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