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, 2016, 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
  70 static unsigned int max_mod_rpcs_per_client = 8;
  71 module_param(max_mod_rpcs_per_client, uint, 0644);
  72 MODULE_PARM_DESC(max_mod_rpcs_per_client, "maximum number of modify RPCs in flight allowed per client");
  73
  74 mdl_mode_t mdt_mdl_lock_modes[] = {
  75         [LCK_MINMODE] = MDL_MINMODE,
  76         [LCK_EX]      = MDL_EX,
  77         [LCK_PW]      = MDL_PW,
  78         [LCK_PR]      = MDL_PR,
  79         [LCK_CW]      = MDL_CW,
  80         [LCK_CR]      = MDL_CR,
  81         [LCK_NL]      = MDL_NL,
  82         [LCK_GROUP]   = MDL_GROUP
  83 };
  84
  85 enum ldlm_mode mdt_dlm_lock_modes[] = {
  86         [MDL_MINMODE]   = LCK_MINMODE,
  87         [MDL_EX]        = LCK_EX,
  88         [MDL_PW]        = LCK_PW,
  89         [MDL_PR]        = LCK_PR,
  90         [MDL_CW]        = LCK_CW,
  91         [MDL_CR]        = LCK_CR,
  92         [MDL_NL]        = LCK_NL,
  93         [MDL_GROUP]     = LCK_GROUP
  94 };
  95
  96 static struct mdt_device *mdt_dev(struct lu_device *d);
  97 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
  98
  99 static const struct lu_object_operations mdt_obj_ops;
 100
 101 /* Slab for MDT object allocation */
 102 static struct kmem_cache *mdt_object_kmem;
 103
 104 /* For HSM restore handles */
 105 struct kmem_cache *mdt_hsm_cdt_kmem;
 106
 107 /* For HSM request handles */
 108 struct kmem_cache *mdt_hsm_car_kmem;
 109
 110 static struct lu_kmem_descr mdt_caches[] = {
 111         {
 112                 .ckd_cache = &mdt_object_kmem,
 113                 .ckd_name  = "mdt_obj",
 114                 .ckd_size  = sizeof(struct mdt_object)
 115         },
 116         {
 117                 .ckd_cache      = &mdt_hsm_cdt_kmem,
 118                 .ckd_name       = "mdt_cdt_restore_handle",
 119                 .ckd_size       = sizeof(struct cdt_restore_handle)
 120         },
 121         {
 122                 .ckd_cache      = &mdt_hsm_car_kmem,
 123                 .ckd_name       = "mdt_cdt_agent_req",
 124                 .ckd_size       = sizeof(struct cdt_agent_req)
 125         },
 126         {
 127                 .ckd_cache = NULL
 128         }
 129 };
 130
 131 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag)
 132 {
 133         if (!rep)
 134                 return 0;
 135         return rep->lock_policy_res1 & op_flag;
 136 }
 137
 138 void mdt_clear_disposition(struct mdt_thread_info *info,
 139                            struct ldlm_reply *rep, __u64 op_flag)
 140 {
 141         if (info) {
 142                 info->mti_opdata &= ~op_flag;
 143                 tgt_opdata_clear(info->mti_env, op_flag);
 144         }
 145         if (rep)
 146                 rep->lock_policy_res1 &= ~op_flag;
 147 }
 148
 149 void mdt_set_disposition(struct mdt_thread_info *info,
 150                          struct ldlm_reply *rep, __u64 op_flag)
 151 {
 152         if (info) {
 153                 info->mti_opdata |= op_flag;
 154                 tgt_opdata_set(info->mti_env, op_flag);
 155         }
 156         if (rep)
 157                 rep->lock_policy_res1 |= op_flag;
 158 }
 159
 160 void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm)
 161 {
 162         lh->mlh_pdo_hash = 0;
 163         lh->mlh_reg_mode = lm;
 164         lh->mlh_rreg_mode = lm;
 165         lh->mlh_type = MDT_REG_LOCK;
 166 }
 167
 168 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode,
 169                        const struct lu_name *lname)
 170 {
 171         lh->mlh_reg_mode = lock_mode;
 172         lh->mlh_pdo_mode = LCK_MINMODE;
 173         lh->mlh_rreg_mode = lock_mode;
 174         lh->mlh_type = MDT_PDO_LOCK;
 175
 176         if (lu_name_is_valid(lname)) {
 177                 lh->mlh_pdo_hash = ll_full_name_hash(NULL, lname->ln_name,
 178                                                      lname->ln_namelen);
 179                 /* XXX Workaround for LU-2856
 180                  *
 181                  * Zero is a valid return value of full_name_hash, but
 182                  * several users of mlh_pdo_hash assume a non-zero
 183                  * hash value. We therefore map zero onto an
 184                  * arbitrary, but consistent value (1) to avoid
 185                  * problems further down the road. */
 186                 if (unlikely(lh->mlh_pdo_hash == 0))
 187                         lh->mlh_pdo_hash = 1;
 188         } else {
 189                 lh->mlh_pdo_hash = 0;
 190         }
 191 }
 192
 193 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 194                               struct mdt_lock_handle *lh)
 195 {
 196         mdl_mode_t mode;
 197         ENTRY;
 198
 199         /*
 200          * Any dir access needs couple of locks:
 201          *
 202          * 1) on part of dir we gonna take lookup/modify;
 203          *
 204          * 2) on whole dir to protect it from concurrent splitting and/or to
 205          * flush client's cache for readdir().
 206          *
 207          * so, for a given mode and object this routine decides what lock mode
 208          * to use for lock #2:
 209          *
 210          * 1) if caller's gonna lookup in dir then we need to protect dir from
 211          * being splitted only - LCK_CR
 212          *
 213          * 2) if caller's gonna modify dir then we need to protect dir from
 214          * being splitted and to flush cache - LCK_CW
 215          *
 216          * 3) if caller's gonna modify dir and that dir seems ready for
 217          * splitting then we need to protect it from any type of access
 218          * (lookup/modify/split) - LCK_EX --bzzz
 219          */
 220
 221         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 222         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 223
 224         /*
 225          * Ask underlaying level its opinion about preferable PDO lock mode
 226          * having access type passed as regular lock mode:
 227          *
 228          * - MDL_MINMODE means that lower layer does not want to specify lock
 229          * mode;
 230          *
 231          * - MDL_NL means that no PDO lock should be taken. This is used in some
 232          * cases. Say, for non-splittable directories no need to use PDO locks
 233          * at all.
 234          */
 235         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 236                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 237
 238         if (mode != MDL_MINMODE) {
 239                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 240         } else {
 241                 /*
 242                  * Lower layer does not want to specify locking mode. We do it
 243                  * our selves. No special protection is needed, just flush
 244                  * client's cache on modification and allow concurrent
 245                  * mondification.
 246                  */
 247                 switch (lh->mlh_reg_mode) {
 248                 case LCK_EX:
 249                         lh->mlh_pdo_mode = LCK_EX;
 250                         break;
 251                 case LCK_PR:
 252                         lh->mlh_pdo_mode = LCK_CR;
 253                         break;
 254                 case LCK_PW:
 255                         lh->mlh_pdo_mode = LCK_CW;
 256                         break;
 257                 default:
 258                         CERROR("Not expected lock type (0x%x)\n",
 259                                (int)lh->mlh_reg_mode);
 260                         LBUG();
 261                 }
 262         }
 263
 264         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 265         EXIT;
 266 }
 267
 268 static int mdt_lookup_fileset(struct mdt_thread_info *info, const char *fileset,
 269                               struct lu_fid *fid)
 270 {
 271         struct mdt_device *mdt = info->mti_mdt;
 272         struct lu_name *lname = &info->mti_name;
 273         char *name = NULL;
 274         struct mdt_object *parent;
 275         u32 mode;
 276         int rc = 0;
 277
 278         LASSERT(!info->mti_cross_ref);
 279
 280         OBD_ALLOC(name, NAME_MAX + 1);
 281         if (name == NULL)
 282                 return -ENOMEM;
 283         lname->ln_name = name;
 284
 285         /*
 286          * We may want to allow this to mount a completely separate
 287          * fileset from the MDT in the future, but keeping it to
 288          * ROOT/ only for now avoid potential security issues.
 289          */
 290         *fid = mdt->mdt_md_root_fid;
 291
 292         while (rc == 0 && fileset != NULL && *fileset != '\0') {
 293                 const char *s1 = fileset;
 294                 const char *s2;
 295
 296                 while (*++s1 == '/')
 297                         ;
 298                 s2 = s1;
 299                 while (*s2 != '/' && *s2 != '\0')
 300                         s2++;
 301
 302                 if (s2 == s1)
 303                         break;
 304
 305                 fileset = s2;
 306
 307                 lname->ln_namelen = s2 - s1;
 308                 if (lname->ln_namelen > NAME_MAX) {
 309                         rc = -EINVAL;
 310                         break;
 311                 }
 312
 313                 /* reject .. as a path component */
 314                 if (lname->ln_namelen == 2 &&
 315                     strncmp(s1, "..", 2) == 0) {
 316                         rc = -EINVAL;
 317                         break;
 318                 }
 319
 320                 strncpy(name, s1, lname->ln_namelen);
 321                 name[lname->ln_namelen] = '\0';
 322
 323                 parent = mdt_object_find(info->mti_env, mdt, fid);
 324                 if (IS_ERR(parent)) {
 325                         rc = PTR_ERR(parent);
 326                         break;
 327                 }
 328                 /* Only got the fid of this obj by name */
 329                 fid_zero(fid);
 330                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
 331                                 fid, &info->mti_spec);
 332                 mdt_object_put(info->mti_env, parent);
 333         }
 334         if (!rc) {
 335                 parent = mdt_object_find(info->mti_env, mdt, fid);
 336                 if (IS_ERR(parent))
 337                         rc = PTR_ERR(parent);
 338                 else {
 339                         mode = lu_object_attr(&parent->mot_obj);
 340                         mdt_object_put(info->mti_env, parent);
 341                         if (!S_ISDIR(mode))
 342                                 rc = -ENOTDIR;
 343                 }
 344         }
 345
 346         OBD_FREE(name, NAME_MAX + 1);
 347
 348         return rc;
 349 }
 350
 351 static int mdt_get_root(struct tgt_session_info *tsi)
 352 {
 353         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
 354         struct mdt_device       *mdt = info->mti_mdt;
 355         struct mdt_body         *repbody;
 356         char                    *fileset = NULL, *buffer = NULL;
 357         int                      rc;
 358         struct obd_export       *exp = info->mti_exp;
 359         char                    *nodemap_fileset;
 360
 361         ENTRY;
 362
 363         rc = mdt_check_ucred(info);
 364         if (rc)
 365                 GOTO(out, rc = err_serious(rc));
 366
 367         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GET_ROOT_PACK))
 368                 GOTO(out, rc = err_serious(-ENOMEM));
 369
 370         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 371         if (req_capsule_get_size(info->mti_pill, &RMF_NAME, RCL_CLIENT) > 0) {
 372                 fileset = req_capsule_client_get(info->mti_pill, &RMF_NAME);
 373                 if (fileset == NULL)
 374                         GOTO(out, rc = err_serious(-EFAULT));
 375         }
 376
 377         nodemap_fileset = nodemap_get_fileset(exp->exp_target_data.ted_nodemap);
 378         if (nodemap_fileset && nodemap_fileset[0]) {
 379                 CDEBUG(D_INFO, "nodemap fileset is %s\n", nodemap_fileset);
 380                 if (fileset) {
 381                         /* consider fileset from client as a sub-fileset
 382                          * of the nodemap one */
 383                         OBD_ALLOC(buffer, PATH_MAX + 1);
 384                         if (buffer == NULL)
 385                                 GOTO(out, rc = err_serious(-ENOMEM));
 386                         if (snprintf(buffer, PATH_MAX + 1, "%s/%s",
 387                                      nodemap_fileset, fileset) >= PATH_MAX + 1)
 388                                 GOTO(out, rc = err_serious(-EINVAL));
 389                         fileset = buffer;
 390                 } else {
 391                         /* enforce fileset as specified in the nodemap */
 392                         fileset = nodemap_fileset;
 393                 }
 394         }
 395
 396         if (fileset) {
 397                 CDEBUG(D_INFO, "Getting fileset %s\n", fileset);
 398                 rc = mdt_lookup_fileset(info, fileset, &repbody->mbo_fid1);
 399                 if (rc < 0)
 400                         GOTO(out, rc = err_serious(rc));
 401         } else {
 402                 repbody->mbo_fid1 = mdt->mdt_md_root_fid;
 403         }
 404         repbody->mbo_valid |= OBD_MD_FLID;
 405
 406         EXIT;
 407 out:
 408         mdt_thread_info_fini(info);
 409         if (buffer)
 410                 OBD_FREE(buffer, PATH_MAX+1);
 411         return rc;
 412 }
 413
 414 static int mdt_statfs(struct tgt_session_info *tsi)
 415 {
 416         struct ptlrpc_request           *req = tgt_ses_req(tsi);
 417         struct mdt_thread_info          *info = tsi2mdt_info(tsi);
 418         struct mdt_device               *mdt = info->mti_mdt;
 419         struct tg_grants_data           *tgd = &mdt->mdt_lut.lut_tgd;
 420         struct ptlrpc_service_part      *svcpt;
 421         struct obd_statfs               *osfs;
 422         int                             rc;
 423
 424         ENTRY;
 425
 426         svcpt = req->rq_rqbd->rqbd_svcpt;
 427
 428         /* This will trigger a watchdog timeout */
 429         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 430                          (MDT_SERVICE_WATCHDOG_FACTOR *
 431                           at_get(&svcpt->scp_at_estimate)) + 1);
 432
 433         rc = mdt_check_ucred(info);
 434         if (rc)
 435                 GOTO(out, rc = err_serious(rc));
 436
 437         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
 438                 GOTO(out, rc = err_serious(-ENOMEM));
 439
 440         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 441         if (!osfs)
 442                 GOTO(out, rc = -EPROTO);
 443
 444         rc = tgt_statfs_internal(tsi->tsi_env, &mdt->mdt_lut, osfs,
 445                                  cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 446                                  NULL);
 447         if (unlikely(rc))
 448                 GOTO(out, rc);
 449
 450         /* at least try to account for cached pages.  its still racy and
 451          * might be under-reporting if clients haven't announced their
 452          * caches with brw recently */
 453         CDEBUG(D_SUPER | D_CACHE, "blocks cached %llu granted %llu"
 454                " pending %llu free %llu avail %llu\n",
 455                tgd->tgd_tot_dirty, tgd->tgd_tot_granted,
 456                tgd->tgd_tot_pending,
 457                osfs->os_bfree << tgd->tgd_blockbits,
 458                osfs->os_bavail << tgd->tgd_blockbits);
 459
 460         osfs->os_bavail -= min_t(u64, osfs->os_bavail,
 461                                  ((tgd->tgd_tot_dirty + tgd->tgd_tot_pending +
 462                                    osfs->os_bsize - 1) >> tgd->tgd_blockbits));
 463
 464         tgt_grant_sanity_check(mdt->mdt_lu_dev.ld_obd, __func__);
 465         CDEBUG(D_CACHE, "%llu blocks: %llu free, %llu avail; "
 466                "%llu objects: %llu free; state %x\n",
 467                osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
 468                osfs->os_files, osfs->os_ffree, osfs->os_state);
 469
 470         if (!exp_grant_param_supp(tsi->tsi_exp) &&
 471             tgd->tgd_blockbits > COMPAT_BSIZE_SHIFT) {
 472                 /* clients which don't support OBD_CONNECT_GRANT_PARAM
 473                  * should not see a block size > page size, otherwise
 474                  * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
 475                  * block size which is the biggest block size known to work
 476                  * with all client's page size. */
 477                 osfs->os_blocks <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
 478                 osfs->os_bfree  <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
 479                 osfs->os_bavail <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
 480                 osfs->os_bsize = 1 << COMPAT_BSIZE_SHIFT;
 481         }
 482         if (rc == 0)
 483                 mdt_counter_incr(req, LPROC_MDT_STATFS);
 484 out:
 485         mdt_thread_info_fini(info);
 486         RETURN(rc);
 487 }
 488
 489 /**
 490  * Pack size attributes into the reply.
 491  */
 492 int mdt_pack_size2body(struct mdt_thread_info *info,
 493                         const struct lu_fid *fid, bool dom_lock)
 494 {
 495         struct mdt_body *b;
 496         struct md_attr *ma = &info->mti_attr;
 497         int dom_stripe;
 498
 499         ENTRY;
 500
 501         LASSERT(ma->ma_attr.la_valid & LA_MODE);
 502
 503         if (!S_ISREG(ma->ma_attr.la_mode) ||
 504             !(ma->ma_valid & MA_LOV && ma->ma_lmm != NULL))
 505                 RETURN(-ENODATA);
 506
 507         dom_stripe = mdt_lmm_dom_entry(ma->ma_lmm);
 508         /* no DoM stripe, no size in reply */
 509         if (dom_stripe == LMM_NO_DOM)
 510                 RETURN(-ENOENT);
 511
 512         /* no DoM lock, no size in reply */
 513         if (!dom_lock)
 514                 RETURN(0);
 515
 516         /* Either DoM lock exists or LMM has only DoM stripe then
 517          * return size on body. */
 518         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 519
 520         mdt_dom_object_size(info->mti_env, info->mti_mdt, fid, b, dom_lock);
 521         RETURN(0);
 522 }
 523
 524 #ifdef CONFIG_FS_POSIX_ACL
 525 /*
 526  * Pack ACL data into the reply. UIDs/GIDs are mapped and filtered by nodemap.
 527  *
 528  * \param       info    thread info object
 529  * \param       repbody reply to pack ACLs into
 530  * \param       o       mdt object of file to examine
 531  * \param       nodemap nodemap of client to reply to
 532  * \retval      0       success
 533  * \retval      -errno  error getting or parsing ACL from disk
 534  */
 535 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
 536                       struct mdt_object *o, struct lu_nodemap *nodemap)
 537 {
 538         const struct lu_env     *env = info->mti_env;
 539         struct md_object        *next = mdt_object_child(o);
 540         struct lu_buf           *buf = &info->mti_buf;
 541         struct mdt_device       *mdt = info->mti_mdt;
 542         int rc;
 543
 544         ENTRY;
 545
 546         buf->lb_buf = req_capsule_server_get(info->mti_pill, &RMF_ACL);
 547         buf->lb_len = req_capsule_get_size(info->mti_pill, &RMF_ACL,
 548                                            RCL_SERVER);
 549         if (buf->lb_len == 0)
 550                 RETURN(0);
 551
 552 again:
 553         rc = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_ACCESS);
 554         if (rc < 0) {
 555                 if (rc == -ENODATA) {
 556                         repbody->mbo_aclsize = 0;
 557                         repbody->mbo_valid |= OBD_MD_FLACL;
 558                         rc = 0;
 559                 } else if (rc == -EOPNOTSUPP) {
 560                         rc = 0;
 561                 } else {
 562                         if (rc == -ERANGE &&
 563                             exp_connect_large_acl(info->mti_exp) &&
 564                             buf->lb_buf != info->mti_big_acl) {
 565                                 if (info->mti_big_acl == NULL) {
 566                                         OBD_ALLOC_LARGE(info->mti_big_acl,
 567                                                         mdt->mdt_max_ea_size);
 568                                         if (info->mti_big_acl == NULL) {
 569                                                 CERROR("%s: unable to grow "
 570                                                        DFID" ACL buffer\n",
 571                                                        mdt_obd_name(mdt),
 572                                                        PFID(mdt_object_fid(o)));
 573                                                 RETURN(-ENOMEM);
 574                                         }
 575
 576                                         info->mti_big_aclsize =
 577                                                         mdt->mdt_max_ea_size;
 578                                 }
 579
 580                                 CDEBUG(D_INODE, "%s: grow the "DFID
 581                                        " ACL buffer to size %d\n",
 582                                        mdt_obd_name(mdt),
 583                                        PFID(mdt_object_fid(o)),
 584                                        mdt->mdt_max_ea_size);
 585
 586                                 buf->lb_buf = info->mti_big_acl;
 587                                 buf->lb_len = info->mti_big_aclsize;
 588
 589                                 goto again;
 590                         }
 591
 592                         CERROR("%s: unable to read "DFID" ACL: rc = %d\n",
 593                                mdt_obd_name(mdt), PFID(mdt_object_fid(o)), rc);
 594                 }
 595         } else {
 596                 if (buf->lb_buf == info->mti_big_acl)
 597                         info->mti_big_acl_used = 1;
 598
 599                 rc = nodemap_map_acl(nodemap, buf->lb_buf,
 600                                      rc, NODEMAP_FS_TO_CLIENT);
 601                 /* if all ACLs mapped out, rc is still >= 0 */
 602                 if (rc < 0) {
 603                         CERROR("%s: nodemap_map_acl unable to parse "DFID
 604                                " ACL: rc = %d\n", mdt_obd_name(mdt),
 605                                PFID(mdt_object_fid(o)), rc);
 606                 } else {
 607                         repbody->mbo_aclsize = rc;
 608                         repbody->mbo_valid |= OBD_MD_FLACL;
 609                         rc = 0;
 610                 }
 611         }
 612
 613         RETURN(rc);
 614 }
 615 #endif
 616
 617 /* XXX Look into layout in MDT layer. */
 618 static inline bool mdt_hsm_is_released(struct lov_mds_md *lmm)
 619 {
 620         struct lov_comp_md_v1   *comp_v1;
 621         struct lov_mds_md       *v1;
 622         int                      i;
 623
 624         if (lmm->lmm_magic == LOV_MAGIC_COMP_V1) {
 625                 comp_v1 = (struct lov_comp_md_v1 *)lmm;
 626
 627                 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
 628                         v1 = (struct lov_mds_md *)((char *)comp_v1 +
 629                                 comp_v1->lcm_entries[i].lcme_offset);
 630                         /* We don't support partial release for now */
 631                         if (!(v1->lmm_pattern & LOV_PATTERN_F_RELEASED))
 632                                 return false;
 633                 }
 634                 return true;
 635         } else {
 636                 return (lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) ?
