lustre/mdt/mdt_handler.c

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