 ma->ma_lmm = info->mti_big_lmm;
 958
 959                         /* NOT return LOV EA with hole to old client. */
 960                         if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
 961                                      LOV_PATTERN_F_HOLE) &&
 962                             !(exp_connect_flags(info->mti_exp) &
 963                               OBD_CONNECT_LFSCK)) {
 964                                 return -EIO;
 965                         } else {
 966                                 ma->ma_lmm_size = rc;
 967                                 ma->ma_valid |= MA_LOV;
 968                         }
 969                 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 970                         if (info->mti_big_lmm_used)
 971                                 ma->ma_lmv = info->mti_big_lmm;
 972
 973                         ma->ma_lmv_size = rc;
 974                         ma->ma_valid |= MA_LMV;
 975                 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 976                         ma->ma_default_lmv_size = rc;
 977                         ma->ma_valid |= MA_LMV_DEF;
 978                 }
 979
 980                 /* Update mdt_max_mdsize so all clients will be aware that */
 981                 if (info->mti_mdt->mdt_max_mdsize < rc)
 982                         info->mti_mdt->mdt_max_mdsize = rc;
 983
 984                 rc = 0;
 985         } else if (rc == -ENODATA) {
 986                 /* no LOV EA */
 987                 rc = 0;
 988         } else if (rc == -ERANGE) {
 989                 /* Default LMV has fixed size, so it must be able to fit
 990                  * in the original buffer */
 991                 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
 992                         return rc;
 993                 rc = mdt_big_xattr_get(info, o, name);
 994                 if (rc > 0) {
 995                         info->mti_big_lmm_used = 1;
 996                         goto got;
 997                 }
 998         }
 999
1000         return rc;
1001 }
1002
1003 int mdt_attr_get_pfid(struct mdt_thread_info *info, struct mdt_object *o,
1004                       struct lu_fid *pfid)
1005 {
1006         struct lu_buf           *buf = &info->mti_buf;
1007         struct link_ea_header   *leh;
1008         struct link_ea_entry    *lee;
1009         int                      rc;
1010         ENTRY;
1011
1012         buf->lb_buf = info->mti_big_lmm;
1013         buf->lb_len = info->mti_big_lmmsize;
1014         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
1015                           buf, XATTR_NAME_LINK);
1016         /* ignore errors, MA_PFID won't be set and it is
1017          * up to the caller to treat this as an error */
1018         if (rc == -ERANGE || buf->lb_len == 0) {
1019                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
1020                 buf->lb_buf = info->mti_big_lmm;
1021                 buf->lb_len = info->mti_big_lmmsize;
1022         }
1023
1024         if (rc < 0)
1025                 RETURN(rc);
1026         if (rc < sizeof(*leh)) {
1027                 CERROR("short LinkEA on "DFID": rc = %d\n",
1028                        PFID(mdt_object_fid(o)), rc);
1029                 RETURN(-ENODATA);
1030         }
1031
1032         leh = (struct link_ea_header *) buf->lb_buf;
1033         lee = (struct link_ea_entry *)(leh + 1);
1034         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
1035                 leh->leh_magic = LINK_EA_MAGIC;
1036                 leh->leh_reccount = __swab32(leh->leh_reccount);
1037                 leh->leh_len = __swab64(leh->leh_len);
1038         }
1039         if (leh->leh_magic != LINK_EA_MAGIC)
1040                 RETURN(-EINVAL);
1041         if (leh->leh_reccount == 0)
1042                 RETURN(-ENODATA);
1043
1044         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
1045         fid_be_to_cpu(pfid, pfid);
1046
1047         RETURN(0);
1048 }
1049
1050 int mdt_attr_get_complex(struct mdt_thread_info *info,
1051                          struct mdt_object *o, struct md_attr *ma)
1052 {
1053         const struct lu_env *env = info->mti_env;
1054         struct md_object    *next = mdt_object_child(o);
1055         struct lu_buf       *buf = &info->mti_buf;
1056         int                  need = ma->ma_need;
1057         int                  rc = 0, rc2;
1058         u32                  mode;
1059         ENTRY;
1060
1061         ma->ma_valid = 0;
1062
1063         if (mdt_object_exists(o) == 0)
1064                 GOTO(out, rc = -ENOENT);
1065         mode = lu_object_attr(&next->mo_lu);
1066
1067         if (need & MA_INODE) {
1068                 ma->ma_need = MA_INODE;
1069                 rc = mo_attr_get(env, next, ma);
1070                 if (rc)
1071                         GOTO(out, rc);
1072
1073                 if (S_ISREG(mode))
1074                         (void) mdt_get_som(info, o, ma);
1075                 ma->ma_valid |= MA_INODE;
1076         }
1077
1078         if (need & MA_PFID) {
1079                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
1080                 if (rc == 0)
1081                         ma->ma_valid |= MA_PFID;
1082                 /* ignore this error, parent fid is not mandatory */
1083                 rc = 0;
1084         }
1085
1086         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
1087                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
1088                 if (rc)
1089                         GOTO(out, rc);
1090         }
1091
1092         if (need & MA_LMV && S_ISDIR(mode)) {
1093                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
1094                 if (rc != 0)
1095                         GOTO(out, rc);
1096         }
1097
1098         if (need & MA_LMV_DEF && S_ISDIR(mode)) {
1099                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
1100                 if (rc != 0)
1101                         GOTO(out, rc);
1102         }
1103
1104         /*
1105          * In the handle of MA_INODE, we may already get the SOM attr.
1106          */
1107         if (need & MA_SOM && S_ISREG(mode) && !(ma->ma_valid & MA_SOM)) {
1108                 rc = mdt_get_som(info, o, ma);
1109                 if (rc != 0)
1110                         GOTO(out, rc);
1111         }
1112
1113         if (need & MA_HSM && S_ISREG(mode)) {
1114                 buf->lb_buf = info->mti_xattr_buf;
1115                 buf->lb_len = sizeof(info->mti_xattr_buf);
1116                 BUILD_BUG_ON(sizeof(struct hsm_attrs) >
1117                              sizeof(info->mti_xattr_buf));
1118                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
1119                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
1120                 if (rc2 == 0)
1121                         ma->ma_valid |= MA_HSM;
1122                 else if (rc2 < 0 && rc2 != -ENODATA)
1123                         GOTO(out, rc = rc2);
1124         }
1125
1126 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1127         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
1128                 buf->lb_buf = ma->ma_acl;
1129                 buf->lb_len = ma->ma_acl_size;
1130                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
1131                 if (rc2 > 0) {
1132                         ma->ma_acl_size = rc2;
1133                         ma->ma_valid |= MA_ACL_DEF;
1134                 } else if (rc2 == -ENODATA) {
1135                         /* no ACLs */
1136                         ma->ma_acl_size = 0;
1137                 } else
1138                         GOTO(out, rc = rc2);
1139         }
1140 #endif
1141 out:
1142         ma->ma_need = need;
1143         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = %#llx ma_lmm=%p\n",
1144                rc, ma->ma_valid, ma->ma_lmm);
1145         RETURN(rc);
1146 }
1147
1148 static int mdt_getattr_internal(struct mdt_thread_info *info,
1149                                 struct mdt_object *o, int ma_need)
1150 {
1151         struct md_object        *next = mdt_object_child(o);
1152         const struct mdt_body   *reqbody = info->mti_body;
1153         struct ptlrpc_request   *req = mdt_info_req(info);
1154         struct md_attr          *ma = &info->mti_attr;
1155         struct lu_attr          *la = &ma->ma_attr;
1156         struct req_capsule      *pill = info->mti_pill;
1157         const struct lu_env     *env = info->mti_env;
1158         struct mdt_body         *repbody;
1159         struct lu_buf           *buffer = &info->mti_buf;
1160         struct obd_export       *exp = info->mti_exp;
1161         int                      rc;
1162         ENTRY;
1163
1164         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
1165                 RETURN(err_serious(-ENOMEM));
1166
1167         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1168
1169         ma->ma_valid = 0;
1170
1171         if (mdt_object_remote(o)) {
1172                 /* This object is located on remote node.*/
1173                 /* Return -ENOTSUPP for old client */
1174                 if (!mdt_is_dne_client(req->rq_export))
1175                         GOTO(out, rc = -ENOTSUPP);
1176
1177                 repbody->mbo_fid1 = *mdt_object_fid(o);
1178                 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
1179                 GOTO(out, rc = 0);
1180         }
1181
1182         if (reqbody->mbo_eadatasize > 0) {
1183                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
1184                 if (buffer->lb_buf == NULL)
1185                         GOTO(out, rc = -EPROTO);
1186                 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
1187                                                       RCL_SERVER);
1188         } else {
1189                 buffer->lb_buf = NULL;
1190                 buffer->lb_len = 0;
1191                 ma_need &= ~(MA_LOV | MA_LMV);
1192                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
1193                        mdt_obd_name(info->mti_mdt),
1194                        req->rq_export->exp_client_uuid.uuid);
1195         }
1196
1197         /* from 2.12.58 intent_getattr pack default LMV in reply */
1198         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1199             ((reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA)) ==
1200                     (OBD_MD_MEA | OBD_MD_DEFAULT_MEA)) &&
1201             req_capsule_has_field(&req->rq_pill, &RMF_DEFAULT_MDT_MD,
1202                                   RCL_SERVER)) {
1203                 ma->ma_lmv = buffer->lb_buf;
1204                 ma->ma_lmv_size = buffer->lb_len;
1205                 ma->ma_default_lmv = req_capsule_server_get(pill,
1206                                                 &RMF_DEFAULT_MDT_MD);
1207                 ma->ma_default_lmv_size = req_capsule_get_size(pill,
1208                                                 &RMF_DEFAULT_MDT_MD,
1209                                                 RCL_SERVER);
1210                 ma->ma_need = MA_INODE;
1211                 if (ma->ma_lmv_size > 0)
1212                         ma->ma_need |= MA_LMV;
1213                 if (ma->ma_default_lmv_size > 0)
1214                         ma->ma_need |= MA_LMV_DEF;
1215         } else if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1216                    (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
1217                 /* If it is dir and client require MEA, then we got MEA */
1218                 /* Assumption: MDT_MD size is enough for lmv size. */
1219                 ma->ma_lmv = buffer->lb_buf;
1220                 ma->ma_lmv_size = buffer->lb_len;
1221                 ma->ma_need = MA_INODE;
1222                 if (ma->ma_lmv_size > 0) {
1223                         if (reqbody->mbo_valid & OBD_MD_MEA) {
1224                                 ma->ma_need |= MA_LMV;
1225                         } else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA) {
1226                                 ma->ma_need |= MA_LMV_DEF;
1227                                 ma->ma_default_lmv = buffer->lb_buf;
1228                                 ma->ma_lmv = NULL;
1229                                 ma->ma_default_lmv_size = buffer->lb_len;
1230                                 ma->ma_lmv_size = 0;
1231                         }
1232                 }
1233         } else {
1234                 ma->ma_lmm = buffer->lb_buf;
1235                 ma->ma_lmm_size = buffer->lb_len;
1236                 ma->ma_need = MA_INODE | MA_HSM;
1237                 if (ma->ma_lmm_size > 0) {
1238                         ma->ma_need |= MA_LOV;
1239                         /* Older clients may crash if they getattr overstriped
1240                          * files
1241                          */
1242                         if (!exp_connect_overstriping(exp) &&
1243                             mdt_lmm_is_overstriping(ma->ma_lmm))
1244                                 RETURN(-EOPNOTSUPP);
1245                 }
1246         }
1247
1248         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1249             reqbody->mbo_valid & OBD_MD_FLDIREA  &&
1250             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
1251                 /* get default stripe info for this dir. */
1252                 ma->ma_need |= MA_LOV_DEF;
1253         }
1254         ma->ma_need |= ma_need;
1255
1256         rc = mdt_attr_get_complex(info, o, ma);
1257         if (unlikely(rc)) {
1258                 CDEBUG(rc == -ENOENT ? D_OTHER : D_ERROR,
1259                        "%s: getattr error for "DFID": rc = %d\n",
1260                        mdt_obd_name(info->mti_mdt),
1261                        PFID(mdt_object_fid(o)), rc);
1262                 RETURN(rc);
1263         }
1264
1265         /* if file is released, check if a restore is running */
1266         if (ma->ma_valid & MA_HSM) {
1267                 repbody->mbo_valid |= OBD_MD_TSTATE;
1268                 if ((ma->ma_hsm.mh_flags & HS_RELEASED) &&
1269                     mdt_hsm_restore_is_running(info, mdt_object_fid(o)))
1270                         repbody->mbo_t_state = MS_RESTORE;
1271         }
1272
1273         if (likely(ma->ma_valid & MA_INODE))
1274                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
1275         else
1276                 RETURN(-EFAULT);
1277
1278         if (mdt_body_has_lov(la, reqbody)) {
1279                 if (ma->ma_valid & MA_LOV) {
1280                         LASSERT(ma->ma_lmm_size);
1281                         repbody->mbo_eadatasize = ma->ma_lmm_size;
1282                         if (S_ISDIR(la->la_mode))
1283                                 repbody->mbo_valid |= OBD_MD_FLDIREA;
1284                         else
1285                                 repbody->mbo_valid |= OBD_MD_FLEASIZE;
1286                         mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
1287                 }
1288                 if (ma->ma_valid & MA_LMV) {
1289                         /* Return -ENOTSUPP for old client */
1290                         if (!mdt_is_striped_client(req->rq_export))
1291                                 RETURN(-ENOTSUPP);
1292
1293                         LASSERT(S_ISDIR(la->la_mode));
1294                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1295                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1296                         repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
1297                 }
1298                 if (ma->ma_valid & MA_LMV_DEF) {
1299                         /* Return -ENOTSUPP for old client */
1300                         if (!mdt_is_striped_client(req->rq_export))
1301                                 RETURN(-ENOTSUPP);
1302                         LASSERT(S_ISDIR(la->la_mode));
1303                         /*
1304                          * when ll_dir_getstripe() gets default LMV, it
1305                          * checks mbo_eadatasize.
1306                          */
1307                         if (!(ma->ma_valid & MA_LMV))
1308                                 repbody->mbo_eadatasize =
1309                                         ma->ma_default_lmv_size;
1310                         repbody->mbo_valid |= (OBD_MD_FLDIREA |
1311                                                OBD_MD_DEFAULT_MEA);
1312                 }
1313         } else if (S_ISLNK(la->la_mode) &&
1314                    reqbody->mbo_valid & OBD_MD_LINKNAME) {
1315                 buffer->lb_buf = ma->ma_lmm;
1316                 /* eadatasize from client includes NULL-terminator, so
1317                  * there is no need to read it */
1318                 buffer->lb_len = reqbody->mbo_eadatasize - 1;
1319                 rc = mo_readlink(env, next, buffer);
1320                 if (unlikely(rc <= 0)) {
1321                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
1322                                mdt_obd_name(info->mti_mdt),
1323                                PFID(mdt_object_fid(o)), rc);
1324                         rc = -EFAULT;
1325                 } else {
1326                         int print_limit = min_t(int, PAGE_SIZE - 128, rc);
1327
1328                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
1329                                 rc -= 2;
1330                         repbody->mbo_valid |= OBD_MD_LINKNAME;
1331                         /* we need to report back size with NULL-terminator
1332                          * because client expects that */
1333                         repbody->mbo_eadatasize = rc + 1;
1334                         if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
1335                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
1336                                        "on "DFID ", expected %d\n",
1337                                        mdt_obd_name(info->mti_mdt),
1338                                        rc, PFID(mdt_object_fid(o)),
1339                                        reqbody->mbo_eadatasize - 1);
1340                         /* NULL terminate */
1341                         ((char *)ma->ma_lmm)[rc] = 0;
1342
1343                         /* If the total CDEBUG() size is larger than a page, it
1344                          * will print a warning to the console, avoid this by
1345                          * printing just the last part of the symlink. */
1346                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
1347                                print_limit < rc ? "..." : "", print_limit,
1348                                (char *)ma->ma_lmm + rc - print_limit, rc);
1349                         rc = 0;
1350                 }
1351         }
1352
1353         if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
1354                 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1355                 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1356                 CDEBUG(D_INODE, "changing the max MD size to %u\n",
1357                        repbody->mbo_max_mdsize);
1358         }
1359
1360 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1361         if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1362                  (reqbody->mbo_valid & OBD_MD_FLACL)) {
1363                 struct lu_nodemap *nodemap = nodemap_get_from_exp(exp);
1364                 if (IS_ERR(nodemap))
1365                         RETURN(PTR_ERR(nodemap));
1366
1367                 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
1368                 nodemap_putref(nodemap);
1369         }
1370 #endif
1371
1372 out:
1373         if (rc == 0)
1374                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
1375
1376         RETURN(rc);
1377 }
1378
1379 static int mdt_getattr(struct tgt_session_info *tsi)
1380 {
1381         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1382         struct mdt_object       *obj = info->mti_object;
1383         struct req_capsule      *pill = info->mti_pill;
1384         struct mdt_body         *reqbody;
1385         struct mdt_body         *repbody;
1386         int rc, rc2;
1387         ENTRY;
1388
1389         if (unlikely(info->mti_object == NULL))
1390                 RETURN(-EPROTO);
1391
1392         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1393         LASSERT(reqbody);
1394         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1395
1396         /* Special case for Data-on-MDT files to get data version */
1397         if (unlikely(reqbody->mbo_valid & OBD_MD_FLDATAVERSION)) {
1398                 rc = mdt_data_version_get(tsi);
1399                 GOTO(out, rc);
1400         }
1401
1402         /* Unlike intent case where we need to pre-fill out buffers early on
1403          * in intent policy for ldlm reasons, here we can have a much better
1404          * guess at EA size by just reading it from disk.
1405          * Exceptions are readdir and (missing) directory striping */
1406         /* Readlink */
1407         if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1408                 /* No easy way to know how long is the symlink, but it cannot
1409                  * be more than PATH_MAX, so we allocate +1 */
1410                 rc = PATH_MAX + 1;
1411         /* A special case for fs ROOT: getattr there might fetch
1412          * default EA for entire fs, not just for this dir!
1413          */
1414         } else if (lu_fid_eq(mdt_object_fid(obj),
1415                              &info->mti_mdt->mdt_md_root_fid) &&
1416                    (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1417                    (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1418                                                                  MDS_GETATTR)) {
1419                 /* Should the default strping be bigger, mdt_fix_reply
1420                  * will reallocate */
1421                 rc = DEF_REP_MD_SIZE;
1422         } else {
1423                 /* Read the actual EA size from disk */
1424                 rc = mdt_attr_get_eabuf_size(info, obj);
1425         }
1426
1427         if (rc < 0)
1428                 GOTO(out, rc = err_serious(rc));
1429
1430         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1431
1432         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
1433          * by default. If the target object has more ACL entries, then
1434          * enlarge the buffer when necessary. */
1435         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
1436                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
1437
1438         rc = req_capsule_server_pack(pill);
1439         if (unlikely(rc != 0))
1440                 GOTO(out, rc = err_serious(rc));
1441
1442         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1443         LASSERT(repbody != NULL);
1444         repbody->mbo_eadatasize = 0;
1445         repbody->mbo_aclsize = 0;
1446
1447         rc = mdt_check_ucred(info);
1448         if (unlikely(rc))
1449                 GOTO(out_shrink, rc);
1450
1451         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1452
1453         rc = mdt_getattr_internal(info, obj, 0);
1454         EXIT;
1455 out_shrink:
1456         mdt_client_compatibility(info);
1457         rc2 = mdt_fix_reply(info);
1458         if (rc == 0)
1459                 rc = rc2;
1460 out:
1461         mdt_thread_info_fini(info);
1462         return rc;
1463 }
1464
1465 /**
1466  * Handler of layout intent RPC requiring the layout modification
1467  *
1468  * \param[in]  info     thread environment
1469  * \param[in]  obj      object
1470  * \param[out] lhc      object ldlm lock handle
1471  * \param[in]  layout   layout change descriptor
1472  *
1473  * \retval 0    on success
1474  * \retval < 0  error code
1475  */
1476 int mdt_layout_change(struct mdt_thread_info *info, struct mdt_object *obj,
1477                       struct mdt_lock_handle *lhc,
1478                       struct md_layout_change *layout)
1479 {
1480         int rc;
1481
1482         ENTRY;
1483
1484         if (!mdt_object_exists(obj))
1485                 RETURN(-ENOENT);
1486
1487         if (!S_ISREG(lu_object_attr(&obj->mot_obj)))
1488                 RETURN(-EINVAL);
1489
1490         rc = mo_permission(info->mti_env, NULL, mdt_object_child(obj), NULL,
1491                            MAY_WRITE);
1492         if (rc)
1493                 RETURN(rc);
1494
1495         rc = mdt_check_resent_lock(info, obj, lhc);
1496         if (rc < 0)
1497                 RETURN(rc);
1498
1499         if (rc > 0) {
1500                 /* not resent */
1501                 mdt_lock_handle_init(lhc);
1502                 mdt_lock_reg_init(lhc, LCK_EX);
1503                 rc = mdt_reint_object_lock(info, obj, lhc, MDS_INODELOCK_LAYOUT,
1504                                            false);
1505                 if (rc)
1506                         RETURN(rc);
1507         }
1508
1509         mutex_lock(&obj->mot_som_mutex);
1510         rc = mo_layout_change(info->mti_env, mdt_object_child(obj), layout);
1511         mutex_unlock(&obj->mot_som_mutex);
1512
1513         if (rc)
1514                 mdt_object_unlock(info, obj, lhc, 1);
1515
1516         RETURN(rc);
1517 }
1518
1519 /**
1520  * Exchange MOF_LOV_CREATED flags between two objects after a
1521  * layout swap. No assumption is made on whether o1 or o2 have
1522  * created objects or not.
1523  *
1524  * \param[in,out] o1    First swap layout object
1525  * \param[in,out] o2    Second swap layout object
1526  */
1527 static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2)
1528 {
1529         unsigned int o1_lov_created = o1->mot_lov_created;
1530
1531         mutex_lock(&o1->mot_lov_mutex);
1532         mutex_lock(&o2->mot_lov_mutex);
1533
1534         o1->mot_lov_created = o2->mot_lov_created;
1535         o2->mot_lov_created = o1_lov_created;
1536
1537         mutex_unlock(&o2->mot_lov_mutex);
1538         mutex_unlock(&o1->mot_lov_mutex);
1539 }
1540
1541 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1542 {
1543         struct mdt_thread_info  *info;
1544         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1545         struct obd_export       *exp = req->rq_export;
1546         struct mdt_object       *o1, *o2, *o;
1547         struct mdt_lock_handle  *lh1, *lh2;
1548         struct mdc_swap_layouts *msl;
1549         int                      rc;
1550         ENTRY;
1551
1552         /* client does not support layout lock, so layout swaping
1553          * is disabled.
1554          * FIXME: there is a problem for old clients which don't support
1555          * layout lock yet. If those clients have already opened the file
1556          * they won't be notified at all so that old layout may still be
1557          * used to do IO. This can be fixed after file release is landed by
1558          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1559         if (!exp_connect_layout(exp))
1560                 RETURN(-EOPNOTSUPP);
1561
1562         info = tsi2mdt_info(tsi);
1563         if (unlikely(info->mti_object == NULL))
1564                 RETURN(-EPROTO);
1565
1566         if (info->mti_dlm_req != NULL)
1567                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1568
1569         o1 = info->mti_object;
1570         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1571                                 &info->mti_body->mbo_fid2);
1572         if (IS_ERR(o))
1573                 GOTO(out, rc = PTR_ERR(o));
1574
1575         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1576                 GOTO(put, rc = -ENOENT);
1577
1578         rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1579         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1580                 GOTO(put, rc);
1581
1582         if (rc < 0)
1583                 swap(o1, o2);
1584
1585         /* permission check. Make sure the calling process having permission
1586          * to write both files. */
1587         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1588                            MAY_WRITE);
1589         if (rc < 0)
1590                 GOTO(put, rc);
1591
1592         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1593                            MAY_WRITE);
1594         if (rc < 0)
1595                 GOTO(put, rc);
1596
1597         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1598         if (msl == NULL)
1599                 GOTO(put, rc = -EPROTO);
1600
1601         lh1 = &info->mti_lh[MDT_LH_NEW];
1602         mdt_lock_reg_init(lh1, LCK_EX);
1603         lh2 = &info->mti_lh[MDT_LH_OLD];
1604         mdt_lock_reg_init(lh2, LCK_EX);
1605
1606         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1607                              MDS_INODELOCK_XATTR);
1608         if (rc < 0)
1609                 GOTO(put, rc);
1610
1611         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1612                              MDS_INODELOCK_XATTR);
1613         if (rc < 0)
1614                 GOTO(unlock1, rc);
1615
1616         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1617                              mdt_object_child(o2), msl->msl_flags);
1618         if (rc < 0)
1619                 GOTO(unlock2, rc);
1620
1621         mdt_swap_lov_flag(o1, o2);
1622
1623 unlock2:
1624         mdt_object_unlock(info, o2, lh2, rc);
1625 unlock1:
1626         mdt_object_unlock(info, o1, lh1, rc);
1627 put:
1628         mdt_object_put(info->mti_env, o);
1629 out:
1630         mdt_thread_info_fini(info);
1631         RETURN(rc);
1632 }
1633
1634 static int mdt_raw_lookup(struct mdt_thread_info *info,
1635                           struct mdt_object *parent,
1636                           const struct lu_name *lname,
1637                           struct ldlm_reply *ldlm_rep)
1638 {
1639         struct lu_fid   *child_fid = &info->mti_tmp_fid1;
1640         int              rc;
1641         ENTRY;
1642
1643         LASSERT(!info->mti_cross_ref);
1644
1645         /* Only got the fid of this obj by name */
1646         fid_zero(child_fid);
1647         rc = mdo_lookup(info->mti_env, mdt_object_child(info->mti_object),
1648                         lname, child_fid, &info->mti_spec);
1649         if (rc == 0) {
1650                 struct mdt_body *repbody;
1651
1652                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1653                 repbody->mbo_fid1 = *child_fid;
1654                 repbody->mbo_valid = OBD_MD_FLID;
1655                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1656         } else if (rc == -ENOENT) {
1657                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1658         }
1659
1660         RETURN(rc);
1661 }
1662
1663 /*
1664  * UPDATE lock should be taken against parent, and be released before exit;
1665  * child_bits lock should be taken against child, and be returned back:
1666  *            (1)normal request should release the child lock;
1667  *            (2)intent request will grant the lock to client.
1668  */
1669 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1670                                  struct mdt_lock_handle *lhc,
1671                                  __u64 child_bits,
1672                                  struct ldlm_reply *ldlm_rep)
1673 {
1674         struct ptlrpc_request  *req = mdt_info_req(info);
1675         struct mdt_body        *reqbody = NULL;
1676         struct mdt_object      *parent = info->mti_object;
1677         struct mdt_object      *child;
1678         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1679         struct lu_name         *lname = NULL;
1680         struct mdt_lock_handle *lhp = NULL;
1681         struct ldlm_lock       *lock;
1682         struct req_capsule *pill = info->mti_pill;
1683         __u64 try_bits = 0;
1684         bool is_resent;
1685         int ma_need = 0;
1686         int rc;
1687
1688         ENTRY;
1689
1690         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1691         LASSERT(ergo(is_resent,
1692                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1693
1694         if (parent == NULL)
1695                 RETURN(-ENOENT);
1696
1697         if (info->mti_cross_ref) {
1698                 /* Only getattr on the child. Parent is on another node. */
1699                 mdt_set_disposition(info, ldlm_rep,
1700                                     DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1701                 child = parent;
1702                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1703                        "ldlm_rep = %p\n",
1704                        PFID(mdt_object_fid(child)), ldlm_rep);
1705
1706                 rc = mdt_check_resent_lock(info, child, lhc);
1707                 if (rc < 0) {
1708                         RETURN(rc);
1709                 } else if (rc > 0) {
1710                         mdt_lock_handle_init(lhc);
1711                         mdt_lock_reg_init(lhc, LCK_PR);
1712
1713                         /*
1714                          * Object's name entry is on another MDS, it will
1715                          * request PERM lock only because LOOKUP lock is owned
1716                          * by the MDS where name entry resides.