 637                         true : false;
 638         }
 639 }
 640
 641 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 642                         const struct lu_attr *attr, const struct lu_fid *fid)
 643 {
 644         struct md_attr *ma = &info->mti_attr;
 645         struct obd_export *exp = info->mti_exp;
 646         struct lu_nodemap *nodemap = NULL;
 647
 648         LASSERT(ma->ma_valid & MA_INODE);
 649
 650         if (attr->la_valid & LA_ATIME) {
 651                 b->mbo_atime = attr->la_atime;
 652                 b->mbo_valid |= OBD_MD_FLATIME;
 653         }
 654         if (attr->la_valid & LA_MTIME) {
 655                 b->mbo_mtime = attr->la_mtime;
 656                 b->mbo_valid |= OBD_MD_FLMTIME;
 657         }
 658         if (attr->la_valid & LA_CTIME) {
 659                 b->mbo_ctime = attr->la_ctime;
 660                 b->mbo_valid |= OBD_MD_FLCTIME;
 661         }
 662         if (attr->la_valid & LA_FLAGS) {
 663                 b->mbo_flags = attr->la_flags;
 664                 b->mbo_valid |= OBD_MD_FLFLAGS;
 665         }
 666         if (attr->la_valid & LA_NLINK) {
 667                 b->mbo_nlink = attr->la_nlink;
 668                 b->mbo_valid |= OBD_MD_FLNLINK;
 669         }
 670         if (attr->la_valid & (LA_UID|LA_GID)) {
 671                 nodemap = nodemap_get_from_exp(exp);
 672                 if (IS_ERR(nodemap))
 673                         goto out;
 674         }
 675         if (attr->la_valid & LA_UID) {
 676                 b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 677                                             NODEMAP_FS_TO_CLIENT,
 678                                             attr->la_uid);
 679                 b->mbo_valid |= OBD_MD_FLUID;
 680         }
 681         if (attr->la_valid & LA_GID) {
 682                 b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 683                                             NODEMAP_FS_TO_CLIENT,
 684                                             attr->la_gid);
 685                 b->mbo_valid |= OBD_MD_FLGID;
 686         }
 687
 688         if (attr->la_valid & LA_PROJID) {
 689                 /* TODO, nodemap for project id */
 690                 b->mbo_projid = attr->la_projid;
 691                 b->mbo_valid |= OBD_MD_FLPROJID;
 692         }
 693
 694         b->mbo_mode = attr->la_mode;
 695         if (attr->la_valid & LA_MODE)
 696                 b->mbo_valid |= OBD_MD_FLMODE;
 697         if (attr->la_valid & LA_TYPE)
 698                 b->mbo_valid |= OBD_MD_FLTYPE;
 699
 700         if (fid != NULL) {
 701                 b->mbo_fid1 = *fid;
 702                 b->mbo_valid |= OBD_MD_FLID;
 703                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid=%#llx\n",
 704                        PFID(fid), b->mbo_nlink, b->mbo_mode, b->mbo_valid);
 705         }
 706
 707         if (!(attr->la_valid & LA_TYPE))
 708                 return;
 709
 710         b->mbo_rdev   = attr->la_rdev;
 711         b->mbo_size   = attr->la_size;
 712         b->mbo_blocks = attr->la_blocks;
 713
 714         if (!S_ISREG(attr->la_mode)) {
 715                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 716         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
 717                 /* means no objects are allocated on osts. */
 718                 LASSERT(!(ma->ma_valid & MA_LOV));
 719                 /* just ignore blocks occupied by extend attributes on MDS */
 720                 b->mbo_blocks = 0;
 721                 /* if no object is allocated on osts, the size on mds is valid.
 722                  * b=22272 */
 723                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 724         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL) {
 725                 if (mdt_hsm_is_released(ma->ma_lmm)) {
 726                         /* A released file stores its size on MDS. */
 727                         /* But return 1 block for released file, unless tools
 728                          * like tar will consider it fully sparse. (LU-3864)
 729                          */
 730                         if (unlikely(b->mbo_size == 0))
 731                                 b->mbo_blocks = 0;
 732                         else
 733                                 b->mbo_blocks = 1;
 734                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 735                 } else if (info->mti_som_valid) { /* som is valid */
 736                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 737                 }
 738         }
 739
 740         if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE))
 741                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 742                        PFID(fid), (unsigned long long)b->mbo_size);
 743
 744 out:
 745         if (!IS_ERR_OR_NULL(nodemap))
 746                 nodemap_putref(nodemap);
 747 }
 748
 749 static inline int mdt_body_has_lov(const struct lu_attr *la,
 750                                    const struct mdt_body *body)
 751 {
 752         return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
 753                (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
 754 }
 755
 756 void mdt_client_compatibility(struct mdt_thread_info *info)
 757 {
 758         struct mdt_body       *body;
 759         struct ptlrpc_request *req = mdt_info_req(info);
 760         struct obd_export     *exp = req->rq_export;
 761         struct md_attr        *ma = &info->mti_attr;
 762         struct lu_attr        *la = &ma->ma_attr;
 763         ENTRY;
 764
 765         if (exp_connect_layout(exp))
 766                 /* the client can deal with 16-bit lmm_stripe_count */
 767                 RETURN_EXIT;
 768
 769         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 770
 771         if (!mdt_body_has_lov(la, body))
 772                 RETURN_EXIT;
 773
 774         /* now we have a reply with a lov for a client not compatible with the
 775          * layout lock so we have to clean the layout generation number */
 776         if (S_ISREG(la->la_mode))
 777                 ma->ma_lmm->lmm_layout_gen = 0;
 778         EXIT;
 779 }
 780
 781 static int mdt_attr_get_eabuf_size(struct mdt_thread_info *info,
 782                                    struct mdt_object *o)
 783 {
 784         const struct lu_env *env = info->mti_env;
 785         int rc, rc2;
 786
 787         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 788                           XATTR_NAME_LOV);
 789
 790         if (rc == -ENODATA)
 791                 rc = 0;
 792
 793         if (rc < 0)
 794                 goto out;
 795
 796         /* Is it a directory? Let's check for the LMV as well */
 797         if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
 798                 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 799                                    XATTR_NAME_LMV);
 800
 801                 if (rc2 == -ENODATA)
 802                         rc2 = mo_xattr_get(env, mdt_object_child(o),
 803                                            &LU_BUF_NULL,
 804                                            XATTR_NAME_DEFAULT_LMV);
 805
 806                 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
 807                         rc = rc2;
 808         }
 809
 810 out:
 811         return rc;
 812 }
 813
 814 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
 815                       const char *name)
 816 {
 817         const struct lu_env *env = info->mti_env;
 818         int rc;
 819         ENTRY;
 820
 821         LASSERT(info->mti_big_lmm_used == 0);
 822         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
 823         if (rc < 0)
 824                 RETURN(rc);
 825
 826         /* big_lmm may need to be grown */
 827         if (info->mti_big_lmmsize < rc) {
 828                 int size = size_roundup_power2(rc);
 829
 830                 if (info->mti_big_lmmsize > 0) {
 831                         /* free old buffer */
 832                         LASSERT(info->mti_big_lmm);
 833                         OBD_FREE_LARGE(info->mti_big_lmm,
 834                                        info->mti_big_lmmsize);
 835                         info->mti_big_lmm = NULL;
 836                         info->mti_big_lmmsize = 0;
 837                 }
 838
 839                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
 840                 if (info->mti_big_lmm == NULL)
 841                         RETURN(-ENOMEM);
 842                 info->mti_big_lmmsize = size;
 843         }
 844         LASSERT(info->mti_big_lmmsize >= rc);
 845
 846         info->mti_buf.lb_buf = info->mti_big_lmm;
 847         info->mti_buf.lb_len = info->mti_big_lmmsize;
 848         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
 849
 850         RETURN(rc);
 851 }
 852
 853 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
 854                    struct md_attr *ma, const char *name)
 855 {
 856         struct md_object *next = mdt_object_child(o);
 857         struct lu_buf    *buf = &info->mti_buf;
 858         int rc;
 859
 860         if (strcmp(name, XATTR_NAME_LOV) == 0) {
 861                 buf->lb_buf = ma->ma_lmm;
 862                 buf->lb_len = ma->ma_lmm_size;
 863                 LASSERT(!(ma->ma_valid & MA_LOV));
 864         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 865                 buf->lb_buf = ma->ma_lmv;
 866                 buf->lb_len = ma->ma_lmv_size;
 867                 LASSERT(!(ma->ma_valid & MA_LMV));
 868         } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 869                 buf->lb_buf = ma->ma_lmv;
 870                 buf->lb_len = ma->ma_lmv_size;
 871                 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
 872         } else {
 873                 return -EINVAL;
 874         }
 875
 876         rc = mo_xattr_get(info->mti_env, next, buf, name);
 877         if (rc > 0) {
 878
 879 got:
 880                 if (strcmp(name, XATTR_NAME_LOV) == 0) {
 881                         if (info->mti_big_lmm_used)
 882                                 ma->ma_lmm = info->mti_big_lmm;
 883
 884                         /* NOT return LOV EA with hole to old client. */
 885                         if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
 886                                      LOV_PATTERN_F_HOLE) &&
 887                             !(exp_connect_flags(info->mti_exp) &
 888                               OBD_CONNECT_LFSCK)) {
 889                                 return -EIO;
 890                         } else {
 891                                 ma->ma_lmm_size = rc;
 892                                 ma->ma_valid |= MA_LOV;
 893                         }
 894                 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 895                         if (info->mti_big_lmm_used)
 896                                 ma->ma_lmv = info->mti_big_lmm;
 897
 898                         ma->ma_lmv_size = rc;
 899                         ma->ma_valid |= MA_LMV;
 900                 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 901                         ma->ma_lmv_size = rc;
 902                         ma->ma_valid |= MA_LMV_DEF;
 903                 }
 904
 905                 /* Update mdt_max_mdsize so all clients will be aware that */
 906                 if (info->mti_mdt->mdt_max_mdsize < rc)
 907                         info->mti_mdt->mdt_max_mdsize = rc;
 908
 909                 rc = 0;
 910         } else if (rc == -ENODATA) {
 911                 /* no LOV EA */
 912                 rc = 0;
 913         } else if (rc == -ERANGE) {
 914                 /* Default LMV has fixed size, so it must be able to fit
 915                  * in the original buffer */
 916                 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
 917                         return rc;
 918                 rc = mdt_big_xattr_get(info, o, name);
 919                 if (rc > 0) {
 920                         info->mti_big_lmm_used = 1;
 921                         goto got;
 922                 }
 923         }
 924
 925         return rc;
 926 }
 927
 928 static int mdt_attr_get_pfid(struct mdt_thread_info *info,
 929                              struct mdt_object *o, struct lu_fid *pfid)
 930 {
 931         struct lu_buf           *buf = &info->mti_buf;
 932         struct link_ea_header   *leh;
 933         struct link_ea_entry    *lee;
 934         int                      rc;
 935         ENTRY;
 936
 937         buf->lb_buf = info->mti_big_lmm;
 938         buf->lb_len = info->mti_big_lmmsize;
 939         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
 940                           buf, XATTR_NAME_LINK);
 941         /* ignore errors, MA_PFID won't be set and it is
 942          * up to the caller to treat this as an error */
 943         if (rc == -ERANGE || buf->lb_len == 0) {
 944                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
 945                 buf->lb_buf = info->mti_big_lmm;
 946                 buf->lb_len = info->mti_big_lmmsize;
 947         }
 948
 949         if (rc < 0)
 950                 RETURN(rc);
 951         if (rc < sizeof(*leh)) {
 952                 CERROR("short LinkEA on "DFID": rc = %d\n",
 953                        PFID(mdt_object_fid(o)), rc);
 954                 RETURN(-ENODATA);
 955         }
 956
 957         leh = (struct link_ea_header *) buf->lb_buf;
 958         lee = (struct link_ea_entry *)(leh + 1);
 959         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
 960                 leh->leh_magic = LINK_EA_MAGIC;
 961                 leh->leh_reccount = __swab32(leh->leh_reccount);
 962                 leh->leh_len = __swab64(leh->leh_len);
 963         }
 964         if (leh->leh_magic != LINK_EA_MAGIC)
 965                 RETURN(-EINVAL);
 966         if (leh->leh_reccount == 0)
 967                 RETURN(-ENODATA);
 968
 969         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
 970         fid_be_to_cpu(pfid, pfid);
 971
 972         RETURN(0);
 973 }
 974
 975 int mdt_attr_get_complex(struct mdt_thread_info *info,
 976                          struct mdt_object *o, struct md_attr *ma)
 977 {
 978         const struct lu_env *env = info->mti_env;
 979         struct md_object    *next = mdt_object_child(o);
 980         struct lu_buf       *buf = &info->mti_buf;
 981         int                  need = ma->ma_need;
 982         int                  rc = 0, rc2;
 983         u32                  mode;
 984         ENTRY;
 985
 986         ma->ma_valid = 0;
 987
 988         if (mdt_object_exists(o) == 0)
 989                 GOTO(out, rc = -ENOENT);
 990         mode = lu_object_attr(&next->mo_lu);
 991
 992         if (need & MA_INODE) {
 993                 ma->ma_need = MA_INODE;
 994                 rc = mo_attr_get(env, next, ma);
 995                 if (rc)
 996                         GOTO(out, rc);
 997
 998                 if (S_ISREG(mode))
 999                         (void) mdt_get_som(info, o, &ma->ma_attr);
1000                 ma->ma_valid |= MA_INODE;
1001         }
1002
1003         if (need & MA_PFID) {
1004                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
1005                 if (rc == 0)
1006                         ma->ma_valid |= MA_PFID;
1007                 /* ignore this error, parent fid is not mandatory */
1008                 rc = 0;
1009         }
1010
1011         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
1012                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
1013                 if (rc)
1014                         GOTO(out, rc);
1015         }
1016
1017         if (need & MA_LMV && S_ISDIR(mode)) {
1018                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
1019                 if (rc != 0)
1020                         GOTO(out, rc);
1021         }
1022
1023         if (need & MA_LMV_DEF && S_ISDIR(mode)) {
1024                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
1025                 if (rc != 0)
1026                         GOTO(out, rc);
1027         }
1028
1029         if (need & MA_HSM && S_ISREG(mode)) {
1030                 buf->lb_buf = info->mti_xattr_buf;
1031                 buf->lb_len = sizeof(info->mti_xattr_buf);
1032                 CLASSERT(sizeof(struct hsm_attrs) <=
1033                          sizeof(info->mti_xattr_buf));
1034                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
1035                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
1036                 if (rc2 == 0)
1037                         ma->ma_valid |= MA_HSM;
1038                 else if (rc2 < 0 && rc2 != -ENODATA)
1039                         GOTO(out, rc = rc2);
1040         }
1041
1042 #ifdef CONFIG_FS_POSIX_ACL
1043         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
1044                 buf->lb_buf = ma->ma_acl;
1045                 buf->lb_len = ma->ma_acl_size;
1046                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
1047                 if (rc2 > 0) {
1048                         ma->ma_acl_size = rc2;
1049                         ma->ma_valid |= MA_ACL_DEF;
1050                 } else if (rc2 == -ENODATA) {
1051                         /* no ACLs */
1052                         ma->ma_acl_size = 0;
1053                 } else
1054                         GOTO(out, rc = rc2);
1055         }
1056 #endif
1057 out:
1058         ma->ma_need = need;
1059         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = %#llx ma_lmm=%p\n",
1060                rc, ma->ma_valid, ma->ma_lmm);
1061         RETURN(rc);
1062 }
1063
1064 static int mdt_getattr_internal(struct mdt_thread_info *info,
1065                                 struct mdt_object *o, int ma_need)
1066 {
1067         struct md_object        *next = mdt_object_child(o);
1068         const struct mdt_body   *reqbody = info->mti_body;
1069         struct ptlrpc_request   *req = mdt_info_req(info);
1070         struct md_attr          *ma = &info->mti_attr;
1071         struct lu_attr          *la = &ma->ma_attr;
1072         struct req_capsule      *pill = info->mti_pill;
1073         const struct lu_env     *env = info->mti_env;
1074         struct mdt_body         *repbody;
1075         struct lu_buf           *buffer = &info->mti_buf;
1076         struct obd_export       *exp = info->mti_exp;
1077         int                      rc;
1078         ENTRY;
1079
1080         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
1081                 RETURN(err_serious(-ENOMEM));
1082
1083         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1084
1085         ma->ma_valid = 0;
1086
1087         if (mdt_object_remote(o)) {
1088                 /* This object is located on remote node.*/
1089                 /* Return -ENOTSUPP for old client */
1090                 if (!mdt_is_dne_client(req->rq_export))
1091                         GOTO(out, rc = -ENOTSUPP);
1092
1093                 repbody->mbo_fid1 = *mdt_object_fid(o);
1094                 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
1095                 GOTO(out, rc = 0);
1096         }
1097
1098         if (reqbody->mbo_eadatasize > 0) {
1099                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
1100                 if (buffer->lb_buf == NULL)
1101                         GOTO(out, rc = -EPROTO);
1102                 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
1103                                                       RCL_SERVER);
1104         } else {
1105                 buffer->lb_buf = NULL;
1106                 buffer->lb_len = 0;
1107                 ma_need &= ~(MA_LOV | MA_LMV);
1108                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
1109                        mdt_obd_name(info->mti_mdt),
1110                        req->rq_export->exp_client_uuid.uuid);
1111         }
1112
1113         /* If it is dir object and client require MEA, then we got MEA */
1114         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1115             (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
1116                 /* Assumption: MDT_MD size is enough for lmv size. */
1117                 ma->ma_lmv = buffer->lb_buf;
1118                 ma->ma_lmv_size = buffer->lb_len;
1119                 ma->ma_need = MA_INODE;
1120                 if (ma->ma_lmv_size > 0) {
1121                         if (reqbody->mbo_valid & OBD_MD_MEA)
1122                                 ma->ma_need |= MA_LMV;
1123                         else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA)
1124                                 ma->ma_need |= MA_LMV_DEF;
1125                 }
1126         } else {
1127                 ma->ma_lmm = buffer->lb_buf;
1128                 ma->ma_lmm_size = buffer->lb_len;
1129                 ma->ma_need = MA_INODE | MA_HSM;
1130                 if (ma->ma_lmm_size > 0)
1131                         ma->ma_need |= MA_LOV;
1132         }
1133
1134         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1135             reqbody->mbo_valid & OBD_MD_FLDIREA  &&
1136             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
1137                 /* get default stripe info for this dir. */
1138                 ma->ma_need |= MA_LOV_DEF;
1139         }
1140         ma->ma_need |= ma_need;
1141
1142         rc = mdt_attr_get_complex(info, o, ma);
1143         if (unlikely(rc)) {
1144                 CDEBUG(rc == -ENOENT ? D_OTHER : D_ERROR,
1145                        "%s: getattr error for "DFID": rc = %d\n",
1146                        mdt_obd_name(info->mti_mdt),
1147                        PFID(mdt_object_fid(o)), rc);
1148                 RETURN(rc);
1149         }
1150
1151         /* if file is released, check if a restore is running */
1152         if (ma->ma_valid & MA_HSM) {
1153                 repbody->mbo_valid |= OBD_MD_TSTATE;
1154                 if ((ma->ma_hsm.mh_flags & HS_RELEASED) &&
1155                     mdt_hsm_restore_is_running(info, mdt_object_fid(o)))
1156                         repbody->mbo_t_state = MS_RESTORE;
1157         }
1158
1159         if (likely(ma->ma_valid & MA_INODE))
1160                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
1161         else
1162                 RETURN(-EFAULT);
1163
1164         if (mdt_body_has_lov(la, reqbody)) {
1165                 if (ma->ma_valid & MA_LOV) {
1166                         LASSERT(ma->ma_lmm_size);
1167                         repbody->mbo_eadatasize = ma->ma_lmm_size;
1168                         if (S_ISDIR(la->la_mode))
1169                                 repbody->mbo_valid |= OBD_MD_FLDIREA;
1170                         else
1171                                 repbody->mbo_valid |= OBD_MD_FLEASIZE;
1172                         mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
1173                 }
1174                 if (ma->ma_valid & MA_LMV) {
1175                         /* Return -ENOTSUPP for old client */
1176                         if (!mdt_is_striped_client(req->rq_export))
1177                                 RETURN(-ENOTSUPP);
1178
1179                         LASSERT(S_ISDIR(la->la_mode));
1180                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1181                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1182                         repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
1183                 }
1184                 if (ma->ma_valid & MA_LMV_DEF) {
1185                         /* Return -ENOTSUPP for old client */
1186                         if (!mdt_is_striped_client(req->rq_export))
1187                                 RETURN(-ENOTSUPP);
1188                         LASSERT(S_ISDIR(la->la_mode));
1189                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1190                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1191                         repbody->mbo_valid |= (OBD_MD_FLDIREA |
1192                                                OBD_MD_DEFAULT_MEA);
1193                 }
1194         } else if (S_ISLNK(la->la_mode) &&
1195                    reqbody->mbo_valid & OBD_MD_LINKNAME) {
1196                 buffer->lb_buf = ma->ma_lmm;
1197                 /* eadatasize from client includes NULL-terminator, so
1198                  * there is no need to read it */
1199                 buffer->lb_len = reqbody->mbo_eadatasize - 1;
1200                 rc = mo_readlink(env, next, buffer);
1201                 if (unlikely(rc <= 0)) {
1202                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
1203                                mdt_obd_name(info->mti_mdt),
1204                                PFID(mdt_object_fid(o)), rc);
1205                         rc = -EFAULT;
1206                 } else {
1207                         int print_limit = min_t(int, PAGE_SIZE - 128, rc);
1208
1209                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
1210                                 rc -= 2;
1211                         repbody->mbo_valid |= OBD_MD_LINKNAME;
1212                         /* we need to report back size with NULL-terminator
1213                          * because client expects that */
1214                         repbody->mbo_eadatasize = rc + 1;
1215                         if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
1216                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
1217                                        "on "DFID ", expected %d\n",
1218                                        mdt_obd_name(info->mti_mdt),
1219                                        rc, PFID(mdt_object_fid(o)),
1220                                        reqbody->mbo_eadatasize - 1);
1221                         /* NULL terminate */
1222                         ((char *)ma->ma_lmm)[rc] = 0;
1223
1224                         /* If the total CDEBUG() size is larger than a page, it
1225                          * will print a warning to the console, avoid this by
1226                          * printing just the last part of the symlink. */
1227                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
1228                                print_limit < rc ? "..." : "", print_limit,
1229                                (char *)ma->ma_lmm + rc - print_limit, rc);
1230                         rc = 0;
1231                 }
1232         }
1233
1234         if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
1235                 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1236                 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1237                 CDEBUG(D_INODE, "changing the max MD size to %u\n",
1238                        repbody->mbo_max_mdsize);
1239         }
1240
1241 #ifdef CONFIG_FS_POSIX_ACL
1242         if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1243                  (reqbody->mbo_valid & OBD_MD_FLACL)) {
1244                 struct lu_nodemap *nodemap = nodemap_get_from_exp(exp);
1245                 if (IS_ERR(nodemap))
1246                         RETURN(PTR_ERR(nodemap));
1247
1248                 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
1249                 nodemap_putref(nodemap);
1250         }
1251 #endif
1252
1253 out:
1254         if (rc == 0)
1255                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
1256
1257         RETURN(rc);
1258 }
1259
1260 static int mdt_getattr(struct tgt_session_info *tsi)
1261 {
1262         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1263         struct mdt_object       *obj = info->mti_object;
1264         struct req_capsule      *pill = info->mti_pill;
1265         struct mdt_body         *reqbody;
1266         struct mdt_body         *repbody;
1267         int rc, rc2;
1268         ENTRY;
1269
1270         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1271         LASSERT(reqbody);
1272         LASSERT(obj != NULL);
1273         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1274
1275         /* Unlike intent case where we need to pre-fill out buffers early on
1276          * in intent policy for ldlm reasons, here we can have a much better
1277          * guess at EA size by just reading it from disk.
