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