1717                          *
1718                          * TODO: it should try layout lock too. - Jinshan
1719                          */
1720                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
1721                                         MDS_INODELOCK_LAYOUT);
1722                         child_bits |= MDS_INODELOCK_PERM;
1723
1724                         rc = mdt_object_lock(info, child, lhc, child_bits);
1725                         if (rc < 0)
1726                                 RETURN(rc);
1727                 }
1728
1729                 /* Finally, we can get attr for child. */
1730                 if (!mdt_object_exists(child)) {
1731                         LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1732                                         &child->mot_obj,
1733                                         "remote object doesn't exist.");
1734                         mdt_object_unlock(info, child, lhc, 1);
1735                         RETURN(-ENOENT);
1736                 }
1737
1738                 rc = mdt_getattr_internal(info, child, 0);
1739                 if (unlikely(rc != 0)) {
1740                         mdt_object_unlock(info, child, lhc, 1);
1741                         RETURN(rc);
1742                 }
1743
1744                 rc = mdt_pack_secctx_in_reply(info, child);
1745                 if (unlikely(rc))
1746                         mdt_object_unlock(info, child, lhc, 1);
1747                 RETURN(rc);
1748         }
1749
1750         lname = &info->mti_name;
1751         mdt_name_unpack(pill, &RMF_NAME, lname, MNF_FIX_ANON);
1752
1753         if (lu_name_is_valid(lname)) {
1754                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1755                        "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1756                        PNAME(lname), ldlm_rep);
1757         } else {
1758                 reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1759                 if (unlikely(reqbody == NULL))
1760                         RETURN(err_serious(-EPROTO));
1761
1762                 *child_fid = reqbody->mbo_fid2;
1763
1764                 if (unlikely(!fid_is_sane(child_fid)))
1765                         RETURN(err_serious(-EINVAL));
1766
1767                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1768                        "ldlm_rep = %p\n",
1769                        PFID(mdt_object_fid(parent)),
1770                        PFID(&reqbody->mbo_fid2), ldlm_rep);
1771         }
1772
1773         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1774
1775         if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1776                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1777                                 &parent->mot_obj,
1778                                 "Parent doesn't exist!");
1779                 RETURN(-ESTALE);
1780         }
1781
1782         if (mdt_object_remote(parent)) {
1783                 CERROR("%s: parent "DFID" is on remote target\n",
1784                        mdt_obd_name(info->mti_mdt),
1785                        PFID(mdt_object_fid(parent)));
1786                 RETURN(-EIO);
1787         }
1788
1789         if (lu_name_is_valid(lname)) {
1790                 /* Always allow to lookup ".." */
1791                 if (unlikely(lname->ln_namelen == 2 &&
1792                              lname->ln_name[0] == '.' &&
1793                              lname->ln_name[1] == '.'))
1794                         info->mti_spec.sp_permitted = 1;
1795
1796                 if (info->mti_body->mbo_valid == OBD_MD_FLID) {
1797                         rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1798
1799                         RETURN(rc);
1800                 }
1801
1802                 /* step 1: lock parent only if parent is a directory */
1803                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1804                         lhp = &info->mti_lh[MDT_LH_PARENT];
1805                         mdt_lock_pdo_init(lhp, LCK_PR, lname);
1806                         rc = mdt_object_lock(info, parent, lhp,
1807                                              MDS_INODELOCK_UPDATE);
1808                         if (unlikely(rc != 0))
1809                                 RETURN(rc);
1810                 }
1811
1812                 /* step 2: lookup child's fid by name */
1813                 fid_zero(child_fid);
1814                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
1815                                 child_fid, &info->mti_spec);
1816                 if (rc == -ENOENT)
1817                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1818
1819                 if (rc != 0)
1820                         GOTO(out_parent, rc);
1821         }
1822
1823         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1824
1825         /*
1826          *step 3: find the child object by fid & lock it.
1827          *        regardless if it is local or remote.
1828          *
1829          *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1830          *      set parent dir fid the same as child fid in getattr by fid case
1831          *      we should not lu_object_find() the object again, could lead
1832          *      to hung if there is a concurrent unlink destroyed the object.
1833          */
1834         if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1835                 mdt_object_get(info->mti_env, parent);
1836                 child = parent;
1837         } else {
1838                 child = mdt_object_find(info->mti_env, info->mti_mdt,
1839                                         child_fid);
1840         }
1841
1842         if (unlikely(IS_ERR(child)))
1843                 GOTO(out_parent, rc = PTR_ERR(child));
1844
1845         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
1846         if (!mdt_object_exists(child)) {
1847                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1848                                 &child->mot_obj,
1849                                 "Object doesn't exist!");
1850                 GOTO(out_child, rc = -ENOENT);
1851         }
1852
1853         rc = mdt_check_resent_lock(info, child, lhc);
1854         if (rc < 0) {
1855                 GOTO(out_child, rc);
1856         } else if (rc > 0) {
1857                 mdt_lock_handle_init(lhc);
1858                 mdt_lock_reg_init(lhc, LCK_PR);
1859
1860                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1861                     !mdt_object_remote(child)) {
1862                         struct md_attr *ma = &info->mti_attr;
1863
1864                         ma->ma_valid = 0;
1865                         ma->ma_need = MA_INODE;
1866                         rc = mdt_attr_get_complex(info, child, ma);
1867                         if (unlikely(rc != 0))
1868                                 GOTO(out_child, rc);
1869
1870                         /* If the file has not been changed for some time, we
1871                          * return not only a LOOKUP lock, but also an UPDATE
1872                          * lock and this might save us RPC on later STAT. For
1873                          * directories, it also let negative dentry cache start
1874                          * working for this dir. */
1875                         if (ma->ma_valid & MA_INODE &&
1876                             ma->ma_attr.la_valid & LA_CTIME &&
1877                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1878                             ma->ma_attr.la_ctime < ktime_get_real_seconds())
1879                                 child_bits |= MDS_INODELOCK_UPDATE;
1880                 }
1881
1882                 /* layout lock must be granted in a best-effort way
1883                  * for IT operations */
1884                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1885                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
1886                     !mdt_object_remote(child) && ldlm_rep != NULL) {
1887                         if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1888                             exp_connect_layout(info->mti_exp)) {
1889                                 /* try to grant layout lock for regular file. */
1890                                 try_bits = MDS_INODELOCK_LAYOUT;
1891                         }
1892                         /* Acquire DOM lock in advance for data-on-mdt file */
1893                         if (child != parent)
1894                                 try_bits |= MDS_INODELOCK_DOM;
1895                 }
1896
1897                 if (try_bits != 0) {
1898                         /* try layout lock, it may fail to be granted due to
1899                          * contention at LOOKUP or UPDATE */
1900                         rc = mdt_object_lock_try(info, child, lhc, &child_bits,
1901                                                  try_bits, false);
1902                         if (child_bits & MDS_INODELOCK_LAYOUT)
1903                                 ma_need |= MA_LOV;
1904                 } else {
1905                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1906                          * client will enqueue the lock to the remote MDT */
1907                         if (mdt_object_remote(child))
1908                                 child_bits &= ~MDS_INODELOCK_UPDATE;
1909                         rc = mdt_object_lock(info, child, lhc, child_bits);
1910                 }
1911                 if (unlikely(rc != 0))
1912                         GOTO(out_child, rc);
1913         }
1914
1915         /* finally, we can get attr for child. */
1916         rc = mdt_getattr_internal(info, child, ma_need);
1917         if (unlikely(rc != 0)) {
1918                 mdt_object_unlock(info, child, lhc, 1);
1919                 GOTO(out_child, rc);
1920         }
1921
1922         rc = mdt_pack_secctx_in_reply(info, child);
1923         if (unlikely(rc)) {
1924                 mdt_object_unlock(info, child, lhc, 1);
1925                 GOTO(out_child, rc);
1926         }
1927
1928         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1929         if (lock) {
1930                 /* Debugging code. */
1931                 LDLM_DEBUG(lock, "Returning lock to client");
1932                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1933                                          &lock->l_resource->lr_name),
1934                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1935                          PLDLMRES(lock->l_resource),
1936                          PFID(mdt_object_fid(child)));
1937
1938                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
1939                     !mdt_object_remote(child) && child != parent) {
1940                         mdt_object_put(info->mti_env, child);
1941                         rc = mdt_pack_size2body(info, child_fid,
1942                                                 &lhc->mlh_reg_lh);
1943                         if (rc != 0 && child_bits & MDS_INODELOCK_DOM) {
1944                                 /* DOM lock was taken in advance but this is
1945                                  * not DoM file. Drop the lock.
1946                                  */
1947                                 lock_res_and_lock(lock);
1948                                 ldlm_inodebits_drop(lock, MDS_INODELOCK_DOM);
1949                                 unlock_res_and_lock(lock);
1950                         }
1951                         LDLM_LOCK_PUT(lock);
1952                         GOTO(out_parent, rc = 0);
1953                 }
1954                 LDLM_LOCK_PUT(lock);
1955         }
1956
1957         EXIT;
1958 out_child:
1959         mdt_object_put(info->mti_env, child);
1960 out_parent:
1961         if (lhp)
1962                 mdt_object_unlock(info, parent, lhp, 1);
1963         return rc;
1964 }
1965
1966 /* normal handler: should release the child lock */
1967 static int mdt_getattr_name(struct tgt_session_info *tsi)
1968 {
1969         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1970         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1971         struct mdt_body        *reqbody;
1972         struct mdt_body        *repbody;
1973         int rc, rc2;
1974         ENTRY;
1975
1976         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1977         LASSERT(reqbody != NULL);
1978         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1979         LASSERT(repbody != NULL);
1980
1981         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1982         repbody->mbo_eadatasize = 0;
1983         repbody->mbo_aclsize = 0;
1984
1985         rc = mdt_init_ucred_intent_getattr(info, reqbody);
1986         if (unlikely(rc))
1987                 GOTO(out_shrink, rc);
1988
1989         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1990         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1991                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1992                 lhc->mlh_reg_lh.cookie = 0;
1993         }
1994         mdt_exit_ucred(info);
1995         EXIT;
1996 out_shrink:
1997         mdt_client_compatibility(info);
1998         rc2 = mdt_fix_reply(info);
1999         if (rc == 0)
2000                 rc = rc2;
2001         mdt_thread_info_fini(info);
2002         return rc;
2003 }
2004
2005 static int mdt_rmfid_unlink(struct mdt_thread_info *info,
2006                             const struct lu_fid *pfid,
2007                             const struct lu_name *name,
2008                             struct mdt_object *obj, s64 ctime)
2009 {
2010         struct lu_fid *child_fid = &info->mti_tmp_fid1;
2011         struct ldlm_enqueue_info *einfo = &info->mti_einfo[0];
2012         struct mdt_device *mdt = info->mti_mdt;
2013         struct md_attr *ma = &info->mti_attr;
2014         struct mdt_lock_handle *parent_lh;
2015         struct mdt_lock_handle *child_lh;
2016         struct mdt_object *pobj;
2017         bool cos_incompat = false;
2018         int rc;
2019         ENTRY;
2020
2021         pobj = mdt_object_find(info->mti_env, mdt, pfid);
2022         if (IS_ERR(pobj))
2023                 GOTO(out, rc = PTR_ERR(pobj));
2024
2025         parent_lh = &info->mti_lh[MDT_LH_PARENT];
2026         mdt_lock_pdo_init(parent_lh, LCK_PW, name);
2027         rc = mdt_object_lock(info, pobj, parent_lh, MDS_INODELOCK_UPDATE);
2028         if (rc != 0)
2029                 GOTO(put_parent, rc);
2030
2031         if (mdt_object_remote(pobj))
2032                 cos_incompat = true;
2033
2034         rc = mdo_lookup(info->mti_env, mdt_object_child(pobj),
2035                         name, child_fid, &info->mti_spec);
2036         if (rc != 0)
2037                 GOTO(unlock_parent, rc);
2038
2039         if (!lu_fid_eq(child_fid, mdt_object_fid(obj)))
2040                 GOTO(unlock_parent, rc = -EREMCHG);
2041
2042         child_lh = &info->mti_lh[MDT_LH_CHILD];
2043         mdt_lock_reg_init(child_lh, LCK_EX);
2044         rc = mdt_reint_striped_lock(info, obj, child_lh,
2045                                     MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE,
2046                                     einfo, cos_incompat);
2047         if (rc != 0)
2048                 GOTO(unlock_parent, rc);
2049
2050         if (atomic_read(&obj->mot_open_count)) {
2051                 CDEBUG(D_OTHER, "object "DFID" open, skip\n",
2052                        PFID(mdt_object_fid(obj)));
2053                 GOTO(unlock_child, rc = -EBUSY);
2054         }
2055
2056         ma->ma_need = 0;
2057         ma->ma_valid = MA_INODE;
2058         ma->ma_attr.la_valid = LA_CTIME;
2059         ma->ma_attr.la_ctime = ctime;
2060
2061         mutex_lock(&obj->mot_lov_mutex);
2062
2063         rc = mdo_unlink(info->mti_env, mdt_object_child(pobj),
2064                         mdt_object_child(obj), name, ma, 0);
2065
2066         mutex_unlock(&obj->mot_lov_mutex);
2067
2068 unlock_child:
2069         mdt_reint_striped_unlock(info, obj, child_lh, einfo, 1);
2070 unlock_parent:
2071         mdt_object_unlock(info, pobj, parent_lh, 1);
2072 put_parent:
2073         mdt_object_put(info->mti_env, pobj);
2074 out:
2075         RETURN(rc);
2076 }
2077
2078 static int mdt_rmfid_check_permission(struct mdt_thread_info *info,
2079                                         struct mdt_object *obj)
2080 {
2081         struct lu_ucred *uc = lu_ucred(info->mti_env);
2082         struct md_attr *ma = &info->mti_attr;
2083         struct lu_attr *la = &ma->ma_attr;
2084         int rc = 0;
2085         ENTRY;
2086
2087         ma->ma_need = MA_INODE;
2088         rc = mo_attr_get(info->mti_env, mdt_object_child(obj), ma);
2089         if (rc)
2090                 GOTO(out, rc);
2091
2092         if (la->la_flags & LUSTRE_IMMUTABLE_FL)
2093                         rc = -EACCES;
2094
2095         if (md_capable(uc, CFS_CAP_DAC_OVERRIDE))
2096                 RETURN(0);
2097         if (uc->uc_fsuid == la->la_uid) {
2098                 if ((la->la_mode & S_IWUSR) == 0)
2099                         rc = -EACCES;
2100         } else if (uc->uc_fsgid == la->la_gid) {
2101                 if ((la->la_mode & S_IWGRP) == 0)
2102                         rc = -EACCES;
2103         } else if ((la->la_mode & S_IWOTH) == 0) {
2104                         rc = -EACCES;
2105         }
2106
2107 out:
2108         RETURN(rc);
2109 }
2110
2111 static int mdt_rmfid_one(struct mdt_thread_info *info, struct lu_fid *fid,
2112                          s64 ctime)
2113 {
2114         struct mdt_device *mdt = info->mti_mdt;
2115         struct mdt_object *obj = NULL;
2116         struct linkea_data ldata = { NULL };
2117         struct lu_buf *buf = &info->mti_big_buf;
2118         struct lu_name *name = &info->mti_name;
2119         struct lu_fid *pfid = &info->mti_tmp_fid1;
2120         struct link_ea_header *leh;
2121         struct link_ea_entry *lee;
2122         int reclen, count, rc = 0;
2123         ENTRY;
2124
2125         if (!fid_is_sane(fid))
2126                 GOTO(out, rc = -EINVAL);
2127
2128         if (!fid_is_namespace_visible(fid))
2129                 GOTO(out, rc = -EINVAL);
2130
2131         obj = mdt_object_find(info->mti_env, mdt, fid);
2132         if (IS_ERR(obj))
2133                 GOTO(out, rc = PTR_ERR(obj));
2134
2135         if (mdt_object_remote(obj))
2136                 GOTO(out, rc = -EREMOTE);
2137         if (!mdt_object_exists(obj) || lu_object_is_dying(&obj->mot_header))
2138                 GOTO(out, rc = -ENOENT);
2139
2140         rc = mdt_rmfid_check_permission(info, obj);
2141         if (rc)
2142                 GOTO(out, rc);
2143
2144         /* take LinkEA */
2145         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
2146         if (!buf->lb_buf)
2147                 GOTO(out, rc = -ENOMEM);
2148
2149         ldata.ld_buf = buf;
2150         rc = mdt_links_read(info, obj, &ldata);
2151         if (rc)
2152                 GOTO(out, rc);
2153
2154         leh = buf->lb_buf;
2155         lee = (struct link_ea_entry *)(leh + 1);
2156         for (count = 0; count < leh->leh_reccount; count++) {
2157                 /* remove every hardlink */
2158                 linkea_entry_unpack(lee, &reclen, name, pfid);
2159                 lee = (struct link_ea_entry *) ((char *)lee + reclen);
2160                 rc = mdt_rmfid_unlink(info, pfid, name, obj, ctime);
2161                 if (rc)
2162                         break;
2163         }
2164
2165 out:
2166         if (obj && !IS_ERR(obj))
2167                 mdt_object_put(info->mti_env, obj);
2168         if (info->mti_big_buf.lb_buf)
2169                 lu_buf_free(&info->mti_big_buf);
2170
2171         RETURN(rc);
2172 }
2173
2174 static int mdt_rmfid(struct tgt_session_info *tsi)
2175 {
2176         struct mdt_thread_info *mti = tsi2mdt_info(tsi);
2177         struct mdt_body *reqbody;
2178         struct lu_fid *fids, *rfids;
2179         int bufsize, rc;
2180         __u32 *rcs;
2181         int i, nr;
2182         ENTRY;
2183
2184         reqbody = req_capsule_client_get(tsi->tsi_pill, &RMF_MDT_BODY);
2185         if (reqbody == NULL)
2186                 RETURN(-EPROTO);
2187         bufsize = req_capsule_get_size(tsi->tsi_pill, &RMF_FID_ARRAY,
2188                                        RCL_CLIENT);
2189         nr = bufsize / sizeof(struct lu_fid);
2190         if (nr * sizeof(struct lu_fid) != bufsize)
2191                 RETURN(-EINVAL);
2192         req_capsule_set_size(tsi->tsi_pill, &RMF_RCS,
2193                              RCL_SERVER, nr * sizeof(__u32));
2194         req_capsule_set_size(tsi->tsi_pill, &RMF_FID_ARRAY,
2195                              RCL_SERVER, nr * sizeof(struct lu_fid));
2196         rc = req_capsule_server_pack(tsi->tsi_pill);
2197         if (rc)
2198                 GOTO(out, rc = err_serious(rc));
2199         fids = req_capsule_client_get(tsi->tsi_pill, &RMF_FID_ARRAY);
2200         if (fids == NULL)
2201                 RETURN(-EPROTO);
2202         rcs = req_capsule_server_get(tsi->tsi_pill, &RMF_RCS);
2203         LASSERT(rcs);
2204         rfids = req_capsule_server_get(tsi->tsi_pill, &RMF_FID_ARRAY);
2205         LASSERT(rfids);
2206
2207         mdt_init_ucred(mti, reqbody);
2208         for (i = 0; i < nr; i++) {
2209                 rfids[i] = fids[i];
2210                 rcs[i] = mdt_rmfid_one(mti, fids + i, reqbody->mbo_ctime);
2211         }
2212         mdt_exit_ucred(mti);
2213
2214 out:
2215         RETURN(rc);
2216 }
2217
2218 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2219                          void *karg, void __user *uarg);
2220
2221 static int mdt_set_info(struct tgt_session_info *tsi)
2222 {
2223         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2224         char                    *key;
2225         void                    *val;
2226         int                      keylen, vallen, rc = 0;
2227
2228         ENTRY;
2229
2230         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
2231         if (key == NULL) {
2232                 DEBUG_REQ(D_HA, req, "no set_info key");
2233                 RETURN(err_serious(-EFAULT));
2234         }
2235
2236         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
2237                                       RCL_CLIENT);
2238
2239         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
2240         if (val == NULL) {
2241                 DEBUG_REQ(D_HA, req, "no set_info val");
2242                 RETURN(err_serious(-EFAULT));
2243         }
2244
2245         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
2246                                       RCL_CLIENT);
2247
2248         /* Swab any part of val you need to here */
2249         if (KEY_IS(KEY_READ_ONLY)) {
2250                 spin_lock(&req->rq_export->exp_lock);
2251                 if (*(__u32 *)val)
2252                         *exp_connect_flags_ptr(req->rq_export) |=
2253                                 OBD_CONNECT_RDONLY;
2254                 else
2255                         *exp_connect_flags_ptr(req->rq_export) &=
2256                                 ~OBD_CONNECT_RDONLY;
2257                 spin_unlock(&req->rq_export->exp_lock);
2258         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2259                 struct changelog_setinfo *cs = val;
2260
2261                 if (vallen != sizeof(*cs)) {
2262                         CERROR("%s: bad changelog_clear setinfo size %d\n",
2263                                tgt_name(tsi->tsi_tgt), vallen);
2264                         RETURN(-EINVAL);
2265                 }
2266                 if (ptlrpc_req_need_swab(req)) {
2267                         __swab64s(&cs->cs_recno);
2268                         __swab32s(&cs->cs_id);
2269                 }
2270
2271                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
2272                                    vallen, val, NULL);
2273         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2274                 if (vallen > 0)
2275                         obd_export_evict_by_nid(req->rq_export->exp_obd, val);
2276         } else {
2277                 RETURN(-EINVAL);
2278         }
2279         RETURN(rc);
2280 }
2281
2282 static int mdt_readpage(struct tgt_session_info *tsi)
2283 {
2284         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
2285         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
2286         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
2287         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
2288         struct mdt_body         *repbody;
2289         int                      rc;
2290         int                      i;
2291
2292         ENTRY;
2293
2294         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
2295                 RETURN(err_serious(-ENOMEM));
2296
2297         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
2298         if (repbody == NULL || reqbody == NULL)
2299                 RETURN(err_serious(-EFAULT));
2300
2301         /*
2302          * prepare @rdpg before calling lower layers and transfer itself. Here
2303          * reqbody->size contains offset of where to start to read and
2304          * reqbody->nlink contains number bytes to read.
2305          */
2306         rdpg->rp_hash = reqbody->mbo_size;
2307         if (rdpg->rp_hash != reqbody->mbo_size) {
2308                 CERROR("Invalid hash: %#llx != %#llx\n",
2309                        rdpg->rp_hash, reqbody->mbo_size);
2310                 RETURN(-EFAULT);
2311         }
2312
2313         rdpg->rp_attrs = reqbody->mbo_mode;
2314         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
2315                 rdpg->rp_attrs |= LUDA_64BITHASH;
2316         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
2317                                 exp_max_brw_size(tsi->tsi_exp));
2318         rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >>
2319                           PAGE_SHIFT;
2320         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
2321         if (rdpg->rp_pages == NULL)
2322                 RETURN(-ENOMEM);
2323
2324         for (i = 0; i < rdpg->rp_npages; ++i) {
2325                 rdpg->rp_pages[i] = alloc_page(GFP_NOFS);
2326                 if (rdpg->rp_pages[i] == NULL)
2327                         GOTO(free_rdpg, rc = -ENOMEM);
2328         }
2329
2330         /* call lower layers to fill allocated pages with directory data */
2331         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
2332         if (rc < 0)
2333                 GOTO(free_rdpg, rc);
2334
2335         /* send pages to client */
2336         rc = tgt_sendpage(tsi, rdpg, rc);
2337
2338         EXIT;
2339 free_rdpg:
2340
2341         for (i = 0; i < rdpg->rp_npages; i++)
2342                 if (rdpg->rp_pages[i] != NULL)
2343                         __free_page(rdpg->rp_pages[i]);
2344         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
2345
2346         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
2347                 RETURN(0);
2348
2349         return rc;
2350 }
2351
2352 static int mdt_fix_attr_ucred(struct mdt_thread_info *info, __u32 op)
2353 {
2354         struct lu_ucred *uc = mdt_ucred_check(info);
2355         struct lu_attr *attr = &info->mti_attr.ma_attr;
2356
2357         if (uc == NULL)
2358                 return -EINVAL;
2359
2360         if (op != REINT_SETATTR) {
2361                 if ((attr->la_valid & LA_UID) && (attr->la_uid != -1))
2362                         attr->la_uid = uc->uc_fsuid;
2363                 /* for S_ISGID, inherit gid from his parent, such work will be
2364                  * done in cmm/mdd layer, here set all cases as uc->uc_fsgid. */
2365                 if ((attr->la_valid & LA_GID) && (attr->la_gid != -1))
2366                         attr->la_gid = uc->uc_fsgid;
2367         }
2368
2369         return 0;
2370 }
2371
2372 static inline bool mdt_is_readonly_open(struct mdt_thread_info *info, __u32 op)
2373 {
2374         return op == REINT_OPEN &&
2375              !(info->mti_spec.sp_cr_flags & (MDS_FMODE_WRITE | MDS_OPEN_CREAT));
2376 }
2377
2378 static void mdt_preset_secctx_size(struct mdt_thread_info *info)
2379 {
2380         struct req_capsule *pill = info->mti_pill;
2381
2382         if (req_capsule_has_field(pill, &RMF_FILE_SECCTX,
2383                                   RCL_SERVER) &&
2384             req_capsule_has_field(pill, &RMF_FILE_SECCTX_NAME,
2385                                   RCL_CLIENT)) {
2386                 if (req_capsule_get_size(pill, &RMF_FILE_SECCTX_NAME,
2387                                          RCL_CLIENT) != 0) {
2388                         /* pre-set size in server part with max size */
2389                         req_capsule_set_size(pill, &RMF_FILE_SECCTX,
2390                                              RCL_SERVER,
2391                                              info->mti_mdt->mdt_max_ea_size);
2392                 } else {
2393                         req_capsule_set_size(pill, &RMF_FILE_SECCTX,
2394                                              RCL_SERVER, 0);
2395                 }
2396         }
2397
2398 }
2399
2400 static int mdt_reint_internal(struct mdt_thread_info *info,
2401                               struct mdt_lock_handle *lhc,
2402                               __u32 op)
2403 {
2404         struct req_capsule      *pill = info->mti_pill;
2405         struct mdt_body         *repbody;
2406         int                      rc = 0, rc2;
2407
2408         ENTRY;
2409
2410         rc = mdt_reint_unpack(info, op);
2411         if (rc != 0) {
2412                 CERROR("Can't unpack reint, rc %d\n", rc);
2413                 RETURN(err_serious(rc));
2414         }
2415
2416
2417         /* check if the file system is set to readonly. O_RDONLY open
2418          * is still allowed even the file system is set to readonly mode */
2419         if (mdt_rdonly(info->mti_exp) && !mdt_is_readonly_open(info, op))
2420                 RETURN(err_serious(-EROFS));
2421
2422         /* for replay (no_create) lmm is not needed, client has it already */
2423         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2424                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2425                                      DEF_REP_MD_SIZE);
2426
2427         /* llog cookies are always 0, the field is kept for compatibility */
2428         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2429                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
2430
2431         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
2432          * by default. If the target object has more ACL entries, then
2433          * enlarge the buffer when necessary. */
2434         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
2435                 req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
2436                                      LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
2437
2438         mdt_preset_secctx_size(info);
2439
2440         rc = req_capsule_server_pack(pill);
2441         if (rc != 0) {
2442                 CERROR("Can't pack response, rc %d\n", rc);
2443                 RETURN(err_serious(rc));
2444         }
2445
2446         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
2447                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
2448                 LASSERT(repbody);
2449                 repbody->mbo_eadatasize = 0;
2450                 repbody->mbo_aclsize = 0;
2451         }
2452
2453         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
2454
2455         /* for replay no cookkie / lmm need, because client have this already */
2456         if (info->mti_spec.no_create)
2457                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2458                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
2459
2460         rc = mdt_init_ucred_reint(info);
2461         if (rc)
2462                 GOTO(out_shrink, rc);
2463
2464         rc = mdt_fix_attr_ucred(info, op);
2465         if (rc != 0)
2466                 GOTO(out_ucred, rc = err_serious(rc));
2467
2468         rc = mdt_check_resent(info, mdt_reconstruct, lhc);
2469         if (rc < 0) {
2470                 GOTO(out_ucred, rc);
2471         } else if (rc == 1) {
2472                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt");
2473                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
2474                 GOTO(out_ucred, rc);
2475         }
2476         rc = mdt_reint_rec(info, lhc);
2477         EXIT;
2478 out_ucred:
2479         mdt_exit_ucred(info);
2480 out_shrink:
2481         mdt_client_compatibility(info);
2482
2483         rc2 = mdt_fix_reply(info);
2484         if (rc == 0)
2485                 rc = rc2;
2486
2487         /*
2488          * Data-on-MDT optimization - read data along with OPEN and return it
2489          * in reply. Do that only if we have both DOM and LAYOUT locks.