1278          * Exceptions are readdir and (missing) directory striping */
1279         /* Readlink */
1280         if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1281                 /* No easy way to know how long is the symlink, but it cannot
1282                  * be more than PATH_MAX, so we allocate +1 */
1283                 rc = PATH_MAX + 1;
1284
1285         /* A special case for fs ROOT: getattr there might fetch
1286          * default EA for entire fs, not just for this dir!
1287          */
1288         } else if (lu_fid_eq(mdt_object_fid(obj),
1289                              &info->mti_mdt->mdt_md_root_fid) &&
1290                    (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1291                    (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1292                                                                  MDS_GETATTR)) {
1293                 /* Should the default strping be bigger, mdt_fix_reply
1294                  * will reallocate */
1295                 rc = DEF_REP_MD_SIZE;
1296         } else {
1297                 /* Read the actual EA size from disk */
1298                 rc = mdt_attr_get_eabuf_size(info, obj);
1299         }
1300
1301         if (rc < 0)
1302                 GOTO(out, rc = err_serious(rc));
1303
1304         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1305
1306         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
1307          * by default. If the target object has more ACL entries, then
1308          * enlarge the buffer when necessary. */
1309         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
1310                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
1311
1312         rc = req_capsule_server_pack(pill);
1313         if (unlikely(rc != 0))
1314                 GOTO(out, rc = err_serious(rc));
1315
1316         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1317         LASSERT(repbody != NULL);
1318         repbody->mbo_eadatasize = 0;
1319         repbody->mbo_aclsize = 0;
1320
1321         rc = mdt_check_ucred(info);
1322         if (unlikely(rc))
1323                 GOTO(out_shrink, rc);
1324
1325         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1326
1327         rc = mdt_getattr_internal(info, obj, 0);
1328         EXIT;
1329 out_shrink:
1330         mdt_client_compatibility(info);
1331         rc2 = mdt_fix_reply(info);
1332         if (rc == 0)
1333                 rc = rc2;
1334 out:
1335         mdt_thread_info_fini(info);
1336         return rc;
1337 }
1338
1339 /**
1340  * Handler of layout intent RPC requiring the layout modification
1341  *
1342  * \param[in] info      thread environment
1343  * \param[in] obj       object
1344  * \param[in] layout    layout intent
1345  * \param[in] buf       buffer containing client's lovea, could be empty
1346  *
1347  * \retval 0    on success
1348  * \retval < 0  error code
1349  */
1350 static int mdt_layout_change(struct mdt_thread_info *info,
1351                              struct mdt_object *obj,
1352                              struct layout_intent *layout,
1353                              const struct lu_buf *buf)
1354 {
1355         struct mdt_lock_handle *lh = &info->mti_lh[MDT_LH_LOCAL];
1356         int rc;
1357         ENTRY;
1358
1359         CDEBUG(D_INFO, "got layout change request from client: "
1360                "opc:%u flags:%#x extent "DEXT"\n",
1361                layout->li_opc, layout->li_flags,
1362                PEXT(&layout->li_extent));
1363
1364         if (layout->li_extent.e_start >= layout->li_extent.e_end) {
1365                 CERROR("Recieved an invalid layout change range "DEXT
1366                        "for "DFID"\n",
1367                        PEXT(&layout->li_extent), PFID(mdt_object_fid(obj)));
1368                 RETURN(-EINVAL);
1369         }
1370
1371         if (!S_ISREG(lu_object_attr(&obj->mot_obj)))
1372                 GOTO(out, rc = -EINVAL);
1373
1374         rc = mo_permission(info->mti_env, NULL, mdt_object_child(obj), NULL,
1375                            MAY_WRITE);
1376         if (rc)
1377                 GOTO(out, rc);
1378
1379         /* take layout lock to prepare layout change */
1380         mdt_lock_reg_init(lh, LCK_EX);
1381         rc = mdt_object_lock(info, obj, lh,
1382                              MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR);
1383         if (rc)
1384                 GOTO(out, rc);
1385
1386         rc = mo_layout_change(info->mti_env, mdt_object_child(obj), layout,
1387                               buf);
1388
1389         mdt_object_unlock(info, obj, lh, 1);
1390 out:
1391         RETURN(rc);
1392 }
1393
1394 /**
1395  * Exchange MOF_LOV_CREATED flags between two objects after a
1396  * layout swap. No assumption is made on whether o1 or o2 have
1397  * created objects or not.
1398  *
1399  * \param[in,out] o1    First swap layout object
1400  * \param[in,out] o2    Second swap layout object
1401  */
1402 static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2)
1403 {
1404         unsigned int o1_lov_created = o1->mot_lov_created;
1405
1406         mutex_lock(&o1->mot_lov_mutex);
1407         mutex_lock(&o2->mot_lov_mutex);
1408
1409         o1->mot_lov_created = o2->mot_lov_created;
1410         o2->mot_lov_created = o1_lov_created;
1411
1412         mutex_unlock(&o2->mot_lov_mutex);
1413         mutex_unlock(&o1->mot_lov_mutex);
1414 }
1415
1416 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1417 {
1418         struct mdt_thread_info  *info;
1419         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1420         struct obd_export       *exp = req->rq_export;
1421         struct mdt_object       *o1, *o2, *o;
1422         struct mdt_lock_handle  *lh1, *lh2;
1423         struct mdc_swap_layouts *msl;
1424         int                      rc;
1425         ENTRY;
1426
1427         /* client does not support layout lock, so layout swaping
1428          * is disabled.
1429          * FIXME: there is a problem for old clients which don't support
1430          * layout lock yet. If those clients have already opened the file
1431          * they won't be notified at all so that old layout may still be
1432          * used to do IO. This can be fixed after file release is landed by
1433          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1434         if (!exp_connect_layout(exp))
1435                 RETURN(-EOPNOTSUPP);
1436
1437         info = tsi2mdt_info(tsi);
1438
1439         if (info->mti_dlm_req != NULL)
1440                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1441
1442         o1 = info->mti_object;
1443         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1444                                 &info->mti_body->mbo_fid2);
1445         if (IS_ERR(o))
1446                 GOTO(out, rc = PTR_ERR(o));
1447
1448         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1449                 GOTO(put, rc = -ENOENT);
1450
1451         rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1452         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1453                 GOTO(put, rc);
1454
1455         if (rc < 0)
1456                 swap(o1, o2);
1457
1458         /* permission check. Make sure the calling process having permission
1459          * to write both files. */
1460         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1461                                 MAY_WRITE);
1462         if (rc < 0)
1463                 GOTO(put, rc);
1464
1465         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1466                                 MAY_WRITE);
1467         if (rc < 0)
1468                 GOTO(put, rc);
1469
1470         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1471         if (msl == NULL)
1472                 GOTO(put, rc = -EPROTO);
1473
1474         lh1 = &info->mti_lh[MDT_LH_NEW];
1475         mdt_lock_reg_init(lh1, LCK_EX);
1476         lh2 = &info->mti_lh[MDT_LH_OLD];
1477         mdt_lock_reg_init(lh2, LCK_EX);
1478
1479         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1480                              MDS_INODELOCK_XATTR);
1481         if (rc < 0)
1482                 GOTO(put, rc);
1483
1484         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1485                              MDS_INODELOCK_XATTR);
1486         if (rc < 0)
1487                 GOTO(unlock1, rc);
1488
1489         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1490                              mdt_object_child(o2), msl->msl_flags);
1491         if (rc < 0)
1492                 GOTO(unlock2, rc);
1493
1494         mdt_swap_lov_flag(o1, o2);
1495
1496 unlock2:
1497         mdt_object_unlock(info, o2, lh2, rc);
1498 unlock1:
1499         mdt_object_unlock(info, o1, lh1, rc);
1500 put:
1501         mdt_object_put(info->mti_env, o);
1502 out:
1503         mdt_thread_info_fini(info);
1504         RETURN(rc);
1505 }
1506
1507 static int mdt_raw_lookup(struct mdt_thread_info *info,
1508                           struct mdt_object *parent,
1509                           const struct lu_name *lname,
1510                           struct ldlm_reply *ldlm_rep)
1511 {
1512         struct lu_fid   *child_fid = &info->mti_tmp_fid1;
1513         int              rc;
1514         ENTRY;
1515
1516         LASSERT(!info->mti_cross_ref);
1517
1518         /* Only got the fid of this obj by name */
1519         fid_zero(child_fid);
1520         rc = mdo_lookup(info->mti_env, mdt_object_child(info->mti_object),
1521                         lname, child_fid, &info->mti_spec);
1522         if (rc == 0) {
1523                 struct mdt_body *repbody;
1524
1525                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1526                 repbody->mbo_fid1 = *child_fid;
1527                 repbody->mbo_valid = OBD_MD_FLID;
1528                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1529         } else if (rc == -ENOENT) {
1530                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1531         }
1532
1533         RETURN(rc);
1534 }
1535
1536 /*
1537  * UPDATE lock should be taken against parent, and be released before exit;
1538  * child_bits lock should be taken against child, and be returned back:
1539  *            (1)normal request should release the child lock;
1540  *            (2)intent request will grant the lock to client.
1541  */
1542 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1543                                  struct mdt_lock_handle *lhc,
1544                                  __u64 child_bits,
1545                                  struct ldlm_reply *ldlm_rep)
1546 {
1547         struct ptlrpc_request  *req = mdt_info_req(info);
1548         struct mdt_body        *reqbody = NULL;
1549         struct mdt_object      *parent = info->mti_object;
1550         struct mdt_object      *child;
1551         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1552         struct lu_name         *lname = NULL;
1553         struct mdt_lock_handle *lhp = NULL;
1554         struct ldlm_lock       *lock;
1555         __u64 try_bits = 0;
1556         bool is_resent;
1557         int ma_need = 0;
1558         int rc;
1559
1560         ENTRY;
1561
1562         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1563         LASSERT(ergo(is_resent,
1564                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1565
1566         if (parent == NULL)
1567                 RETURN(-ENOENT);
1568
1569         if (info->mti_cross_ref) {
1570                 /* Only getattr on the child. Parent is on another node. */
1571                 mdt_set_disposition(info, ldlm_rep,
1572                                     DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1573                 child = parent;
1574                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1575                        "ldlm_rep = %p\n",
1576                        PFID(mdt_object_fid(child)), ldlm_rep);
1577
1578                 rc = mdt_check_resent_lock(info, child, lhc);
1579                 if (rc < 0) {
1580                         RETURN(rc);
1581                 } else if (rc > 0) {
1582                         mdt_lock_handle_init(lhc);
1583                         mdt_lock_reg_init(lhc, LCK_PR);
1584
1585                         /*
1586                          * Object's name is on another MDS, no lookup or layout
1587                          * lock is needed here but update lock is.
1588                          */
1589                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
1590                                         MDS_INODELOCK_LAYOUT);
1591                         child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1592
1593                         rc = mdt_object_lock(info, child, lhc, child_bits);
1594                         if (rc < 0)
1595                                 RETURN(rc);
1596                 }
1597
1598                 /* Finally, we can get attr for child. */
1599                 if (!mdt_object_exists(child)) {
1600                         LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1601                                         &child->mot_obj,
1602                                         "remote object doesn't exist.");
1603                         mdt_object_unlock(info, child, lhc, 1);
1604                         RETURN(-ENOENT);
1605                 }
1606
1607                 rc = mdt_getattr_internal(info, child, 0);
1608                 if (unlikely(rc != 0))
1609                         mdt_object_unlock(info, child, lhc, 1);
1610
1611                 RETURN(rc);
1612         }
1613
1614         lname = &info->mti_name;
1615         mdt_name_unpack(info->mti_pill, &RMF_NAME, lname, MNF_FIX_ANON);
1616
1617         if (lu_name_is_valid(lname)) {
1618                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1619                        "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1620                        PNAME(lname), ldlm_rep);
1621         } else {
1622                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1623                 if (unlikely(reqbody == NULL))
1624                         RETURN(err_serious(-EPROTO));
1625
1626                 *child_fid = reqbody->mbo_fid2;
1627
1628                 if (unlikely(!fid_is_sane(child_fid)))
1629                         RETURN(err_serious(-EINVAL));
1630
1631                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1632                        "ldlm_rep = %p\n",
1633                        PFID(mdt_object_fid(parent)),
1634                        PFID(&reqbody->mbo_fid2), ldlm_rep);
1635         }
1636
1637         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1638
1639         if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1640                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1641                                 &parent->mot_obj,
1642                                 "Parent doesn't exist!");
1643                 RETURN(-ESTALE);
1644         }
1645
1646         if (mdt_object_remote(parent)) {
1647                 CERROR("%s: parent "DFID" is on remote target\n",
1648                        mdt_obd_name(info->mti_mdt),
1649                        PFID(mdt_object_fid(parent)));
1650                 RETURN(-EIO);
1651         }
1652
1653         if (lu_name_is_valid(lname)) {
1654                 /* Always allow to lookup ".." */
1655                 if (unlikely(lname->ln_namelen == 2 &&
1656                              lname->ln_name[0] == '.' &&
1657                              lname->ln_name[1] == '.'))
1658                         info->mti_spec.sp_permitted = 1;
1659
1660                 if (info->mti_body->mbo_valid == OBD_MD_FLID) {
1661                         rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1662
1663                         RETURN(rc);
1664                 }
1665
1666                 /* step 1: lock parent only if parent is a directory */
1667                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1668                         lhp = &info->mti_lh[MDT_LH_PARENT];
1669                         mdt_lock_pdo_init(lhp, LCK_PR, lname);
1670                         rc = mdt_object_lock(info, parent, lhp,
1671                                              MDS_INODELOCK_UPDATE);
1672                         if (unlikely(rc != 0))
1673                                 RETURN(rc);
1674                 }
1675
1676                 /* step 2: lookup child's fid by name */
1677                 fid_zero(child_fid);
1678                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
1679                                 child_fid, &info->mti_spec);
1680                 if (rc == -ENOENT)
1681                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1682
1683                 if (rc != 0)
1684                         GOTO(out_parent, rc);
1685         }
1686
1687         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1688
1689         /*
1690          *step 3: find the child object by fid & lock it.
1691          *        regardless if it is local or remote.
1692          *
1693          *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1694          *      set parent dir fid the same as child fid in getattr by fid case
1695          *      we should not lu_object_find() the object again, could lead
1696          *      to hung if there is a concurrent unlink destroyed the object.
1697          */
1698         if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1699                 mdt_object_get(info->mti_env, parent);
1700                 child = parent;
1701         } else {
1702                 child = mdt_object_find(info->mti_env, info->mti_mdt,
1703                                         child_fid);
1704         }
1705
1706         if (unlikely(IS_ERR(child)))
1707                 GOTO(out_parent, rc = PTR_ERR(child));
1708
1709         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
1710         if (!mdt_object_exists(child)) {
1711                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1712                                 &child->mot_obj,
1713                                 "Object doesn't exist!");
1714                 GOTO(out_child, rc = -ENOENT);
1715         }
1716
1717         rc = mdt_check_resent_lock(info, child, lhc);
1718         if (rc < 0) {
1719                 GOTO(out_child, rc);
1720         } else if (rc > 0) {
1721                 mdt_lock_handle_init(lhc);
1722                 mdt_lock_reg_init(lhc, LCK_PR);
1723
1724                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1725                       mdt_object_exists(child) && !mdt_object_remote(child)) {
1726                         struct md_attr *ma = &info->mti_attr;
1727
1728                         ma->ma_valid = 0;
1729                         ma->ma_need = MA_INODE;
1730                         rc = mdt_attr_get_complex(info, child, ma);
1731                         if (unlikely(rc != 0))
1732                                 GOTO(out_child, rc);
1733
1734                         /* If the file has not been changed for some time, we
1735                          * return not only a LOOKUP lock, but also an UPDATE
1736                          * lock and this might save us RPC on later STAT. For
1737                          * directories, it also let negative dentry cache start
1738                          * working for this dir. */
1739                         if (ma->ma_valid & MA_INODE &&
1740                             ma->ma_attr.la_valid & LA_CTIME &&
1741                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1742                                 ma->ma_attr.la_ctime < cfs_time_current_sec())
1743                                 child_bits |= MDS_INODELOCK_UPDATE;
1744                 }
1745
1746                 /* layout lock must be granted in a best-effort way
1747                  * for IT operations */
1748                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1749                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
1750                     !mdt_object_remote(child) && ldlm_rep != NULL) {
1751                         if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1752                             exp_connect_layout(info->mti_exp)) {
1753                                 /* try to grant layout lock for regular file. */
1754                                 try_bits = MDS_INODELOCK_LAYOUT;
1755                         }
1756                         /* Acquire DOM lock in advance for data-on-mdt file */
1757                         if (child != parent)
1758                                 try_bits |= MDS_INODELOCK_DOM;
1759                 }
1760
1761                 if (try_bits != 0) {
1762                         /* try layout lock, it may fail to be granted due to
1763                          * contention at LOOKUP or UPDATE */
1764                         rc = mdt_object_lock_try(info, child, lhc, &child_bits,
1765                                                  try_bits, false);
1766                         if (child_bits & MDS_INODELOCK_LAYOUT)
1767                                 ma_need |= MA_LOV;
1768                 } else {
1769                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1770                          * client will enqueue the lock to the remote MDT */
1771                         if (mdt_object_remote(child))
1772                                 child_bits &= ~MDS_INODELOCK_UPDATE;
1773                         rc = mdt_object_lock(info, child, lhc, child_bits);
1774                 }
1775                 if (unlikely(rc != 0))
1776                         GOTO(out_child, rc);
1777         }
1778
1779         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1780
1781         /* finally, we can get attr for child. */
1782         rc = mdt_getattr_internal(info, child, ma_need);
1783         if (unlikely(rc != 0)) {
1784                 mdt_object_unlock(info, child, lhc, 1);
1785         } else if (lock) {
1786                 /* Debugging code. */
1787                 LDLM_DEBUG(lock, "Returning lock to client");
1788                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1789                                          &lock->l_resource->lr_name),
1790                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1791                          PLDLMRES(lock->l_resource),
1792                          PFID(mdt_object_fid(child)));
1793
1794                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
1795                     mdt_object_exists(child) && !mdt_object_remote(child) &&
1796                     child != parent) {
1797                         LDLM_LOCK_PUT(lock);
1798                         mdt_object_put(info->mti_env, child);
1799                         /* NB: call the mdt_pack_size2body always after
1800                          * mdt_object_put(), that is why this speacial
1801                          * exit path is used. */
1802                         rc = mdt_pack_size2body(info, child_fid,
1803                                                 child_bits & MDS_INODELOCK_DOM);
1804                         if (rc != 0 && child_bits & MDS_INODELOCK_DOM) {
1805                                 /* DOM lock was taken in advance but this is
1806                                  * not DoM file. Drop the lock. */
1807                                 lock_res_and_lock(lock);
1808                                 ldlm_inodebits_drop(lock, MDS_INODELOCK_DOM);
1809                                 unlock_res_and_lock(lock);
1810                         }
1811
1812                         GOTO(out_parent, rc = 0);
1813                 }
1814         }
1815         if (lock)
1816                 LDLM_LOCK_PUT(lock);
1817
1818         EXIT;
1819 out_child:
1820         mdt_object_put(info->mti_env, child);
1821 out_parent:
1822         if (lhp)
1823                 mdt_object_unlock(info, parent, lhp, 1);
1824         return rc;
1825 }
1826
1827 /* normal handler: should release the child lock */
1828 static int mdt_getattr_name(struct tgt_session_info *tsi)
1829 {
1830         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1831         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1832         struct mdt_body        *reqbody;
1833         struct mdt_body        *repbody;
1834         int rc, rc2;
1835         ENTRY;
1836
1837         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1838         LASSERT(reqbody != NULL);
1839         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1840         LASSERT(repbody != NULL);
1841
1842         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1843         repbody->mbo_eadatasize = 0;
1844         repbody->mbo_aclsize = 0;
1845
1846         rc = mdt_init_ucred_intent_getattr(info, reqbody);
1847         if (unlikely(rc))
1848                 GOTO(out_shrink, rc);
1849
1850         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1851         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1852                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1853                 lhc->mlh_reg_lh.cookie = 0;
1854         }
1855         mdt_exit_ucred(info);
1856         EXIT;
1857 out_shrink:
1858         mdt_client_compatibility(info);
1859         rc2 = mdt_fix_reply(info);
1860         if (rc == 0)
1861                 rc = rc2;
1862         mdt_thread_info_fini(info);
1863         return rc;
1864 }
1865
1866 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1867                          void *karg, void __user *uarg);
1868
1869 static int mdt_set_info(struct tgt_session_info *tsi)
1870 {
1871         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1872         char                    *key;
1873         void                    *val;
1874         int                      keylen, vallen, rc = 0;
1875
1876         ENTRY;
1877
1878         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
1879         if (key == NULL) {
1880                 DEBUG_REQ(D_HA, req, "no set_info key");
1881                 RETURN(err_serious(-EFAULT));
1882         }
1883
1884         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
1885                                       RCL_CLIENT);
1886
1887         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
1888         if (val == NULL) {
1889                 DEBUG_REQ(D_HA, req, "no set_info val");
1890                 RETURN(err_serious(-EFAULT));
1891         }
1892
1893         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
1894                                       RCL_CLIENT);
1895
1896         /* Swab any part of val you need to here */
1897         if (KEY_IS(KEY_READ_ONLY)) {
1898                 spin_lock(&req->rq_export->exp_lock);
1899                 if (*(__u32 *)val)
1900                         *exp_connect_flags_ptr(req->rq_export) |=
1901                                 OBD_CONNECT_RDONLY;
1902                 else
1903                         *exp_connect_flags_ptr(req->rq_export) &=
1904                                 ~OBD_CONNECT_RDONLY;
1905                 spin_unlock(&req->rq_export->exp_lock);
1906         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1907                 struct changelog_setinfo *cs = val;
1908
1909                 if (vallen != sizeof(*cs)) {
1910                         CERROR("%s: bad changelog_clear setinfo size %d\n",
1911                                tgt_name(tsi->tsi_tgt), vallen);
1912                         RETURN(-EINVAL);
1913                 }
1914                 if (ptlrpc_req_need_swab(req)) {
1915                         __swab64s(&cs->cs_recno);
1916                         __swab32s(&cs->cs_id);
1917                 }
1918
1919                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
1920                                    vallen, val, NULL);
1921         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
1922                 if (vallen > 0)
1923                         obd_export_evict_by_nid(req->rq_export->exp_obd, val);
1924         } else {
1925                 RETURN(-EINVAL);
1926         }
1927         RETURN(rc);
1928 }
1929
1930 static int mdt_readpage(struct tgt_session_info *tsi)
1931 {
1932         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
1933         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
1934         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
1935         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
1936         struct mdt_body         *repbody;
1937         int                      rc;
1938         int                      i;
1939
1940         ENTRY;
1941
1942         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1943                 RETURN(err_serious(-ENOMEM));
1944
1945         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
1946         if (repbody == NULL || reqbody == NULL)
1947                 RETURN(err_serious(-EFAULT));
1948
1949         /*
1950          * prepare @rdpg before calling lower layers and transfer itself. Here
1951          * reqbody->size contains offset of where to start to read and
1952          * reqbody->nlink contains number bytes to read.
