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