2490          */
2491         if (rc == 0 && op == REINT_OPEN && !req_is_replay(pill->rc_req) &&
2492             info->mti_attr.ma_lmm != NULL &&
2493             mdt_lmm_dom_entry(info->mti_attr.ma_lmm) == LMM_DOM_ONLY) {
2494                 rc = mdt_dom_read_on_open(info, info->mti_mdt,
2495                                           &lhc->mlh_reg_lh);
2496         }
2497
2498         return rc;
2499 }
2500
2501 static long mdt_reint_opcode(struct ptlrpc_request *req,
2502                              const struct req_format **fmt)
2503 {
2504         struct mdt_device       *mdt;
2505         struct mdt_rec_reint    *rec;
2506         long                     opc;
2507
2508         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
2509         if (rec != NULL) {
2510                 opc = rec->rr_opcode;
2511                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
2512                 if (opc < REINT_MAX && fmt[opc] != NULL)
2513                         req_capsule_extend(&req->rq_pill, fmt[opc]);
2514                 else {
2515                         mdt = mdt_exp2dev(req->rq_export);
2516                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
2517                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
2518                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
2519                         opc = err_serious(-EFAULT);
2520                 }
2521         } else {
2522                 opc = err_serious(-EFAULT);
2523         }
2524         return opc;
2525 }
2526
2527 static int mdt_reint(struct tgt_session_info *tsi)
2528 {
2529         long opc;
2530         int  rc;
2531         static const struct req_format *reint_fmts[REINT_MAX] = {
2532                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
2533                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
2534                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
2535                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
2536                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
2537                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
2538                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
2539                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
2540                 [REINT_MIGRATE]  = &RQF_MDS_REINT_MIGRATE,
2541                 [REINT_RESYNC]   = &RQF_MDS_REINT_RESYNC,
2542         };
2543
2544         ENTRY;
2545
2546         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
2547         if (opc >= 0) {
2548                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2549                 /*
2550                  * No lock possible here from client to pass it to reint code
2551                  * path.
2552                  */
2553                 rc = mdt_reint_internal(info, NULL, opc);
2554                 mdt_thread_info_fini(info);
2555         } else {
2556                 rc = opc;
2557         }
2558
2559         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
2560         RETURN(rc);
2561 }
2562
2563 /* this should sync the whole device */
2564 int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
2565 {
2566         struct dt_device *dt = mdt->mdt_bottom;
2567         int rc;
2568         ENTRY;
2569
2570         rc = dt->dd_ops->dt_sync(env, dt);
2571         RETURN(rc);
2572 }
2573
2574 /* this should sync this object */
2575 static int mdt_object_sync(const struct lu_env *env, struct obd_export *exp,
2576                            struct mdt_object *mo)
2577 {
2578         int rc = 0;
2579
2580         ENTRY;
2581
2582         if (!mdt_object_exists(mo)) {
2583                 CWARN("%s: non existing object "DFID": rc = %d\n",
2584                       exp->exp_obd->obd_name, PFID(mdt_object_fid(mo)),
2585                       -ESTALE);
2586                 RETURN(-ESTALE);
2587         }
2588
2589         if (S_ISREG(lu_object_attr(&mo->mot_obj))) {
2590                 struct lu_target *tgt = tgt_ses_info(env)->tsi_tgt;
2591                 dt_obj_version_t version;
2592
2593                 version = dt_version_get(env, mdt_obj2dt(mo));
2594                 if (version > tgt->lut_obd->obd_last_committed)
2595                         rc = mo_object_sync(env, mdt_object_child(mo));
2596         } else {
2597                 rc = mo_object_sync(env, mdt_object_child(mo));
2598         }
2599
2600         RETURN(rc);
2601 }
2602
2603 static int mdt_sync(struct tgt_session_info *tsi)
2604 {
2605         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2606         struct req_capsule      *pill = tsi->tsi_pill;
2607         struct mdt_body         *body;
2608         int                      rc;
2609
2610         ENTRY;
2611
2612         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
2613                 RETURN(err_serious(-ENOMEM));
2614
2615         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
2616                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
2617         } else {
2618                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2619
2620                 if (unlikely(info->mti_object == NULL))
2621                         RETURN(-EPROTO);
2622
2623                 /* sync an object */
2624                 rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp,
2625                                      info->mti_object);
2626                 if (rc == 0) {
2627                         const struct lu_fid *fid;
2628                         struct lu_attr *la = &info->mti_attr.ma_attr;
2629
2630                         info->mti_attr.ma_need = MA_INODE;
2631                         info->mti_attr.ma_valid = 0;
2632                         rc = mdt_attr_get_complex(info, info->mti_object,
2633                                                   &info->mti_attr);
2634                         if (rc == 0) {
2635                                 body = req_capsule_server_get(pill,
2636                                                               &RMF_MDT_BODY);
2637                                 fid = mdt_object_fid(info->mti_object);
2638                                 mdt_pack_attr2body(info, body, la, fid);
2639                         }
2640                 }
2641                 mdt_thread_info_fini(info);
2642         }
2643         if (rc == 0)
2644                 mdt_counter_incr(req, LPROC_MDT_SYNC);
2645
2646         RETURN(rc);
2647 }
2648
2649 static int mdt_data_sync(struct tgt_session_info *tsi)
2650 {
2651         struct mdt_thread_info *info;
2652         struct mdt_device *mdt = mdt_exp2dev(tsi->tsi_exp);
2653         struct ost_body *body = tsi->tsi_ost_body;
2654         struct ost_body *repbody;
2655         struct mdt_object *mo = NULL;
2656         struct md_attr *ma;
2657         int rc = 0;
2658
2659         ENTRY;
2660
2661         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2662
2663         /* if no fid is specified then do nothing,
2664          * device sync is done via MDS_SYNC */
2665         if (fid_is_zero(&tsi->tsi_fid))
2666                 RETURN(0);
2667
2668         mo = mdt_object_find(tsi->tsi_env, mdt, &tsi->tsi_fid);
2669         if (IS_ERR(mo))
2670                 RETURN(PTR_ERR(mo));
2671
2672         rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp, mo);
2673         if (rc)
2674                 GOTO(put, rc);
2675
2676         repbody->oa.o_oi = body->oa.o_oi;
2677         repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
2678
2679         info = tsi2mdt_info(tsi);
2680         ma = &info->mti_attr;
2681         ma->ma_need = MA_INODE;
2682         ma->ma_valid = 0;
2683         rc = mdt_attr_get_complex(info, mo, ma);
2684         if (rc == 0)
2685                 obdo_from_la(&repbody->oa, &ma->ma_attr, VALID_FLAGS);
2686         else
2687                 rc = 0;
2688         mdt_thread_info_fini(info);
2689
2690         EXIT;
2691 put:
2692         if (mo != NULL)
2693                 mdt_object_put(tsi->tsi_env, mo);
2694         return rc;
2695 }
2696
2697 /*
2698  * Handle quota control requests to consult current usage/limit, but also
2699  * to configure quota enforcement
2700  */
2701 static int mdt_quotactl(struct tgt_session_info *tsi)
2702 {
2703         struct obd_export       *exp  = tsi->tsi_exp;
2704         struct req_capsule      *pill = tsi->tsi_pill;
2705         struct obd_quotactl     *oqctl, *repoqc;
2706         int                      id, rc;
2707         struct mdt_device       *mdt = mdt_exp2dev(exp);
2708         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2709         struct lu_nodemap       *nodemap;
2710         ENTRY;
2711
2712         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
2713         if (oqctl == NULL)
2714                 RETURN(err_serious(-EPROTO));
2715
2716         rc = req_capsule_server_pack(pill);
2717         if (rc)
2718                 RETURN(err_serious(rc));
2719
2720         nodemap = nodemap_get_from_exp(exp);
2721         if (IS_ERR(nodemap))
2722                 RETURN(PTR_ERR(nodemap));
2723
2724         switch (oqctl->qc_cmd) {
2725                 /* master quotactl */
2726         case Q_SETINFO:
2727         case Q_SETQUOTA:
2728         case LUSTRE_Q_SETDEFAULT:
2729                 if (!nodemap_can_setquota(nodemap))
2730                         GOTO(out_nodemap, rc = -EPERM);
2731                 /* fallthrough */
2732         case Q_GETINFO:
2733         case Q_GETQUOTA:
2734         case LUSTRE_Q_GETDEFAULT:
2735                 if (qmt == NULL)
2736                         GOTO(out_nodemap, rc = -EOPNOTSUPP);
2737                 /* slave quotactl */
2738                 /* fallthrough */
2739         case Q_GETOINFO:
2740         case Q_GETOQUOTA:
2741                 break;
2742         default:
2743                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2744                 GOTO(out_nodemap, rc = -EFAULT);
2745         }
2746
2747         id = oqctl->qc_id;
2748         switch (oqctl->qc_type) {
2749         case USRQUOTA:
2750                 id = nodemap_map_id(nodemap, NODEMAP_UID,
2751                                     NODEMAP_CLIENT_TO_FS, id);
2752                 break;
2753         case GRPQUOTA:
2754                 id = nodemap_map_id(nodemap, NODEMAP_GID,
2755                                     NODEMAP_CLIENT_TO_FS, id);
2756                 break;
2757         case PRJQUOTA:
2758                 /* todo: check/map project id */
2759                 id = oqctl->qc_id;
2760                 break;
2761         default:
2762                 GOTO(out_nodemap, rc = -EOPNOTSUPP);
2763         }
2764         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2765         if (repoqc == NULL)
2766                 GOTO(out_nodemap, rc = err_serious(-EFAULT));
2767
2768         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA)
2769                 barrier_exit(tsi->tsi_tgt->lut_bottom);
2770
2771         if (oqctl->qc_id != id)
2772                 swap(oqctl->qc_id, id);
2773
2774         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA) {
2775                 if (unlikely(!barrier_entry(tsi->tsi_tgt->lut_bottom)))
2776                         RETURN(-EINPROGRESS);
2777         }
2778
2779         switch (oqctl->qc_cmd) {
2780
2781         case Q_GETINFO:
2782         case Q_SETINFO:
2783         case Q_SETQUOTA:
2784         case Q_GETQUOTA:
2785         case LUSTRE_Q_SETDEFAULT:
2786         case LUSTRE_Q_GETDEFAULT:
2787                 /* forward quotactl request to QMT */
2788                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2789                 break;
2790
2791         case Q_GETOINFO:
2792         case Q_GETOQUOTA:
2793                 /* slave quotactl */
2794                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2795                                    oqctl);
2796                 break;
2797
2798         default:
2799                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2800                 GOTO(out_nodemap, rc = -EFAULT);
2801         }
2802
2803         if (oqctl->qc_id != id)
2804                 swap(oqctl->qc_id, id);
2805
2806         *repoqc = *oqctl;
2807
2808         EXIT;
2809
2810 out_nodemap:
2811         nodemap_putref(nodemap);
2812
2813         return rc;
2814 }
2815
2816 /** clone llog ctxt from child (mdd)
2817  * This allows remote llog (replicator) access.
2818  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2819  * context was originally set up, or we can handle them directly.
2820  * I choose the latter, but that means I need any llog
2821  * contexts set up by child to be accessable by the mdt.  So we clone the
2822  * context into our context list here.
2823  */
2824 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2825                                int idx)
2826 {
2827         struct md_device  *next = mdt->mdt_child;
2828         struct llog_ctxt *ctxt;
2829         int rc;
2830
2831         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2832                 return 0;
2833
2834         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2835         if (rc || ctxt == NULL) {
2836                 return 0;
2837         }
2838
2839         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2840         if (rc)
2841                 CERROR("Can't set mdt ctxt %d\n", rc);
2842
2843         return rc;
2844 }
2845
2846 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2847                                  struct mdt_device *mdt, int idx)
2848 {
2849         struct llog_ctxt *ctxt;
2850
2851         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2852         if (ctxt == NULL)
2853                 return 0;
2854         /* Put once for the get we just did, and once for the clone */
2855         llog_ctxt_put(ctxt);
2856         llog_ctxt_put(ctxt);
2857         return 0;
2858 }
2859
2860 /*
2861  * sec context handlers
2862  */
2863 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2864 {
2865         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2866
2867         return 0;
2868 }
2869
2870 /*
2871  * quota request handlers
2872  */
2873 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2874 {
2875         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
2876         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2877         int                      rc;
2878         ENTRY;
2879
2880         if (qmt == NULL)
2881                 RETURN(err_serious(-EOPNOTSUPP));
2882
2883         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2884         RETURN(rc);
2885 }
2886
2887 struct mdt_object *mdt_object_new(const struct lu_env *env,
2888                                   struct mdt_device *d,
2889                                   const struct lu_fid *f)
2890 {
2891         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2892         struct lu_object *o;
2893         struct mdt_object *m;
2894         ENTRY;
2895
2896         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2897         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2898         if (unlikely(IS_ERR(o)))
2899                 m = (struct mdt_object *)o;
2900         else
2901                 m = mdt_obj(o);
2902         RETURN(m);
2903 }
2904
2905 struct mdt_object *mdt_object_find(const struct lu_env *env,
2906                                    struct mdt_device *d,
2907                                    const struct lu_fid *f)
2908 {
2909         struct lu_object *o;
2910         struct mdt_object *m;
2911         ENTRY;
2912
2913         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2914         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2915         if (unlikely(IS_ERR(o)))
2916                 m = (struct mdt_object *)o;
2917         else
2918                 m = mdt_obj(o);
2919
2920         RETURN(m);
2921 }
2922
2923 /**
2924  * Asyncronous commit for mdt device.
2925  *
2926  * Pass asynchonous commit call down the MDS stack.
2927  *
2928  * \param env environment
2929  * \param mdt the mdt device
2930  */
2931 static void mdt_device_commit_async(const struct lu_env *env,
2932                                     struct mdt_device *mdt)
2933 {
2934         struct dt_device *dt = mdt->mdt_bottom;
2935         int rc;
2936         ENTRY;
2937
2938         rc = dt->dd_ops->dt_commit_async(env, dt);
2939         if (unlikely(rc != 0))
2940                 CWARN("%s: async commit start failed: rc = %d\n",
2941                       mdt_obd_name(mdt), rc);
2942         atomic_inc(&mdt->mdt_async_commit_count);
2943         EXIT;
2944 }
2945
2946 /**
2947  * Mark the lock as "synchonous".
2948  *
2949  * Mark the lock to deffer transaction commit to the unlock time.
2950  *
2951  * \param lock the lock to mark as "synchonous"
2952  *
2953  * \see mdt_is_lock_sync
2954  * \see mdt_save_lock
2955  */
2956 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2957 {
2958         lock->l_ast_data = (void*)1;
2959 }
2960
2961 /**
2962  * Check whehter the lock "synchonous" or not.
2963  *
2964  * \param lock the lock to check
2965  * \retval 1 the lock is "synchonous"
2966  * \retval 0 the lock isn't "synchronous"
2967  *
2968  * \see mdt_set_lock_sync
2969  * \see mdt_save_lock
2970  */
2971 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2972 {
2973         return lock->l_ast_data != NULL;
2974 }
2975
2976 /**
2977  * Blocking AST for mdt locks.
2978  *
2979  * Starts transaction commit if in case of COS lock conflict or
2980  * deffers such a commit to the mdt_save_lock.
2981  *
2982  * \param lock the lock which blocks a request or cancelling lock
2983  * \param desc unused
2984  * \param data unused
2985  * \param flag indicates whether this cancelling or blocking callback
2986  * \retval 0
2987  * \see ldlm_blocking_ast_nocheck
2988  */
2989 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2990                      void *data, int flag)
2991 {
2992         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2993         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2994         struct ldlm_cb_set_arg *arg = data;
2995         bool commit_async = false;
2996         int rc;
2997         ENTRY;
2998
2999         if (flag == LDLM_CB_CANCELING)
3000                 RETURN(0);
3001
3002         lock_res_and_lock(lock);
3003         if (lock->l_blocking_ast != mdt_blocking_ast) {
3004                 unlock_res_and_lock(lock);
3005                 RETURN(0);
3006         }
3007
3008         /* A blocking ast may be sent from ldlm_lock_decref_internal
3009          * when the last reference to a local lock was released and
3010          * during blocking event from ldlm_work_bl_ast_lock().
3011          * The 'data' parameter is l_ast_data in the first case and
3012          * callback arguments in the second one. Distinguish them by that.
3013          */
3014         if (!data || data == lock->l_ast_data || !arg->bl_desc)
3015                 goto skip_cos_checks;
3016
3017         if (lock->l_req_mode & (LCK_PW | LCK_EX)) {
3018                 if (mdt_cos_is_enabled(mdt)) {
3019                         if (!arg->bl_desc->bl_same_client)
3020                                 mdt_set_lock_sync(lock);
3021                 } else if (mdt_slc_is_enabled(mdt) &&
3022                            arg->bl_desc->bl_cos_incompat) {
3023                         mdt_set_lock_sync(lock);
3024                         /*
3025                          * we may do extra commit here, but there is a small
3026                          * window to miss a commit: lock was unlocked (saved),
3027                          * then a conflict lock queued and we come here, but
3028                          * REP-ACK not received, so lock was not converted to
3029                          * COS mode yet.
3030                          * Fortunately this window is quite small, so the
3031                          * extra commit should be rare (not to say distributed
3032                          * operation is rare too).
3033                          */
3034                         commit_async = true;
3035                 }
3036         } else if (lock->l_req_mode == LCK_COS) {
3037                 commit_async = true;
3038         }
3039
3040 skip_cos_checks:
3041         rc = ldlm_blocking_ast_nocheck(lock);
3042
3043         if (commit_async) {
3044                 struct lu_env env;
3045
3046                 rc = lu_env_init(&env, LCT_LOCAL);
3047                 if (unlikely(rc != 0))
3048                         CWARN("%s: lu_env initialization failed, cannot "
3049                               "start asynchronous commit: rc = %d\n",
3050                               obd->obd_name, rc);
3051                 else
3052                         mdt_device_commit_async(&env, mdt);
3053                 lu_env_fini(&env);
3054         }
3055         RETURN(rc);
3056 }
3057
3058 /*
3059  * Blocking AST for cross-MDT lock
3060  *
3061  * Discard lock from uncommitted_slc_locks and cancel it.
3062  *
3063  * \param lock  the lock which blocks a request or cancelling lock
3064  * \param desc  unused
3065  * \param data  unused
3066  * \param flag  indicates whether this cancelling or blocking callback
3067  * \retval      0 on success
3068  * \retval      negative number on error
3069  */
3070 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
3071                             void *data, int flag)
3072 {
3073         int rc = 0;
3074         ENTRY;
3075
3076         switch (flag) {
3077         case LDLM_CB_BLOCKING: {
3078                 struct lustre_handle lockh;
3079
3080                 ldlm_lock2handle(lock, &lockh);
3081                 rc = ldlm_cli_cancel(&lockh,
3082                         ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
3083                 if (rc < 0) {
3084                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
3085                         RETURN(rc);
3086                 }
3087                 break;
3088         }
3089         case LDLM_CB_CANCELING: {
3090                 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
3091                 struct mdt_device *mdt =
3092                                 mdt_dev(obd->obd_lu_dev->ld_site->ls_top_dev);
3093
3094                 LDLM_DEBUG(lock, "Revoke remote lock\n");
3095
3096                 /* discard slc lock here so that it can be cleaned anytime,
3097                  * especially for cleanup_resource() */
3098                 tgt_discard_slc_lock(&mdt->mdt_lut, lock);
3099
3100                 /* once we cache lock, l_ast_data is set to mdt_object */
3101                 if (lock->l_ast_data != NULL) {
3102                         struct mdt_object *mo = lock->l_ast_data;
3103                         struct lu_env env;
3104
3105                         rc = lu_env_init(&env, LCT_MD_THREAD);
3106                         if (unlikely(rc != 0)) {
3107                                 CWARN("%s: lu_env initialization failed, object %p "DFID" is leaked!: rc = %d\n",
3108                                       obd->obd_name, mo,
3109                                       PFID(mdt_object_fid(mo)), rc);
3110                                 RETURN(rc);
3111                         }
3112
3113                         if (lock->l_policy_data.l_inodebits.bits &
3114                             (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)) {
3115                                 rc = mo_invalidate(&env, mdt_object_child(mo));
3116                                 mo->mot_cache_attr = 0;
3117                         }
3118                         mdt_object_put(&env, mo);
3119                         lu_env_fini(&env);
3120                 }
3121                 break;
3122         }
3123         default:
3124                 LBUG();
3125         }
3126
3127         RETURN(rc);
3128 }
3129
3130 int mdt_check_resent_lock(struct mdt_thread_info *info,
3131                           struct mdt_object *mo,
3132                           struct mdt_lock_handle *lhc)
3133 {
3134         /* the lock might already be gotten in ldlm_handle_enqueue() */
3135         if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
3136                 struct ptlrpc_request *req = mdt_info_req(info);
3137                 struct ldlm_lock      *lock;
3138
3139                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
3140                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
3141                 if (lock == NULL) {
3142                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
3143                          * a resend case, however, it could be already destroyed
3144                          * due to client eviction or a raced cancel RPC. */
3145                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx",
3146                                           lhc->mlh_reg_lh.cookie);
3147                         RETURN(-ESTALE);
3148                 }
3149
3150                 if (!fid_res_name_eq(mdt_object_fid(mo),
3151                                      &lock->l_resource->lr_name)) {
3152                         CWARN("%s: Although resent, but still not "
3153                               "get child lock:"DFID"\n",
3154                               info->mti_exp->exp_obd->obd_name,
3155                               PFID(mdt_object_fid(mo)));
3156                         LDLM_LOCK_PUT(lock);
3157                         RETURN(-EPROTO);
3158                 }
3159                 LDLM_LOCK_PUT(lock);
3160                 return 0;
3161         }
3162         return 1;
3163 }
3164
3165 static void mdt_remote_object_lock_created_cb(struct ldlm_lock *lock)
3166 {
3167         mdt_object_get(NULL, lock->l_ast_data);
3168 }
3169
3170 int mdt_remote_object_lock_try(struct mdt_thread_info *mti,
3171                                struct mdt_object *o, const struct lu_fid *fid,
3172                                struct lustre_handle *lh, enum ldlm_mode mode,
3173                                __u64 *ibits, __u64 trybits, bool cache)
3174 {
3175         struct ldlm_enqueue_info *einfo = &mti->mti_remote_einfo;
3176         union ldlm_policy_data *policy = &mti->mti_policy;
3177         struct ldlm_res_id *res_id = &mti->mti_res_id;
3178         int rc = 0;
3179         ENTRY;
3180
3181         LASSERT(mdt_object_remote(o));
3182
3183         fid_build_reg_res_name(fid, res_id);
3184
3185         memset(einfo, 0, sizeof(*einfo));
3186         einfo->ei_type = LDLM_IBITS;
3187         einfo->ei_mode = mode;
3188         einfo->ei_cb_bl = mdt_remote_blocking_ast;
3189         einfo->ei_cb_cp = ldlm_completion_ast;
3190         einfo->ei_enq_slave = 0;
3191         einfo->ei_res_id = res_id;
3192
3193         if (cache) {
3194                 /*
3195                  * if we cache lock, couple lock with mdt_object, so that object
3196                  * can be easily found in lock ASTs.
3197                  */
3198                 einfo->ei_cbdata = o;
3199                 einfo->ei_cb_created = mdt_remote_object_lock_created_cb;
3200         }
3201
3202         memset(policy, 0, sizeof(*policy));
3203         policy->l_inodebits.bits = *ibits;
3204         policy->l_inodebits.try_bits = trybits;
3205
3206         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
3207                             policy);
3208
3209         /* Return successfully acquired bits to a caller */
3210         if (rc == 0) {
3211                 struct ldlm_lock *lock = ldlm_handle2lock(lh);
3212
3213                 LASSERT(lock);
3214                 *ibits = lock->l_policy_data.l_inodebits.bits;
3215                 LDLM_LOCK_PUT(lock);
3216         }
3217         RETURN(rc);
3218 }
3219
3220 int mdt_remote_object_lock(struct mdt_thread_info *mti, struct mdt_object *o,
3221                            const struct lu_fid *fid, struct lustre_handle *lh,
3222                            enum ldlm_mode mode, __u64 ibits, bool cache)
3223 {
3224         return mdt_remote_object_lock_try(mti, o, fid, lh, mode, &ibits, 0,
3225                                           cache);
3226 }
3227
3228 static int mdt_object_local_lock(struct mdt_thread_info *info,
3229                                  struct mdt_object *o,
3230                                  struct mdt_lock_handle *lh, __u64 *ibits,
3231                                  __u64 trybits, bool cos_incompat)
3232 {
3233         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
3234         union ldlm_policy_data *policy = &info->mti_policy;
3235         struct ldlm_res_id *res_id = &info->mti_res_id;
3236         __u64 dlmflags = 0, *cookie = NULL;
3237         int rc;
3238         ENTRY;
3239
3240         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
3241         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
3242         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
3243         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
3244
3245         if (cos_incompat) {
3246                 LASSERT(lh->mlh_reg_mode == LCK_PW ||
3247                         lh->mlh_reg_mode == LCK_EX);
3248                 dlmflags |= LDLM_FL_COS_INCOMPAT;
3249         } else if (mdt_cos_is_enabled(info->mti_mdt)) {
3250                 dlmflags |= LDLM_FL_COS_ENABLED;
3251         }
3252
3253         /* Only enqueue LOOKUP lock for remote object */
3254         LASSERT(ergo(mdt_object_remote(o), *ibits == MDS_INODELOCK_LOOKUP));
3255
3256         /* Lease lock are granted with LDLM_FL_CANCEL_ON_BLOCK */
3257         if (lh->mlh_type == MDT_REG_LOCK && lh->mlh_reg_mode == LCK_EX &&
3258             *ibits == MDS_INODELOCK_OPEN)
3259                 dlmflags |= LDLM_FL_CANCEL_ON_BLOCK;
3260
3261         if (lh->mlh_type == MDT_PDO_LOCK) {
3262                 /* check for exists after object is locked */
3263                 if (mdt_object_exists(o) == 0) {
3264                         /* Non-existent object shouldn't have PDO lock */
3265                         RETURN(-ESTALE);
3266                 } else {
3267                         /* Non-dir object shouldn't have PDO lock */
3268                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
3269                                 RETURN(-ENOTDIR);
3270                 }
3271         }
3272
3273         fid_build_reg_res_name(mdt_object_fid(o), res_id);
3274         dlmflags |= LDLM_FL_ATOMIC_CB;
3275
3276         if (info->mti_exp)
3277                 cookie = &info->mti_exp->exp_handle.h_cookie;
3278
3279         /*
3280          * Take PDO lock on whole directory and build correct @res_id for lock
3281          * on part of directory.