1953          */
1954         rdpg->rp_hash = reqbody->mbo_size;
1955         if (rdpg->rp_hash != reqbody->mbo_size) {
1956                 CERROR("Invalid hash: %#llx != %#llx\n",
1957                        rdpg->rp_hash, reqbody->mbo_size);
1958                 RETURN(-EFAULT);
1959         }
1960
1961         rdpg->rp_attrs = reqbody->mbo_mode;
1962         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
1963                 rdpg->rp_attrs |= LUDA_64BITHASH;
1964         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
1965                                 exp_max_brw_size(tsi->tsi_exp));
1966         rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >>
1967                           PAGE_SHIFT;
1968         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1969         if (rdpg->rp_pages == NULL)
1970                 RETURN(-ENOMEM);
1971
1972         for (i = 0; i < rdpg->rp_npages; ++i) {
1973                 rdpg->rp_pages[i] = alloc_page(GFP_NOFS);
1974                 if (rdpg->rp_pages[i] == NULL)
1975                         GOTO(free_rdpg, rc = -ENOMEM);
1976         }
1977
1978         /* call lower layers to fill allocated pages with directory data */
1979         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
1980         if (rc < 0)
1981                 GOTO(free_rdpg, rc);
1982
1983         /* send pages to client */
1984         rc = tgt_sendpage(tsi, rdpg, rc);
1985
1986         EXIT;
1987 free_rdpg:
1988
1989         for (i = 0; i < rdpg->rp_npages; i++)
1990                 if (rdpg->rp_pages[i] != NULL)
1991                         __free_page(rdpg->rp_pages[i]);
1992         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1993
1994         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1995                 RETURN(0);
1996
1997         return rc;
1998 }
1999
2000 static int mdt_fix_attr_ucred(struct mdt_thread_info *info, __u32 op)
2001 {
2002         struct lu_ucred *uc = mdt_ucred_check(info);
2003         struct lu_attr *attr = &info->mti_attr.ma_attr;
2004
2005         if (uc == NULL)
2006                 return -EINVAL;
2007
2008         if (op != REINT_SETATTR) {
2009                 if ((attr->la_valid & LA_UID) && (attr->la_uid != -1))
2010                         attr->la_uid = uc->uc_fsuid;
2011                 /* for S_ISGID, inherit gid from his parent, such work will be
2012                  * done in cmm/mdd layer, here set all cases as uc->uc_fsgid. */
2013                 if ((attr->la_valid & LA_GID) && (attr->la_gid != -1))
2014                         attr->la_gid = uc->uc_fsgid;
2015         }
2016
2017         return 0;
2018 }
2019
2020 static int mdt_reint_internal(struct mdt_thread_info *info,
2021                               struct mdt_lock_handle *lhc,
2022                               __u32 op)
2023 {
2024         struct req_capsule      *pill = info->mti_pill;
2025         struct mdt_body         *repbody;
2026         int                      rc = 0, rc2;
2027
2028         ENTRY;
2029
2030         rc = mdt_reint_unpack(info, op);
2031         if (rc != 0) {
2032                 CERROR("Can't unpack reint, rc %d\n", rc);
2033                 RETURN(err_serious(rc));
2034         }
2035
2036         /* for replay (no_create) lmm is not needed, client has it already */
2037         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2038                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2039                                      DEF_REP_MD_SIZE);
2040
2041         /* llog cookies are always 0, the field is kept for compatibility */
2042         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2043                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
2044
2045         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
2046          * by default. If the target object has more ACL entries, then
2047          * enlarge the buffer when necessary. */
2048         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
2049                 req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
2050                                      LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
2051
2052         rc = req_capsule_server_pack(pill);
2053         if (rc != 0) {
2054                 CERROR("Can't pack response, rc %d\n", rc);
2055                 RETURN(err_serious(rc));
2056         }
2057
2058         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
2059                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
2060                 LASSERT(repbody);
2061                 repbody->mbo_eadatasize = 0;
2062                 repbody->mbo_aclsize = 0;
2063         }
2064
2065         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
2066
2067         /* for replay no cookkie / lmm need, because client have this already */
2068         if (info->mti_spec.no_create)
2069                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2070                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
2071
2072         rc = mdt_init_ucred_reint(info);
2073         if (rc)
2074                 GOTO(out_shrink, rc);
2075
2076         rc = mdt_fix_attr_ucred(info, op);
2077         if (rc != 0)
2078                 GOTO(out_ucred, rc = err_serious(rc));
2079
2080         rc = mdt_check_resent(info, mdt_reconstruct, lhc);
2081         if (rc < 0) {
2082                 GOTO(out_ucred, rc);
2083         } else if (rc == 1) {
2084                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
2085                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
2086                 GOTO(out_ucred, rc);
2087         }
2088         rc = mdt_reint_rec(info, lhc);
2089         EXIT;
2090 out_ucred:
2091         mdt_exit_ucred(info);
2092 out_shrink:
2093         mdt_client_compatibility(info);
2094         rc2 = mdt_fix_reply(info);
2095         if (rc == 0)
2096                 rc = rc2;
2097         return rc;
2098 }
2099
2100 static long mdt_reint_opcode(struct ptlrpc_request *req,
2101                              const struct req_format **fmt)
2102 {
2103         struct mdt_device       *mdt;
2104         struct mdt_rec_reint    *rec;
2105         long                     opc;
2106
2107         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
2108         if (rec != NULL) {
2109                 opc = rec->rr_opcode;
2110                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
2111                 if (opc < REINT_MAX && fmt[opc] != NULL)
2112                         req_capsule_extend(&req->rq_pill, fmt[opc]);
2113                 else {
2114                         mdt = mdt_exp2dev(req->rq_export);
2115                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
2116                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
2117                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
2118                         opc = err_serious(-EFAULT);
2119                 }
2120         } else {
2121                 opc = err_serious(-EFAULT);
2122         }
2123         return opc;
2124 }
2125
2126 static int mdt_reint(struct tgt_session_info *tsi)
2127 {
2128         long opc;
2129         int  rc;
2130         static const struct req_format *reint_fmts[REINT_MAX] = {
2131                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
2132                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
2133                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
2134                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
2135                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
2136                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
2137                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
2138                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
2139                 [REINT_MIGRATE]  = &RQF_MDS_REINT_RENAME
2140         };
2141
2142         ENTRY;
2143
2144         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
2145         if (opc >= 0) {
2146                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2147                 /*
2148                  * No lock possible here from client to pass it to reint code
2149                  * path.
2150                  */
2151                 rc = mdt_reint_internal(info, NULL, opc);
2152                 mdt_thread_info_fini(info);
2153         } else {
2154                 rc = opc;
2155         }
2156
2157         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
2158         RETURN(rc);
2159 }
2160
2161 /* this should sync the whole device */
2162 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
2163 {
2164         struct dt_device *dt = mdt->mdt_bottom;
2165         int rc;
2166         ENTRY;
2167
2168         rc = dt->dd_ops->dt_sync(env, dt);
2169         RETURN(rc);
2170 }
2171
2172 /* this should sync this object */
2173 static int mdt_object_sync(const struct lu_env *env, struct obd_export *exp,
2174                            struct mdt_object *mo)
2175 {
2176         int rc;
2177
2178         ENTRY;
2179
2180         if (!mdt_object_exists(mo)) {
2181                 CWARN("%s: non existing object "DFID": rc = %d\n",
2182                       exp->exp_obd->obd_name, PFID(mdt_object_fid(mo)),
2183                       -ESTALE);
2184                 RETURN(-ESTALE);
2185         }
2186
2187         rc = mo_object_sync(env, mdt_object_child(mo));
2188
2189         RETURN(rc);
2190 }
2191
2192 static int mdt_sync(struct tgt_session_info *tsi)
2193 {
2194         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2195         struct req_capsule      *pill = tsi->tsi_pill;
2196         struct mdt_body         *body;
2197         int                      rc;
2198
2199         ENTRY;
2200
2201         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
2202                 RETURN(err_serious(-ENOMEM));
2203
2204         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
2205                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
2206         } else {
2207                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2208
2209                 /* sync an object */
2210                 rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp,
2211                                      info->mti_object);
2212                 if (rc == 0) {
2213                         const struct lu_fid *fid;
2214                         struct lu_attr *la = &info->mti_attr.ma_attr;
2215
2216                         info->mti_attr.ma_need = MA_INODE;
2217                         info->mti_attr.ma_valid = 0;
2218                         rc = mdt_attr_get_complex(info, info->mti_object,
2219                                                   &info->mti_attr);
2220                         if (rc == 0) {
2221                                 body = req_capsule_server_get(pill,
2222                                                               &RMF_MDT_BODY);
2223                                 fid = mdt_object_fid(info->mti_object);
2224                                 mdt_pack_attr2body(info, body, la, fid);
2225                         }
2226                 }
2227                 mdt_thread_info_fini(info);
2228         }
2229         if (rc == 0)
2230                 mdt_counter_incr(req, LPROC_MDT_SYNC);
2231
2232         RETURN(rc);
2233 }
2234
2235 static int mdt_data_sync(struct tgt_session_info *tsi)
2236 {
2237         struct mdt_thread_info *info;
2238         struct mdt_device *mdt = mdt_exp2dev(tsi->tsi_exp);
2239         struct ost_body *body = tsi->tsi_ost_body;
2240         struct ost_body *repbody;
2241         struct mdt_object *mo = NULL;
2242         struct md_attr *ma;
2243         int rc = 0;
2244
2245         ENTRY;
2246
2247         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2248
2249         /* if no fid is specified then do nothing,
2250          * device sync is done via MDS_SYNC */
2251         if (fid_is_zero(&tsi->tsi_fid))
2252                 RETURN(0);
2253
2254         mo = mdt_object_find(tsi->tsi_env, mdt, &tsi->tsi_fid);
2255         if (IS_ERR(mo))
2256                 RETURN(PTR_ERR(mo));
2257
2258         rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp, mo);
2259         if (rc)
2260                 GOTO(put, rc);
2261
2262         repbody->oa.o_oi = body->oa.o_oi;
2263         repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
2264
2265         info = tsi2mdt_info(tsi);
2266         ma = &info->mti_attr;
2267         ma->ma_need = MA_INODE;
2268         ma->ma_valid = 0;
2269         rc = mdt_attr_get_complex(info, mo, ma);
2270         if (rc == 0)
2271                 obdo_from_la(&repbody->oa, &ma->ma_attr, VALID_FLAGS);
2272         else
2273                 rc = 0;
2274         mdt_thread_info_fini(info);
2275
2276         EXIT;
2277 put:
2278         if (mo != NULL)
2279                 mdt_object_put(tsi->tsi_env, mo);
2280         return rc;
2281 }
2282
2283 /*
2284  * Handle quota control requests to consult current usage/limit, but also
2285  * to configure quota enforcement
2286  */
2287 static int mdt_quotactl(struct tgt_session_info *tsi)
2288 {
2289         struct obd_export       *exp  = tsi->tsi_exp;
2290         struct req_capsule      *pill = tsi->tsi_pill;
2291         struct obd_quotactl     *oqctl, *repoqc;
2292         int                      id, rc;
2293         struct mdt_device       *mdt = mdt_exp2dev(exp);
2294         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2295         struct lu_nodemap       *nodemap;
2296         ENTRY;
2297
2298         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
2299         if (oqctl == NULL)
2300                 RETURN(err_serious(-EPROTO));
2301
2302         rc = req_capsule_server_pack(pill);
2303         if (rc)
2304                 RETURN(err_serious(rc));
2305
2306         nodemap = nodemap_get_from_exp(exp);
2307         if (IS_ERR(nodemap))
2308                 RETURN(PTR_ERR(nodemap));
2309
2310         switch (oqctl->qc_cmd) {
2311                 /* master quotactl */
2312         case Q_SETINFO:
2313         case Q_SETQUOTA:
2314                 if (!nodemap_can_setquota(nodemap))
2315                         GOTO(out_nodemap, rc = -EPERM);
2316         case Q_GETINFO:
2317         case Q_GETQUOTA:
2318                 if (qmt == NULL)
2319                         GOTO(out_nodemap, rc = -EOPNOTSUPP);
2320                 /* slave quotactl */
2321         case Q_GETOINFO:
2322         case Q_GETOQUOTA:
2323                 break;
2324         default:
2325                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2326                 GOTO(out_nodemap, rc = -EFAULT);
2327         }
2328
2329         id = oqctl->qc_id;
2330         switch (oqctl->qc_type) {
2331         case USRQUOTA:
2332                 id = nodemap_map_id(nodemap, NODEMAP_UID,
2333                                     NODEMAP_CLIENT_TO_FS, id);
2334                 break;
2335         case GRPQUOTA:
2336                 id = nodemap_map_id(nodemap, NODEMAP_GID,
2337                                     NODEMAP_CLIENT_TO_FS, id);
2338                 break;
2339         case PRJQUOTA:
2340                 /* todo: check/map project id */
2341                 id = oqctl->qc_id;
2342                 break;
2343         default:
2344                 GOTO(out_nodemap, rc = -EOPNOTSUPP);
2345         }
2346         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2347         if (repoqc == NULL)
2348                 GOTO(out_nodemap, rc = err_serious(-EFAULT));
2349
2350         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA)
2351                 barrier_exit(tsi->tsi_tgt->lut_bottom);
2352
2353         if (oqctl->qc_id != id)
2354                 swap(oqctl->qc_id, id);
2355
2356         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA) {
2357                 if (unlikely(!barrier_entry(tsi->tsi_tgt->lut_bottom)))
2358                         RETURN(-EINPROGRESS);
2359         }
2360
2361         switch (oqctl->qc_cmd) {
2362
2363         case Q_GETINFO:
2364         case Q_SETINFO:
2365         case Q_SETQUOTA:
2366         case Q_GETQUOTA:
2367                 /* forward quotactl request to QMT */
2368                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2369                 break;
2370
2371         case Q_GETOINFO:
2372         case Q_GETOQUOTA:
2373                 /* slave quotactl */
2374                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2375                                    oqctl);
2376                 break;
2377
2378         default:
2379                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2380                 GOTO(out_nodemap, rc = -EFAULT);
2381         }
2382
2383         if (oqctl->qc_id != id)
2384                 swap(oqctl->qc_id, id);
2385
2386         *repoqc = *oqctl;
2387
2388         EXIT;
2389
2390 out_nodemap:
2391         nodemap_putref(nodemap);
2392
2393         return rc;
2394 }
2395
2396 /** clone llog ctxt from child (mdd)
2397  * This allows remote llog (replicator) access.
2398  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2399  * context was originally set up, or we can handle them directly.
2400  * I choose the latter, but that means I need any llog
2401  * contexts set up by child to be accessable by the mdt.  So we clone the
2402  * context into our context list here.
2403  */
2404 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2405                                int idx)
2406 {
2407         struct md_device  *next = mdt->mdt_child;
2408         struct llog_ctxt *ctxt;
2409         int rc;
2410
2411         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2412                 return 0;
2413
2414         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2415         if (rc || ctxt == NULL) {
2416                 return 0;
2417         }
2418
2419         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2420         if (rc)
2421                 CERROR("Can't set mdt ctxt %d\n", rc);
2422
2423         return rc;
2424 }
2425
2426 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2427                                  struct mdt_device *mdt, int idx)
2428 {
2429         struct llog_ctxt *ctxt;
2430
2431         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2432         if (ctxt == NULL)
2433                 return 0;
2434         /* Put once for the get we just did, and once for the clone */
2435         llog_ctxt_put(ctxt);
2436         llog_ctxt_put(ctxt);
2437         return 0;
2438 }
2439
2440 /*
2441  * sec context handlers
2442  */
2443 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2444 {
2445         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2446
2447         return 0;
2448 }
2449
2450 /*
2451  * quota request handlers
2452  */
2453 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2454 {
2455         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
2456         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2457         int                      rc;
2458         ENTRY;
2459
2460         if (qmt == NULL)
2461                 RETURN(err_serious(-EOPNOTSUPP));
2462
2463         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2464         RETURN(rc);
2465 }
2466
2467 struct mdt_object *mdt_object_new(const struct lu_env *env,
2468                                   struct mdt_device *d,
2469                                   const struct lu_fid *f)
2470 {
2471         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2472         struct lu_object *o;
2473         struct mdt_object *m;
2474         ENTRY;
2475
2476         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2477         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2478         if (unlikely(IS_ERR(o)))
2479                 m = (struct mdt_object *)o;
2480         else
2481                 m = mdt_obj(o);
2482         RETURN(m);
2483 }
2484
2485 struct mdt_object *mdt_object_find(const struct lu_env *env,
2486                                    struct mdt_device *d,
2487                                    const struct lu_fid *f)
2488 {
2489         struct lu_object *o;
2490         struct mdt_object *m;
2491         ENTRY;
2492
2493         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2494         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2495         if (unlikely(IS_ERR(o)))
2496                 m = (struct mdt_object *)o;
2497         else
2498                 m = mdt_obj(o);
2499
2500         RETURN(m);
2501 }
2502
2503 /**
2504  * Asyncronous commit for mdt device.
2505  *
2506  * Pass asynchonous commit call down the MDS stack.
2507  *
2508  * \param env environment
2509  * \param mdt the mdt device
2510  */
2511 static void mdt_device_commit_async(const struct lu_env *env,
2512                                     struct mdt_device *mdt)
2513 {
2514         struct dt_device *dt = mdt->mdt_bottom;
2515         int rc;
2516         ENTRY;
2517
2518         rc = dt->dd_ops->dt_commit_async(env, dt);
2519         if (unlikely(rc != 0))
2520                 CWARN("%s: async commit start failed: rc = %d\n",
2521                       mdt_obd_name(mdt), rc);
2522         atomic_inc(&mdt->mdt_async_commit_count);
2523         EXIT;
2524 }
2525
2526 /**
2527  * Mark the lock as "synchonous".
2528  *
2529  * Mark the lock to deffer transaction commit to the unlock time.