3282          */
3283         if (lh->mlh_pdo_hash != 0) {
3284                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
3285                 mdt_lock_pdo_mode(info, o, lh);
3286                 if (lh->mlh_pdo_mode != LCK_NL) {
3287                         /*
3288                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
3289                          * is never going to be sent to client and we do not
3290                          * want it slowed down due to possible cancels.
3291                          */
3292                         policy->l_inodebits.bits =
3293                                 *ibits & MDS_INODELOCK_UPDATE;
3294                         policy->l_inodebits.try_bits =
3295                                 trybits & MDS_INODELOCK_UPDATE;
3296                         /* at least one of them should be set */
3297                         LASSERT(policy->l_inodebits.bits |
3298                                 policy->l_inodebits.try_bits);
3299                         rc = mdt_fid_lock(info->mti_env, ns, &lh->mlh_pdo_lh,
3300                                           lh->mlh_pdo_mode, policy, res_id,
3301                                           dlmflags, cookie);
3302                         if (unlikely(rc != 0))
3303                                 GOTO(out_unlock, rc);
3304                 }
3305
3306                 /*
3307                  * Finish res_id initializing by name hash marking part of
3308                  * directory which is taking modification.
3309                  */
3310                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
3311         }
3312
3313         policy->l_inodebits.bits = *ibits;
3314         policy->l_inodebits.try_bits = trybits;
3315
3316         /*
3317          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
3318          * going to be sent to client. If it is - mdt_intent_policy() path will
3319          * fix it up and turn FL_LOCAL flag off.
3320          */
3321         rc = mdt_fid_lock(info->mti_env, ns, &lh->mlh_reg_lh, lh->mlh_reg_mode,
3322                           policy, res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
3323                           cookie);
3324 out_unlock:
3325         if (rc != 0)
3326                 mdt_object_unlock(info, o, lh, 1);
3327         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
3328                    lh->mlh_pdo_hash != 0 &&
3329                    (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
3330                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
3331
3332         /* Return successfully acquired bits to a caller */
3333         if (rc == 0) {
3334                 struct ldlm_lock *lock = ldlm_handle2lock(&lh->mlh_reg_lh);
3335
3336                 LASSERT(lock);
3337                 *ibits = lock->l_policy_data.l_inodebits.bits;
3338                 LDLM_LOCK_PUT(lock);
3339         }
3340         RETURN(rc);
3341 }
3342
3343 static int
3344 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
3345                          struct mdt_lock_handle *lh, __u64 *ibits,
3346                          __u64 trybits, bool cos_incompat)
3347 {
3348         struct mdt_lock_handle *local_lh = NULL;
3349         int rc;
3350         ENTRY;
3351
3352         if (!mdt_object_remote(o)) {
3353                 rc = mdt_object_local_lock(info, o, lh, ibits, trybits,
3354                                            cos_incompat);
3355                 RETURN(rc);
3356         }
3357
3358         /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
3359         *ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
3360                     MDS_INODELOCK_XATTR);
3361
3362         /* Only enqueue LOOKUP lock for remote object */
3363         if (*ibits & MDS_INODELOCK_LOOKUP) {
3364                 __u64 local = MDS_INODELOCK_LOOKUP;
3365
3366                 rc = mdt_object_local_lock(info, o, lh, &local, 0,
3367                                            cos_incompat);
3368                 if (rc != ELDLM_OK)
3369                         RETURN(rc);
3370
3371                 local_lh = lh;
3372         }
3373
3374         if ((*ibits | trybits) & MDS_INODELOCK_UPDATE) {
3375                 /* Sigh, PDO needs to enqueue 2 locks right now, but
3376                  * enqueue RPC can only request 1 lock, to avoid extra
3377                  * RPC, so it will instead enqueue EX lock for remote
3378                  * object anyway XXX*/
3379                 if (lh->mlh_type == MDT_PDO_LOCK &&
3380                     lh->mlh_pdo_hash != 0) {
3381                         CDEBUG(D_INFO,
3382                                "%s: "DFID" convert PDO lock to EX lock.\n",
3383                                mdt_obd_name(info->mti_mdt),
3384                                PFID(mdt_object_fid(o)));
3385                         lh->mlh_pdo_hash = 0;
3386                         lh->mlh_rreg_mode = LCK_EX;
3387                         lh->mlh_type = MDT_REG_LOCK;
3388                 }
3389
3390                 rc = mdt_remote_object_lock_try(info, o, mdt_object_fid(o),
3391                                                 &lh->mlh_rreg_lh,
3392                                                 lh->mlh_rreg_mode,
3393                                                 ibits, trybits, false);
3394                 if (rc != ELDLM_OK) {
3395                         if (local_lh != NULL)
3396                                 mdt_object_unlock(info, o, local_lh, rc);
3397                         RETURN(rc);
3398                 }
3399         }
3400
3401         /* other components like LFSCK can use lockless access
3402          * and populate cache, so we better invalidate it */
3403         mo_invalidate(info->mti_env, mdt_object_child(o));
3404
3405         RETURN(0);
3406 }
3407
3408 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
3409                     struct mdt_lock_handle *lh, __u64 ibits)
3410 {
3411         return mdt_object_lock_internal(info, o, lh, &ibits, 0, false);
3412 }
3413
3414 int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
3415                           struct mdt_lock_handle *lh, __u64 ibits,
3416                           bool cos_incompat)
3417 {
3418         LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
3419         return mdt_object_lock_internal(info, o, lh, &ibits, 0,
3420                                         cos_incompat);
3421 }
3422
3423 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
3424                         struct mdt_lock_handle *lh, __u64 *ibits,
3425                         __u64 trybits, bool cos_incompat)
3426 {
3427         bool trylock_only = *ibits == 0;
3428         int rc;
3429
3430         LASSERT(!(*ibits & trybits));
3431         rc = mdt_object_lock_internal(info, o, lh, ibits, trybits,
3432                                       cos_incompat);
3433         if (rc && trylock_only) { /* clear error for try ibits lock only */
3434                 LASSERT(*ibits == 0);
3435                 rc = 0;
3436         }
3437         return rc;
3438 }
3439
3440 /**
3441  * Save a lock within request object.
3442  *
3443  * Keep the lock referenced until whether client ACK or transaction
3444  * commit happens or release the lock immediately depending on input
3445  * parameters. If COS is ON, a write lock is converted to COS lock
3446  * before saving.
3447  *
3448  * \param info thead info object
3449  * \param h lock handle
3450  * \param mode lock mode
3451  * \param decref force immediate lock releasing
3452  */
3453 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
3454                    enum ldlm_mode mode, int decref)
3455 {
3456         ENTRY;
3457
3458         if (lustre_handle_is_used(h)) {
3459                 if (decref || !info->mti_has_trans ||
3460                     !(mode & (LCK_PW | LCK_EX))) {
3461                         mdt_fid_unlock(h, mode);
3462                 } else {
3463                         struct mdt_device *mdt = info->mti_mdt;
3464                         struct ldlm_lock *lock = ldlm_handle2lock(h);
3465                         struct ptlrpc_request *req = mdt_info_req(info);
3466                         bool cos = mdt_cos_is_enabled(mdt);
3467                         bool convert_lock = !cos && mdt_slc_is_enabled(mdt);
3468
3469                         LASSERTF(lock != NULL, "no lock for cookie %#llx\n",
3470                                  h->cookie);
3471
3472                         /* there is no request if mdt_object_unlock() is called
3473                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
3474                         if (likely(req != NULL)) {
3475                                 LDLM_DEBUG(lock, "save lock request %p reply "
3476                                         "state %p transno %lld\n", req,
3477                                         req->rq_reply_state, req->rq_transno);
3478                                 if (cos) {
3479                                         ldlm_lock_mode_downgrade(lock, LCK_COS);
3480                                         mode = LCK_COS;
3481                                 }
3482                                 if (req->rq_export->exp_disconnected)
3483                                         mdt_fid_unlock(h, mode);
3484                                 else
3485                                         ptlrpc_save_lock(req, h, mode, cos,
3486                                                          convert_lock);
3487                         } else {
3488                                 mdt_fid_unlock(h, mode);
3489                         }
3490                         if (mdt_is_lock_sync(lock)) {
3491                                 CDEBUG(D_HA, "found sync-lock,"
3492                                        " async commit started\n");
3493                                 mdt_device_commit_async(info->mti_env,
3494                                                         mdt);
3495                         }
3496                         LDLM_LOCK_PUT(lock);
3497                 }
3498                 h->cookie = 0ull;
3499         }
3500
3501         EXIT;
3502 }
3503
3504 /**
3505  * Save cross-MDT lock in uncommitted_slc_locks
3506  *
3507  * Keep the lock referenced until transaction commit happens or release the lock
3508  * immediately depending on input parameters.
3509  *
3510  * \param info thead info object
3511  * \param h lock handle
3512  * \param mode lock mode
3513  * \param decref force immediate lock releasing
3514  */
3515 static void mdt_save_remote_lock(struct mdt_thread_info *info,
3516                                  struct mdt_object *o, struct lustre_handle *h,
3517                                  enum ldlm_mode mode, int decref)
3518 {
3519         ENTRY;
3520
3521         if (lustre_handle_is_used(h)) {
3522                 struct ldlm_lock *lock = ldlm_handle2lock(h);
3523
3524                 if (o != NULL &&
3525                     (lock->l_policy_data.l_inodebits.bits &
3526                      (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)))
3527                         mo_invalidate(info->mti_env, mdt_object_child(o));
3528
3529                 if (decref || !info->mti_has_trans ||
3530                     !(mode & (LCK_PW | LCK_EX))) {
3531                         ldlm_lock_decref_and_cancel(h, mode);
3532                         LDLM_LOCK_PUT(lock);
3533                 } else {
3534                         struct ptlrpc_request *req = mdt_info_req(info);
3535
3536                         LASSERT(req != NULL);
3537                         tgt_save_slc_lock(&info->mti_mdt->mdt_lut, lock,
3538                                           req->rq_transno);
3539                         ldlm_lock_decref(h, mode);
3540                 }
3541                 h->cookie = 0ull;
3542         }
3543
3544         EXIT;
3545 }
3546
3547 /**
3548  * Unlock mdt object.
3549  *
3550  * Immeditely release the regular lock and the PDO lock or save the
3551  * lock in request and keep them referenced until client ACK or
3552  * transaction commit.
3553  *
3554  * \param info thread info object
3555  * \param o mdt object
3556  * \param lh mdt lock handle referencing regular and PDO locks
3557  * \param decref force immediate lock releasing
3558  *
3559  * XXX o is not used and may be NULL, see hsm_cdt_request_completed().
3560  */
3561 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
3562                        struct mdt_lock_handle *lh, int decref)
3563 {
3564         ENTRY;
3565
3566         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
3567         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
3568         mdt_save_remote_lock(info, o, &lh->mlh_rreg_lh, lh->mlh_rreg_mode,
3569                              decref);
3570
3571         EXIT;
3572 }
3573
3574 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
3575                                         const struct lu_fid *f,
3576                                         struct mdt_lock_handle *lh,
3577                                         __u64 ibits)
3578 {
3579         struct mdt_object *o;
3580
3581         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
3582         if (!IS_ERR(o)) {
3583                 int rc;
3584
3585                 rc = mdt_object_lock(info, o, lh, ibits);
3586                 if (rc != 0) {
3587                         mdt_object_put(info->mti_env, o);
3588                         o = ERR_PTR(rc);
3589                 }
3590         }
3591         return o;
3592 }
3593
3594 void mdt_object_unlock_put(struct mdt_thread_info * info,
3595                            struct mdt_object * o,
3596                            struct mdt_lock_handle *lh,
3597                            int decref)
3598 {
3599         mdt_object_unlock(info, o, lh, decref);
3600         mdt_object_put(info->mti_env, o);
3601 }
3602
3603 /*
3604  * Generic code handling requests that have struct mdt_body passed in:
3605  *
3606  *  - extract mdt_body from request and save it in @info, if present;
3607  *
3608  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
3609  *  @info;
3610  *
3611  *  - if HAS_BODY flag is set for this request type check whether object
3612  *  actually exists on storage (lu_object_exists()).
3613  *
3614  */
3615 static int mdt_body_unpack(struct mdt_thread_info *info,
3616                            enum tgt_handler_flags flags)
3617 {
3618         const struct mdt_body    *body;
3619         struct mdt_object        *obj;
3620         const struct lu_env      *env;
3621         struct req_capsule       *pill;
3622         int                       rc;
3623         ENTRY;
3624
3625         env = info->mti_env;
3626         pill = info->mti_pill;
3627
3628         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
3629         if (body == NULL)
3630                 RETURN(-EFAULT);
3631
3632         if (!(body->mbo_valid & OBD_MD_FLID))
3633                 RETURN(0);
3634
3635         if (!fid_is_sane(&body->mbo_fid1)) {
3636                 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
3637                 RETURN(-EINVAL);
3638         }
3639
3640         obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
3641         if (!IS_ERR(obj)) {
3642                 if ((flags & HAS_BODY) && !mdt_object_exists(obj)) {
3643                         mdt_object_put(env, obj);
3644                         rc = -ENOENT;
3645                 } else {
3646                         info->mti_object = obj;
3647                         rc = 0;
3648                 }
3649         } else
3650                 rc = PTR_ERR(obj);
3651
3652         RETURN(rc);
3653 }
3654
3655 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info,
3656                                    enum tgt_handler_flags flags)
3657 {
3658         struct req_capsule *pill = info->mti_pill;
3659         int rc;
3660
3661         ENTRY;
3662
3663         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
3664                 rc = mdt_body_unpack(info, flags);
3665         else
3666                 rc = 0;
3667
3668         if (rc == 0 && (flags & HAS_REPLY)) {
3669                 /* Pack reply. */
3670                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
3671                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
3672                                              DEF_REP_MD_SIZE);
3673                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
3674                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
3675                                              RCL_SERVER, 0);
3676
3677                 /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
3678                  * by default. If the target object has more ACL entries, then
3679                  * enlarge the buffer when necessary. */
3680                 if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
3681                         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
3682                                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
3683
3684                 mdt_preset_secctx_size(info);
3685
3686                 rc = req_capsule_server_pack(pill);
3687                 if (rc)
3688                         CWARN("%s: cannot pack response: rc = %d\n",
3689                                       mdt_obd_name(info->mti_mdt), rc);
3690         }
3691         RETURN(rc);
3692 }
3693
3694 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
3695 {
3696         lh->mlh_type = MDT_NUL_LOCK;
3697         lh->mlh_reg_lh.cookie = 0ull;
3698         lh->mlh_reg_mode = LCK_MINMODE;
3699         lh->mlh_pdo_lh.cookie = 0ull;
3700         lh->mlh_pdo_mode = LCK_MINMODE;
3701         lh->mlh_rreg_lh.cookie = 0ull;
3702         lh->mlh_rreg_mode = LCK_MINMODE;
3703 }
3704
3705 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
3706 {
3707         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
3708         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
3709 }
3710
3711 /*
3712  * Initialize fields of struct mdt_thread_info. Other fields are left in
3713  * uninitialized state, because it's too expensive to zero out whole
3714  * mdt_thread_info (> 1K) on each request arrival.
3715  */
3716 void mdt_thread_info_init(struct ptlrpc_request *req,
3717                           struct mdt_thread_info *info)
3718 {
3719         int i;
3720
3721         info->mti_pill = &req->rq_pill;
3722
3723         /* lock handle */
3724         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3725                 mdt_lock_handle_init(&info->mti_lh[i]);
3726
3727         /* mdt device: it can be NULL while CONNECT */
3728         if (req->rq_export) {
3729                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
3730                 info->mti_exp = req->rq_export;
3731         } else
3732                 info->mti_mdt = NULL;
3733         info->mti_env = req->rq_svc_thread->t_env;
3734         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
3735
3736         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
3737         info->mti_big_buf = LU_BUF_NULL;
3738         info->mti_body = NULL;
3739         info->mti_object = NULL;
3740         info->mti_dlm_req = NULL;
3741         info->mti_has_trans = 0;
3742         info->mti_cross_ref = 0;
3743         info->mti_opdata = 0;
3744         info->mti_big_lmm_used = 0;
3745         info->mti_big_acl_used = 0;
3746         info->mti_som_valid = 0;
3747
3748         info->mti_spec.no_create = 0;
3749         info->mti_spec.sp_rm_entry = 0;
3750         info->mti_spec.sp_permitted = 0;
3751         info->mti_spec.sp_migrate_close = 0;
3752
3753         info->mti_spec.u.sp_ea.eadata = NULL;
3754         info->mti_spec.u.sp_ea.eadatalen = 0;
3755 }
3756
3757 void mdt_thread_info_fini(struct mdt_thread_info *info)
3758 {
3759         int i;
3760
3761         if (info->mti_object != NULL) {
3762                 mdt_object_put(info->mti_env, info->mti_object);
3763                 info->mti_object = NULL;
3764         }
3765
3766         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3767                 mdt_lock_handle_fini(&info->mti_lh[i]);
3768         info->mti_env = NULL;
3769         info->mti_pill = NULL;
3770         info->mti_exp = NULL;
3771         info->mti_mdt = NULL;
3772
3773         if (unlikely(info->mti_big_buf.lb_buf != NULL))
3774                 lu_buf_free(&info->mti_big_buf);
3775 }
3776
3777 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
3778 {
3779         struct mdt_thread_info  *mti;
3780
3781         mti = mdt_th_info(tsi->tsi_env);
3782         LASSERT(mti != NULL);
3783
3784         mdt_thread_info_init(tgt_ses_req(tsi), mti);
3785         if (tsi->tsi_corpus != NULL) {
3786                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
3787                 lu_object_get(tsi->tsi_corpus);
3788         }
3789         mti->mti_body = tsi->tsi_mdt_body;
3790         mti->mti_dlm_req = tsi->tsi_dlm_req;
3791
3792         return mti;
3793 }
3794
3795 static int mdt_tgt_connect(struct tgt_session_info *tsi)
3796 {
3797         if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
3798             cfs_fail_val ==
3799             tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id)
3800                 schedule_timeout_uninterruptible(cfs_time_seconds(3));
3801
3802         return tgt_connect(tsi);
3803 }
3804
3805 static int mdt_intent_glimpse(enum ldlm_intent_flags it_opc,
3806                               struct mdt_thread_info *info,
3807                               struct ldlm_lock **lockp, __u64 flags)
3808 {
3809         return mdt_glimpse_enqueue(info, info->mti_mdt->mdt_namespace,
3810                                    lockp, flags);
3811 }
3812 static int mdt_intent_brw(enum ldlm_intent_flags it_opc,
3813                           struct mdt_thread_info *info,
3814                           struct ldlm_lock **lockp, __u64 flags)
3815 {
3816         return mdt_brw_enqueue(info, info->mti_mdt->mdt_namespace,
3817                                lockp, flags);
3818 }
3819
3820 int mdt_intent_lock_replace(struct mdt_thread_info *info,
3821                             struct ldlm_lock **lockp,
3822                             struct mdt_lock_handle *lh,
3823                             __u64 flags, int result)
3824 {
3825         struct ptlrpc_request  *req = mdt_info_req(info);
3826         struct ldlm_lock       *lock = *lockp;
3827         struct ldlm_lock       *new_lock;
3828
3829         /* If possible resent found a lock, @lh is set to its handle */
3830         new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3831
3832         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3833                 lh->mlh_reg_lh.cookie = 0;
3834                 RETURN(0);
3835         }
3836
3837         if (new_lock == NULL && (flags & LDLM_FL_RESENT)) {
3838                 /* Lock is pinned by ldlm_handle_enqueue0() as it is
3839                  * a resend case, however, it could be already destroyed
3840                  * due to client eviction or a raced cancel RPC. */
3841                 LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx\n",
3842                                   lh->mlh_reg_lh.cookie);
3843                 lh->mlh_reg_lh.cookie = 0;
3844                 RETURN(-ESTALE);
3845         }
3846
3847         LASSERTF(new_lock != NULL,
3848                  "lockh %#llx flags %#llx : rc = %d\n",
3849                  lh->mlh_reg_lh.cookie, flags, result);
3850
3851         /*
3852          * If we've already given this lock to a client once, then we should
3853          * have no readers or writers.  Otherwise, we should have one reader
3854          * _or_ writer ref (which will be zeroed below) before returning the
3855          * lock to a client.
3856          */
3857         if (new_lock->l_export == req->rq_export) {
3858                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3859         } else {
3860                 LASSERT(new_lock->l_export == NULL);
3861                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3862         }
3863
3864         *lockp = new_lock;
3865
3866         if (new_lock->l_export == req->rq_export) {
3867                 /*
3868                  * Already gave this to the client, which means that we
3869                  * reconstructed a reply.
3870                  */
3871                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3872                         MSG_RESENT);
3873
3874                 LDLM_LOCK_RELEASE(new_lock);
3875                 lh->mlh_reg_lh.cookie = 0;
3876                 RETURN(ELDLM_LOCK_REPLACED);
3877         }
3878
3879         /*
3880          * Fixup the lock to be given to the client.
3881          */
3882         lock_res_and_lock(new_lock);
3883         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3884          * possible blocking AST. */
3885         while (new_lock->l_readers > 0) {
3886                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3887                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3888                 new_lock->l_readers--;
3889         }
3890         while (new_lock->l_writers > 0) {
3891                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3892                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3893                 new_lock->l_writers--;
3894         }
3895
3896         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3897         new_lock->l_blocking_ast = lock->l_blocking_ast;
3898         new_lock->l_completion_ast = lock->l_completion_ast;
3899         if (ldlm_has_dom(new_lock))
3900                 new_lock->l_glimpse_ast = ldlm_server_glimpse_ast;
3901         new_lock->l_remote_handle = lock->l_remote_handle;
3902         new_lock->l_flags &= ~LDLM_FL_LOCAL;
3903
3904         unlock_res_and_lock(new_lock);
3905
3906         cfs_hash_add(new_lock->l_export->exp_lock_hash,
3907                      &new_lock->l_remote_handle,
3908                      &new_lock->l_exp_hash);
3909
3910         LDLM_LOCK_RELEASE(new_lock);
3911         lh->mlh_reg_lh.cookie = 0;
3912
3913         RETURN(ELDLM_LOCK_REPLACED);
3914 }
3915
3916 void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3917                              struct ldlm_lock *new_lock,
3918                              struct mdt_lock_handle *lh, __u64 flags)
3919 {
3920         struct ptlrpc_request  *req = mdt_info_req(info);
3921         struct ldlm_request    *dlmreq;
3922
3923         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3924                 return;
3925
3926         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3927
3928         /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
3929          * lock was found by ldlm_handle_enqueue(); if so @lh must be
3930          * initialized. */
3931         if (flags & LDLM_FL_RESENT) {
3932                 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
3933                 lh->mlh_reg_mode = new_lock->l_granted_mode;
3934
3935                 LDLM_DEBUG(new_lock, "Restoring lock cookie");
3936                 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie %#llx",
3937                           lh->mlh_reg_lh.cookie);
3938                 return;
3939         }
3940
3941         /*
3942          * If the xid matches, then we know this is a resent request, and allow
3943          * it. (It's probably an OPEN, for which we don't send a lock.
3944          */
3945         if (req_can_reconstruct(req, NULL) != 0)
3946                 return;
3947
3948         /*
3949          * This remote handle isn't enqueued, so we never received or processed
3950          * this request.  Clear MSG_RESENT, because it can be handled like any
3951          * normal request now.
3952          */
3953         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3954
3955         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle %#llx",
3956                   dlmreq->lock_handle[0].cookie);
3957 }
3958
3959 static int mdt_intent_getxattr(enum ldlm_intent_flags it_opc,
3960                                struct mdt_thread_info *info,
3961                                struct ldlm_lock **lockp,
3962                                __u64 flags)
3963 {
3964         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3965         struct ldlm_reply      *ldlm_rep = NULL;
3966         int rc;
3967         ENTRY;
3968
3969         /*
3970          * Initialize lhc->mlh_reg_lh either from a previously granted lock
3971          * (for the resend case) or a new lock. Below we will use it to
3972          * replace the original lock.
3973          */
3974         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3975         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3976                 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3977                 rc = mdt_object_lock(info, info->mti_object, lhc,
3978                                      MDS_INODELOCK_XATTR);
3979                 if (rc)
3980                         return rc;
3981         }
3982
3983         rc = mdt_getxattr(info);
3984
3985         if (mdt_info_req(info)->rq_repmsg != NULL)
3986                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3987
3988         if (ldlm_rep == NULL ||
3989             OBD_FAIL_CHECK(OBD_FAIL_MDS_XATTR_REP)) {
3990                 mdt_object_unlock(info,  info->mti_object, lhc, 1);
3991                 if (is_serious(rc))
3992                         RETURN(rc);
3993                 else
3994                         RETURN(err_serious(-EFAULT));
3995         }
3996
3997         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3998
3999         /* This is left for interop instead of adding a new interop flag.
4000          * LU-7433 */
4001 #if LUSTRE_VERSION_CODE > OBD_OCD_VERSION(3, 0, 0, 0)
4002         if (ldlm_rep->lock_policy_res2) {
4003                 mdt_object_unlock(info, info->mti_object, lhc, 1);
4004                 RETURN(ELDLM_LOCK_ABORTED);
4005         }
4006 #endif
4007
4008         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4009         RETURN(rc);
4010 }
4011
4012 static int mdt_intent_getattr(enum ldlm_intent_flags it_opc,
4013                               struct mdt_thread_info *info,
4014                               struct ldlm_lock **lockp,
4015                               __u64 flags)
4016 {
4017         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4018         __u64                   child_bits;
4019         struct ldlm_reply      *ldlm_rep;
4020         struct mdt_body        *reqbody;
4021         struct mdt_body        *repbody;
4022         int                     rc, rc2;
4023         ENTRY;
4024
4025         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
4026         LASSERT(reqbody);
4027
4028         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
4029         LASSERT(repbody);
4030
4031         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
4032         repbody->mbo_eadatasize = 0;
4033         repbody->mbo_aclsize = 0;
4034
4035         switch (it_opc) {
4036         case IT_LOOKUP:
4037                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
4038                 break;
4039         case IT_GETATTR:
4040                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
4041                              MDS_INODELOCK_PERM;
4042                 break;
4043         default:
4044                 CERROR("%s: unsupported intent %#x\n",
4045                        mdt_obd_name(info->mti_mdt), (unsigned int)it_opc);
4046                 GOTO(out_shrink, rc = -EINVAL);
4047         }
4048
4049         rc = mdt_init_ucred_intent_getattr(info, reqbody);
4050         if (rc)
4051                 GOTO(out_shrink, rc);
4052
4053         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4054         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
4055
4056         /* Get lock from request for possible resent case. */
4057         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4058
4059         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
4060         ldlm_rep->lock_policy_res2 = clear_serious(rc);
4061
4062         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
4063                 ldlm_rep->lock_policy_res2 = 0;
4064         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
4065             ldlm_rep->lock_policy_res2) {
4066                 lhc->mlh_reg_lh.cookie = 0ull;
4067                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
4068         }
4069
4070         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4071         EXIT;
4072 out_ucred:
4073         mdt_exit_ucred(info);
4074 out_shrink:
4075         mdt_client_compatibility(info);
4076         rc2 = mdt_fix_reply(info);
4077         if (rc == 0)
4078                 rc = rc2;
4079         return rc;
4080 }
4081
4082 static int mdt_intent_layout(enum ldlm_intent_flags it_opc,
4083                              struct mdt_thread_info *info,
4084                              struct ldlm_lock **lockp,
4085                              __u64 flags)
4086 {
4087         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4088         struct md_layout_change layout = { .mlc_opc = MD_LAYOUT_NOP };
4089         struct layout_intent *intent;
4090         struct ldlm_reply *ldlm_rep;
4091         struct lu_fid *fid = &info->mti_tmp_fid2;
4092         struct mdt_object *obj = NULL;
4093         int layout_size = 0;
4094         struct lu_buf *buf = &layout.mlc_buf;
4095         int rc = 0;
4096
4097         ENTRY;
4098
4099         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
4100
4101         intent = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
4102         if (intent == NULL)
4103                 RETURN(-EPROTO);
4104
4105         CDEBUG(D_INFO, DFID "got layout change request from client: "
4106                "opc:%u flags:%#x extent "DEXT"\n",
4107                PFID(fid), intent->li_opc, intent->li_flags,
4108                PEXT(&intent->li_extent));
4109
4110         switch (intent->li_opc) {
4111         case LAYOUT_INTENT_TRUNC:
4112         case LAYOUT_INTENT_WRITE:
4113                 layout.mlc_opc = MD_LAYOUT_WRITE;
4114                 layout.mlc_intent = intent;
4115                 break;
4116         case LAYOUT_INTENT_ACCESS:
4117                 break;
4118         case LAYOUT_INTENT_READ:
4119         case LAYOUT_INTENT_GLIMPSE:
4120         case LAYOUT_INTENT_RELEASE:
4121         case LAYOUT_INTENT_RESTORE:
4122                 CERROR("%s: Unsupported layout intent opc %d\n",
4123                        mdt_obd_name(info->mti_mdt), intent->li_opc);
4124                 RETURN(-ENOTSUPP);
4125         default:
4126                 CERROR("%s: Unknown layout intent opc %d\n",
4127                        mdt_obd_name(info->mti_mdt), intent->li_opc);
4128                 RETURN(-EINVAL);
4129         }
4130
4131         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
4132         if (IS_ERR(obj))
4133                 RETURN(PTR_ERR(obj));
4134
4135         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
4136                 /* if layout is going to be changed don't use the current EA
4137                  * size but the maximum one. That buffer will be shrinked
4138                  * to the actual size in req_capsule_shrink() before reply.
4139                  */
4140                 if (layout.mlc_opc == MD_LAYOUT_WRITE) {
4141                         layout_size = info->mti_mdt->mdt_max_mdsize;
4142                 } else {
4143                         layout_size = mdt_attr_get_eabuf_size(info, obj);
4144                         if (layout_size < 0)
4145                                 GOTO(out, rc = layout_size);
4146
4147                         if (layout_size > info->mti_mdt->mdt_max_mdsize)
4148                                 info->mti_mdt->mdt_max_mdsize = layout_size;
4149                 }
4150                 CDEBUG(D_INFO, "%s: layout_size %d\n",
4151                        mdt_obd_name(info->mti_mdt), layout_size);
4152         }
4153
4154         /*
4155          * set reply buffer size, so that ldlm_handle_enqueue0()->
4156          * ldlm_lvbo_fill() will fill the reply buffer with lovea.
4157          */
4158         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
4159                              layout_size);
4160         rc = req_capsule_server_pack(info->mti_pill);
4161         if (rc)
4162                 GOTO(out, rc);
4163
4164         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4165         if (!ldlm_rep)
4166                 GOTO(out, rc = -EPROTO);
4167
4168         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
4169
4170         /* take lock in ldlm_lock_enqueue() for LAYOUT_INTENT_ACCESS */
4171         if (layout.mlc_opc == MD_LAYOUT_NOP)
4172                 GOTO(out, rc = 0);
4173
4174         rc = mdt_check_resent(info, mdt_reconstruct_generic, lhc);
4175         if (rc < 0)
4176                 GOTO(out, rc);
4177         if (rc == 1) {
4178                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
4179                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
4180                 GOTO(out, rc);
4181         }
4182
4183         buf->lb_buf = NULL;
4184         buf->lb_len = 0;
4185         if (unlikely(req_is_replay(mdt_info_req(info)))) {
4186                 buf->lb_buf = req_capsule_client_get(info->mti_pill,
4187                                                      &RMF_EADATA);
4188                 buf->lb_len = req_capsule_get_size(info->mti_pill,
4189                                                      &RMF_EADATA, RCL_CLIENT);
4190                 /*
4191                  * If it's a replay of layout write intent RPC, the client has
4192                  * saved the extended lovea when it get reply then.
4193                  */
4194                 if (buf->lb_len > 0)
4195                         mdt_fix_lov_magic(info, buf->lb_buf);
4196         }
4197
4198         /* Get lock from request for possible resent case. */
4199         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4200         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
4201
4202         /*
4203          * Instantiate some layout components, if @buf contains lovea, then it's
4204          * a replay of the layout intent write RPC.
4205          */
4206         rc = mdt_layout_change(info, obj, lhc, &layout);
4207         ldlm_rep->lock_policy_res2 = clear_serious(rc);
4208
4209         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
4210                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4211                 if (rc == ELDLM_LOCK_REPLACED &&
4212                     (*lockp)->l_granted_mode == LCK_EX)
4213                         ldlm_lock_mode_downgrade(*lockp, LCK_CR);
4214         }
4215
4216         EXIT;
4217 out:
4218         mdt_object_put(info->mti_env, obj);
4219         return rc;
4220 }
4221
4222 static int mdt_intent_open(enum ldlm_intent_flags it_opc,
4223                            struct mdt_thread_info *info,
4224                            struct ldlm_lock **lockp,
4225                            __u64 flags)
4226 {
4227         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4228         struct ldlm_reply      *rep = NULL;
4229         long                    opc;
4230         int                     rc;
4231
4232         static const struct req_format *intent_fmts[REINT_MAX] = {
4233                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
4234                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
4235         };
4236
4237         ENTRY;
4238
4239         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
4240         if (opc < 0)
4241                 RETURN(opc);
4242
4243         /* Get lock from request for possible resent case. */
4244         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4245
4246         rc = mdt_reint_internal(info, lhc, opc);
4247
4248         /* Check whether the reply has been packed successfully. */
4249         if (mdt_info_req(info)->rq_repmsg != NULL)
4250                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4251         if (rep == NULL) {
4252                 if (is_serious(rc))
4253                         RETURN(rc);
4254                 else
4255                         RETURN(err_serious(-EFAULT));
4256         }
4257
4258         /* MDC expects this in any case */
4259         if (rc != 0)
4260                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
4261
4262         /* the open lock or the lock for cross-ref object should be
4263          * returned to the client */
4264         if (lustre_handle_is_used(&lhc->mlh_reg_lh) &&
4265             (rc == 0 || rc == -MDT_EREMOTE_OPEN)) {
4266                 rep->lock_policy_res2 = 0;
4267                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4268                 RETURN(rc);
4269         }
4270
4271         rep->lock_policy_res2 = clear_serious(rc);
4272
4273         if (rep->lock_policy_res2 == -ENOENT &&
4274             mdt_get_disposition(rep, DISP_LOOKUP_NEG) &&
4275             !mdt_get_disposition(rep, DISP_OPEN_CREATE))
4276                 rep->lock_policy_res2 = 0;
4277
4278         lhc->mlh_reg_lh.cookie = 0ull;
4279         if (rc == -ENOTCONN || rc == -ENODEV ||
4280             rc == -EOVERFLOW) { /**< if VBR failure then return error */
4281                 /*
4282                  * If it is the disconnect error (ENODEV & ENOCONN), the error
4283                  * will be returned by rq_status, and client at ptlrpc layer
4284                  * will detect this, then disconnect, reconnect the import
4285                  * immediately, instead of impacting the following the rpc.
4286                  */
4287                 RETURN(rc);
4288         }
4289         /*
4290          * For other cases, the error will be returned by intent, and client
4291          * will retrieve the result from intent.
4292          */
4293         RETURN(ELDLM_LOCK_ABORTED);
4294 }
4295
4296 static int mdt_intent_opc(enum ldlm_intent_flags it_opc,
4297                           struct mdt_thread_info *info,
4298                           struct ldlm_lock **lockp,
4299                           u64 flags /* LDLM_FL_* */)
4300 {
4301         struct req_capsule *pill = info->mti_pill;
4302         struct ptlrpc_request *req = mdt_info_req(info);
4303         const struct req_format *it_format;
4304         int (*it_handler)(enum ldlm_intent_flags,
4305                           struct mdt_thread_info *,
4306                           struct ldlm_lock **,
4307                           u64);
4308         enum tgt_handler_flags it_handler_flags = 0;
4309         struct ldlm_reply *rep;
4310         bool check_mdt_object = false;
4311         int rc;
4312         ENTRY;
4313
4314         switch (it_opc) {
4315         case IT_OPEN:
4316         case IT_OPEN|IT_CREAT:
4317                 /*
4318                  * OCREAT is not a IS_MUTABLE request since the file may
4319                  * already exist. We do the extra check of
4320                  * OBD_CONNECT_RDONLY in mdt_reint_open() when we
4321                  * really need to create the object.
4322                  */
4323                 it_format = &RQF_LDLM_INTENT;
4324                 it_handler = &mdt_intent_open;
4325                 break;
4326         case IT_GETATTR:
4327                 check_mdt_object = true;
4328                 /* fallthrough */
4329         case IT_LOOKUP:
4330                 it_format = &RQF_LDLM_INTENT_GETATTR;
4331                 it_handler = &mdt_intent_getattr;
4332                 it_handler_flags = HAS_REPLY;
4333                 break;
4334         case IT_GETXATTR:
4335                 check_mdt_object = true;
4336                 it_format = &RQF_LDLM_INTENT_GETXATTR;
4337                 it_handler = &mdt_intent_getxattr;
4338                 it_handler_flags = HAS_BODY;
4339                 break;
4340         case IT_LAYOUT:
4341                 it_format = &RQF_LDLM_INTENT_LAYOUT;
4342                 it_handler = &mdt_intent_layout;
4343                 break;
4344         case IT_GLIMPSE:
4345                 it_format = &RQF_LDLM_INTENT;
4346                 it_handler = &mdt_intent_glimpse;
4347                 break;
4348         case IT_BRW:
4349                 it_format = &RQF_LDLM_INTENT;
4350                 it_handler = &mdt_intent_brw;
4351                 break;
4352         case IT_QUOTA_DQACQ:
4353         case IT_QUOTA_CONN: {
4354                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
4355
4356                 if (qmt == NULL)
4357                         RETURN(-EOPNOTSUPP);
4358
4359                 if (mdt_rdonly(req->rq_export))
4360                         RETURN(-EROFS);
4361
4362                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
4363                 /* pass the request to quota master */
4364                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
4365                                                  mdt_info_req(info), lockp,
4366                                                  flags);
4367                 RETURN(rc);
4368         }
4369         default:
4370                 CERROR("%s: unknown intent code %#x\n",
4371                        mdt_obd_name(info->mti_mdt), it_opc);
4372                 RETURN(-EPROTO);
4373         }
4374
4375         req_capsule_extend(pill, it_format);
4376
4377         rc = mdt_unpack_req_pack_rep(info, it_handler_flags);
4378         if (rc < 0)
4379                 RETURN(rc);
4380
4381         if (unlikely(info->mti_object == NULL && check_mdt_object))
4382                 RETURN(-EPROTO);
4383
4384         if (it_handler_flags & IS_MUTABLE && mdt_rdonly(req->rq_export))
4385                 RETURN(-EROFS);
4386
4387         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_INTENT_DELAY, 10);
4388
4389         /* execute policy */
4390         rc = (*it_handler)(it_opc, info, lockp, flags);
4391
4392         /* Check whether the reply has been packed successfully. */
4393         if (req->rq_repmsg != NULL) {
4394                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4395                 rep->lock_policy_res2 =
4396                         ptlrpc_status_hton(rep->lock_policy_res2);
4397         }
4398
4399         RETURN(rc);
4400 }
4401
4402 static void mdt_ptlrpc_stats_update(struct ptlrpc_request *req,
4403                                     enum ldlm_intent_flags it_opc)
4404 {
4405         struct lprocfs_stats *srv_stats = ptlrpc_req2svc(req)->srv_stats;
4406
4407         /* update stats when IT code is known */
4408         if (srv_stats != NULL)
4409                 lprocfs_counter_incr(srv_stats,
4410                                 PTLRPC_LAST_CNTR + (it_opc == IT_GLIMPSE ?
4411                                 LDLM_GLIMPSE_ENQUEUE : LDLM_IBITS_ENQUEUE));
4412 }
4413
4414 static int mdt_intent_policy(const struct lu_env *env,
4415                              struct ldlm_namespace *ns,
4416                              struct ldlm_lock **lockp,
4417                              void *req_cookie,
4418                              enum ldlm_mode mode,
4419                              __u64 flags, void *data)
4420 {
4421         struct tgt_session_info *tsi;
4422         struct mdt_thread_info  *info;
4423         struct ptlrpc_request   *req  =  req_cookie;
4424         struct ldlm_intent      *it;
4425         struct req_capsule      *pill;
4426         const struct ldlm_lock_desc *ldesc;
4427         int rc;
4428
4429         ENTRY;
4430
4431         LASSERT(req != NULL);
4432
4433         tsi = tgt_ses_info(env);
4434
4435         info = tsi2mdt_info(tsi);
4436         LASSERT(info != NULL);
4437         pill = info->mti_pill;
4438         LASSERT(pill->rc_req == req);
4439         ldesc = &info->mti_dlm_req->lock_desc;
4440
4441         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
4442                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
4443                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
4444                 if (it != NULL) {
4445                         mdt_ptlrpc_stats_update(req, it->opc);
4446                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
4447                         if (rc == 0)
4448                                 rc = ELDLM_OK;
4449
4450                         /* Lock without inodebits makes no sense and will oops
4451                          * later in ldlm. Let's check it now to see if we have
4452                          * ibits corrupted somewhere in mdt_intent_opc().
4453                          * The case for client miss to set ibits has been
4454                          * processed by others. */
4455                         LASSERT(ergo(ldesc->l_resource.lr_type == LDLM_IBITS,
4456                                 ldesc->l_policy_data.l_inodebits.bits != 0));
4457                 } else {
4458                         rc = err_serious(-EFAULT);
4459                 }
4460         } else {
4461                 /* No intent was provided */
4462                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
4463                 rc = req_capsule_server_pack(pill);
4464                 if (rc)
4465                         rc = err_serious(rc);
4466         }
4467         mdt_thread_info_fini(info);
4468         RETURN(rc);
4469 }
4470
4471 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
4472 {
4473         struct seq_server_site  *ss = mdt_seq_site(mdt);
4474
4475         if (ss->ss_node_id == 0)
4476                 return;
4477
4478         if (ss->ss_client_seq != NULL) {
4479                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
4480                 ss->ss_client_seq->lcs_exp = NULL;
4481         }
4482
4483         if (ss->ss_server_fld != NULL) {
4484                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
4485                 ss->ss_server_fld->lsf_control_exp = NULL;
4486         }
4487 }
4488
4489 static void mdt_seq_fini_cli(struct mdt_device *mdt)
4490 {
4491         struct seq_server_site *ss = mdt_seq_site(mdt);
4492
4493         if (ss == NULL)
4494                 return;
4495
4496         if (ss->ss_server_seq != NULL)
4497                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
4498 }
4499
4500 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
4501 {
4502         mdt_seq_fini_cli(mdt);
4503         mdt_deregister_seq_exp(mdt);
4504
4505         return seq_site_fini(env, mdt_seq_site(mdt));
4506 }
4507
4508 /**
4509  * It will retrieve its FLDB entries from MDT0, and it only happens
4510  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
4511  * and it needs to refresh FLDB from the MDT0.
4512  **/
4513 static int mdt_register_lwp_callback(void *data)
4514 {
4515         struct lu_env           env;
4516         struct mdt_device       *mdt = data;
4517         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
4518         int                     rc;
4519         ENTRY;
4520
4521         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
4522
4523         rc = lu_env_init(&env, LCT_MD_THREAD);
4524         if (rc < 0) {
4525                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
4526                 RETURN(rc);
4527         }
4528
4529         /* Allocate new sequence now to avoid creating local transaction
4530          * in the normal transaction process */
4531         rc = seq_server_check_and_alloc_super(&env,
4532                                               mdt_seq_site(mdt)->ss_server_seq);
4533         if (rc < 0)
4534                 GOTO(out, rc);
4535
4536         if (fld->lsf_new) {
4537                 rc = fld_update_from_controller(&env, fld);
4538                 if (rc != 0) {
4539                         CERROR("%s: cannot update controller: rc = %d\n",
4540                                mdt_obd_name(mdt), rc);
4541                         GOTO(out, rc);
4542                 }
4543         }
4544 out:
4545         lu_env_fini(&env);
4546         RETURN(rc);
4547 }
4548
4549 static int mdt_register_seq_exp(struct mdt_device *mdt)
4550 {
4551         struct seq_server_site  *ss = mdt_seq_site(mdt);
4552         char                    *lwp_name = NULL;
4553         int                     rc;
4554
4555         if (ss->ss_node_id == 0)
4556                 return 0;
4557
4558         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
4559         if (lwp_name == NULL)
4560                 GOTO(out_free, rc = -ENOMEM);
4561
4562         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
4563         if (rc != 0)
4564                 GOTO(out_free, rc);
4565
4566         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
4567                                       NULL, NULL);
4568         if (rc != 0)
4569                 GOTO(out_free, rc);
4570
4571         rc = lustre_register_lwp_item(lwp_name,
4572                                       &ss->ss_server_fld->lsf_control_exp,
4573                                       mdt_register_lwp_callback, mdt);
4574         if (rc != 0) {
4575                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
4576                 ss->ss_client_seq->lcs_exp = NULL;
4577                 GOTO(out_free, rc);
4578         }
4579 out_free:
4580         if (lwp_name != NULL)
4581                 OBD_FREE(lwp_name, MAX_OBD_NAME);
4582
4583         return rc;
4584 }
4585
4586 /*
4587  * Init client sequence manager which is used by local MDS to talk to sequence
4588  * controller on remote node.
4589  */
4590 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
4591 {
4592         struct seq_server_site *ss = mdt_seq_site(mdt);
4593         char *prefix;
4594         ENTRY;
4595
4596         /* check if this is adding the first MDC and controller is not yet
4597          * initialized. */
4598         OBD_ALLOC_PTR(ss->ss_client_seq);
4599         if (ss->ss_client_seq == NULL)
4600                 RETURN(-ENOMEM);
4601
4602         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
4603         if (prefix == NULL) {
4604                 OBD_FREE_PTR(ss->ss_client_seq);
4605                 ss->ss_client_seq = NULL;
4606                 RETURN(-ENOMEM);
4607         }
4608
4609         /* Note: seq_client_fini will be called in seq_site_fini */
4610         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
4611         seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
4612                         prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
4613                                                             NULL);
4614         OBD_FREE(prefix, MAX_OBD_NAME + 5);
4615
4616         RETURN(seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq));
4617 }
4618
4619 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
4620 {
4621         struct seq_server_site  *ss;
4622         int                     rc;
4623         ENTRY;
4624
4625         ss = mdt_seq_site(mdt);
4626         /* init sequence controller server(MDT0) */
4627         if (ss->ss_node_id == 0) {
4628                 OBD_ALLOC_PTR(ss->ss_control_seq);
4629                 if (ss->ss_control_seq == NULL)
4630                         RETURN(-ENOMEM);
4631
4632                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
4633                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
4634                                      ss);
4635                 if (rc)
4636                         GOTO(out_seq_fini, rc);
4637         }
4638
4639         /* Init normal sequence server */
4640         OBD_ALLOC_PTR(ss->ss_server_seq);
4641         if (ss->ss_server_seq == NULL)
4642                 GOTO(out_seq_fini, rc = -ENOMEM);
4643
4644         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
4645                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
4646         if (rc)
4647                 GOTO(out_seq_fini, rc);
4648
4649         /* init seq client for seq server to talk to seq controller(MDT0) */
4650         rc = mdt_seq_init_cli(env, mdt);
4651         if (rc != 0)
4652                 GOTO(out_seq_fini, rc);
4653
4654         if (ss->ss_node_id != 0)
4655                 /* register controller export through lwp */
4656                 rc = mdt_register_seq_exp(mdt);
4657
4658         EXIT;
4659 out_seq_fini:
4660         if (rc)
4661                 mdt_seq_fini(env, mdt);
4662
4663         return rc;
4664 }
4665
4666 /*
4667  * FLD wrappers
4668  */
4669 static int mdt_fld_fini(const struct lu_env *env,
4670                         struct mdt_device *m)
4671 {
4672         struct seq_server_site *ss = mdt_seq_site(m);
4673         ENTRY;
4674
4675         if (ss && ss->ss_server_fld) {
4676                 fld_server_fini(env, ss->ss_server_fld);
4677                 OBD_FREE_PTR(ss->ss_server_fld);
4678                 ss->ss_server_fld = NULL;
4679         }
4680
4681         RETURN(0);
4682 }
4683
4684 static int mdt_fld_init(const struct lu_env *env,
4685                         const char *uuid,
4686                         struct mdt_device *m)
4687 {
4688         struct seq_server_site *ss;
4689         int rc;
4690         ENTRY;
4691
4692         ss = mdt_seq_site(m);
4693
4694         OBD_ALLOC_PTR(ss->ss_server_fld);
4695         if (ss->ss_server_fld == NULL)
4696                 RETURN(rc = -ENOMEM);
4697
4698         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
4699                              LU_SEQ_RANGE_MDT);
4700         if (rc) {
4701                 OBD_FREE_PTR(ss->ss_server_fld);
4702                 ss->ss_server_fld = NULL;
4703                 RETURN(rc);
4704         }
4705
4706         RETURN(0);
4707 }
4708
4709 static void mdt_stack_pre_fini(const struct lu_env *env,
4710                            struct mdt_device *m, struct lu_device *top)
4711 {
4712         struct lustre_cfg_bufs  *bufs;
4713         struct lustre_cfg       *lcfg;
4714         struct mdt_thread_info  *info;
4715         ENTRY;
4716
4717         LASSERT(top);
4718
4719         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4720         LASSERT(info != NULL);
4721
4722         bufs = &info->mti_u.bufs;
4723
4724         LASSERT(m->mdt_child_exp);
4725         LASSERT(m->mdt_child_exp->exp_obd);
4726
4727         /* process cleanup, pass mdt obd name to get obd umount flags */
4728         /* XXX: this is needed because all layers are referenced by
4729          * objects (some of them are pinned by osd, for example *
4730          * the proper solution should be a model where object used
4731          * by osd only doesn't have mdt/mdd slices -bzzz */
4732         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4733         lustre_cfg_bufs_set_string(bufs, 1, NULL);
4734         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4735         if (!lcfg)
4736                 RETURN_EXIT;
4737         lustre_cfg_init(lcfg, LCFG_PRE_CLEANUP, bufs);
4738
4739         top->ld_ops->ldo_process_config(env, top, lcfg);
4740         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4741         EXIT;
4742 }
4743
4744 static void mdt_stack_fini(const struct lu_env *env,
4745                            struct mdt_device *m, struct lu_device *top)
4746 {
4747         struct obd_device       *obd = mdt2obd_dev(m);
4748         struct lustre_cfg_bufs  *bufs;
4749         struct lustre_cfg       *lcfg;
4750         struct mdt_thread_info  *info;
4751         char                     flags[3] = "";
4752         ENTRY;
4753
4754         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4755         LASSERT(info != NULL);
4756
4757         lu_dev_del_linkage(top->ld_site, top);
4758
4759         lu_site_purge(env, top->ld_site, -1);
4760
4761         bufs = &info->mti_u.bufs;
4762         /* process cleanup, pass mdt obd name to get obd umount flags */
4763         /* another purpose is to let all layers to release their objects */
4764         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4765         if (obd->obd_force)
4766                 strcat(flags, "F");
4767         if (obd->obd_fail)
4768                 strcat(flags, "A");
4769         lustre_cfg_bufs_set_string(bufs, 1, flags);
4770         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4771         if (!lcfg)
4772                 RETURN_EXIT;
4773         lustre_cfg_init(lcfg, LCFG_CLEANUP, bufs);
4774
4775         LASSERT(top);
4776         top->ld_ops->ldo_process_config(env, top, lcfg);
4777         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4778
4779         lu_site_purge(env, top->ld_site, -1);
4780
4781         m->mdt_child = NULL;
4782         m->mdt_bottom = NULL;
4783
4784         obd_disconnect(m->mdt_child_exp);
4785         m->mdt_child_exp = NULL;
4786
4787         obd_disconnect(m->mdt_bottom_exp);
4788         m->mdt_child_exp = NULL;
4789 }
4790
4791 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
4792                                const char *next, struct obd_export **exp)
4793 {
4794         struct obd_connect_data *data = NULL;
4795         struct obd_device       *obd;
4796         int                      rc;
4797         ENTRY;
4798
4799         OBD_ALLOC_PTR(data);
4800         if (data == NULL)
4801                 GOTO(out, rc = -ENOMEM);
4802
4803         obd = class_name2obd(next);
4804         if (obd == NULL) {
4805                 CERROR("%s: can't locate next device: %s\n",
4806                        mdt_obd_name(m), next);
4807                 GOTO(out, rc = -ENOTCONN);
4808         }
4809
4810         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4811         data->ocd_version = LUSTRE_VERSION_CODE;
4812
4813         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
4814         if (rc) {
4815                 CERROR("%s: cannot connect to next dev %s (%d)\n",
4816                        mdt_obd_name(m), next, rc);
4817                 GOTO(out, rc);
4818         }
4819
4820 out:
4821         if (data)
4822                 OBD_FREE_PTR(data);
4823         RETURN(rc);
4824 }
4825
4826 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
4827                           struct lustre_cfg *cfg)
4828 {
4829         char                   *dev = lustre_cfg_string(cfg, 0);
4830         int                     rc, name_size, uuid_size;
4831         char                   *name, *uuid, *p;
4832         struct lustre_cfg_bufs *bufs;
4833         struct lustre_cfg      *lcfg;
4834         struct obd_device      *obd;
4835         struct lustre_profile  *lprof;
4836         struct lu_site         *site;
4837         ENTRY;
4838
4839         /* in 1.8 we had the only device in the stack - MDS.