2530  *
2531  * \param lock the lock to mark as "synchonous"
2532  *
2533  * \see mdt_is_lock_sync
2534  * \see mdt_save_lock
2535  */
2536 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2537 {
2538         lock->l_ast_data = (void*)1;
2539 }
2540
2541 /**
2542  * Check whehter the lock "synchonous" or not.
2543  *
2544  * \param lock the lock to check
2545  * \retval 1 the lock is "synchonous"
2546  * \retval 0 the lock isn't "synchronous"
2547  *
2548  * \see mdt_set_lock_sync
2549  * \see mdt_save_lock
2550  */
2551 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2552 {
2553         return lock->l_ast_data != NULL;
2554 }
2555
2556 /**
2557  * Blocking AST for mdt locks.
2558  *
2559  * Starts transaction commit if in case of COS lock conflict or
2560  * deffers such a commit to the mdt_save_lock.
2561  *
2562  * \param lock the lock which blocks a request or cancelling lock
2563  * \param desc unused
2564  * \param data unused
2565  * \param flag indicates whether this cancelling or blocking callback
2566  * \retval 0
2567  * \see ldlm_blocking_ast_nocheck
2568  */
2569 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2570                      void *data, int flag)
2571 {
2572         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2573         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2574         bool commit_async = false;
2575         int rc;
2576         ENTRY;
2577
2578         if (flag == LDLM_CB_CANCELING)
2579                 RETURN(0);
2580
2581         lock_res_and_lock(lock);
2582         if (lock->l_blocking_ast != mdt_blocking_ast) {
2583                 unlock_res_and_lock(lock);
2584                 RETURN(0);
2585         }
2586         /* There is no lock conflict if l_blocking_lock == NULL,
2587          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2588          * when the last reference to a local lock was released */
2589         if (lock->l_req_mode & (LCK_PW | LCK_EX) &&
2590             lock->l_blocking_lock != NULL) {
2591                 if (mdt_cos_is_enabled(mdt)) {
2592                         if (lock->l_client_cookie !=
2593                             lock->l_blocking_lock->l_client_cookie)
2594                                 mdt_set_lock_sync(lock);
2595                 } else if (mdt_slc_is_enabled(mdt) &&
2596                            ldlm_is_cos_incompat(lock->l_blocking_lock)) {
2597                         mdt_set_lock_sync(lock);
2598                         /*
2599                          * we may do extra commit here, but there is a small
2600                          * window to miss a commit: lock was unlocked (saved),
2601                          * then a conflict lock queued and we come here, but
2602                          * REP-ACK not received, so lock was not converted to
2603                          * COS mode yet.
2604                          * Fortunately this window is quite small, so the
2605                          * extra commit should be rare (not to say distributed
2606                          * operation is rare too).
2607                          */
2608                         commit_async = true;
2609                 }
2610         } else if (lock->l_req_mode == LCK_COS &&
2611                    lock->l_blocking_lock != NULL) {
2612                 commit_async = true;
2613         }
2614
2615         rc = ldlm_blocking_ast_nocheck(lock);
2616
2617         if (commit_async) {
2618                 struct lu_env env;
2619
2620                 rc = lu_env_init(&env, LCT_LOCAL);
2621                 if (unlikely(rc != 0))
2622                         CWARN("%s: lu_env initialization failed, cannot "
2623                               "start asynchronous commit: rc = %d\n",
2624                               obd->obd_name, rc);
2625                 else
2626                         mdt_device_commit_async(&env, mdt);
2627                 lu_env_fini(&env);
2628         }
2629         RETURN(rc);
2630 }
2631
2632 /*
2633  * Blocking AST for cross-MDT lock
2634  *
2635  * Discard lock from uncommitted_slc_locks and cancel it.
2636  *
2637  * \param lock  the lock which blocks a request or cancelling lock
2638  * \param desc  unused
2639  * \param data  unused
2640  * \param flag  indicates whether this cancelling or blocking callback
2641  * \retval      0 on success
2642  * \retval      negative number on error
2643  */
2644 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2645                             void *data, int flag)
2646 {
2647         int rc = 0;
2648         ENTRY;
2649
2650         switch (flag) {
2651         case LDLM_CB_BLOCKING: {
2652                 struct lustre_handle lockh;
2653
2654                 ldlm_lock2handle(lock, &lockh);
2655                 rc = ldlm_cli_cancel(&lockh,
2656                         ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
2657                 if (rc < 0) {
2658                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2659                         RETURN(rc);
2660                 }
2661                 break;
2662         }
2663         case LDLM_CB_CANCELING: {
2664                 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2665                 struct mdt_device *mdt =
2666                                 mdt_dev(obd->obd_lu_dev->ld_site->ls_top_dev);
2667
2668                 LDLM_DEBUG(lock, "Revoke remote lock\n");
2669
2670                 /* discard slc lock here so that it can be cleaned anytime,
2671                  * especially for cleanup_resource() */
2672                 tgt_discard_slc_lock(&mdt->mdt_lut, lock);
2673
2674                 /* once we cache lock, l_ast_data is set to mdt_object */
2675                 if (lock->l_ast_data != NULL) {
2676                         struct mdt_object *mo = lock->l_ast_data;
2677                         struct lu_env env;
2678
2679                         rc = lu_env_init(&env, LCT_MD_THREAD);
2680                         if (unlikely(rc != 0)) {
2681                                 CWARN("%s: lu_env initialization failed, object"
2682                                       "%p "DFID" is leaked!\n",
2683                                       obd->obd_name, mo,
2684                                       PFID(mdt_object_fid(mo)));
2685                                 RETURN(rc);
2686                         }
2687
2688                         if (lock->l_policy_data.l_inodebits.bits &
2689                             (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)) {
2690                                 rc = mo_invalidate(&env, mdt_object_child(mo));
2691                                 mo->mot_cache_attr = 0;
2692                         }
2693                         mdt_object_put(&env, mo);
2694                         lu_env_fini(&env);
2695                 }
2696                 break;
2697         }
2698         default:
2699                 LBUG();
2700         }
2701
2702         RETURN(rc);
2703 }
2704
2705 int mdt_check_resent_lock(struct mdt_thread_info *info,
2706                           struct mdt_object *mo,
2707                           struct mdt_lock_handle *lhc)
2708 {
2709         /* the lock might already be gotten in ldlm_handle_enqueue() */
2710         if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
2711                 struct ptlrpc_request *req = mdt_info_req(info);
2712                 struct ldlm_lock      *lock;
2713
2714                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2715                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
2716                 if (lock == NULL) {
2717                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
2718                          * a resend case, however, it could be already destroyed
2719                          * due to client eviction or a raced cancel RPC. */
2720                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx",
2721                                           lhc->mlh_reg_lh.cookie);
2722                         RETURN(-ESTALE);
2723                 }
2724
2725                 if (!fid_res_name_eq(mdt_object_fid(mo),
2726                                      &lock->l_resource->lr_name)) {
2727                         CWARN("%s: Although resent, but still not "
2728                               "get child lock:"DFID"\n",
2729                               info->mti_exp->exp_obd->obd_name,
2730                               PFID(mdt_object_fid(mo)));
2731                         LDLM_LOCK_PUT(lock);
2732                         RETURN(-EPROTO);
2733                 }
2734                 LDLM_LOCK_PUT(lock);
2735                 return 0;
2736         }
2737         return 1;
2738 }
2739
2740 int mdt_remote_object_lock_try(struct mdt_thread_info *mti,
2741                                struct mdt_object *o, const struct lu_fid *fid,
2742                                struct lustre_handle *lh, enum ldlm_mode mode,
2743                                __u64 *ibits, __u64 trybits, bool cache)
2744 {
2745         struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
2746         union ldlm_policy_data *policy = &mti->mti_policy;
2747         struct ldlm_res_id *res_id = &mti->mti_res_id;
2748         int rc = 0;
2749         ENTRY;
2750
2751         LASSERT(mdt_object_remote(o));
2752
2753         fid_build_reg_res_name(fid, res_id);
2754
2755         memset(einfo, 0, sizeof(*einfo));
2756         einfo->ei_type = LDLM_IBITS;
2757         einfo->ei_mode = mode;
2758         einfo->ei_cb_bl = mdt_remote_blocking_ast;
2759         einfo->ei_cb_cp = ldlm_completion_ast;
2760         einfo->ei_enq_slave = 0;
2761         einfo->ei_res_id = res_id;
2762
2763         if (cache) {
2764                 /*
2765                  * if we cache lock, couple lock with mdt_object, so that object
2766                  * can be easily found in lock ASTs.
2767                  */
2768                 mdt_object_get(mti->mti_env, o);
2769                 einfo->ei_cbdata = o;
2770         }
2771
2772
2773         memset(policy, 0, sizeof(*policy));
2774         policy->l_inodebits.bits = *ibits;
2775         policy->l_inodebits.try_bits = trybits;
2776
2777         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2778                             policy);
2779         if (rc < 0 && cache) {
2780                 mdt_object_put(mti->mti_env, o);
2781                 einfo->ei_cbdata = NULL;
2782         }
2783
2784         /* Return successfully acquired bits to a caller */
2785         if (rc == 0) {
2786                 struct ldlm_lock *lock = ldlm_handle2lock(lh);
2787
2788                 LASSERT(lock);
2789                 *ibits = lock->l_policy_data.l_inodebits.bits;
2790                 LDLM_LOCK_PUT(lock);
2791         }
2792         RETURN(rc);
2793 }
2794
2795 int mdt_remote_object_lock(struct mdt_thread_info *mti, struct mdt_object *o,
2796                            const struct lu_fid *fid, struct lustre_handle *lh,
2797                            enum ldlm_mode mode, __u64 ibits, bool cache)
2798 {
2799         return mdt_remote_object_lock_try(mti, o, fid, lh, mode, &ibits, 0,
2800                                           cache);
2801 }
2802
2803 static int mdt_object_local_lock(struct mdt_thread_info *info,
2804                                  struct mdt_object *o,
2805                                  struct mdt_lock_handle *lh, __u64 *ibits,
2806                                  __u64 trybits, bool cos_incompat)
2807 {
2808         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2809         union ldlm_policy_data *policy = &info->mti_policy;
2810         struct ldlm_res_id *res_id = &info->mti_res_id;
2811         __u64 dlmflags = 0;
2812         int rc;
2813         ENTRY;
2814
2815         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2816         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2817         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2818         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2819
2820         if (cos_incompat) {
2821                 LASSERT(lh->mlh_reg_mode == LCK_PW ||
2822                         lh->mlh_reg_mode == LCK_EX);
2823                 dlmflags |= LDLM_FL_COS_INCOMPAT;
2824         } else if (mdt_cos_is_enabled(info->mti_mdt)) {
2825                 dlmflags |= LDLM_FL_COS_ENABLED;
2826         }
2827
2828         /* Only enqueue LOOKUP lock for remote object */
2829         if (mdt_object_remote(o)) {
2830                 LASSERT(*ibits == MDS_INODELOCK_LOOKUP);
2831         }
2832
2833         if (lh->mlh_type == MDT_PDO_LOCK) {
2834                 /* check for exists after object is locked */
2835                 if (mdt_object_exists(o) == 0) {
2836                         /* Non-existent object shouldn't have PDO lock */
2837                         RETURN(-ESTALE);
2838                 } else {
2839                         /* Non-dir object shouldn't have PDO lock */
2840                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2841                                 RETURN(-ENOTDIR);
2842                 }
2843         }
2844
2845
2846         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2847         dlmflags |= LDLM_FL_ATOMIC_CB;
2848
2849         /*
2850          * Take PDO lock on whole directory and build correct @res_id for lock
2851          * on part of directory.
2852          */
2853         if (lh->mlh_pdo_hash != 0) {
2854                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2855                 mdt_lock_pdo_mode(info, o, lh);
2856                 if (lh->mlh_pdo_mode != LCK_NL) {
2857                         /*
2858                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2859                          * is never going to be sent to client and we do not
2860                          * want it slowed down due to possible cancels.
2861                          */
2862                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2863                         policy->l_inodebits.try_bits = 0;
2864                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2865                                           policy, res_id, dlmflags,
2866                                           info->mti_exp == NULL ? NULL :
2867                                           &info->mti_exp->exp_handle.h_cookie);
2868                         if (unlikely(rc != 0))
2869                                 GOTO(out_unlock, rc);
2870                 }
2871
2872                 /*
2873                  * Finish res_id initializing by name hash marking part of
2874                  * directory which is taking modification.
2875                  */
2876                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2877         }
2878
2879         policy->l_inodebits.bits = *ibits;
2880         policy->l_inodebits.try_bits = trybits;
2881
2882         /*
2883          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2884          * going to be sent to client. If it is - mdt_intent_policy() path will
2885          * fix it up and turn FL_LOCAL flag off.
2886          */
2887         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2888                           res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2889                           info->mti_exp == NULL ? NULL :
2890                           &info->mti_exp->exp_handle.h_cookie);
2891 out_unlock:
2892         if (rc != 0)
2893                 mdt_object_unlock(info, o, lh, 1);
2894         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2895                    lh->mlh_pdo_hash != 0 &&
2896                    (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
2897                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2898
2899         /* Return successfully acquired bits to a caller */
2900         if (rc == 0) {
2901                 struct ldlm_lock *lock = ldlm_handle2lock(&lh->mlh_reg_lh);
2902
2903                 LASSERT(lock);
2904                 *ibits = lock->l_policy_data.l_inodebits.bits;
2905                 LDLM_LOCK_PUT(lock);
2906         }
2907         RETURN(rc);
2908 }
2909
2910 static int
2911 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
2912                          struct mdt_lock_handle *lh, __u64 *ibits,
2913                          __u64 trybits, bool cos_incompat)
2914 {
2915         struct mdt_lock_handle *local_lh = NULL;
2916         int rc;
2917         ENTRY;
2918
2919         if (!mdt_object_remote(o)) {
2920                 rc = mdt_object_local_lock(info, o, lh, ibits, trybits,
2921                                            cos_incompat);
2922                 RETURN(rc);
2923         }
2924
2925         /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
2926         *ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
2927                     MDS_INODELOCK_XATTR);
2928
2929         /* Only enqueue LOOKUP lock for remote object */
2930         if (*ibits & MDS_INODELOCK_LOOKUP) {
2931                 __u64 local = MDS_INODELOCK_LOOKUP;
2932
2933                 rc = mdt_object_local_lock(info, o, lh, &local, 0,
2934                                            cos_incompat);
2935                 if (rc != ELDLM_OK)
2936                         RETURN(rc);
2937
2938                 local_lh = lh;
2939         }
2940
2941         if ((*ibits | trybits) & MDS_INODELOCK_UPDATE) {
2942                 /* Sigh, PDO needs to enqueue 2 locks right now, but
2943                  * enqueue RPC can only request 1 lock, to avoid extra
2944                  * RPC, so it will instead enqueue EX lock for remote
2945                  * object anyway XXX*/
2946                 if (lh->mlh_type == MDT_PDO_LOCK &&
2947                     lh->mlh_pdo_hash != 0) {
2948                         CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
2949                                "EX lock.\n", mdt_obd_name(info->mti_mdt),
2950                                PFID(mdt_object_fid(o)));
2951                         lh->mlh_pdo_hash = 0;
2952                         lh->mlh_rreg_mode = LCK_EX;
2953                         lh->mlh_type = MDT_REG_LOCK;
2954                 }
2955
2956                 rc = mdt_remote_object_lock_try(info, o, mdt_object_fid(o),
2957                                                 &lh->mlh_rreg_lh,
2958                                                 lh->mlh_rreg_mode,
2959                                                 ibits, trybits, false);
2960                 if (rc != ELDLM_OK) {
2961                         if (local_lh != NULL)
2962                                 mdt_object_unlock(info, o, local_lh, rc);
2963                         RETURN(rc);
2964                 }
2965         }
2966
2967         RETURN(0);
2968 }
2969
2970 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2971                     struct mdt_lock_handle *lh, __u64 ibits)
2972 {
2973         return mdt_object_lock_internal(info, o, lh, &ibits, 0, false);
2974 }
2975
2976 int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2977                           struct mdt_lock_handle *lh, __u64 ibits,
2978                           bool cos_incompat)
2979 {
2980         LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
2981         return mdt_object_lock_internal(info, o, lh, &ibits, 0,
2982                                         cos_incompat);
2983 }
2984
2985 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
2986                         struct mdt_lock_handle *lh, __u64 *ibits,
2987                         __u64 trybits, bool cos_incompat)
2988 {
2989         return mdt_object_lock_internal(info, o, lh, ibits, trybits,
2990                                         cos_incompat);
2991 }
2992
2993 /**
2994  * Save a lock within request object.
2995  *
2996  * Keep the lock referenced until whether client ACK or transaction
2997  * commit happens or release the lock immediately depending on input
2998  * parameters. If COS is ON, a write lock is converted to COS lock
2999  * before saving.
3000  *
3001  * \param info thead info object
3002  * \param h lock handle
3003  * \param mode lock mode
3004  * \param decref force immediate lock releasing
3005  */
3006 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
3007                    enum ldlm_mode mode, int decref)
3008 {
3009         ENTRY;
3010
3011         if (lustre_handle_is_used(h)) {
3012                 if (decref || !info->mti_has_trans ||
3013                     !(mode & (LCK_PW | LCK_EX))) {
3014                         mdt_fid_unlock(h, mode);
3015                 } else {
3016                         struct mdt_device *mdt = info->mti_mdt;
3017                         struct ldlm_lock *lock = ldlm_handle2lock(h);
3018                         struct ptlrpc_request *req = mdt_info_req(info);
3019                         bool cos = mdt_cos_is_enabled(mdt);
3020                         bool convert_lock = !cos && mdt_slc_is_enabled(mdt);
3021
3022                         LASSERTF(lock != NULL, "no lock for cookie %#llx\n",
3023                                  h->cookie);
3024
3025                         /* there is no request if mdt_object_unlock() is called
3026                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
3027                         if (likely(req != NULL)) {
3028                                 LDLM_DEBUG(lock, "save lock request %p reply "
3029                                         "state %p transno %lld\n", req,
3030                                         req->rq_reply_state, req->rq_transno);
3031                                 if (cos) {
3032                                         ldlm_lock_downgrade(lock, LCK_COS);
3033                                         mode = LCK_COS;
3034                                 }
3035                                 if (req->rq_export->exp_disconnected)
3036                                         mdt_fid_unlock(h, mode);
3037                                 else
3038                                         ptlrpc_save_lock(req, h, mode, cos,
3039                                                          convert_lock);
3040                         } else {
3041                                 mdt_fid_unlock(h, mode);
3042                         }
3043                         if (mdt_is_lock_sync(lock)) {
3044                                 CDEBUG(D_HA, "found sync-lock,"
3045                                        " async commit started\n");
3046                                 mdt_device_commit_async(info->mti_env,
3047                                                         mdt);
3048                         }
3049                         LDLM_LOCK_PUT(lock);
3050                 }
3051                 h->cookie = 0ull;
3052         }
3053
3054         EXIT;
3055 }
3056
3057 /**
3058  * Save cross-MDT lock in uncommitted_slc_locks
3059  *
3060  * Keep the lock referenced until transaction commit happens or release the lock
3061  * immediately depending on input parameters.
3062  *
3063  * \param info thead info object
3064  * \param h lock handle
3065  * \param mode lock mode
3066  * \param decref force immediate lock releasing
3067  */
3068 static void mdt_save_remote_lock(struct mdt_thread_info *info,
3069                                  struct mdt_object *o, struct lustre_handle *h,
3070                                  enum ldlm_mode mode, int decref)
3071 {
3072         ENTRY;
3073
3074         if (lustre_handle_is_used(h)) {
3075                 struct ldlm_lock *lock = ldlm_handle2lock(h);
3076
3077                 if (o != NULL &&
3078                     (lock->l_policy_data.l_inodebits.bits &
3079                      (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)))
3080                         mo_invalidate(info->mti_env, mdt_object_child(o));
3081
3082                 if (decref || !info->mti_has_trans ||
3083                     !(mode & (LCK_PW | LCK_EX))) {
3084                         ldlm_lock_decref_and_cancel(h, mode);
3085                         LDLM_LOCK_PUT(lock);
3086                 } else {
3087                         struct ptlrpc_request *req = mdt_info_req(info);
3088
3089                         LASSERT(req != NULL);
3090                         tgt_save_slc_lock(&info->mti_mdt->mdt_lut, lock,
3091                                           req->rq_transno);
3092                         ldlm_lock_decref(h, mode);
3093                 }
3094                 h->cookie = 0ull;
3095         }
3096
3097         EXIT;
3098 }
3099
3100 /**
3101  * Unlock mdt object.
3102  *
3103  * Immeditely release the regular lock and the PDO lock or save the
3104  * lock in request and keep them referenced until client ACK or
3105  * transaction commit.
3106  *
3107  * \param info thread info object
3108  * \param o mdt object
3109  * \param lh mdt lock handle referencing regular and PDO locks
3110  * \param decref force immediate lock releasing
3111  *
3112  * XXX o is not used and may be NULL, see hsm_cdt_request_completed().
3113  */
3114 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
3115                        struct mdt_lock_handle *lh, int decref)
3116 {
3117         ENTRY;
3118
3119         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
3120         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
3121         mdt_save_remote_lock(info, o, &lh->mlh_rreg_lh, lh->mlh_rreg_mode,
3122                              decref);
3123
3124         EXIT;
3125 }
3126
3127 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
3128                                         const struct lu_fid *f,
3129                                         struct mdt_lock_handle *lh,
3130                                         __u64 ibits)
3131 {
3132         struct mdt_object *o;
3133
3134         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
3135         if (!IS_ERR(o)) {
3136                 int rc;
3137
3138                 rc = mdt_object_lock(info, o, lh, ibits);
3139                 if (rc != 0) {
3140                         mdt_object_put(info->mti_env, o);
3141                         o = ERR_PTR(rc);
3142                 }
3143         }
3144         return o;
3145 }
3146
3147 void mdt_object_unlock_put(struct mdt_thread_info * info,
3148                            struct mdt_object * o,
3149                            struct mdt_lock_handle *lh,
3150                            int decref)
3151 {
3152         mdt_object_unlock(info, o, lh, decref);
3153         mdt_object_put(info->mti_env, o);
3154 }
3155
3156 /*
3157  * Generic code handling requests that have struct mdt_body passed in:
3158  *
3159  *  - extract mdt_body from request and save it in @info, if present;
3160  *
3161  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
3162  *  @info;
3163  *
3164  *  - if HABEO_CORPUS flag is set for this request type check whether object
3165  *  actually exists on storage (lu_object_exists()).