4840          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
4841          * in 2.3 OSD is instantiated by obd_mount.c, so we need
4842          * to generate names and setup MDT, MDD. MDT will be using
4843          * generated name to connect to MDD. for MDD the next device
4844          * will be LOD with name taken from so called "profile" which
4845          * is generated by mount_option line
4846          *
4847          * 1.8 MGS generates config. commands like this:
4848          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
4849          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
4850          * 2.0 MGS generates config. commands like this:
4851          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
4852          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4853          *                    3:lustre-MDT0000-mdtlov  4:f
4854          *
4855          * we generate MDD name from MDT one, just replacing T with D
4856          *
4857          * after all the preparations, the logical equivalent will be
4858          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
4859          *                    3:lustre-MDT0000-mdtlov  4:f
4860          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4861          *                    3:lustre-MDD0000  4:f
4862          *
4863          *  notice we build the stack from down to top: MDD first, then MDT */
4864
4865         name_size = MAX_OBD_NAME;
4866         uuid_size = MAX_OBD_NAME;
4867
4868         OBD_ALLOC(name, name_size);
4869         OBD_ALLOC(uuid, uuid_size);
4870         if (name == NULL || uuid == NULL)
4871                 GOTO(cleanup_mem, rc = -ENOMEM);
4872
4873         OBD_ALLOC_PTR(bufs);
4874         if (!bufs)
4875                 GOTO(cleanup_mem, rc = -ENOMEM);
4876
4877         strcpy(name, dev);
4878         p = strstr(name, "-MDT");
4879         if (p == NULL)
4880                 GOTO(free_bufs, rc = -ENOMEM);
4881         p[3] = 'D';
4882
4883         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
4884
4885         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4886         if (lprof == NULL || lprof->lp_dt == NULL) {
4887                 CERROR("can't find the profile: %s\n",
4888                        lustre_cfg_string(cfg, 0));
4889                 GOTO(free_bufs, rc = -EINVAL);
4890         }
4891
4892         lustre_cfg_bufs_reset(bufs, name);
4893         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
4894         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4895         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4896
4897         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4898         if (!lcfg)
4899                 GOTO(put_profile, rc = -ENOMEM);
4900         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
4901
4902         rc = class_attach(lcfg);
4903         if (rc)
4904                 GOTO(lcfg_cleanup, rc);
4905
4906         obd = class_name2obd(name);
4907         if (!obd) {
4908                 CERROR("Can not find obd %s (%s in config)\n",
4909                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
4910                 GOTO(lcfg_cleanup, rc = -EINVAL);
4911         }
4912
4913         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4914
4915         lustre_cfg_bufs_reset(bufs, name);
4916         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4917         lustre_cfg_bufs_set_string(bufs, 2, dev);
4918         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4919
4920         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4921         if (!lcfg)
4922                 GOTO(class_detach, rc = -ENOMEM);
4923         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
4924
4925         rc = class_setup(obd, lcfg);
4926         if (rc)
4927                 GOTO(class_detach, rc);
4928
4929         /* connect to MDD we just setup */
4930         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
4931         if (rc)
4932                 GOTO(class_detach, rc);
4933
4934         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
4935         LASSERT(site);
4936         LASSERT(mdt_lu_site(mdt) == NULL);
4937         mdt->mdt_lu_dev.ld_site = site;
4938         site->ls_top_dev = &mdt->mdt_lu_dev;
4939         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
4940
4941         /* now connect to bottom OSD */
4942         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
4943         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
4944         if (rc)
4945                 GOTO(class_detach, rc);
4946         mdt->mdt_bottom =
4947                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
4948
4949         rc = lu_env_refill((struct lu_env *)env);
4950         if (rc != 0)
4951                 CERROR("Failure to refill session: '%d'\n", rc);
4952
4953         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
4954
4955         EXIT;
4956 class_detach:
4957         if (rc)
4958                 class_detach(obd, lcfg);
4959 lcfg_cleanup:
4960         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4961 put_profile:
4962         class_put_profile(lprof);
4963 free_bufs:
4964         OBD_FREE_PTR(bufs);
4965 cleanup_mem:
4966         if (name)
4967                 OBD_FREE(name, name_size);
4968         if (uuid)
4969                 OBD_FREE(uuid, uuid_size);
4970         RETURN(rc);
4971 }
4972
4973 /* setup quota master target on MDT0 */
4974 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
4975                           struct lustre_cfg *cfg)
4976 {
4977         struct obd_device       *obd;
4978         char                    *dev = lustre_cfg_string(cfg, 0);
4979         char                    *qmtname, *uuid, *p;
4980         struct lustre_cfg_bufs  *bufs;
4981         struct lustre_cfg       *lcfg;
4982         struct lustre_profile   *lprof;
4983         struct obd_connect_data *data;
4984         int                      rc;
4985         ENTRY;
4986
4987         LASSERT(mdt->mdt_qmt_exp == NULL);
4988         LASSERT(mdt->mdt_qmt_dev == NULL);
4989
4990         /* quota master is on MDT0 only for now */
4991         if (mdt->mdt_seq_site.ss_node_id != 0)
4992                 RETURN(0);
4993
4994         /* MGS generates config commands which look as follows:
4995          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4996          *                    3:lustre-MDT0000-mdtlov  4:f
4997          *
4998          * We generate the QMT name from the MDT one, just replacing MD with QM
4999          * after all the preparations, the logical equivalent will be:
5000          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
5001          *                    3:lustre-MDT0000-osd  4:f */
5002         OBD_ALLOC(qmtname, MAX_OBD_NAME);
5003         OBD_ALLOC(uuid, UUID_MAX);
5004         OBD_ALLOC_PTR(bufs);
5005         OBD_ALLOC_PTR(data);
5006         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
5007                 GOTO(cleanup_mem, rc = -ENOMEM);
5008
5009         strcpy(qmtname, dev);
5010         p = strstr(qmtname, "-MDT");
5011         if (p == NULL)
5012                 GOTO(cleanup_mem, rc = -ENOMEM);
5013         /* replace MD with QM */
5014         p[1] = 'Q';
5015         p[2] = 'M';
5016
5017         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
5018
5019         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
5020         if (lprof == NULL || lprof->lp_dt == NULL) {
5021                 CERROR("can't find profile for %s\n",
5022                        lustre_cfg_string(cfg, 0));
5023                 GOTO(cleanup_mem, rc = -EINVAL);
5024         }
5025
5026         lustre_cfg_bufs_reset(bufs, qmtname);
5027         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
5028         lustre_cfg_bufs_set_string(bufs, 2, uuid);
5029         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
5030
5031         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5032         if (!lcfg)
5033                 GOTO(put_profile, rc = -ENOMEM);
5034         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
5035
5036         rc = class_attach(lcfg);
5037         if (rc)
5038                 GOTO(lcfg_cleanup, rc);
5039
5040         obd = class_name2obd(qmtname);
5041         if (!obd) {
5042                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
5043                        lustre_cfg_string(cfg, 0));
5044                 GOTO(lcfg_cleanup, rc = -EINVAL);
5045         }
5046
5047         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5048
5049         lustre_cfg_bufs_reset(bufs, qmtname);
5050         lustre_cfg_bufs_set_string(bufs, 1, uuid);
5051         lustre_cfg_bufs_set_string(bufs, 2, dev);
5052
5053         /* for quota, the next device should be the OSD device */
5054         lustre_cfg_bufs_set_string(bufs, 3,
5055                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
5056
5057         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5058         if (!lcfg)
5059                 GOTO(class_detach, rc = -ENOMEM);
5060         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
5061
5062         rc = class_setup(obd, lcfg);
5063         if (rc)
5064                 GOTO(class_detach, rc);
5065
5066         mdt->mdt_qmt_dev = obd->obd_lu_dev;
5067
5068         /* configure local quota objects */
5069         if (OBD_FAIL_CHECK(OBD_FAIL_QUOTA_INIT))
5070                 rc = -EBADF;
5071         else
5072                 rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
5073                                                            &mdt->mdt_lu_dev,
5074                                                            mdt->mdt_qmt_dev);
5075         if (rc)
5076                 GOTO(class_cleanup, rc);
5077
5078         /* connect to quota master target */
5079         data->ocd_connect_flags = OBD_CONNECT_VERSION;
5080         data->ocd_version = LUSTRE_VERSION_CODE;
5081         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
5082                          data, NULL);
5083         if (rc) {
5084                 CERROR("cannot connect to quota master device %s (%d)\n",
5085                        qmtname, rc);
5086                 GOTO(class_cleanup, rc);
5087         }
5088
5089         EXIT;
5090 class_cleanup:
5091         if (rc) {
5092                 class_manual_cleanup(obd);
5093                 mdt->mdt_qmt_dev = NULL;
5094                 GOTO(lcfg_cleanup, rc);
5095         }
5096 class_detach:
5097         if (rc)
5098                 class_detach(obd, lcfg);
5099 lcfg_cleanup:
5100         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5101 put_profile:
5102         class_put_profile(lprof);
5103 cleanup_mem:
5104         if (bufs)
5105                 OBD_FREE_PTR(bufs);
5106         if (qmtname)
5107                 OBD_FREE(qmtname, MAX_OBD_NAME);
5108         if (uuid)
5109                 OBD_FREE(uuid, UUID_MAX);
5110         if (data)
5111                 OBD_FREE_PTR(data);
5112         return rc;
5113 }
5114
5115 /* Shutdown quota master target associated with mdt */
5116 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
5117 {
5118         ENTRY;
5119
5120         if (mdt->mdt_qmt_exp == NULL)
5121                 RETURN_EXIT;
5122         LASSERT(mdt->mdt_qmt_dev != NULL);
5123
5124         /* the qmt automatically shuts down when the mdt disconnects */
5125         obd_disconnect(mdt->mdt_qmt_exp);
5126         mdt->mdt_qmt_exp = NULL;
5127         mdt->mdt_qmt_dev = NULL;
5128         EXIT;
5129 }
5130
5131 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
5132  * for now. This will be removed along with converting rest of MDT code
5133  * to use tgt_session_info */
5134 static int mdt_tgt_getxattr(struct tgt_session_info *tsi)
5135 {
5136         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
5137         int                      rc;
5138
5139         if (unlikely(info->mti_object == NULL))
5140                 return -EPROTO;
5141
5142         rc = mdt_getxattr(info);
5143
5144         mdt_thread_info_fini(info);
5145         return rc;
5146 }
5147
5148 #define OBD_FAIL_OST_READ_NET   OBD_FAIL_OST_BRW_NET
5149 #define OBD_FAIL_OST_WRITE_NET  OBD_FAIL_OST_BRW_NET
5150 #define OST_BRW_READ    OST_READ
5151 #define OST_BRW_WRITE   OST_WRITE
5152
5153 static struct tgt_handler mdt_tgt_handlers[] = {
5154 TGT_RPC_HANDLER(MDS_FIRST_OPC,
5155                 0,                      MDS_CONNECT,    mdt_tgt_connect,
5156                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
5157 TGT_RPC_HANDLER(MDS_FIRST_OPC,
5158                 0,                      MDS_DISCONNECT, tgt_disconnect,
5159                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
5160 TGT_RPC_HANDLER(MDS_FIRST_OPC,
5161                 HAS_REPLY,              MDS_SET_INFO,   mdt_set_info,
5162                 &RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION),
5163 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
5164 TGT_MDT_HDL(HAS_REPLY,          MDS_GET_ROOT,   mdt_get_root),
5165 TGT_MDT_HDL(HAS_BODY,           MDS_GETATTR,    mdt_getattr),
5166 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_GETATTR_NAME,
5167                                                         mdt_getattr_name),
5168 TGT_MDT_HDL(HAS_BODY,           MDS_GETXATTR,   mdt_tgt_getxattr),
5169 TGT_MDT_HDL(HAS_REPLY,          MDS_STATFS,     mdt_statfs),
5170 TGT_MDT_HDL(IS_MUTABLE,         MDS_REINT,      mdt_reint),
5171 TGT_MDT_HDL(HAS_BODY,           MDS_CLOSE,      mdt_close),
5172 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_READPAGE,   mdt_readpage),
5173 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_SYNC,       mdt_sync),
5174 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
5175 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_PROGRESS,
5176                                                         mdt_hsm_progress),
5177 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_CT_REGISTER,
5178                                                         mdt_hsm_ct_register),
5179 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_CT_UNREGISTER,
5180                                                         mdt_hsm_ct_unregister),
5181 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_STATE_GET,
5182                                                         mdt_hsm_state_get),
5183 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_STATE_SET,
5184                                                         mdt_hsm_state_set),
5185 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_ACTION,       mdt_hsm_action),
5186 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_REQUEST,
5187                                                         mdt_hsm_request),
5188 TGT_MDT_HDL(HAS_KEY | HAS_BODY | HAS_REPLY | IS_MUTABLE,
5189             MDS_SWAP_LAYOUTS,
5190             mdt_swap_layouts),
5191 TGT_MDT_HDL(IS_MUTABLE,         MDS_RMFID,      mdt_rmfid),
5192 };
5193
5194 static struct tgt_handler mdt_io_ops[] = {
5195 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY, OST_BRW_READ, tgt_brw_read,
5196                                                         mdt_hp_brw),
5197 TGT_OST_HDL_HP(HAS_BODY | IS_MUTABLE,    OST_BRW_WRITE, tgt_brw_write,
5198                                                         mdt_hp_brw),
5199 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY | IS_MUTABLE,
5200                                          OST_PUNCH,     mdt_punch_hdl,
5201                                                         mdt_hp_punch),
5202 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SYNC,     mdt_data_sync),
5203 };
5204
5205 static struct tgt_handler mdt_sec_ctx_ops[] = {
5206 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
5207 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
5208 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
5209 };
5210
5211 static struct tgt_handler mdt_quota_ops[] = {
5212 TGT_QUOTA_HDL(HAS_REPLY,                QUOTA_DQACQ,      mdt_quota_dqacq),
5213 };
5214
5215 static struct tgt_opc_slice mdt_common_slice[] = {
5216         {
5217                 .tos_opc_start  = MDS_FIRST_OPC,
5218                 .tos_opc_end    = MDS_LAST_OPC,
5219                 .tos_hs         = mdt_tgt_handlers
5220         },
5221         {
5222                 .tos_opc_start  = OBD_FIRST_OPC,
5223                 .tos_opc_end    = OBD_LAST_OPC,
5224                 .tos_hs         = tgt_obd_handlers
5225         },
5226         {
5227                 .tos_opc_start  = LDLM_FIRST_OPC,
5228                 .tos_opc_end    = LDLM_LAST_OPC,
5229                 .tos_hs         = tgt_dlm_handlers
5230         },
5231         {
5232                 .tos_opc_start  = SEC_FIRST_OPC,
5233                 .tos_opc_end    = SEC_LAST_OPC,
5234                 .tos_hs         = mdt_sec_ctx_ops
5235         },
5236         {
5237                 .tos_opc_start  = OUT_UPDATE_FIRST_OPC,
5238                 .tos_opc_end    = OUT_UPDATE_LAST_OPC,
5239                 .tos_hs         = tgt_out_handlers
5240         },
5241         {
5242                 .tos_opc_start  = FLD_FIRST_OPC,
5243                 .tos_opc_end    = FLD_LAST_OPC,
5244                 .tos_hs         = fld_handlers
5245         },
5246         {
5247                 .tos_opc_start  = SEQ_FIRST_OPC,
5248                 .tos_opc_end    = SEQ_LAST_OPC,
5249                 .tos_hs         = seq_handlers
5250         },
5251         {
5252                 .tos_opc_start  = QUOTA_DQACQ,
5253                 .tos_opc_end    = QUOTA_LAST_OPC,
5254                 .tos_hs         = mdt_quota_ops
5255         },
5256         {
5257                 .tos_opc_start  = LLOG_FIRST_OPC,
5258                 .tos_opc_end    = LLOG_LAST_OPC,
5259                 .tos_hs         = tgt_llog_handlers
5260         },
5261         {
5262                 .tos_opc_start  = LFSCK_FIRST_OPC,
5263                 .tos_opc_end    = LFSCK_LAST_OPC,
5264                 .tos_hs         = tgt_lfsck_handlers
5265         },
5266         {
5267                 .tos_opc_start  = OST_FIRST_OPC,
5268                 .tos_opc_end    = OST_LAST_OPC,
5269                 .tos_hs         = mdt_io_ops
5270         },
5271         {
5272                 .tos_hs         = NULL
5273         }
5274 };
5275
5276 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
5277 {
5278         struct md_device *next = m->mdt_child;
5279         struct lu_device *d = &m->mdt_lu_dev;
5280         struct obd_device *obd = mdt2obd_dev(m);
5281         struct lfsck_stop stop;
5282
5283         ENTRY;
5284         stop.ls_status = LS_PAUSED;
5285         stop.ls_flags = 0;
5286         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
5287
5288         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
5289         ping_evictor_stop();
5290
5291         /* Remove the HSM /proc entry so the coordinator cannot be
5292          * restarted by a user while it's shutting down.
5293          */
5294         mdt_hsm_cdt_stop(m);
5295
5296         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
5297         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
5298
5299         if (m->mdt_namespace != NULL)
5300                 ldlm_namespace_free_prior(m->mdt_namespace, NULL,
5301                                           d->ld_obd->obd_force);
5302
5303         obd_exports_barrier(obd);
5304         obd_zombie_barrier();
5305
5306         mdt_quota_fini(env, m);
5307
5308         cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
5309
5310         /* Calling the cleanup functions in the same order as in the mdt_init0
5311          * error path
5312          */
5313         mdt_tunables_fini(m);
5314
5315         target_recovery_fini(obd);
5316         upcall_cache_cleanup(m->mdt_identity_cache);
5317         m->mdt_identity_cache = NULL;
5318
5319         mdt_fs_cleanup(env, m);
5320
5321         tgt_fini(env, &m->mdt_lut);
5322
5323         mdt_hsm_cdt_fini(m);
5324
5325         if (m->mdt_los != NULL) {
5326                 local_oid_storage_fini(env, m->mdt_los);
5327                 m->mdt_los = NULL;
5328         }
5329
5330         if (m->mdt_namespace != NULL) {
5331                 ldlm_namespace_free_post(m->mdt_namespace);
5332                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
5333         }
5334
5335         if (m->mdt_md_root != NULL) {
5336                 mdt_object_put(env, m->mdt_md_root);
5337                 m->mdt_md_root = NULL;
5338         }
5339
5340         mdt_seq_fini(env, m);
5341
5342         mdt_fld_fini(env, m);
5343
5344         /*
5345          * Finish the stack
5346          */
5347         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
5348
5349         LASSERT(atomic_read(&d->ld_ref) == 0);
5350
5351         server_put_mount(mdt_obd_name(m), true);
5352
5353         EXIT;
5354 }
5355
5356 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
5357
5358 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
5359                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
5360 {
5361         const struct dt_device_param *dt_conf;
5362         struct mdt_thread_info *info;
5363         struct obd_device *obd;
5364         const char *dev = lustre_cfg_string(cfg, 0);
5365         const char *num = lustre_cfg_string(cfg, 2);
5366         struct tg_grants_data *tgd = &m->mdt_lut.lut_tgd;
5367         struct lustre_mount_info *lmi = NULL;
5368         struct lustre_sb_info *lsi;
5369         struct lu_site *s;
5370         struct seq_server_site *ss_site;
5371         const char *identity_upcall = "NONE";
5372         struct md_device *next;
5373         struct lu_fid fid;
5374         int rc;
5375         long node_id;
5376         mntopt_t mntopts;
5377         ENTRY;
5378
5379         lu_device_init(&m->mdt_lu_dev, ldt);
5380         /*
5381          * Environment (env) might be missing mdt_thread_key values at that
5382          * point, if device is allocated when mdt_thread_key is in QUIESCENT
5383          * mode.
5384          *
5385          * Usually device allocation path doesn't use module key values, but
5386          * mdt has to do a lot of work here, so allocate key value.
5387          */
5388         rc = lu_env_refill((struct lu_env *)env);
5389         if (rc != 0)
5390                 RETURN(rc);
5391
5392         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5393         LASSERT(info != NULL);
5394
5395         obd = class_name2obd(dev);
5396         LASSERT(obd != NULL);
5397
5398         m->mdt_max_mdsize = MAX_MD_SIZE_OLD;
5399         m->mdt_opts.mo_evict_tgt_nids = 1;
5400         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
5401
5402         lmi = server_get_mount(dev);
5403         if (lmi == NULL) {
5404                 CERROR("Cannot get mount info for %s!\n", dev);
5405                 RETURN(-EFAULT);
5406         } else {
5407                 lsi = s2lsi(lmi->lmi_sb);
5408                 LASSERT(lsi->lsi_lmd);
5409                 /* CMD is supported only in IAM mode */
5410                 LASSERT(num);
5411                 node_id = simple_strtol(num, NULL, 10);
5412                 obd->u.obt.obt_magic = OBT_MAGIC;
5413                 if (lsi->lsi_lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
5414                         m->mdt_skip_lfsck = 1;
5415         }
5416
5417         /* DoM files get IO lock at open by default */
5418         m->mdt_opts.mo_dom_lock = ALWAYS_DOM_LOCK_ON_OPEN;
5419         /* DoM files are read at open and data is packed in the reply */
5420         m->mdt_opts.mo_dom_read_open = 1;
5421
5422         m->mdt_squash.rsi_uid = 0;
5423         m->mdt_squash.rsi_gid = 0;
5424         INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
5425         spin_lock_init(&m->mdt_squash.rsi_lock);
5426         spin_lock_init(&m->mdt_lock);
5427         m->mdt_enable_remote_dir = 1;
5428         m->mdt_enable_striped_dir = 1;
5429         m->mdt_enable_dir_migration = 1;
5430         m->mdt_enable_remote_dir_gid = 0;
5431         m->mdt_enable_chprojid_gid = 0;
5432         m->mdt_enable_remote_rename = 1;
5433
5434         atomic_set(&m->mdt_mds_mds_conns, 0);
5435         atomic_set(&m->mdt_async_commit_count, 0);
5436
5437         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
5438         m->mdt_lu_dev.ld_obd = obd;
5439         /* Set this lu_device to obd for error handling purposes. */
5440         obd->obd_lu_dev = &m->mdt_lu_dev;
5441
5442         /* init the stack */
5443         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
5444         if (rc) {
5445                 CERROR("%s: Can't init device stack, rc %d\n",
5446                        mdt_obd_name(m), rc);
5447                 GOTO(err_lmi, rc);
5448         }
5449
5450         s = mdt_lu_site(m);
5451         ss_site = mdt_seq_site(m);
5452         s->ld_seq_site = ss_site;
5453         ss_site->ss_lu = s;
5454
5455         /* set server index */
5456         ss_site->ss_node_id = node_id;
5457
5458         /* failover is the default
5459          * FIXME: we do not failout mds0/mgs, which may cause some problems.
5460          * assumed whose ss_node_id == 0 XXX
5461          * */
5462         obd->obd_replayable = 1;
5463         /* No connection accepted until configurations will finish */
5464         obd->obd_no_conn = 1;
5465
5466         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
5467                 char *str = lustre_cfg_string(cfg, 4);
5468                 if (strchr(str, 'n')) {
5469                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
5470                         obd->obd_replayable = 0;
5471                 }
5472         }
5473
5474         rc = mdt_fld_init(env, mdt_obd_name(m), m);
5475         if (rc)
5476                 GOTO(err_fini_stack, rc);
5477
5478         rc = mdt_seq_init(env, m);
5479         if (rc)
5480                 GOTO(err_fini_fld, rc);
5481
5482         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
5483                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
5484         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
5485                                               LDLM_NAMESPACE_SERVER,
5486                                               LDLM_NAMESPACE_GREEDY,
5487                                               LDLM_NS_TYPE_MDT);
5488         if (m->mdt_namespace == NULL)
5489                 GOTO(err_fini_seq, rc = -ENOMEM);
5490
5491         m->mdt_namespace->ns_lvbp = m;
5492         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
5493
5494         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
5495         /* set obd_namespace for compatibility with old code */
5496         obd->obd_namespace = m->mdt_namespace;
5497
5498         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
5499                       OBD_FAIL_MDS_ALL_REQUEST_NET,
5500                       OBD_FAIL_MDS_ALL_REPLY_NET);
5501         if (rc)
5502                 GOTO(err_free_ns, rc);
5503
5504         /* Amount of available space excluded from granting and reserved
5505          * for metadata. It is in percentage and 50% is default value. */
5506         tgd->tgd_reserved_pcnt = 50;
5507
5508         if (ONE_MB_BRW_SIZE < (1U << tgd->tgd_blockbits))
5509                 m->mdt_brw_size = 1U << tgd->tgd_blockbits;
5510         else
5511                 m->mdt_brw_size = ONE_MB_BRW_SIZE;
5512
5513         rc = mdt_fs_setup(env, m, obd, lsi);
5514         if (rc)
5515                 GOTO(err_tgt, rc);
5516
5517         fid.f_seq = FID_SEQ_LOCAL_NAME;
5518         fid.f_oid = 1;
5519         fid.f_ver = 0;
5520         rc = local_oid_storage_init(env, m->mdt_bottom, &fid, &m->mdt_los);
5521         if (rc != 0)
5522                 GOTO(err_fs_cleanup, rc);
5523
5524         rc = mdt_hsm_cdt_init(m);
5525         if (rc != 0) {
5526                 CERROR("%s: error initializing coordinator, rc %d\n",
5527                        mdt_obd_name(m), rc);
5528                 GOTO(err_los_fini, rc);
5529         }
5530
5531         tgt_adapt_sptlrpc_conf(&m->mdt_lut);
5532
5533         next = m->mdt_child;
5534         dt_conf = next->md_ops->mdo_dtconf_get(env, next);
5535
5536         mntopts = dt_conf->ddp_mntopts;
5537
5538         if (mntopts & MNTOPT_USERXATTR)
5539                 m->mdt_opts.mo_user_xattr = 1;
5540         else
5541                 m->mdt_opts.mo_user_xattr = 0;
5542
5543         m->mdt_max_ea_size = dt_conf->ddp_max_ea_size;
5544
5545         if (mntopts & MNTOPT_ACL)
5546                 m->mdt_opts.mo_acl = 1;
5547         else
5548                 m->mdt_opts.mo_acl = 0;
5549
5550         /* XXX: to support suppgid for ACL, we enable identity_upcall
5551          * by default, otherwise, maybe got unexpected -EACCESS. */
5552         if (m->mdt_opts.mo_acl)
5553                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
5554
5555         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
5556                                                 identity_upcall,
5557                                                 &mdt_identity_upcall_cache_ops);
5558         if (IS_ERR(m->mdt_identity_cache)) {
5559                 rc = PTR_ERR(m->mdt_identity_cache);
5560                 m->mdt_identity_cache = NULL;
5561                 GOTO(err_free_hsm, rc);
5562         }
5563
5564         rc = mdt_tunables_init(m, dev);
5565         if (rc) {
5566                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
5567                 GOTO(err_recovery, rc);
5568         }
5569
5570         rc = mdt_quota_init(env, m, cfg);
5571         if (rc)
5572                 GOTO(err_procfs, rc);
5573
5574         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
5575         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
5576                            "mdt_ldlm_client", m->mdt_ldlm_client);
5577
5578         ping_evictor_start();
5579
5580         /* recovery will be started upon mdt_prepare()
5581          * when the whole stack is complete and ready
5582          * to serve the requests */
5583
5584         /* Reduce the initial timeout on an MDS because it doesn't need such
5585          * a long timeout as an OST does. Adaptive timeouts will adjust this
5586          * value appropriately. */
5587         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
5588                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
5589
5590         if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_LOCAL_RECOV))
5591                 m->mdt_lut.lut_local_recovery = 1;
5592
5593         RETURN(0);
5594 err_procfs:
5595         mdt_tunables_fini(m);
5596 err_recovery:
5597         upcall_cache_cleanup(m->mdt_identity_cache);
5598         m->mdt_identity_cache = NULL;
5599 err_free_hsm:
5600         mdt_hsm_cdt_fini(m);
5601 err_los_fini:
5602         local_oid_storage_fini(env, m->mdt_los);
5603         m->mdt_los = NULL;
5604 err_fs_cleanup:
5605         mdt_fs_cleanup(env, m);
5606 err_tgt:
5607         /* keep recoverable clients */
5608         obd->obd_fail = 1;
5609         target_recovery_fini(obd);
5610         obd_exports_barrier(obd);
5611         obd_zombie_barrier();
5612         tgt_fini(env, &m->mdt_lut);
5613 err_free_ns:
5614         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
5615         obd->obd_namespace = m->mdt_namespace = NULL;
5616 err_fini_seq:
5617         mdt_seq_fini(env, m);
5618 err_fini_fld:
5619         mdt_fld_fini(env, m);
5620 err_fini_stack:
5621         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
5622 err_lmi:
5623         if (lmi)
5624                 server_put_mount(dev, true);
5625         return(rc);
5626 }
5627
5628 /* For interoperability, the left element is old parameter, the right one
5629  * is the new version of the parameter, if some parameter is deprecated,
5630  * the new version should be set as NULL. */
5631 static struct cfg_interop_param mdt_interop_param[] = {
5632         { "mdt.group_upcall",   NULL },
5633         { "mdt.quota_type",     NULL },
5634         { "mdd.quota_type",     NULL },
5635         { "mdt.som",            NULL },
5636         { "mdt.rootsquash",     "mdt.root_squash" },
5637         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
5638         { NULL }
5639 };
5640
5641 /* used by MGS to process specific configurations */
5642 static int mdt_process_config(const struct lu_env *env,
5643                               struct lu_device *d, struct lustre_cfg *cfg)
5644 {
5645         struct mdt_device *m = mdt_dev(d);
5646         struct md_device *md_next = m->mdt_child;
5647         struct lu_device *next = md2lu_dev(md_next);
5648         int rc;
5649         ENTRY;
5650
5651         switch (cfg->lcfg_command) {
5652         case LCFG_PARAM: {
5653                 struct obd_device *obd = d->ld_obd;
5654                 /* For interoperability */
5655                 struct cfg_interop_param *ptr = NULL;
5656                 struct lustre_cfg *old_cfg = NULL;
5657                 char *param = NULL;
5658                 ssize_t count;
5659
5660                 param = lustre_cfg_string(cfg, 1);
5661                 if (param == NULL) {
5662                         CERROR("param is empty\n");
5663                         rc = -EINVAL;
5664                         break;
5665                 }
5666
5667                 ptr = class_find_old_param(param, mdt_interop_param);
5668                 if (ptr != NULL) {
5669                         if (ptr->new_param == NULL) {
5670                                 rc = 0;