3166  *
3167  */
3168 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
3169 {
3170         const struct mdt_body    *body;
3171         struct mdt_object        *obj;
3172         const struct lu_env      *env;
3173         struct req_capsule       *pill;
3174         int                       rc;
3175         ENTRY;
3176
3177         env = info->mti_env;
3178         pill = info->mti_pill;
3179
3180         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
3181         if (body == NULL)
3182                 RETURN(-EFAULT);
3183
3184         if (!(body->mbo_valid & OBD_MD_FLID))
3185                 RETURN(0);
3186
3187         if (!fid_is_sane(&body->mbo_fid1)) {
3188                 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
3189                 RETURN(-EINVAL);
3190         }
3191
3192         obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
3193         if (!IS_ERR(obj)) {
3194                 if ((flags & HABEO_CORPUS) && !mdt_object_exists(obj)) {
3195                         mdt_object_put(env, obj);
3196                         rc = -ENOENT;
3197                 } else {
3198                         info->mti_object = obj;
3199                         rc = 0;
3200                 }
3201         } else
3202                 rc = PTR_ERR(obj);
3203
3204         RETURN(rc);
3205 }
3206
3207 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
3208 {
3209         struct req_capsule *pill = info->mti_pill;
3210         int rc;
3211         ENTRY;
3212
3213         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
3214                 rc = mdt_body_unpack(info, flags);
3215         else
3216                 rc = 0;
3217
3218         if (rc == 0 && (flags & HABEO_REFERO)) {
3219                 /* Pack reply. */
3220                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
3221                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
3222                                              DEF_REP_MD_SIZE);
3223                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
3224                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
3225                                              RCL_SERVER, 0);
3226
3227                 /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
3228                  * by default. If the target object has more ACL entries, then
3229                  * enlarge the buffer when necessary. */
3230                 if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
3231                         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
3232                                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
3233
3234                 rc = req_capsule_server_pack(pill);
3235         }
3236         RETURN(rc);
3237 }
3238
3239 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
3240 {
3241         lh->mlh_type = MDT_NUL_LOCK;
3242         lh->mlh_reg_lh.cookie = 0ull;
3243         lh->mlh_reg_mode = LCK_MINMODE;
3244         lh->mlh_pdo_lh.cookie = 0ull;
3245         lh->mlh_pdo_mode = LCK_MINMODE;
3246         lh->mlh_rreg_lh.cookie = 0ull;
3247         lh->mlh_rreg_mode = LCK_MINMODE;
3248 }
3249
3250 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
3251 {
3252         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
3253         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
3254 }
3255
3256 /*
3257  * Initialize fields of struct mdt_thread_info. Other fields are left in
3258  * uninitialized state, because it's too expensive to zero out whole
3259  * mdt_thread_info (> 1K) on each request arrival.
3260  */
3261 void mdt_thread_info_init(struct ptlrpc_request *req,
3262                           struct mdt_thread_info *info)
3263 {
3264         int i;
3265
3266         info->mti_pill = &req->rq_pill;
3267
3268         /* lock handle */
3269         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3270                 mdt_lock_handle_init(&info->mti_lh[i]);
3271
3272         /* mdt device: it can be NULL while CONNECT */
3273         if (req->rq_export) {
3274                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
3275                 info->mti_exp = req->rq_export;
3276         } else
3277                 info->mti_mdt = NULL;
3278         info->mti_env = req->rq_svc_thread->t_env;
3279         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
3280
3281         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
3282         info->mti_big_buf = LU_BUF_NULL;
3283         info->mti_body = NULL;
3284         info->mti_object = NULL;
3285         info->mti_dlm_req = NULL;
3286         info->mti_has_trans = 0;
3287         info->mti_cross_ref = 0;
3288         info->mti_opdata = 0;
3289         info->mti_big_lmm_used = 0;
3290         info->mti_big_acl_used = 0;
3291         info->mti_som_valid = 0;
3292
3293         info->mti_spec.no_create = 0;
3294         info->mti_spec.sp_rm_entry = 0;
3295         info->mti_spec.sp_permitted = 0;
3296         info->mti_spec.sp_migrate_close = 0;
3297
3298         info->mti_spec.u.sp_ea.eadata = NULL;
3299         info->mti_spec.u.sp_ea.eadatalen = 0;
3300 }
3301
3302 void mdt_thread_info_fini(struct mdt_thread_info *info)
3303 {
3304         int i;
3305
3306         if (info->mti_object != NULL) {
3307                 mdt_object_put(info->mti_env, info->mti_object);
3308                 info->mti_object = NULL;
3309         }
3310
3311         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3312                 mdt_lock_handle_fini(&info->mti_lh[i]);
3313         info->mti_env = NULL;
3314         info->mti_pill = NULL;
3315         info->mti_exp = NULL;
3316
3317         if (unlikely(info->mti_big_buf.lb_buf != NULL))
3318                 lu_buf_free(&info->mti_big_buf);
3319 }
3320
3321 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
3322 {
3323         struct mdt_thread_info  *mti;
3324
3325         mti = mdt_th_info(tsi->tsi_env);
3326         LASSERT(mti != NULL);
3327
3328         mdt_thread_info_init(tgt_ses_req(tsi), mti);
3329         if (tsi->tsi_corpus != NULL) {
3330                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
3331                 lu_object_get(tsi->tsi_corpus);
3332         }
3333         mti->mti_body = tsi->tsi_mdt_body;
3334         mti->mti_dlm_req = tsi->tsi_dlm_req;
3335
3336         return mti;
3337 }
3338
3339 static int mdt_tgt_connect(struct tgt_session_info *tsi)
3340 {
3341         if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
3342             cfs_fail_val ==
3343             tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id) {
3344                 set_current_state(TASK_UNINTERRUPTIBLE);
3345                 schedule_timeout(msecs_to_jiffies(3 * MSEC_PER_SEC));
3346         }
3347
3348         return tgt_connect(tsi);
3349 }
3350
3351 enum mdt_it_code {
3352         MDT_IT_OPEN,
3353         MDT_IT_OCREAT,
3354         MDT_IT_CREATE,
3355         MDT_IT_GETATTR,
3356         MDT_IT_READDIR,
3357         MDT_IT_LOOKUP,
3358         MDT_IT_UNLINK,
3359         MDT_IT_TRUNC,
3360         MDT_IT_GETXATTR,
3361         MDT_IT_LAYOUT,
3362         MDT_IT_QUOTA,
3363         MDT_IT_GLIMPSE,
3364         MDT_IT_BRW,
3365         MDT_IT_NR
3366 };
3367
3368 static int mdt_intent_getattr(enum mdt_it_code opcode,
3369                               struct mdt_thread_info *info,
3370                               struct ldlm_lock **, __u64);
3371
3372 static int mdt_intent_getxattr(enum mdt_it_code opcode,
3373                                 struct mdt_thread_info *info,
3374                                 struct ldlm_lock **lockp,
3375                                 __u64 flags);
3376
3377 static int mdt_intent_layout(enum mdt_it_code opcode,
3378                              struct mdt_thread_info *info,
3379                              struct ldlm_lock **,
3380                              __u64);
3381 static int mdt_intent_reint(enum mdt_it_code opcode,
3382                             struct mdt_thread_info *info,
3383                             struct ldlm_lock **,
3384                             __u64);
3385 static int mdt_intent_glimpse(enum mdt_it_code opcode,
3386                               struct mdt_thread_info *info,
3387                               struct ldlm_lock **lockp, __u64 flags)
3388 {
3389         return mdt_glimpse_enqueue(info, info->mti_mdt->mdt_namespace,
3390                                    lockp, flags);
3391 }
3392 static int mdt_intent_brw(enum mdt_it_code opcode,
3393                           struct mdt_thread_info *info,
3394                           struct ldlm_lock **lockp, __u64 flags)
3395 {
3396         return mdt_brw_enqueue(info, info->mti_mdt->mdt_namespace,
3397                                lockp, flags);
3398 }
3399
3400 static struct mdt_it_flavor {
3401         const struct req_format *it_fmt;
3402         __u32                    it_flags;
3403         int                    (*it_act)(enum mdt_it_code ,
3404                                          struct mdt_thread_info *,
3405                                          struct ldlm_lock **,
3406                                          __u64);
3407         long                     it_reint;
3408 } mdt_it_flavor[] = {
3409         [MDT_IT_OPEN]     = {
3410                 .it_fmt   = &RQF_LDLM_INTENT,
3411                 /*.it_flags = HABEO_REFERO,*/
3412                 .it_flags = 0,
3413                 .it_act   = mdt_intent_reint,
3414                 .it_reint = REINT_OPEN
3415         },
3416         [MDT_IT_OCREAT]   = {
3417                 .it_fmt   = &RQF_LDLM_INTENT,
3418                 /*
3419                  * OCREAT is not a MUTABOR request as if the file
3420                  * already exists.
3421                  * We do the extra check of OBD_CONNECT_RDONLY in
3422                  * mdt_reint_open() when we really need to create
3423                  * the object.
3424                  */
3425                 .it_flags = 0,
3426                 .it_act   = mdt_intent_reint,
3427                 .it_reint = REINT_OPEN
3428         },
3429         [MDT_IT_CREATE]   = {
3430                 .it_fmt   = &RQF_LDLM_INTENT,
3431                 .it_flags = MUTABOR,
3432                 .it_act   = mdt_intent_reint,
3433                 .it_reint = REINT_CREATE
3434         },
3435         [MDT_IT_GETATTR]  = {
3436                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3437                 .it_flags = HABEO_REFERO,
3438                 .it_act   = mdt_intent_getattr
3439         },
3440         [MDT_IT_READDIR]  = {
3441                 .it_fmt   = NULL,
3442                 .it_flags = 0,
3443                 .it_act   = NULL
3444         },
3445         [MDT_IT_LOOKUP]   = {
3446                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3447                 .it_flags = HABEO_REFERO,
3448                 .it_act   = mdt_intent_getattr
3449         },
3450         [MDT_IT_UNLINK]   = {
3451                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
3452                 .it_flags = MUTABOR,
3453                 .it_act   = NULL,
3454                 .it_reint = REINT_UNLINK
3455         },
3456         [MDT_IT_TRUNC]    = {
3457                 .it_fmt   = NULL,
3458                 .it_flags = MUTABOR,
3459                 .it_act   = NULL
3460         },
3461         [MDT_IT_GETXATTR] = {
3462                 .it_fmt   = &RQF_LDLM_INTENT_GETXATTR,
3463                 .it_flags = HABEO_CORPUS,
3464                 .it_act   = mdt_intent_getxattr
3465         },
3466         [MDT_IT_LAYOUT] = {
3467                 .it_fmt   = &RQF_LDLM_INTENT_LAYOUT,
3468                 .it_flags = 0,
3469                 .it_act   = mdt_intent_layout
3470         },
3471         [MDT_IT_GLIMPSE] = {
3472                 .it_fmt = &RQF_LDLM_INTENT,
3473                 .it_flags = 0,
3474                 .it_act = mdt_intent_glimpse,
3475         },
3476         [MDT_IT_BRW] = {
3477                 .it_fmt = &RQF_LDLM_INTENT,
3478                 .it_flags = 0,
3479                 .it_act = mdt_intent_brw,
3480         },
3481
3482 };
3483
3484 int mdt_intent_lock_replace(struct mdt_thread_info *info,
3485                             struct ldlm_lock **lockp,
3486                             struct mdt_lock_handle *lh,
3487                             __u64 flags, int result)
3488 {
3489         struct ptlrpc_request  *req = mdt_info_req(info);
3490         struct ldlm_lock       *lock = *lockp;
3491         struct ldlm_lock       *new_lock;
3492
3493         /* If possible resent found a lock, @lh is set to its handle */
3494         new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3495
3496         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3497                 lh->mlh_reg_lh.cookie = 0;
3498                 RETURN(0);
3499         }
3500
3501         if (new_lock == NULL && (flags & LDLM_FL_RESENT)) {
3502                 /* Lock is pinned by ldlm_handle_enqueue0() as it is
3503                  * a resend case, however, it could be already destroyed
3504                  * due to client eviction or a raced cancel RPC. */
3505                 LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx\n",
3506                                   lh->mlh_reg_lh.cookie);
3507                 lh->mlh_reg_lh.cookie = 0;
3508                 RETURN(-ESTALE);
3509         }
3510
3511         LASSERTF(new_lock != NULL,
3512                  "lockh %#llx flags %#llx : rc = %d\n",
3513                  lh->mlh_reg_lh.cookie, flags, result);
3514
3515         /*
3516          * If we've already given this lock to a client once, then we should
3517          * have no readers or writers.  Otherwise, we should have one reader
3518          * _or_ writer ref (which will be zeroed below) before returning the
3519          * lock to a client.
3520          */
3521         if (new_lock->l_export == req->rq_export) {
3522                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3523         } else {
3524                 LASSERT(new_lock->l_export == NULL);
3525                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3526         }
3527
3528         *lockp = new_lock;
3529
3530         if (new_lock->l_export == req->rq_export) {
3531                 /*
3532                  * Already gave this to the client, which means that we
3533                  * reconstructed a reply.
3534                  */
3535                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3536                         MSG_RESENT);
3537
3538                 LDLM_LOCK_RELEASE(new_lock);
3539                 lh->mlh_reg_lh.cookie = 0;
3540                 RETURN(ELDLM_LOCK_REPLACED);
3541         }
3542
3543         /*
3544          * Fixup the lock to be given to the client.
3545          */
3546         lock_res_and_lock(new_lock);
3547         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3548          * possible blocking AST. */
3549         while (new_lock->l_readers > 0) {
3550                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3551                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3552                 new_lock->l_readers--;
3553         }
3554         while (new_lock->l_writers > 0) {
3555                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3556                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3557                 new_lock->l_writers--;
3558         }
3559
3560         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3561         new_lock->l_blocking_ast = lock->l_blocking_ast;
3562         new_lock->l_completion_ast = lock->l_completion_ast;
3563         if (ldlm_has_dom(new_lock))
3564                 new_lock->l_glimpse_ast = ldlm_server_glimpse_ast;
3565         new_lock->l_remote_handle = lock->l_remote_handle;
3566         new_lock->l_flags &= ~LDLM_FL_LOCAL;
3567
3568         unlock_res_and_lock(new_lock);
3569
3570         cfs_hash_add(new_lock->l_export->exp_lock_hash,
3571                      &new_lock->l_remote_handle,
3572                      &new_lock->l_exp_hash);
3573
3574         LDLM_LOCK_RELEASE(new_lock);
3575         lh->mlh_reg_lh.cookie = 0;
3576
3577         RETURN(ELDLM_LOCK_REPLACED);
3578 }
3579
3580 void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3581                              struct ldlm_lock *new_lock,
3582                              struct mdt_lock_handle *lh, __u64 flags)
3583 {
3584         struct ptlrpc_request  *req = mdt_info_req(info);
3585         struct ldlm_request    *dlmreq;
3586
3587         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3588                 return;
3589
3590         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3591
3592         /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
3593          * lock was found by ldlm_handle_enqueue(); if so @lh must be
3594          * initialized. */
3595         if (flags & LDLM_FL_RESENT) {
3596                 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
3597                 lh->mlh_reg_mode = new_lock->l_granted_mode;
3598
3599                 LDLM_DEBUG(new_lock, "Restoring lock cookie");
3600                 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie %#llx",
3601                           lh->mlh_reg_lh.cookie);
3602                 return;
3603         }
3604
3605         /*
3606          * If the xid matches, then we know this is a resent request, and allow
3607          * it. (It's probably an OPEN, for which we don't send a lock.
3608          */
3609         if (req_can_reconstruct(req, NULL))
3610                 return;
3611
3612         /*
3613          * This remote handle isn't enqueued, so we never received or processed
3614          * this request.  Clear MSG_RESENT, because it can be handled like any
3615          * normal request now.
3616          */
3617         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3618
3619         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle %#llx",
3620                   dlmreq->lock_handle[0].cookie);
3621 }
3622
3623 static int mdt_intent_getxattr(enum mdt_it_code opcode,
3624                                 struct mdt_thread_info *info,
3625                                 struct ldlm_lock **lockp,
3626                                 __u64 flags)
3627 {
3628         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3629         struct ldlm_reply      *ldlm_rep = NULL;
3630         int rc;
3631         ENTRY;
3632
3633         /*
3634          * Initialize lhc->mlh_reg_lh either from a previously granted lock
3635          * (for the resend case) or a new lock. Below we will use it to
3636          * replace the original lock.
3637          */
3638         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3639         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3640                 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3641                 rc = mdt_object_lock(info, info->mti_object, lhc,
3642                                      MDS_INODELOCK_XATTR);
3643                 if (rc)
3644                         return rc;
3645         }
3646
3647         rc = mdt_getxattr(info);
3648
3649         if (mdt_info_req(info)->rq_repmsg != NULL)
3650                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3651
3652         if (ldlm_rep == NULL ||
3653             OBD_FAIL_CHECK(OBD_FAIL_MDS_XATTR_REP)) {
3654                 mdt_object_unlock(info,  info->mti_object, lhc, 1);
3655                 RETURN(err_serious(-EFAULT));
3656         }
3657
3658         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3659
3660         /* This is left for interop instead of adding a new interop flag.
3661          * LU-7433 */
3662 #if LUSTRE_VERSION_CODE > OBD_OCD_VERSION(3, 0, 0, 0)
3663         if (ldlm_rep->lock_policy_res2) {
3664                 mdt_object_unlock(info, info->mti_object, lhc, 1);
3665                 RETURN(ELDLM_LOCK_ABORTED);
3666         }
3667 #endif
3668
3669         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3670         RETURN(rc);
3671 }
3672
3673 static int mdt_intent_getattr(enum mdt_it_code opcode,
3674                               struct mdt_thread_info *info,
3675                               struct ldlm_lock **lockp,
3676                               __u64 flags)
3677 {
3678         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3679         __u64                   child_bits;
3680         struct ldlm_reply      *ldlm_rep;
3681         struct mdt_body        *reqbody;
3682         struct mdt_body        *repbody;
3683         int                     rc, rc2;
3684         ENTRY;
3685
3686         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3687         LASSERT(reqbody);
3688
3689         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3690         LASSERT(repbody);
3691
3692         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
3693         repbody->mbo_eadatasize = 0;
3694         repbody->mbo_aclsize = 0;
3695
3696         switch (opcode) {
3697         case MDT_IT_LOOKUP:
3698                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3699                 break;
3700         case MDT_IT_GETATTR:
3701                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3702                              MDS_INODELOCK_PERM;
3703                 break;
3704         default:
3705                 CERROR("Unsupported intent (%d)\n", opcode);
3706                 GOTO(out_shrink, rc = -EINVAL);
3707         }
3708
3709         rc = mdt_init_ucred_intent_getattr(info, reqbody);
3710         if (rc)
3711                 GOTO(out_shrink, rc);
3712
3713         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3714         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3715
3716         /* Get lock from request for possible resent case. */
3717         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3718
3719         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3720         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3721
3722         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3723                 ldlm_rep->lock_policy_res2 = 0;
3724         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3725             ldlm_rep->lock_policy_res2) {
3726                 lhc->mlh_reg_lh.cookie = 0ull;
3727                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3728         }
3729
3730         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3731         EXIT;
3732 out_ucred:
3733         mdt_exit_ucred(info);
3734 out_shrink:
3735         mdt_client_compatibility(info);
3736         rc2 = mdt_fix_reply(info);
3737         if (rc == 0)
3738                 rc = rc2;
3739         return rc;
3740 }
3741
3742 static int mdt_intent_layout(enum mdt_it_code opcode,
3743                              struct mdt_thread_info *info,
3744                              struct ldlm_lock **lockp,
3745                              __u64 flags)
3746 {
3747         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_LAYOUT];
3748         struct layout_intent *layout;
3749         struct lu_fid *fid;
3750         struct mdt_object *obj = NULL;
3751         bool layout_change = false;
3752         int layout_size = 0;
3753         int rc = 0;
3754         ENTRY;
3755
3756         if (opcode != MDT_IT_LAYOUT) {
3757                 CERROR("%s: Unknown intent (%d)\n", mdt_obd_name(info->mti_mdt),
3758                         opcode);
3759                 RETURN(-EINVAL);
3760         }
3761
3762         layout = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3763         if (layout == NULL)
3764                 RETURN(-EPROTO);
3765
3766         switch (layout->li_opc) {
3767         case LAYOUT_INTENT_TRUNC:
3768         case LAYOUT_INTENT_WRITE:
3769                 layout_change = true;
3770                 break;
3771         case LAYOUT_INTENT_ACCESS:
3772                 break;
3773         case LAYOUT_INTENT_READ:
3774         case LAYOUT_INTENT_GLIMPSE:
3775         case LAYOUT_INTENT_RELEASE:
3776         case LAYOUT_INTENT_RESTORE:
3777                 CERROR("%s: Unsupported layout intent opc %d\n",
3778                        mdt_obd_name(info->mti_mdt), layout->li_opc);
3779                 rc = -ENOTSUPP;
3780                 break;
3781         default:
3782                 CERROR("%s: Unknown layout intent opc %d\n",
3783                        mdt_obd_name(info->mti_mdt), layout->li_opc);
3784                 rc = -EINVAL;
3785                 break;
3786         }
3787         if (rc < 0)
3788                 RETURN(rc);
3789
3790         fid = &info->mti_tmp_fid2;
3791         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3792
3793         /* Get lock from request for possible resent case. */
3794         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3795
3796         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3797         if (IS_ERR(obj))
3798                 GOTO(out, rc = PTR_ERR(obj));
3799
3800         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3801                 layout_size = mdt_attr_get_eabuf_size(info, obj);
3802                 if (layout_size < 0)
3803                         GOTO(out_obj, rc = layout_size);
3804
3805                 if (layout_size > info->mti_mdt->mdt_max_mdsize)
3806                         info->mti_mdt->mdt_max_mdsize = layout_size;
3807         }
3808
3809         /*
3810          * set reply buffer size, so that ldlm_handle_enqueue0()->
3811          * ldlm_lvbo_fill() will fill the reply buffer with lovea.
3812          */
3813         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3814         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
3815                              layout_size);
3816         rc = req_capsule_server_pack(info->mti_pill);
3817         if (rc)
3818                 GOTO(out_obj, rc);
3819
3820
3821         if (layout_change) {
3822                 struct lu_buf *buf = &info->mti_buf;
3823
3824                 /**
3825                  * mdt_layout_change is a reint operation, when the request
3826                  * is resent, layout write shouldn't reprocess it again.
3827                  */
3828                 rc = mdt_check_resent(info, mdt_reconstruct_generic, lhc);
3829                 if (rc)
3830                         GOTO(out_obj, rc = rc < 0 ? rc : 0);
3831
3832                 /**
3833                  * There is another resent case: the client's job has been
3834                  * done by another client, referring lod_declare_layout_change
3835                  * -EALREADY case, and it became a operation w/o transaction,
3836                  * so we should not do the layout change, otherwise
3837                  * mdt_layout_change() will try to cancel the granted server
3838                  * CR lock whose remote counterpart is still in hold on the
3839                  * client, and a deadlock ensues.
3840                  */
3841                 rc = mdt_check_resent_lock(info, obj, lhc);
3842                 if (rc <= 0)
3843                         GOTO(out_obj, rc);
3844
3845                 buf->lb_buf = NULL;
3846                 buf->lb_len = 0;
3847                 if (unlikely(req_is_replay(mdt_info_req(info)))) {
3848                         buf->lb_buf = req_capsule_client_get(info->mti_pill,
3849                                         &RMF_EADATA);
3850                         buf->lb_len = req_capsule_get_size(info->mti_pill,
3851                                         &RMF_EADATA, RCL_CLIENT);
3852                         /*
3853                          * If it's a replay of layout write intent RPC, the
3854                          * client has saved the extended lovea when
3855                          * it get reply then.
3856                          */
3857                         if (buf->lb_len > 0)
3858                                 mdt_fix_lov_magic(info, buf->lb_buf);
3859                 }
3860
3861                 /*
3862                  * Instantiate some layout components, if @buf contains
3863                  * lovea, then it's a replay of the layout intent write
3864                  * RPC.
3865                  */
3866                 rc = mdt_layout_change(info, obj, layout, buf);
3867                 if (rc)
3868                         GOTO(out_obj, rc);
3869         }
3870 out_obj:
3871         mdt_object_put(info->mti_env, obj);
3872
3873         if (rc == 0 && lustre_handle_is_used(&lhc->mlh_reg_lh))
3874                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3875
3876 out:
3877         lhc->mlh_reg_lh.cookie = 0;
3878
3879         return rc;
3880 }
3881
3882 static int mdt_intent_reint(enum mdt_it_code opcode,
3883                             struct mdt_thread_info *info,
3884                             struct ldlm_lock **lockp,
3885                             __u64 flags)
3886 {
3887         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3888         struct ldlm_reply      *rep = NULL;
3889         long                    opc;
3890         int                     rc;
3891
3892         static const struct req_format *intent_fmts[REINT_MAX] = {
3893                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3894                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3895         };
3896
3897         ENTRY;
3898
3899         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
3900         if (opc < 0)
3901                 RETURN(opc);
3902
3903         if (mdt_it_flavor[opcode].it_reint != opc) {
3904                 CERROR("Reint code %ld doesn't match intent: %d\n",
3905                        opc, opcode);
3906                 RETURN(err_serious(-EPROTO));
3907         }
3908
3909         /* Get lock from request for possible resent case. */
3910         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3911
3912         rc = mdt_reint_internal(info, lhc, opc);
3913
3914         /* Check whether the reply has been packed successfully. */
3915         if (mdt_info_req(info)->rq_repmsg != NULL)
3916                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3917         if (rep == NULL)
3918                 RETURN(err_serious(-EFAULT));
3919
3920         /* MDC expects this in any case */
3921         if (rc != 0)
3922                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3923
3924         /* the open lock or the lock for cross-ref object should be
3925          * returned to the client */
3926         if (lustre_handle_is_used(&lhc->mlh_reg_lh) &&
3927             (rc == 0 || rc == -MDT_EREMOTE_OPEN)) {
3928                 rep->lock_policy_res2 = 0;
3929                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3930                 RETURN(rc);
3931         }
3932
3933         rep->lock_policy_res2 = clear_serious(rc);
3934
3935         if (rep->lock_policy_res2 == -ENOENT &&
3936             mdt_get_disposition(rep, DISP_LOOKUP_NEG) &&
3937             !mdt_get_disposition(rep, DISP_OPEN_CREATE))
3938                 rep->lock_policy_res2 = 0;
3939
3940         lhc->mlh_reg_lh.cookie = 0ull;
3941         if (rc == -ENOTCONN || rc == -ENODEV ||
3942             rc == -EOVERFLOW) { /**< if VBR failure then return error */
3943                 /*
3944                  * If it is the disconnect error (ENODEV & ENOCONN), the error
3945                  * will be returned by rq_status, and client at ptlrpc layer
3946                  * will detect this, then disconnect, reconnect the import
3947                  * immediately, instead of impacting the following the rpc.
3948                  */
3949                 RETURN(rc);
3950         }
3951         /*
3952          * For other cases, the error will be returned by intent, and client
3953          * will retrieve the result from intent.
3954          */
3955         RETURN(ELDLM_LOCK_ABORTED);
3956 }
3957
3958 static int mdt_intent_code(enum ldlm_intent_flags itcode)
3959 {
3960         int rc;
3961
3962         switch (itcode) {
3963         case IT_OPEN:
3964                 rc = MDT_IT_OPEN;
3965                 break;
3966         case IT_OPEN|IT_CREAT:
3967                 rc = MDT_IT_OCREAT;
3968                 break;
3969         case IT_CREAT:
3970                 rc = MDT_IT_CREATE;
3971                 break;
3972         case IT_READDIR:
3973                 rc = MDT_IT_READDIR;
3974                 break;
3975         case IT_GETATTR:
3976                 rc = MDT_IT_GETATTR;
3977                 break;
3978         case IT_LOOKUP:
3979                 rc = MDT_IT_LOOKUP;
3980                 break;
3981         case IT_UNLINK:
3982                 rc = MDT_IT_UNLINK;
3983                 break;
3984         case IT_TRUNC:
3985                 rc = MDT_IT_TRUNC;
3986                 break;
3987         case IT_GETXATTR:
3988                 rc = MDT_IT_GETXATTR;
3989                 break;
3990         case IT_LAYOUT:
3991                 rc = MDT_IT_LAYOUT;
3992                 break;
3993         case IT_QUOTA_DQACQ:
3994         case IT_QUOTA_CONN:
3995                 rc = MDT_IT_QUOTA;
3996                 break;
3997         case IT_GLIMPSE:
3998                 rc = MDT_IT_GLIMPSE;
3999                 break;
4000         case IT_BRW:
4001                 rc = MDT_IT_BRW;
4002                 break;
4003         default:
4004                 CERROR("Unknown intent opcode: 0x%08x\n", itcode);
4005                 rc = -EINVAL;
4006                 break;
4007         }
4008         return rc;
4009 }
4010
4011 static int mdt_intent_opc(enum ldlm_intent_flags itopc,
4012                           struct mdt_thread_info *info,
4013                           struct ldlm_lock **lockp, __u64 flags)
4014 {
4015         struct req_capsule      *pill = info->mti_pill;
4016         struct ptlrpc_request   *req = mdt_info_req(info);
4017         struct mdt_it_flavor    *flv;
4018         int opc;
4019         int rc;
4020         ENTRY;
4021
4022         opc = mdt_intent_code(itopc);
4023         if (opc < 0)
4024                 RETURN(-EINVAL);
4025
4026         if (opc == MDT_IT_QUOTA) {
4027                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
4028
4029                 if (qmt == NULL)
4030                         RETURN(-EOPNOTSUPP);
4031
4032                 if (mdt_rdonly(req->rq_export))
4033                         RETURN(-EROFS);
4034
4035                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
4036                 /* pass the request to quota master */
4037                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
4038                                                  mdt_info_req(info), lockp,
4039                                                  flags);
4040                 RETURN(rc);
4041         }
4042
4043         flv = &mdt_it_flavor[opc];
4044         if (flv->it_fmt != NULL)
4045                 req_capsule_extend(pill, flv->it_fmt);
4046
4047         rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
4048         if (rc < 0)
4049                 RETURN(rc);
4050
4051         if (flv->it_flags & MUTABOR && mdt_rdonly(req->rq_export))
4052                 RETURN(-EROFS);
4053
4054         if (flv->it_act != NULL) {
4055                 struct ldlm_reply *rep;
4056
4057                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_INTENT_DELAY, 10);
4058
4059                 /* execute policy */
4060                 rc = flv->it_act(opc, info, lockp, flags);
4061
4062                 /* Check whether the reply has been packed successfully. */
4063                 if (req->rq_repmsg != NULL) {
4064                         rep = req_capsule_server_get(info->mti_pill,
4065                                                      &RMF_DLM_REP);
4066                         rep->lock_policy_res2 =
4067                                 ptlrpc_status_hton(rep->lock_policy_res2);
4068                 }
4069         }
4070
4071         RETURN(rc);
4072 }
4073
4074 static void mdt_ptlrpc_stats_update(struct ptlrpc_request *req,
4075                                     enum ldlm_intent_flags it_opc)
4076 {
4077         struct lprocfs_stats *srv_stats = ptlrpc_req2svc(req)->srv_stats;
4078
4079         /* update stats when IT code is known */
4080         if (srv_stats != NULL)
4081                 lprocfs_counter_incr(srv_stats,
4082                                 PTLRPC_LAST_CNTR + (it_opc == IT_GLIMPSE ?
4083                                 LDLM_GLIMPSE_ENQUEUE : LDLM_IBITS_ENQUEUE));
4084 }
4085
4086 static int mdt_intent_policy(struct ldlm_namespace *ns,
4087                              struct ldlm_lock **lockp, void *req_cookie,
4088                              enum ldlm_mode mode, __u64 flags, void *data)
4089 {
4090         struct tgt_session_info *tsi;
4091         struct mdt_thread_info  *info;
4092         struct ptlrpc_request   *req  =  req_cookie;
4093         struct ldlm_intent      *it;
4094         struct req_capsule      *pill;
4095         const struct ldlm_lock_desc *ldesc;
4096         int rc;
4097
4098         ENTRY;
4099
4100         LASSERT(req != NULL);
4101
4102         tsi = tgt_ses_info(req->rq_svc_thread->t_env);
4103
4104         info = tsi2mdt_info(tsi);
4105         LASSERT(info != NULL);
4106         pill = info->mti_pill;
4107         LASSERT(pill->rc_req == req);
4108         ldesc = &info->mti_dlm_req->lock_desc;
4109
4110         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
4111                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
4112                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
4113                 if (it != NULL) {
4114                         mdt_ptlrpc_stats_update(req, it->opc);
4115                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
4116                         if (rc == 0)
4117                                 rc = ELDLM_OK;
4118
4119                         /* Lock without inodebits makes no sense and will oops
4120                          * later in ldlm. Let's check it now to see if we have
4121                          * ibits corrupted somewhere in mdt_intent_opc().
4122                          * The case for client miss to set ibits has been
4123                          * processed by others. */
4124                         LASSERT(ergo(ldesc->l_resource.lr_type == LDLM_IBITS,
4125                                 ldesc->l_policy_data.l_inodebits.bits != 0));
4126                 } else {
4127                         rc = err_serious(-EFAULT);
4128                 }
4129         } else {
4130                 /* No intent was provided */
4131                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
4132                 rc = req_capsule_server_pack(pill);
4133                 if (rc)
4134                         rc = err_serious(rc);
4135         }
4136         mdt_thread_info_fini(info);
4137         RETURN(rc);
4138 }
4139
4140 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
4141 {
4142         struct seq_server_site  *ss = mdt_seq_site(mdt);
4143
4144         if (ss->ss_node_id == 0)
4145                 return;
4146
4147         if (ss->ss_client_seq != NULL) {
4148                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
4149                 ss->ss_client_seq->lcs_exp = NULL;
4150         }
4151
4152         if (ss->ss_server_fld != NULL) {
4153                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
4154                 ss->ss_server_fld->lsf_control_exp = NULL;
4155         }
4156 }
4157
4158 static void mdt_seq_fini_cli(struct mdt_device *mdt)
4159 {
4160         struct seq_server_site *ss = mdt_seq_site(mdt);
4161
4162         if (ss == NULL)
4163                 return;
4164
4165         if (ss->ss_server_seq != NULL)
4166                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
4167 }
4168
4169 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
4170 {
4171         mdt_seq_fini_cli(mdt);
4172         mdt_deregister_seq_exp(mdt);
4173
4174         return seq_site_fini(env, mdt_seq_site(mdt));
4175 }
4176
4177 /**
4178  * It will retrieve its FLDB entries from MDT0, and it only happens
4179  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
4180  * and it needs to refresh FLDB from the MDT0.
4181  **/
4182 static int mdt_register_lwp_callback(void *data)
4183 {
4184         struct lu_env           env;
4185         struct mdt_device       *mdt = data;
4186         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
4187         int                     rc;
4188         ENTRY;
4189
4190         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
4191
4192         rc = lu_env_init(&env, LCT_MD_THREAD);
4193         if (rc < 0) {
4194                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
4195                 RETURN(rc);
4196         }
4197
4198         /* Allocate new sequence now to avoid creating local transaction
4199          * in the normal transaction process */
4200         rc = seq_server_check_and_alloc_super(&env,
4201                                               mdt_seq_site(mdt)->ss_server_seq);
4202         if (rc < 0)
4203                 GOTO(out, rc);
4204
4205         if (fld->lsf_new) {
4206                 rc = fld_update_from_controller(&env, fld);
4207                 if (rc != 0) {
4208                         CERROR("%s: cannot update controller: rc = %d\n",
4209                                mdt_obd_name(mdt), rc);
4210                         GOTO(out, rc);
4211                 }
4212         }
4213 out:
4214         lu_env_fini(&env);
4215         RETURN(rc);
4216 }
4217
4218 static int mdt_register_seq_exp(struct mdt_device *mdt)
4219 {
4220         struct seq_server_site  *ss = mdt_seq_site(mdt);
4221         char                    *lwp_name = NULL;
4222         int                     rc;
4223
4224         if (ss->ss_node_id == 0)
4225                 return 0;
4226
4227         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
4228         if (lwp_name == NULL)
4229                 GOTO(out_free, rc = -ENOMEM);
4230
4231         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
4232         if (rc != 0)
4233                 GOTO(out_free, rc);
4234
4235         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
4236                                       NULL, NULL);
4237         if (rc != 0)
4238                 GOTO(out_free, rc);
4239
4240         rc = lustre_register_lwp_item(lwp_name,
4241                                       &ss->ss_server_fld->lsf_control_exp,
4242                                       mdt_register_lwp_callback, mdt);
4243         if (rc != 0) {
4244                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
4245                 ss->ss_client_seq->lcs_exp = NULL;
4246                 GOTO(out_free, rc);
4247         }
4248 out_free:
4249         if (lwp_name != NULL)
4250                 OBD_FREE(lwp_name, MAX_OBD_NAME);
4251
4252         return rc;
4253 }
4254
4255 /*
4256  * Init client sequence manager which is used by local MDS to talk to sequence
4257  * controller on remote node.
4258  */
4259 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
4260 {
4261         struct seq_server_site  *ss = mdt_seq_site(mdt);
4262         int                     rc;
4263         char                    *prefix;
4264         ENTRY;
4265
4266         /* check if this is adding the first MDC and controller is not yet
4267          * initialized. */
4268         OBD_ALLOC_PTR(ss->ss_client_seq);
4269         if (ss->ss_client_seq == NULL)
4270                 RETURN(-ENOMEM);
4271
4272         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
4273         if (prefix == NULL) {
4274                 OBD_FREE_PTR(ss->ss_client_seq);
4275                 ss->ss_client_seq = NULL;
4276                 RETURN(-ENOMEM);
4277         }
4278
4279         /* Note: seq_client_fini will be called in seq_site_fini */
4280         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
4281         rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
4282                              prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
4283                                                             NULL);
4284         OBD_FREE(prefix, MAX_OBD_NAME + 5);
4285         if (rc != 0) {
4286                 OBD_FREE_PTR(ss->ss_client_seq);
4287                 ss->ss_client_seq = NULL;
4288                 RETURN(rc);
4289         }
4290
4291         rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
4292
4293         RETURN(rc);
4294 }
4295
4296 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
4297 {
4298         struct seq_server_site  *ss;
4299         int                     rc;
4300         ENTRY;
4301
4302         ss = mdt_seq_site(mdt);
4303         /* init sequence controller server(MDT0) */
4304         if (ss->ss_node_id == 0) {
4305                 OBD_ALLOC_PTR(ss->ss_control_seq);
4306                 if (ss->ss_control_seq == NULL)
4307                         RETURN(-ENOMEM);
4308
4309                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
4310                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
4311                                      ss);
4312                 if (rc)
4313                         GOTO(out_seq_fini, rc);
4314         }
4315
4316         /* Init normal sequence server */
4317         OBD_ALLOC_PTR(ss->ss_server_seq);
4318         if (ss->ss_server_seq == NULL)
4319                 GOTO(out_seq_fini, rc = -ENOMEM);
4320
4321         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
4322                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
4323         if (rc)
4324                 GOTO(out_seq_fini, rc);
4325
4326         /* init seq client for seq server to talk to seq controller(MDT0) */
4327         rc = mdt_seq_init_cli(env, mdt);
4328         if (rc != 0)
4329                 GOTO(out_seq_fini, rc);
4330
4331         if (ss->ss_node_id != 0)
4332                 /* register controller export through lwp */
4333                 rc = mdt_register_seq_exp(mdt);
4334
4335         EXIT;
4336 out_seq_fini:
4337         if (rc)
4338                 mdt_seq_fini(env, mdt);
4339
4340         return rc;
4341 }
4342
4343 /*
4344  * FLD wrappers
4345  */
4346 static int mdt_fld_fini(const struct lu_env *env,
4347                         struct mdt_device *m)
4348 {
4349         struct seq_server_site *ss = mdt_seq_site(m);
4350         ENTRY;
4351
4352         if (ss && ss->ss_server_fld) {
4353                 fld_server_fini(env, ss->ss_server_fld);
4354                 OBD_FREE_PTR(ss->ss_server_fld);
4355                 ss->ss_server_fld = NULL;
4356         }
4357
4358         RETURN(0);
4359 }
4360
4361 static int mdt_fld_init(const struct lu_env *env,
4362                         const char *uuid,
4363                         struct mdt_device *m)
4364 {
4365         struct seq_server_site *ss;
4366         int rc;
4367         ENTRY;
4368
4369         ss = mdt_seq_site(m);
4370
4371         OBD_ALLOC_PTR(ss->ss_server_fld);
4372         if (ss->ss_server_fld == NULL)
4373                 RETURN(rc = -ENOMEM);
4374
4375         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
4376                              LU_SEQ_RANGE_MDT);
4377         if (rc) {
4378                 OBD_FREE_PTR(ss->ss_server_fld);
4379                 ss->ss_server_fld = NULL;
4380                 RETURN(rc);
4381         }
4382
4383         RETURN(0);
4384 }
4385
4386 static void mdt_stack_pre_fini(const struct lu_env *env,
4387                            struct mdt_device *m, struct lu_device *top)
4388 {
4389         struct lustre_cfg_bufs  *bufs;
4390         struct lustre_cfg       *lcfg;
4391         struct mdt_thread_info  *info;
4392         ENTRY;
4393
4394         LASSERT(top);
4395
4396         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4397         LASSERT(info != NULL);
4398
4399         bufs = &info->mti_u.bufs;
4400
4401         LASSERT(m->mdt_child_exp);
4402         LASSERT(m->mdt_child_exp->exp_obd);
4403
4404         /* process cleanup, pass mdt obd name to get obd umount flags */
4405         /* XXX: this is needed because all layers are referenced by
4406          * objects (some of them are pinned by osd, for example *
4407          * the proper solution should be a model where object used
4408          * by osd only doesn't have mdt/mdd slices -bzzz */
4409         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4410         lustre_cfg_bufs_set_string(bufs, 1, NULL);
4411         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4412         if (!lcfg)
4413                 RETURN_EXIT;
4414         lustre_cfg_init(lcfg, LCFG_PRE_CLEANUP, bufs);
4415
4416         top->ld_ops->ldo_process_config(env, top, lcfg);
4417         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4418         EXIT;
4419 }
4420
4421 static void mdt_stack_fini(const struct lu_env *env,
4422                            struct mdt_device *m, struct lu_device *top)
4423 {
4424         struct obd_device       *obd = mdt2obd_dev(m);
4425         struct lustre_cfg_bufs  *bufs;
4426         struct lustre_cfg       *lcfg;
4427         struct mdt_thread_info  *info;
4428         char                     flags[3] = "";
4429         ENTRY;
4430
4431         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4432         LASSERT(info != NULL);
4433
4434         lu_dev_del_linkage(top->ld_site, top);
4435
4436         lu_site_purge(env, top->ld_site, -1);
4437
4438         bufs = &info->mti_u.bufs;
4439         /* process cleanup, pass mdt obd name to get obd umount flags */
4440         /* another purpose is to let all layers to release their objects */
4441         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4442         if (obd->obd_force)
4443                 strcat(flags, "F");
4444         if (obd->obd_fail)
4445                 strcat(flags, "A");
4446         lustre_cfg_bufs_set_string(bufs, 1, flags);
4447         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4448         if (!lcfg)
4449                 RETURN_EXIT;
4450         lustre_cfg_init(lcfg, LCFG_CLEANUP, bufs);
4451
4452         LASSERT(top);
4453         top->ld_ops->ldo_process_config(env, top, lcfg);
4454         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4455
4456         lu_site_purge(env, top->ld_site, -1);
4457
4458         m->mdt_child = NULL;
4459         m->mdt_bottom = NULL;
4460
4461         obd_disconnect(m->mdt_child_exp);
4462         m->mdt_child_exp = NULL;
4463
4464         obd_disconnect(m->mdt_bottom_exp);
4465         m->mdt_child_exp = NULL;
4466 }
4467
4468 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
4469                                const char *next, struct obd_export **exp)
4470 {
4471         struct obd_connect_data *data = NULL;
4472         struct obd_device       *obd;
4473         int                      rc;
4474         ENTRY;
4475
4476         OBD_ALLOC_PTR(data);
4477         if (data == NULL)
4478                 GOTO(out, rc = -ENOMEM);
4479
4480         obd = class_name2obd(next);
4481         if (obd == NULL) {
4482                 CERROR("%s: can't locate next device: %s\n",
4483                        mdt_obd_name(m), next);
4484                 GOTO(out, rc = -ENOTCONN);
4485         }
4486
4487         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4488         data->ocd_version = LUSTRE_VERSION_CODE;
4489
4490         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
4491         if (rc) {
4492                 CERROR("%s: cannot connect to next dev %s (%d)\n",
4493                        mdt_obd_name(m), next, rc);
4494                 GOTO(out, rc);
4495         }
4496
4497 out:
4498         if (data)
4499                 OBD_FREE_PTR(data);
4500         RETURN(rc);
4501 }
4502
4503 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
4504                           struct lustre_cfg *cfg)
4505 {
4506         char                   *dev = lustre_cfg_string(cfg, 0);
4507         int                     rc, name_size, uuid_size;
4508         char                   *name, *uuid, *p;
4509         struct lustre_cfg_bufs *bufs;
4510         struct lustre_cfg      *lcfg;
4511         struct obd_device      *obd;
4512         struct lustre_profile  *lprof;
4513         struct lu_site         *site;
4514         ENTRY;
4515
4516         /* in 1.8 we had the only device in the stack - MDS.