5671                                 CWARN("For interoperability, skip this %s."
5672                                       " It is obsolete.\n", ptr->old_param);
5673                                 break;
5674                         }
5675
5676                         CWARN("Found old param %s, changed it to %s.\n",
5677                               ptr->old_param, ptr->new_param);
5678
5679                         old_cfg = cfg;
5680                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
5681                         if (IS_ERR(cfg)) {
5682                                 rc = PTR_ERR(cfg);
5683                                 break;
5684                         }
5685                 }
5686
5687                 count = class_modify_config(cfg, PARAM_MDT,
5688                                             &obd->obd_kset.kobj);
5689                 if (count < 0) {
5690                         struct coordinator *cdt = &m->mdt_coordinator;
5691
5692                         /* is it an HSM var ? */
5693                         count = class_modify_config(cfg, PARAM_HSM,
5694                                                     &cdt->cdt_hsm_kobj);
5695                         if (count < 0)
5696                                 /* we don't understand; pass it on */
5697                                 rc = next->ld_ops->ldo_process_config(env, next,
5698                                                                       cfg);
5699                         else
5700                                 rc = count > 0 ? 0 : count;
5701                 } else {
5702                         rc = count > 0 ? 0 : count;
5703                 }
5704
5705                 if (old_cfg)
5706                         OBD_FREE(cfg, lustre_cfg_len(cfg->lcfg_bufcount,
5707                                                      cfg->lcfg_buflens));
5708                 break;
5709         }
5710         default:
5711                 /* others are passed further */
5712                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
5713                 break;
5714         }
5715         RETURN(rc);
5716 }
5717
5718 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
5719                                           const struct lu_object_header *hdr,
5720                                           struct lu_device *d)
5721 {
5722         struct mdt_object *mo;
5723
5724         ENTRY;
5725
5726         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
5727         if (mo != NULL) {
5728                 struct lu_object *o;
5729                 struct lu_object_header *h;
5730
5731                 o = &mo->mot_obj;
5732                 h = &mo->mot_header;
5733                 lu_object_header_init(h);
5734                 lu_object_init(o, h, d);
5735                 lu_object_add_top(h, o);
5736                 o->lo_ops = &mdt_obj_ops;
5737                 spin_lock_init(&mo->mot_write_lock);
5738                 mutex_init(&mo->mot_som_mutex);
5739                 mutex_init(&mo->mot_lov_mutex);
5740                 init_rwsem(&mo->mot_dom_sem);
5741                 init_rwsem(&mo->mot_open_sem);
5742                 atomic_set(&mo->mot_open_count, 0);
5743                 RETURN(o);
5744         }
5745         RETURN(NULL);
5746 }
5747
5748 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
5749                            const struct lu_object_conf *unused)
5750 {
5751         struct mdt_device *d = mdt_dev(o->lo_dev);
5752         struct lu_device  *under;
5753         struct lu_object  *below;
5754         int                rc = 0;
5755         ENTRY;
5756
5757         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
5758                PFID(lu_object_fid(o)));
5759
5760         under = &d->mdt_child->md_lu_dev;
5761         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
5762         if (below != NULL) {
5763                 lu_object_add(o, below);
5764         } else
5765                 rc = -ENOMEM;
5766
5767         RETURN(rc);
5768 }
5769
5770 static void mdt_object_free_rcu(struct rcu_head *head)
5771 {
5772         struct mdt_object *mo = container_of(head, struct mdt_object,
5773                                              mot_header.loh_rcu);
5774
5775         kmem_cache_free(mdt_object_kmem, mo);
5776 }
5777
5778 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
5779 {
5780         struct mdt_object *mo = mdt_obj(o);
5781         struct lu_object_header *h;
5782         ENTRY;
5783
5784         h = o->lo_header;
5785         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
5786                PFID(lu_object_fid(o)));
5787
5788         LASSERT(atomic_read(&mo->mot_open_count) == 0);
5789         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
5790
5791         lu_object_fini(o);
5792         lu_object_header_fini(h);
5793         OBD_FREE_PRE(mo, sizeof(*mo), "slab-freed");
5794         call_rcu(&mo->mot_header.loh_rcu, mdt_object_free_rcu);
5795
5796         EXIT;
5797 }
5798
5799 static int mdt_object_print(const struct lu_env *env, void *cookie,
5800                             lu_printer_t p, const struct lu_object *o)
5801 {
5802         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
5803
5804         return (*p)(env, cookie,
5805                     LUSTRE_MDT_NAME"-object@%p(%s %s, writecount=%d)",
5806                     mdto, mdto->mot_lov_created ? "lov_created" : "",
5807                     mdto->mot_cache_attr ? "cache_attr" : "",
5808                     mdto->mot_write_count);
5809 }
5810
5811 static int mdt_prepare(const struct lu_env *env,
5812                 struct lu_device *pdev,
5813                 struct lu_device *cdev)
5814 {
5815         struct mdt_device *mdt = mdt_dev(cdev);
5816         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
5817         struct obd_device *obd = cdev->ld_obd;
5818         int rc;
5819
5820         ENTRY;
5821
5822         LASSERT(obd);
5823
5824         rc = next->ld_ops->ldo_prepare(env, cdev, next);
5825         if (rc)
5826                 RETURN(rc);
5827
5828         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
5829         if (rc)
5830                 RETURN(rc);
5831
5832         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
5833         if (rc)
5834                 RETURN(rc);
5835
5836         rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
5837         /* The LFSCK instance is registered just now, so it must be there when
5838          * register the namespace to such instance. */
5839         LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
5840
5841         if (mdt->mdt_seq_site.ss_node_id == 0) {
5842                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
5843                                                          &mdt->mdt_md_root_fid);
5844                 if (rc)
5845                         RETURN(rc);
5846         }
5847
5848         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
5849
5850         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
5851         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
5852         LASSERT(obd->obd_no_conn);
5853         spin_lock(&obd->obd_dev_lock);
5854         obd->obd_no_conn = 0;
5855         spin_unlock(&obd->obd_dev_lock);
5856
5857         if (obd->obd_recovering == 0)
5858                 mdt_postrecov(env, mdt);
5859
5860         RETURN(rc);
5861 }
5862
5863 const struct lu_device_operations mdt_lu_ops = {
5864         .ldo_object_alloc   = mdt_object_alloc,
5865         .ldo_process_config = mdt_process_config,
5866         .ldo_prepare        = mdt_prepare,
5867 };
5868
5869 static const struct lu_object_operations mdt_obj_ops = {
5870         .loo_object_init    = mdt_object_init,
5871         .loo_object_free    = mdt_object_free,
5872         .loo_object_print   = mdt_object_print
5873 };
5874
5875 static int mdt_obd_set_info_async(const struct lu_env *env,
5876                                   struct obd_export *exp,
5877                                   __u32 keylen, void *key,
5878                                   __u32 vallen, void *val,
5879                                   struct ptlrpc_request_set *set)
5880 {
5881         int rc;
5882
5883         ENTRY;
5884
5885         if (KEY_IS(KEY_SPTLRPC_CONF)) {
5886                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp));
5887                 RETURN(rc);
5888         }
5889
5890         RETURN(0);
5891 }
5892
5893 /**
5894  * Match client and server connection feature flags.
5895  *
5896  * Compute the compatibility flags for a connection request based on
5897  * features mutually supported by client and server.
5898  *
5899  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
5900  * must not be updated here, otherwise a partially initialized value may
5901  * be exposed. After the connection request is successfully processed,
5902  * the top-level MDT connect request handler atomically updates the export
5903  * connect flags from the obd_connect_data::ocd_connect_flags field of the
5904  * reply. \see mdt_connect().
5905  *
5906  * Before 2.7.50 clients will send a struct obd_connect_data_v1 rather than a
5907  * full struct obd_connect_data. So care must be taken when accessing fields
5908  * that are not present in struct obd_connect_data_v1. See LU-16.
5909  *
5910  * \param exp   the obd_export associated with this client/target pair
5911  * \param mdt   the target device for the connection
5912  * \param data  stores data for this connect request
5913  *
5914  * \retval 0       success
5915  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
5916  * \retval -EBADE  client and server feature requirements are incompatible
5917  */
5918 static int mdt_connect_internal(const struct lu_env *env,
5919                                 struct obd_export *exp,
5920                                 struct mdt_device *mdt,
5921                                 struct obd_connect_data *data, bool reconnect)
5922 {
5923         const char *obd_name = mdt_obd_name(mdt);
5924         LASSERT(data != NULL);
5925
5926         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
5927
5928         if (mdt->mdt_bottom->dd_rdonly &&
5929             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5930             !(data->ocd_connect_flags & OBD_CONNECT_RDONLY))
5931                 RETURN(-EACCES);
5932
5933         if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
5934                 data->ocd_connect_flags2 &= MDT_CONNECT_SUPPORTED2;
5935
5936         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
5937
5938         if (!mdt->mdt_opts.mo_acl)
5939                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
5940
5941         if (!mdt->mdt_opts.mo_user_xattr)
5942                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
5943
5944         if (OCD_HAS_FLAG(data, BRW_SIZE)) {
5945                 data->ocd_brw_size = min(data->ocd_brw_size,
5946                                          mdt->mdt_brw_size);
5947                 if (data->ocd_brw_size == 0) {
5948                         CERROR("%s: cli %s/%p ocd_connect_flags: %#llx "
5949                                "ocd_version: %x ocd_grant: %d ocd_index: %u "
5950                                "ocd_brw_size unexpectedly zero, network data "
5951                                "corruption? Refusing to connect this client\n",
5952                                obd_name, exp->exp_client_uuid.uuid,
5953                                exp, data->ocd_connect_flags, data->ocd_version,
5954                                data->ocd_grant, data->ocd_index);
5955                         return -EPROTO;
5956                 }
5957         }
5958
5959         if (OCD_HAS_FLAG(data, GRANT_PARAM)) {
5960                 struct dt_device_param *ddp = &mdt->mdt_lut.lut_dt_conf;
5961
5962                 /* client is reporting its page size, for future use */
5963                 exp->exp_target_data.ted_pagebits = data->ocd_grant_blkbits;
5964                 data->ocd_grant_blkbits  = mdt->mdt_lut.lut_tgd.tgd_blockbits;
5965                 /* ddp_inodespace may not be power-of-two value, eg. for ldiskfs
5966                  * it's LDISKFS_DIR_REC_LEN(20) = 28. */
5967                 data->ocd_grant_inobits = fls(ddp->ddp_inodespace - 1);
5968                 /* ocd_grant_tax_kb is in 1K byte blocks */
5969                 data->ocd_grant_tax_kb = ddp->ddp_extent_tax >> 10;
5970                 data->ocd_grant_max_blks = ddp->ddp_max_extent_blks;
5971         }
5972
5973         /* Save connect_data we have so far because tgt_grant_connect()
5974          * uses it to calculate grant, and we want to save the client
5975          * version before it is overwritten by LUSTRE_VERSION_CODE. */
5976         exp->exp_connect_data = *data;
5977         if (OCD_HAS_FLAG(data, GRANT))
5978                 tgt_grant_connect(env, exp, data, !reconnect);
5979
5980         if (OCD_HAS_FLAG(data, MAXBYTES))
5981                 data->ocd_maxbytes = mdt->mdt_lut.lut_dt_conf.ddp_maxbytes;
5982
5983         /* NB: Disregard the rule against updating
5984          * exp_connect_data.ocd_connect_flags in this case, since
5985          * tgt_client_new() needs to know if this is a lightweight
5986          * connection, and it is safe to expose this flag before
5987          * connection processing completes. */
5988         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
5989                 spin_lock(&exp->exp_lock);
5990                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
5991                 spin_unlock(&exp->exp_lock);
5992         }
5993
5994         data->ocd_version = LUSTRE_VERSION_CODE;
5995
5996         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
5997                 CWARN("%s: MDS requires FID support, but client not\n",
5998                       obd_name);
5999                 return -EBADE;
6000         }
6001
6002         if (OCD_HAS_FLAG(data, PINGLESS)) {
6003                 if (ptlrpc_pinger_suppress_pings()) {
6004                         spin_lock(&exp->exp_obd->obd_dev_lock);
6005                         list_del_init(&exp->exp_obd_chain_timed);
6006                         spin_unlock(&exp->exp_obd->obd_dev_lock);
6007                 } else {
6008                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
6009                 }
6010         }
6011
6012         data->ocd_max_easize = mdt->mdt_max_ea_size;
6013
6014         /* NB: Disregard the rule against updating
6015          * exp_connect_data.ocd_connect_flags in this case, since
6016          * tgt_client_new() needs to know if this is client supports
6017          * multiple modify RPCs, and it is safe to expose this flag before
6018          * connection processing completes. */
6019         if (data->ocd_connect_flags & OBD_CONNECT_MULTIMODRPCS) {
6020                 data->ocd_maxmodrpcs = max_mod_rpcs_per_client;
6021                 spin_lock(&exp->exp_lock);
6022                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_MULTIMODRPCS;
6023                 spin_unlock(&exp->exp_lock);
6024         }
6025
6026         if (OCD_HAS_FLAG(data, CKSUM)) {
6027                 __u32 cksum_types = data->ocd_cksum_types;
6028
6029                 /* The client set in ocd_cksum_types the checksum types it
6030                  * supports. We have to mask off the algorithms that we don't
6031                  * support */
6032                 data->ocd_cksum_types &=
6033                         obd_cksum_types_supported_server(obd_name);
6034
6035                 if (unlikely(data->ocd_cksum_types == 0)) {
6036                         CERROR("%s: Connect with checksum support but no "
6037                                "ocd_cksum_types is set\n",
6038                                exp->exp_obd->obd_name);
6039                         RETURN(-EPROTO);
6040                 }
6041
6042                 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
6043                        "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
6044                        cksum_types, data->ocd_cksum_types);
6045         } else {
6046                 /* This client does not support OBD_CONNECT_CKSUM
6047                  * fall back to CRC32 */
6048                 CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
6049                        "OBD_CONNECT_CKSUM, CRC32 will be used\n",
6050                        exp->exp_obd->obd_name, obd_export_nid2str(exp));
6051         }
6052
6053         return 0;
6054 }
6055
6056 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
6057                                    struct mdt_thread_info *info,
6058                                    struct mdt_file_data *mfd)
6059 {
6060         struct lu_context ses;
6061         int rc;
6062         ENTRY;
6063
6064         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
6065         if (rc)
6066                 RETURN(rc);
6067
6068         env->le_ses = &ses;
6069         lu_context_enter(&ses);
6070
6071         mdt_ucred(info)->uc_valid = UCRED_OLD;
6072         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
6073
6074         lu_context_exit(&ses);
6075         lu_context_fini(&ses);
6076         env->le_ses = NULL;
6077
6078         RETURN(rc);
6079 }
6080
6081 static int mdt_export_cleanup(struct obd_export *exp)
6082 {
6083         LIST_HEAD(closing_list);
6084         struct mdt_export_data  *med = &exp->exp_mdt_data;
6085         struct obd_device       *obd = exp->exp_obd;
6086         struct mdt_device       *mdt;
6087         struct mdt_thread_info  *info;
6088         struct lu_env            env;
6089         struct mdt_file_data    *mfd, *n;
6090         int rc = 0;
6091         ENTRY;
6092
6093         spin_lock(&med->med_open_lock);
6094         while (!list_empty(&med->med_open_head)) {
6095                 struct list_head *tmp = med->med_open_head.next;
6096                 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
6097
6098                 /* Remove mfd handle so it can't be found again.
6099                  * We are consuming the mfd_list reference here. */
6100                 class_handle_unhash(&mfd->mfd_open_handle);
6101                 list_move_tail(&mfd->mfd_list, &closing_list);
6102         }
6103         spin_unlock(&med->med_open_lock);
6104         mdt = mdt_dev(obd->obd_lu_dev);
6105         LASSERT(mdt != NULL);
6106
6107         rc = lu_env_init(&env, LCT_MD_THREAD);
6108         if (rc)
6109                 RETURN(rc);
6110
6111         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6112         LASSERT(info != NULL);
6113         memset(info, 0, sizeof *info);
6114         info->mti_env = &env;
6115         info->mti_mdt = mdt;
6116         info->mti_exp = exp;
6117
6118         if (!list_empty(&closing_list)) {
6119                 struct md_attr *ma = &info->mti_attr;
6120
6121                 /* Close any open files (which may also cause orphan
6122                  * unlinking). */
6123                 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
6124                         list_del_init(&mfd->mfd_list);
6125                         ma->ma_need = ma->ma_valid = 0;
6126
6127                         /* This file is being closed due to an eviction, it
6128                          * could have been modified and now dirty regarding to
6129                          * HSM archive, check this!
6130                          * The logic here is to mark a file dirty if there's a
6131                          * chance it was dirtied before the client was evicted,
6132                          * so that we don't have to wait for a release attempt
6133                          * before finding out the file was actually dirty and
6134                          * fail the release. Aggressively marking it dirty here
6135                          * will cause the policy engine to attempt to
6136                          * re-archive it; when rearchiving, we can compare the
6137                          * current version to the HSM data_version and make the
6138                          * archive request into a noop if it's not actually
6139                          * dirty.
6140                          */
6141                         if (mfd->mfd_open_flags & MDS_FMODE_WRITE)
6142                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
6143
6144                         /* Don't unlink orphan on failover umount, LU-184 */
6145                         if (exp->exp_flags & OBD_OPT_FAILOVER) {
6146                                 ma->ma_valid = MA_FLAGS;
6147                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
6148                         }
6149                         ma->ma_valid |= MA_FORCE_LOG;
6150                         mdt_mfd_close(info, mfd);
6151                 }
6152         }
6153         info->mti_mdt = NULL;
6154         /* cleanup client slot early */
6155         /* Do not erase record for recoverable client. */
6156         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
6157                 tgt_client_del(&env, exp);
6158         lu_env_fini(&env);
6159
6160         RETURN(rc);
6161 }
6162
6163 static inline void mdt_enable_slc(struct mdt_device *mdt)
6164 {
6165         if (mdt->mdt_lut.lut_sync_lock_cancel == SYNC_LOCK_CANCEL_NEVER)
6166                 mdt->mdt_lut.lut_sync_lock_cancel = SYNC_LOCK_CANCEL_BLOCKING;
6167 }
6168
6169 static inline void mdt_disable_slc(struct mdt_device *mdt)
6170 {
6171         if (mdt->mdt_lut.lut_sync_lock_cancel == SYNC_LOCK_CANCEL_BLOCKING)
6172                 mdt->mdt_lut.lut_sync_lock_cancel = SYNC_LOCK_CANCEL_NEVER;
6173 }
6174
6175 static int mdt_obd_disconnect(struct obd_export *exp)
6176 {
6177         int rc;
6178
6179         ENTRY;
6180
6181         LASSERT(exp);
6182         class_export_get(exp);
6183
6184         if (!(exp->exp_flags & OBD_OPT_FORCE))
6185                 tgt_grant_sanity_check(exp->exp_obd, __func__);
6186
6187         if ((exp_connect_flags(exp) & OBD_CONNECT_MDS_MDS) &&
6188             !(exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT)) {
6189                 struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
6190
6191                 if (atomic_dec_and_test(&mdt->mdt_mds_mds_conns))
6192                         mdt_disable_slc(mdt);
6193         }
6194
6195         rc = server_disconnect_export(exp);
6196         if (rc != 0)
6197                 CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
6198
6199         tgt_grant_discard(exp);
6200
6201         rc = mdt_export_cleanup(exp);
6202         nodemap_del_member(exp);
6203         class_export_put(exp);
6204         RETURN(rc);
6205 }
6206
6207 /* mds_connect copy */
6208 static int mdt_obd_connect(const struct lu_env *env,
6209                            struct obd_export **exp, struct obd_device *obd,
6210                            struct obd_uuid *cluuid,
6211                            struct obd_connect_data *data,
6212                            void *localdata)
6213 {
6214         struct obd_export       *lexp;
6215         struct lustre_handle    conn = { 0 };
6216         struct mdt_device       *mdt;
6217         int                      rc;
6218         lnet_nid_t              *client_nid = localdata;
6219         ENTRY;
6220
6221         LASSERT(env != NULL);
6222         LASSERT(data != NULL);
6223
6224         if (!exp || !obd || !cluuid)
6225                 RETURN(-EINVAL);
6226
6227         mdt = mdt_dev(obd->obd_lu_dev);
6228
6229         if ((data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
6230             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
6231                 atomic_inc(&mdt->mdt_mds_mds_conns);
6232                 mdt_enable_slc(mdt);
6233         }
6234
6235         /*
6236          * first, check whether the stack is ready to handle requests
6237          * XXX: probably not very appropriate method is used now
6238          *      at some point we should find a better one
6239          */
6240         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) &&
6241             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) &&
6242             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
6243                 rc = obd_get_info(env, mdt->mdt_child_exp,
6244                                   sizeof(KEY_OSP_CONNECTED),
6245                                   KEY_OSP_CONNECTED, NULL, NULL);
6246                 if (rc)
6247                         RETURN(-EAGAIN);
6248                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
6249         }
6250
6251         rc = class_connect(&conn, obd, cluuid);
6252         if (rc)
6253                 RETURN(rc);
6254
6255         lexp = class_conn2export(&conn);
6256         LASSERT(lexp != NULL);
6257
6258         rc = nodemap_add_member(*client_nid, lexp);
6259         if (rc != 0 && rc != -EEXIST)
6260                 GOTO(out, rc);
6261
6262         rc = mdt_connect_internal(env, lexp, mdt, data, false);
6263         if (rc == 0) {
6264                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
6265
6266                 LASSERT(lcd);
6267                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
6268                 rc = tgt_client_new(env, lexp);
6269                 if (rc == 0)
6270                         mdt_export_stats_init(obd, lexp, localdata);
6271         }
6272 out:
6273         if (rc != 0) {
6274                 class_disconnect(lexp);
6275                 nodemap_del_member(lexp);
6276                 *exp = NULL;
6277         } else {
6278                 *exp = lexp;
6279                 /* Because we do not want this export to be evicted by pinger,
6280                  * let's not add this export to the timed chain list. */
6281                 if (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) {
6282                         spin_lock(&lexp->exp_obd->obd_dev_lock);
6283                         list_del_init(&lexp->exp_obd_chain_timed);
6284                         spin_unlock(&lexp->exp_obd->obd_dev_lock);
6285                 }
6286         }
6287
6288         RETURN(rc);
6289 }
6290
6291 static int mdt_obd_reconnect(const struct lu_env *env,
6292                              struct obd_export *exp, struct obd_device *obd,
6293                              struct obd_uuid *cluuid,
6294                              struct obd_connect_data *data,
6295                              void *localdata)
6296 {
6297         lnet_nid_t             *client_nid = localdata;
6298         int                     rc;
6299         ENTRY;
6300
6301         if (exp == NULL || obd == NULL || cluuid == NULL)
6302                 RETURN(-EINVAL);
6303
6304         rc = nodemap_add_member(*client_nid, exp);
6305         if (rc != 0 && rc != -EEXIST)
6306                 RETURN(rc);
6307
6308         rc = mdt_connect_internal(env, exp, mdt_dev(obd->obd_lu_dev), data,
6309                                   true);
6310         if (rc == 0)
6311                 mdt_export_stats_init(obd, exp, localdata);
6312         else
6313                 nodemap_del_member(exp);
6314
6315         RETURN(rc);
6316 }
6317
6318 /* FIXME: Can we avoid using these two interfaces? */
6319 static int mdt_init_export(struct obd_export *exp)
6320 {
6321         struct mdt_export_data *med = &exp->exp_mdt_data;
6322         int                     rc;
6323         ENTRY;
6324
6325         INIT_LIST_HEAD(&med->med_open_head);
6326         spin_lock_init(&med->med_open_lock);
6327         spin_lock(&exp->exp_lock);
6328         exp->exp_connecting = 1;
6329         spin_unlock(&exp->exp_lock);
6330
6331         OBD_ALLOC(exp->exp_used_slots,
6332                   BITS_TO_LONGS(OBD_MAX_RIF_MAX) * sizeof(long));
6333         if (exp->exp_used_slots == NULL)
6334                 RETURN(-ENOMEM);
6335
6336         /* self-export doesn't need client data and ldlm initialization */
6337         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
6338                                      &exp->exp_client_uuid)))
6339                 RETURN(0);
6340
6341         rc = tgt_client_alloc(exp);
6342         if (rc)
6343                 GOTO(err, rc);
6344
6345         rc = ldlm_init_export(exp);
6346         if (rc)
6347                 GOTO(err_free, rc);
6348
6349         RETURN(rc);
6350
6351 err_free:
6352         tgt_client_free(exp);
6353 err:
6354         OBD_FREE(exp->exp_used_slots,
6355                  BITS_TO_LONGS(OBD_MAX_RIF_MAX) * sizeof(long));
6356         exp->exp_used_slots = NULL;
6357
6358         CERROR("%s: Failed to initialize export: rc = %d\n",
6359                exp->exp_obd->obd_name, rc);
6360         return rc;
6361 }
6362
6363 static int mdt_destroy_export(struct obd_export *exp)
6364 {
6365         ENTRY;
6366
6367         target_destroy_export(exp);
6368         if (exp->exp_used_slots)
6369                 OBD_FREE(exp->exp_used_slots,
6370                          BITS_TO_LONGS(OBD_MAX_RIF_MAX) * sizeof(long));
6371
6372         /* destroy can be called from failed obd_setup, so
6373          * checking uuid is safer than obd_self_export */
6374         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
6375                                      &exp->exp_client_uuid)))
6376                 RETURN(0);
6377
6378         ldlm_destroy_export(exp);
6379         tgt_client_free(exp);
6380
6381         LASSERT(list_empty(&exp->exp_outstanding_replies));
6382         LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
6383
6384         /*
6385          * discard grants once we're sure no more
6386          * interaction with the client is possible
6387          */
6388         tgt_grant_discard(exp);
6389         if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
6390                 exp->exp_obd->u.obt.obt_lut->lut_tgd.tgd_tot_granted_clients--;
6391
6392         if (!(exp->exp_flags & OBD_OPT_FORCE))
6393                 tgt_grant_sanity_check(exp->exp_obd, __func__);
6394
6395         RETURN(0);
6396 }
6397
6398 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
6399                    struct linkea_data *ldata)
6400 {
6401         int rc;
6402
6403         LASSERT(ldata->ld_buf->lb_buf != NULL);
6404
6405         if (!mdt_object_exists(mdt_obj))
6406                 return -ENODATA;
6407
6408         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6409                           ldata->ld_buf, XATTR_NAME_LINK);
6410         if (rc == -ERANGE) {
6411                 /* Buf was too small, figure out what we need. */
6412                 lu_buf_free(ldata->ld_buf);
6413                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6414                                   ldata->ld_buf, XATTR_NAME_LINK);
6415                 if (rc < 0)
6416                         return rc;
6417                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
6418                 if (ldata->ld_buf->lb_buf == NULL)
6419                         return -ENOMEM;
6420                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6421                                   ldata->ld_buf, XATTR_NAME_LINK);
6422         }
6423         if (rc < 0)
6424                 return rc;
6425
6426         return linkea_init_with_rec(ldata);
6427 }
6428
6429 /**
6430  * Given an MDT object, try to look up the full path to the object.
6431  * Part of the MDT layer implementation of lfs fid2path.
6432  *
6433  * \param[in]     info  Per-thread common data shared by MDT level handlers.
6434  * \param[in]     obj   Object to do path lookup of
6435  * \param[in,out] fp    User-provided struct to store path information
6436  * \param[in]     root_fid Root FID of current path should reach
6437  *
6438  * \retval 0 Lookup successful, path information stored in fp
6439  * \retval -EAGAIN Lookup failed, usually because object is being moved
6440  * \retval negative errno if there was a problem
6441  */
6442 static int mdt_path_current(struct mdt_thread_info *info,
6443                             struct mdt_object *obj,
6444                             struct getinfo_fid2path *fp,
6445                             struct lu_fid *root_fid)
6446 {
6447         struct mdt_device       *mdt = info->mti_mdt;
6448         struct mdt_object       *mdt_obj;
6449         struct link_ea_header   *leh;
6450         struct link_ea_entry    *lee;
6451         struct lu_name          *tmpname = &info->mti_name;
6452         struct lu_fid           *tmpfid = &info->mti_tmp_fid1;
6453         struct lu_buf           *buf = &info->mti_big_buf;
6454         char                    *ptr;
6455         int                     reclen;
6456         struct linkea_data      ldata = { NULL };
6457         int                     rc = 0;
6458         bool                    first = true;
6459         ENTRY;
6460
6461         /* temp buffer for path element, the buffer will be finally freed
6462          * in mdt_thread_info_fini */
6463         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
6464         if (buf->lb_buf == NULL)
6465                 RETURN(-ENOMEM);
6466
6467         ldata.ld_buf = buf;
6468         ptr = fp->gf_u.gf_path + fp->gf_pathlen - 1;
6469         *ptr = 0;
6470         --ptr;
6471         *tmpfid = fp->gf_fid = *mdt_object_fid(obj);
6472
6473         while (!lu_fid_eq(root_fid, &fp->gf_fid)) {
6474                 struct lu_buf           lmv_buf;
6475
6476                 if (!lu_fid_eq(root_fid, &mdt->mdt_md_root_fid) &&
6477                     lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid))
6478                         GOTO(out, rc = -ENOENT);
6479
6480                 if (lu_fid_eq(mdt_object_fid(obj), tmpfid)) {
6481                         mdt_obj = obj;
6482                         mdt_object_get(info->mti_env, mdt_obj);
6483                 } else {
6484                         mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
6485                         if (IS_ERR(mdt_obj))
6486                                 GOTO(out, rc = PTR_ERR(mdt_obj));
6487                 }
6488
6489                 if (!mdt_object_exists(mdt_obj)) {
6490                         mdt_object_put(info->mti_env, mdt_obj);
6491                         GOTO(out, rc = -ENOENT);
6492                 }
6493
6494                 if (mdt_object_remote(mdt_obj)) {
6495                         mdt_object_put(info->mti_env, mdt_obj);
6496                         GOTO(remote_out, rc = -EREMOTE);
6497                 }
6498
6499                 rc = mdt_links_read(info, mdt_obj, &ldata);
6500                 if (rc != 0) {
6501                         mdt_object_put(info->mti_env, mdt_obj);
6502                         GOTO(out, rc);
6503                 }
6504
6505                 leh = buf->lb_buf;
6506                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
6507                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
6508                 /* If set, use link #linkno for path lookup, otherwise use
6509                    link #0.  Only do this for the final path element. */
6510                 if (first && fp->gf_linkno < leh->leh_reccount) {
6511                         int count;
6512                         for (count = 0; count < fp->gf_linkno; count++) {
6513                                 lee = (struct link_ea_entry *)
6514                                      ((char *)lee + reclen);
6515                                 linkea_entry_unpack(lee, &reclen, tmpname,
6516                                                     tmpfid);
6517                         }
6518                         if (fp->gf_linkno < leh->leh_reccount - 1)
6519                                 /* indicate to user there are more links */
6520                                 fp->gf_linkno++;
6521                 }
6522
6523                 lmv_buf.lb_buf = info->mti_xattr_buf;
6524                 lmv_buf.lb_len = sizeof(info->mti_xattr_buf);
6525                 /* Check if it is slave stripes */
6526                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6527                                   &lmv_buf, XATTR_NAME_LMV);
6528                 mdt_object_put(info->mti_env, mdt_obj);
6529                 if (rc > 0) {
6530                         union lmv_mds_md *lmm = lmv_buf.lb_buf;
6531
6532                         /* For slave stripes, get its master */
6533                         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE) {
6534                                 fp->gf_fid = *tmpfid;
6535                                 continue;
6536                         }
6537                 } else if (rc < 0 && rc != -ENODATA) {
6538                         GOTO(out, rc);
6539                 }
6540
6541                 rc = 0;
6542
6543                 /* Pack the name in the end of the buffer */
6544                 ptr -= tmpname->ln_namelen;
6545                 if (ptr - 1 <= fp->gf_u.gf_path)
6546                         GOTO(out, rc = -EOVERFLOW);
6547                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
6548                 *(--ptr) = '/';
6549
6550                 /* keep the last resolved fid to the client, so the
6551                  * client will build the left path on another MDT for
6552                  * remote object */
6553                 fp->gf_fid = *tmpfid;
6554
6555                 first = false;
6556         }
6557
6558 remote_out:
6559         ptr++; /* skip leading / */
6560         memmove(fp->gf_u.gf_path, ptr,
6561                 fp->gf_u.gf_path + fp->gf_pathlen - ptr);
6562
6563 out:
6564         RETURN(rc);
6565 }
6566
6567 /**
6568  * Given an MDT object, use mdt_path_current to get the path.
6569  * Essentially a wrapper to retry mdt_path_current a set number of times
6570  * if -EAGAIN is returned (usually because an object is being moved).
6571  *
6572  * Part of the MDT layer implementation of lfs fid2path.
6573  *
6574  * \param[in]     info  Per-thread common data shared by mdt level handlers.
6575  * \param[in]     obj   Object to do path lookup of
6576  * \param[in,out] fp    User-provided struct for arguments and to store path
6577  *                      information
6578  *
6579  * \retval 0 Lookup successful, path information stored in fp
6580  * \retval negative errno if there was a problem
6581  */
6582 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
6583                     struct getinfo_fid2path *fp, struct lu_fid *root_fid)
6584 {
6585         struct mdt_device       *mdt = info->mti_mdt;
6586         int                     tries = 3;
6587         int                     rc = -EAGAIN;
6588         ENTRY;
6589
6590         if (fp->gf_pathlen < 3)
6591                 RETURN(-EOVERFLOW);
6592
6593         if (root_fid == NULL)
6594                 root_fid = &mdt->mdt_md_root_fid;
6595
6596         if (lu_fid_eq(root_fid, mdt_object_fid(obj))) {
6597                 fp->gf_u.gf_path[0] = '\0';
6598                 RETURN(0);
6599         }
6600
6601         /* Retry multiple times in case file is being moved */
6602         while (tries-- && rc == -EAGAIN)
6603                 rc = mdt_path_current(info, obj, fp, root_fid);
6604
6605         RETURN(rc);
6606 }
6607
6608 /**
6609  * Get the full path of the provided FID, as of changelog record recno.
6610  *
6611  * This checks sanity and looks up object for user provided FID
6612  * before calling the actual path lookup code.
6613  *
6614  * Part of the MDT layer implementation of lfs fid2path.
6615  *
6616  * \param[in]     info  Per-thread common data shared by mdt level handlers.
6617  * \param[in,out] fp    User-provided struct for arguments and to store path
6618  *                      information
6619  *
6620  * \retval 0 Lookup successful, path information and recno stored in fp
6621  * \retval -ENOENT, object does not exist
6622  * \retval negative errno if there was a problem
6623  */
6624 static int mdt_fid2path(struct mdt_thread_info *info,
6625                         struct lu_fid *root_fid,
6626                         struct getinfo_fid2path *fp)
6627 {
6628         struct mdt_device *mdt = info->mti_mdt;
6629         struct mdt_object *obj;
6630         int    rc;
6631         ENTRY;
6632
6633         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
6634                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
6635
6636         if (!fid_is_sane(&fp->gf_fid))
6637                 RETURN(-EINVAL);
6638
6639         if (!fid_is_namespace_visible(&fp->gf_fid)) {
6640                 CDEBUG(D_INFO, "%s: "DFID" is invalid, f_seq should be >= %#llx"
6641                        ", or f_oid != 0, or f_ver == 0\n", mdt_obd_name(mdt),
6642                        PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
6643                 RETURN(-EINVAL);
6644         }
6645
6646         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
6647         if (IS_ERR(obj)) {
6648                 rc = PTR_ERR(obj);
6649                 CDEBUG(D_IOCTL, "cannot find "DFID": rc = %d\n",
6650                        PFID(&fp->gf_fid), rc);
6651                 RETURN(rc);
6652         }
6653
6654         if (mdt_object_remote(obj))
6655                 rc = -EREMOTE;
6656         else if (!mdt_object_exists(obj))
6657                 rc = -ENOENT;
6658         else
6659                 rc = 0;
6660
6661         if (rc < 0) {
6662                 mdt_object_put(info->mti_env, obj);
6663                 CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n",
6664                        PFID(&fp->gf_fid), rc);
6665                 RETURN(rc);
6666         }
6667
6668         rc = mdt_path(info, obj, fp, root_fid);
6669
6670         CDEBUG(D_INFO, "fid "DFID", path %s recno %#llx linkno %u\n",
6671                PFID(&fp->gf_fid), fp->gf_u.gf_path,
6672                fp->gf_recno, fp->gf_linkno);
6673
6674         mdt_object_put(info->mti_env, obj);
6675
6676         RETURN(rc);
6677 }
6678
6679 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, int keylen,
6680                             void *val, int vallen)
6681 {
6682         struct getinfo_fid2path *fpout, *fpin;
6683         struct lu_fid *root_fid = NULL;
6684         int rc = 0;
6685
6686         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
6687         fpout = val;
6688
6689         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
6690                 lustre_swab_fid2path(fpin);
6691
6692         memcpy(fpout, fpin, sizeof(*fpin));
6693         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
6694                 RETURN(-EINVAL);
6695
6696         if (keylen >= cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*fpin) +
6697                       sizeof(struct lu_fid)) {
6698                 /* client sent its root FID, which is normally fileset FID */
6699                 root_fid = fpin->gf_u.gf_root_fid;
6700                 if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
6701                         lustre_swab_lu_fid(root_fid);
6702
6703                 if (root_fid != NULL && !fid_is_sane(root_fid))
6704                         RETURN(-EINVAL);
6705         }
6706
6707         rc = mdt_fid2path(info, root_fid, fpout);
6708         RETURN(rc);
6709 }
6710
6711 int mdt_get_info(struct tgt_session_info *tsi)
6712 {
6713         char    *key;
6714         int      keylen;
6715         __u32   *vallen;
6716         void    *valout;
6717         int      rc;
6718
6719         ENTRY;
6720
6721         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
6722         if (key == NULL) {
6723                 CDEBUG(D_IOCTL, "No GETINFO key\n");
6724                 RETURN(err_serious(-EFAULT));
6725         }
6726         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
6727                                       RCL_CLIENT);
6728
6729         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
6730         if (vallen == NULL) {
6731                 CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n",
6732                                 tgt_name(tsi->tsi_tgt));
6733                 RETURN(err_serious(-EFAULT));
6734         }
6735
6736         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
6737                              *vallen);
6738         rc = req_capsule_server_pack(tsi->tsi_pill);
6739         if (rc)
6740                 RETURN(err_serious(rc));
6741
6742         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
6743         if (valout == NULL) {
6744                 CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n",
6745                                 tgt_name(tsi->tsi_tgt));
6746                 RETURN(err_serious(-EFAULT));
6747         }
6748
6749         if (KEY_IS(KEY_FID2PATH)) {
6750                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
6751
6752                 rc = mdt_rpc_fid2path(info, key, keylen, valout, *vallen);
6753                 mdt_thread_info_fini(info);
6754         } else {
6755                 rc = -EINVAL;
6756         }
6757         RETURN(rc);
6758 }
6759
6760 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
6761 {
6762         struct obd_ioctl_data *data = karg;
6763         struct lu_fid *fid;
6764         __u64 version;
6765         struct mdt_object *obj;
6766         struct mdt_lock_handle  *lh;
6767         int rc;
6768         ENTRY;
6769
6770         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
6771             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
6772                 RETURN(-EINVAL);
6773
6774         fid = (struct lu_fid *)data->ioc_inlbuf1;
6775
6776         if (!fid_is_sane(fid))
6777                 RETURN(-EINVAL);
6778
6779         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
6780
6781         lh = &mti->mti_lh[MDT_LH_PARENT];
6782         mdt_lock_reg_init(lh, LCK_CR);
6783
6784         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
6785         if (IS_ERR(obj))
6786                 RETURN(PTR_ERR(obj));
6787
6788         if (mdt_object_remote(obj)) {
6789                 rc = -EREMOTE;
6790                 /**
6791                  * before calling version get the correct MDS should be
6792                  * fid, this is error to find remote object here
6793                  */
6794                 CERROR("nonlocal object "DFID"\n", PFID(fid));
6795         } else if (!mdt_object_exists(obj)) {
6796                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
6797                 rc = -ENOENT;
6798         } else {
6799                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
6800                *(__u64 *)data->ioc_inlbuf2 = version;
6801                 rc = 0;
6802         }
6803         mdt_object_unlock_put(mti, obj, lh, 1);
6804         RETURN(rc);
6805 }
6806
6807 /* ioctls on obd dev */
6808 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
6809                          void *karg, void __user *uarg)
6810 {
6811         struct lu_env      env;
6812         struct obd_device *obd = exp->exp_obd;
6813         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
6814         struct dt_device  *dt = mdt->mdt_bottom;
6815         int rc;
6816
6817         ENTRY;
6818         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
6819         rc = lu_env_init(&env, LCT_MD_THREAD);
6820         if (rc)
6821                 RETURN(rc);
6822
6823         switch (cmd) {
6824         case OBD_IOC_SYNC:
6825                 rc = mdt_device_sync(&env, mdt);
6826                 break;
6827         case OBD_IOC_SET_READONLY:
6828                 rc = dt_sync(&env, dt);
6829                 if (rc == 0)
6830                         rc = dt_ro(&env, dt);
6831                 break;
6832         case OBD_IOC_ABORT_RECOVERY:
6833                 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
6834                 obd->obd_abort_recovery = 1;
6835                 target_stop_recovery_thread(obd);
6836                 rc = 0;
6837                 break;
6838         case OBD_IOC_CHANGELOG_REG:
6839         case OBD_IOC_CHANGELOG_DEREG:
6840         case OBD_IOC_CHANGELOG_CLEAR:
6841                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(&env,
6842                                                            mdt->mdt_child,
6843                                                            cmd, len, karg);
6844                 break;
6845         case OBD_IOC_START_LFSCK: {
6846                 struct md_device *next = mdt->mdt_child;
6847                 struct obd_ioctl_data *data = karg;
6848                 struct lfsck_start_param lsp;
6849
6850                 if (unlikely(data == NULL)) {
6851                         rc = -EINVAL;
6852                         break;
6853                 }
6854
6855                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
6856                 lsp.lsp_index_valid = 0;
6857                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
6858                 break;
6859         }
6860         case OBD_IOC_STOP_LFSCK: {
6861                 struct md_device        *next = mdt->mdt_child;
6862                 struct obd_ioctl_data   *data = karg;
6863                 struct lfsck_stop        stop;
6864
6865                 stop.ls_status = LS_STOPPED;
6866                 /* Old lfsck utils may pass NULL @stop. */
6867                 if (data->ioc_inlbuf1 == NULL)
6868                         stop.ls_flags = 0;
6869                 else
6870                         stop.ls_flags =
6871                         ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
6872
6873                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
6874                 break;
6875         }
6876         case OBD_IOC_QUERY_LFSCK: {
6877                 struct md_device        *next = mdt->mdt_child;
6878                 struct obd_ioctl_data   *data = karg;
6879
6880                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0,
6881                                                  data->ioc_inlbuf1);
6882                 break;
6883         }
6884         case OBD_IOC_GET_OBJ_VERSION: {
6885                 struct mdt_thread_info *mti;
6886                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6887                 memset(mti, 0, sizeof *mti);
6888                 mti->mti_env = &env;
6889                 mti->mti_mdt = mdt;
6890                 mti->mti_exp = exp;
6891
6892                 rc = mdt_ioc_version_get(mti, karg);
6893                 break;
6894         }
6895         case OBD_IOC_CATLOGLIST: {
6896                 struct mdt_thread_info *mti;
6897
6898                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6899                 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
6900                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
6901                                        &mti->mti_tmp_fid1);
6902                 break;
6903          }
6904         default:
6905                 rc = -EOPNOTSUPP;
6906                 CERROR("%s: Not supported cmd = %d, rc = %d\n",
6907                         mdt_obd_name(mdt), cmd, rc);
6908         }
6909
6910         lu_env_fini(&env);
6911         RETURN(rc);
6912 }
6913
6914 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
6915 {
6916         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
6917         int rc;
6918         ENTRY;
6919
6920         if (!mdt->mdt_skip_lfsck && !mdt->mdt_bottom->dd_rdonly) {
6921                 struct lfsck_start_param lsp;
6922
6923                 lsp.lsp_start = NULL;
6924                 lsp.lsp_index_valid = 0;
6925                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
6926                                                            OBD_IOC_START_LFSCK,
6927                                                            0, &lsp);
6928                 if (rc != 0 && rc != -EALREADY)
6929                         CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
6930                               mdt_obd_name(mdt), rc);
6931         }
6932
6933         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
6934         RETURN(rc);
6935 }
6936
6937 static int mdt_obd_postrecov(struct obd_device *obd)
6938 {
6939         struct lu_env env;
6940         int rc;
6941
6942         rc = lu_env_init(&env, LCT_MD_THREAD);
6943         if (rc)
6944                 RETURN(rc);
6945         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
6946         lu_env_fini(&env);
6947         return rc;
6948 }
6949
6950 static const struct obd_ops mdt_obd_device_ops = {
6951         .o_owner          = THIS_MODULE,
6952         .o_set_info_async = mdt_obd_set_info_async,
6953         .o_connect        = mdt_obd_connect,
6954         .o_reconnect      = mdt_obd_reconnect,
6955         .o_disconnect     = mdt_obd_disconnect,
6956         .o_init_export    = mdt_init_export,
6957         .o_destroy_export = mdt_destroy_export,
6958         .o_iocontrol      = mdt_iocontrol,
6959         .o_postrecov      = mdt_obd_postrecov,
6960         /* Data-on-MDT IO methods */
6961         .o_preprw         = mdt_obd_preprw,
6962         .o_commitrw       = mdt_obd_commitrw,
6963 };
6964
6965 static struct lu_device* mdt_device_fini(const struct lu_env *env,
6966                                          struct lu_device *d)
6967 {
6968         struct mdt_device *m = mdt_dev(d);
6969         ENTRY;
6970
6971         mdt_fini(env, m);
6972         RETURN(NULL);
6973 }
6974
6975 static struct lu_device *mdt_device_free(const struct lu_env *env,
6976                                          struct lu_device *d)
6977 {
6978         struct mdt_device *m = mdt_dev(d);
6979         ENTRY;
6980
6981         lu_device_fini(&m->mdt_lu_dev);
6982         OBD_FREE_PTR(m);
6983
6984         RETURN(NULL);
6985 }
6986
6987 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
6988                                           struct lu_device_type *t,
6989                                           struct lustre_cfg *cfg)
6990 {
6991         struct lu_device  *l;
6992         struct mdt_device *m;
6993
6994         OBD_ALLOC_PTR(m);
6995         if (m != NULL) {
6996                 int rc;
6997
6998                 l = &m->mdt_lu_dev;
6999                 rc = mdt_init0(env, m, t, cfg);
7000                 if (rc != 0) {
7001                         mdt_device_free(env, l);
7002                         l = ERR_PTR(rc);
7003                         return l;
7004                 }
7005         } else
7006                 l = ERR_PTR(-ENOMEM);
7007         return l;
7008 }
7009
7010 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
7011 LU_KEY_INIT(mdt, struct mdt_thread_info);
7012
7013 static void mdt_key_fini(const struct lu_context *ctx,
7014                          struct lu_context_key *key, void* data)
7015 {
7016         struct mdt_thread_info *info = data;
7017
7018         if (info->mti_big_lmm) {
7019                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
7020                 info->mti_big_lmm = NULL;
7021                 info->mti_big_lmmsize = 0;
7022         }
7023
7024         if (info->mti_big_acl) {
7025                 OBD_FREE_LARGE(info->mti_big_acl, info->mti_big_aclsize);
7026                 info->mti_big_acl = NULL;
7027                 info->mti_big_aclsize = 0;
7028         }
7029
7030         OBD_FREE_PTR(info);
7031 }
7032
7033 /* context key: mdt_thread_key */
7034 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
7035
7036 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
7037 {
7038         return lu_ucred(info->mti_env);
7039 }
7040
7041 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
7042 {
7043         return lu_ucred_check(info->mti_env);
7044 }
7045
7046 /**
7047  * Enable/disable COS (Commit On Sharing).
7048  *
7049  * Set/Clear the COS flag in mdt options.
7050  *
7051  * \param mdt mdt device
7052  * \param val 0 disables COS, other values enable COS
7053  */
7054 void mdt_enable_cos(struct mdt_device *mdt, bool val)
7055 {
7056         struct lu_env env;
7057         int rc;
7058
7059         mdt->mdt_opts.mo_cos = val;
7060         rc = lu_env_init(&env, LCT_LOCAL);
7061         if (unlikely(rc != 0)) {
7062                 CWARN("%s: lu_env initialization failed, cannot "
7063                       "sync: rc = %d\n", mdt_obd_name(mdt), rc);
7064                 return;
7065         }
7066         mdt_device_sync(&env, mdt);
7067         lu_env_fini(&env);
7068 }
7069
7070 /**
7071  * Check COS (Commit On Sharing) status.
7072  *
7073  * Return COS flag status.
7074  *
7075  * \param mdt mdt device
7076  */
7077 int mdt_cos_is_enabled(struct mdt_device *mdt)
7078 {
7079         return mdt->mdt_opts.mo_cos != 0;
7080 }
7081
7082 static struct lu_device_type_operations mdt_device_type_ops = {
7083         .ldto_device_alloc = mdt_device_alloc,
7084         .ldto_device_free  = mdt_device_free,
7085         .ldto_device_fini  = mdt_device_fini
7086 };
7087
7088 static struct lu_device_type mdt_device_type = {
7089         .ldt_tags     = LU_DEVICE_MD,
7090         .ldt_name     = LUSTRE_MDT_NAME,
7091         .ldt_ops      = &mdt_device_type_ops,
7092         .ldt_ctx_tags = LCT_MD_THREAD
7093 };
7094
7095 static int __init mdt_init(void)
7096 {
7097         int rc;
7098
7099         BUILD_BUG_ON(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") !=
7100                      FID_NOBRACE_LEN + 1);
7101         BUILD_BUG_ON(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") !=
7102                      FID_LEN + 1);
7103         rc = lu_kmem_init(mdt_caches);
7104         if (rc)
7105                 return rc;
7106
7107         rc = mds_mod_init();
7108         if (rc)
7109                 GOTO(lu_fini, rc);
7110
7111         rc = class_register_type(&mdt_obd_device_ops, NULL, true, NULL,
7112                                  LUSTRE_MDT_NAME, &mdt_device_type);
7113         if (rc)
7114                 GOTO(mds_fini, rc);
7115 lu_fini:
7116         if (rc)
7117                 lu_kmem_fini(mdt_caches);
7118 mds_fini:
7119         if (rc)
7120                 mds_mod_exit();
7121         return rc;
7122 }
7123
7124 static void __exit mdt_exit(void)
7125 {
7126         class_unregister_type(LUSTRE_MDT_NAME);
7127         mds_mod_exit();
7128         lu_kmem_fini(mdt_caches);
7129 }
7130
7131 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
7132 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
7133 MODULE_VERSION(LUSTRE_VERSION_STRING);
7134 MODULE_LICENSE("GPL");
7135
7136 module_init(mdt_init);
7137 module_exit(mdt_exit);