4517          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
4518          * in 2.3 OSD is instantiated by obd_mount.c, so we need
4519          * to generate names and setup MDT, MDD. MDT will be using
4520          * generated name to connect to MDD. for MDD the next device
4521          * will be LOD with name taken from so called "profile" which
4522          * is generated by mount_option line
4523          *
4524          * 1.8 MGS generates config. commands like this:
4525          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
4526          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
4527          * 2.0 MGS generates config. commands like this:
4528          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
4529          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4530          *                    3:lustre-MDT0000-mdtlov  4:f
4531          *
4532          * we generate MDD name from MDT one, just replacing T with D
4533          *
4534          * after all the preparations, the logical equivalent will be
4535          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
4536          *                    3:lustre-MDT0000-mdtlov  4:f
4537          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4538          *                    3:lustre-MDD0000  4:f
4539          *
4540          *  notice we build the stack from down to top: MDD first, then MDT */
4541
4542         name_size = MAX_OBD_NAME;
4543         uuid_size = MAX_OBD_NAME;
4544
4545         OBD_ALLOC(name, name_size);
4546         OBD_ALLOC(uuid, uuid_size);
4547         if (name == NULL || uuid == NULL)
4548                 GOTO(cleanup_mem, rc = -ENOMEM);
4549
4550         OBD_ALLOC_PTR(bufs);
4551         if (!bufs)
4552                 GOTO(cleanup_mem, rc = -ENOMEM);
4553
4554         strcpy(name, dev);
4555         p = strstr(name, "-MDT");
4556         if (p == NULL)
4557                 GOTO(free_bufs, rc = -ENOMEM);
4558         p[3] = 'D';
4559
4560         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
4561
4562         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4563         if (lprof == NULL || lprof->lp_dt == NULL) {
4564                 CERROR("can't find the profile: %s\n",
4565                        lustre_cfg_string(cfg, 0));
4566                 GOTO(free_bufs, rc = -EINVAL);
4567         }
4568
4569         lustre_cfg_bufs_reset(bufs, name);
4570         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
4571         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4572         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4573
4574         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4575         if (!lcfg)
4576                 GOTO(put_profile, rc = -ENOMEM);
4577         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
4578
4579         rc = class_attach(lcfg);
4580         if (rc)
4581                 GOTO(lcfg_cleanup, rc);
4582
4583         obd = class_name2obd(name);
4584         if (!obd) {
4585                 CERROR("Can not find obd %s (%s in config)\n",
4586                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
4587                 GOTO(lcfg_cleanup, rc = -EINVAL);
4588         }
4589
4590         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4591
4592         lustre_cfg_bufs_reset(bufs, name);
4593         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4594         lustre_cfg_bufs_set_string(bufs, 2, dev);
4595         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4596
4597         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4598         if (!lcfg)
4599                 GOTO(class_detach, rc = -ENOMEM);
4600         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
4601
4602         rc = class_setup(obd, lcfg);
4603         if (rc)
4604                 GOTO(class_detach, rc);
4605
4606         /* connect to MDD we just setup */
4607         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
4608         if (rc)
4609                 GOTO(class_detach, rc);
4610
4611         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
4612         LASSERT(site);
4613         LASSERT(mdt_lu_site(mdt) == NULL);
4614         mdt->mdt_lu_dev.ld_site = site;
4615         site->ls_top_dev = &mdt->mdt_lu_dev;
4616         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
4617
4618         /* now connect to bottom OSD */
4619         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
4620         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
4621         if (rc)
4622                 GOTO(class_detach, rc);
4623         mdt->mdt_bottom =
4624                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
4625
4626         rc = lu_env_refill((struct lu_env *)env);
4627         if (rc != 0)
4628                 CERROR("Failure to refill session: '%d'\n", rc);
4629
4630         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
4631
4632         EXIT;
4633 class_detach:
4634         if (rc)
4635                 class_detach(obd, lcfg);
4636 lcfg_cleanup:
4637         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4638 put_profile:
4639         class_put_profile(lprof);
4640 free_bufs:
4641         OBD_FREE_PTR(bufs);
4642 cleanup_mem:
4643         if (name)
4644                 OBD_FREE(name, name_size);
4645         if (uuid)
4646                 OBD_FREE(uuid, uuid_size);
4647         RETURN(rc);
4648 }
4649
4650 /* setup quota master target on MDT0 */
4651 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
4652                           struct lustre_cfg *cfg)
4653 {
4654         struct obd_device       *obd;
4655         char                    *dev = lustre_cfg_string(cfg, 0);
4656         char                    *qmtname, *uuid, *p;
4657         struct lustre_cfg_bufs  *bufs;
4658         struct lustre_cfg       *lcfg;
4659         struct lustre_profile   *lprof;
4660         struct obd_connect_data *data;
4661         int                      rc;
4662         ENTRY;
4663
4664         LASSERT(mdt->mdt_qmt_exp == NULL);
4665         LASSERT(mdt->mdt_qmt_dev == NULL);
4666
4667         /* quota master is on MDT0 only for now */
4668         if (mdt->mdt_seq_site.ss_node_id != 0)
4669                 RETURN(0);
4670
4671         /* MGS generates config commands which look as follows:
4672          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4673          *                    3:lustre-MDT0000-mdtlov  4:f
4674          *
4675          * We generate the QMT name from the MDT one, just replacing MD with QM
4676          * after all the preparations, the logical equivalent will be:
4677          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
4678          *                    3:lustre-MDT0000-osd  4:f */
4679         OBD_ALLOC(qmtname, MAX_OBD_NAME);
4680         OBD_ALLOC(uuid, UUID_MAX);
4681         OBD_ALLOC_PTR(bufs);
4682         OBD_ALLOC_PTR(data);
4683         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
4684                 GOTO(cleanup_mem, rc = -ENOMEM);
4685
4686         strcpy(qmtname, dev);
4687         p = strstr(qmtname, "-MDT");
4688         if (p == NULL)
4689                 GOTO(cleanup_mem, rc = -ENOMEM);
4690         /* replace MD with QM */
4691         p[1] = 'Q';
4692         p[2] = 'M';
4693
4694         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
4695
4696         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4697         if (lprof == NULL || lprof->lp_dt == NULL) {
4698                 CERROR("can't find profile for %s\n",
4699                        lustre_cfg_string(cfg, 0));
4700                 GOTO(cleanup_mem, rc = -EINVAL);
4701         }
4702
4703         lustre_cfg_bufs_reset(bufs, qmtname);
4704         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
4705         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4706         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4707
4708         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4709         if (!lcfg)
4710                 GOTO(put_profile, rc = -ENOMEM);
4711         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
4712
4713         rc = class_attach(lcfg);
4714         if (rc)
4715                 GOTO(lcfg_cleanup, rc);
4716
4717         obd = class_name2obd(qmtname);
4718         if (!obd) {
4719                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
4720                        lustre_cfg_string(cfg, 0));
4721                 GOTO(lcfg_cleanup, rc = -EINVAL);
4722         }
4723
4724         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4725
4726         lustre_cfg_bufs_reset(bufs, qmtname);
4727         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4728         lustre_cfg_bufs_set_string(bufs, 2, dev);
4729
4730         /* for quota, the next device should be the OSD device */
4731         lustre_cfg_bufs_set_string(bufs, 3,
4732                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
4733
4734         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4735         if (!lcfg)
4736                 GOTO(class_detach, rc = -ENOMEM);
4737         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
4738
4739         rc = class_setup(obd, lcfg);
4740         if (rc)
4741                 GOTO(class_detach, rc);
4742
4743         mdt->mdt_qmt_dev = obd->obd_lu_dev;
4744
4745         /* configure local quota objects */
4746         rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
4747                                                    &mdt->mdt_lu_dev,
4748                                                    mdt->mdt_qmt_dev);
4749         if (rc)
4750                 GOTO(class_cleanup, rc);
4751
4752         /* connect to quota master target */
4753         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4754         data->ocd_version = LUSTRE_VERSION_CODE;
4755         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
4756                          data, NULL);
4757         if (rc) {
4758                 CERROR("cannot connect to quota master device %s (%d)\n",
4759                        qmtname, rc);
4760                 GOTO(class_cleanup, rc);
4761         }
4762
4763         EXIT;
4764 class_cleanup:
4765         if (rc) {
4766                 class_manual_cleanup(obd);
4767                 mdt->mdt_qmt_dev = NULL;
4768         }
4769 class_detach:
4770         if (rc)
4771                 class_detach(obd, lcfg);
4772 lcfg_cleanup:
4773         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4774 put_profile:
4775         class_put_profile(lprof);
4776 cleanup_mem:
4777         if (bufs)
4778                 OBD_FREE_PTR(bufs);
4779         if (qmtname)
4780                 OBD_FREE(qmtname, MAX_OBD_NAME);
4781         if (uuid)
4782                 OBD_FREE(uuid, UUID_MAX);
4783         if (data)
4784                 OBD_FREE_PTR(data);
4785         return rc;
4786 }
4787
4788 /* Shutdown quota master target associated with mdt */
4789 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
4790 {
4791         ENTRY;
4792
4793         if (mdt->mdt_qmt_exp == NULL)
4794                 RETURN_EXIT;
4795         LASSERT(mdt->mdt_qmt_dev != NULL);
4796
4797         /* the qmt automatically shuts down when the mdt disconnects */
4798         obd_disconnect(mdt->mdt_qmt_exp);
4799         mdt->mdt_qmt_exp = NULL;
4800         mdt->mdt_qmt_dev = NULL;
4801         EXIT;
4802 }
4803
4804 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
4805  * for now. This will be removed along with converting rest of MDT code
4806  * to use tgt_session_info */
4807 static int mdt_tgt_getxattr(struct tgt_session_info *tsi)
4808 {
4809         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
4810         int                      rc;
4811
4812         rc = mdt_getxattr(info);
4813
4814         mdt_thread_info_fini(info);
4815         return rc;
4816 }
4817
4818 #define OBD_FAIL_OST_READ_NET   OBD_FAIL_OST_BRW_NET
4819 #define OBD_FAIL_OST_WRITE_NET  OBD_FAIL_OST_BRW_NET
4820 #define OST_BRW_READ    OST_READ
4821 #define OST_BRW_WRITE   OST_WRITE
4822
4823 static struct tgt_handler mdt_tgt_handlers[] = {
4824 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4825                 0,                      MDS_CONNECT,    mdt_tgt_connect,
4826                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
4827 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4828                 0,                      MDS_DISCONNECT, tgt_disconnect,
4829                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
4830 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4831                 HABEO_REFERO,           MDS_SET_INFO,   mdt_set_info,
4832                 &RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION),
4833 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
4834 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_GET_ROOT,   mdt_get_root),
4835 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETATTR,    mdt_getattr),
4836 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_GETATTR_NAME,
4837                                                         mdt_getattr_name),
4838 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETXATTR,   mdt_tgt_getxattr),
4839 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_STATFS,     mdt_statfs),
4840 TGT_MDT_HDL(0           | MUTABOR,      MDS_REINT,      mdt_reint),
4841 TGT_MDT_HDL(HABEO_CORPUS,               MDS_CLOSE,      mdt_close),
4842 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_READPAGE,   mdt_readpage),
4843 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_SYNC,       mdt_sync),
4844 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
4845 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_PROGRESS,
4846                                                         mdt_hsm_progress),
4847 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_REGISTER,
4848                                                         mdt_hsm_ct_register),
4849 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_UNREGISTER,
4850                                                         mdt_hsm_ct_unregister),
4851 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_STATE_GET,
4852                                                         mdt_hsm_state_get),
4853 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_STATE_SET,
4854                                                         mdt_hsm_state_set),
4855 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_ACTION, mdt_hsm_action),
4856 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_REQUEST,
4857                                                         mdt_hsm_request),
4858 TGT_MDT_HDL(HABEO_CLAVIS | HABEO_CORPUS | HABEO_REFERO | MUTABOR,
4859             MDS_SWAP_LAYOUTS,
4860             mdt_swap_layouts),
4861 };
4862
4863 static struct tgt_handler mdt_io_ops[] = {
4864 TGT_OST_HDL(HABEO_CORPUS | HABEO_REFERO, OST_BRW_READ,  tgt_brw_read),
4865 TGT_OST_HDL(HABEO_CORPUS | MUTABOR,      OST_BRW_WRITE, tgt_brw_write),
4866 TGT_OST_HDL(HABEO_CORPUS | HABEO_REFERO | MUTABOR,
4867                                          OST_PUNCH,     mdt_punch_hdl),
4868 TGT_OST_HDL(HABEO_CORPUS | HABEO_REFERO, OST_SYNC,      mdt_data_sync),
4869 };
4870
4871 static struct tgt_handler mdt_sec_ctx_ops[] = {
4872 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
4873 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
4874 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
4875 };
4876
4877 static struct tgt_handler mdt_quota_ops[] = {
4878 TGT_QUOTA_HDL(HABEO_REFERO,             QUOTA_DQACQ,      mdt_quota_dqacq),
4879 };
4880
4881 static struct tgt_opc_slice mdt_common_slice[] = {
4882         {
4883                 .tos_opc_start  = MDS_FIRST_OPC,
4884                 .tos_opc_end    = MDS_LAST_OPC,
4885                 .tos_hs         = mdt_tgt_handlers
4886         },
4887         {
4888                 .tos_opc_start  = OBD_FIRST_OPC,
4889                 .tos_opc_end    = OBD_LAST_OPC,
4890                 .tos_hs         = tgt_obd_handlers
4891         },
4892         {
4893                 .tos_opc_start  = LDLM_FIRST_OPC,
4894                 .tos_opc_end    = LDLM_LAST_OPC,
4895                 .tos_hs         = tgt_dlm_handlers
4896         },
4897         {
4898                 .tos_opc_start  = SEC_FIRST_OPC,
4899                 .tos_opc_end    = SEC_LAST_OPC,
4900                 .tos_hs         = mdt_sec_ctx_ops
4901         },
4902         {
4903                 .tos_opc_start  = OUT_UPDATE_FIRST_OPC,
4904                 .tos_opc_end    = OUT_UPDATE_LAST_OPC,
4905                 .tos_hs         = tgt_out_handlers
4906         },
4907         {
4908                 .tos_opc_start  = FLD_FIRST_OPC,
4909                 .tos_opc_end    = FLD_LAST_OPC,
4910                 .tos_hs         = fld_handlers
4911         },
4912         {
4913                 .tos_opc_start  = SEQ_FIRST_OPC,
4914                 .tos_opc_end    = SEQ_LAST_OPC,
4915                 .tos_hs         = seq_handlers
4916         },
4917         {
4918                 .tos_opc_start  = QUOTA_DQACQ,
4919                 .tos_opc_end    = QUOTA_LAST_OPC,
4920                 .tos_hs         = mdt_quota_ops
4921         },
4922         {
4923                 .tos_opc_start  = LLOG_FIRST_OPC,
4924                 .tos_opc_end    = LLOG_LAST_OPC,
4925                 .tos_hs         = tgt_llog_handlers
4926         },
4927         {
4928                 .tos_opc_start  = LFSCK_FIRST_OPC,
4929                 .tos_opc_end    = LFSCK_LAST_OPC,
4930                 .tos_hs         = tgt_lfsck_handlers
4931         },
4932         {
4933                 .tos_opc_start  = OST_FIRST_OPC,
4934                 .tos_opc_end    = OST_LAST_OPC,
4935                 .tos_hs         = mdt_io_ops
4936         },
4937         {
4938                 .tos_hs         = NULL
4939         }
4940 };
4941
4942 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4943 {
4944         struct md_device *next = m->mdt_child;
4945         struct lu_device *d = &m->mdt_lu_dev;
4946         struct obd_device *obd = mdt2obd_dev(m);
4947         struct lfsck_stop stop;
4948
4949         ENTRY;
4950         stop.ls_status = LS_PAUSED;
4951         stop.ls_flags = 0;
4952         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
4953
4954         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
4955         ping_evictor_stop();
4956
4957         /* Remove the HSM /proc entry so the coordinator cannot be
4958          * restarted by a user while it's shutting down. */
4959         hsm_cdt_procfs_fini(m);
4960         mdt_hsm_cdt_stop(m);
4961
4962         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
4963         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4964
4965         if (m->mdt_namespace != NULL)
4966                 ldlm_namespace_free_prior(m->mdt_namespace, NULL,
4967                                           d->ld_obd->obd_force);
4968
4969         obd_exports_barrier(obd);
4970         obd_zombie_barrier();
4971
4972         mdt_quota_fini(env, m);
4973
4974         cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
4975
4976         /* Calling the cleanup functions in the same order as in the mdt_init0
4977          * error path
4978          */
4979         mdt_procfs_fini(m);
4980
4981         target_recovery_fini(obd);
4982         upcall_cache_cleanup(m->mdt_identity_cache);
4983         m->mdt_identity_cache = NULL;
4984
4985         mdt_fs_cleanup(env, m);
4986
4987         tgt_fini(env, &m->mdt_lut);
4988
4989         mdt_hsm_cdt_fini(m);
4990
4991         if (m->mdt_namespace != NULL) {
4992                 ldlm_namespace_free_post(m->mdt_namespace);
4993                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4994         }
4995
4996         if (m->mdt_md_root != NULL) {
4997                 mdt_object_put(env, m->mdt_md_root);
4998                 m->mdt_md_root = NULL;
4999         }
5000
5001         mdt_seq_fini(env, m);
5002
5003         mdt_fld_fini(env, m);
5004
5005         /*
5006          * Finish the stack
5007          */
5008         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
5009
5010         LASSERT(atomic_read(&d->ld_ref) == 0);
5011
5012         server_put_mount(mdt_obd_name(m), true);
5013
5014         EXIT;
5015 }
5016
5017 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
5018
5019 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
5020                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
5021 {
5022         const struct dt_device_param *dt_conf;
5023         struct mdt_thread_info *info;
5024         struct obd_device *obd;
5025         const char *dev = lustre_cfg_string(cfg, 0);
5026         const char *num = lustre_cfg_string(cfg, 2);
5027         struct tg_grants_data *tgd = &m->mdt_lut.lut_tgd;
5028         struct lustre_mount_info *lmi = NULL;
5029         struct lustre_sb_info *lsi;
5030         struct lu_site *s;
5031         struct seq_server_site *ss_site;
5032         const char *identity_upcall = "NONE";
5033         struct md_device *next;
5034         int rc;
5035         long node_id;
5036         mntopt_t mntopts;
5037         ENTRY;
5038
5039         lu_device_init(&m->mdt_lu_dev, ldt);
5040         /*
5041          * Environment (env) might be missing mdt_thread_key values at that
5042          * point, if device is allocated when mdt_thread_key is in QUIESCENT
5043          * mode.
5044          *
5045          * Usually device allocation path doesn't use module key values, but
5046          * mdt has to do a lot of work here, so allocate key value.
5047          */
5048         rc = lu_env_refill((struct lu_env *)env);
5049         if (rc != 0)
5050                 RETURN(rc);
5051
5052         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5053         LASSERT(info != NULL);
5054
5055         obd = class_name2obd(dev);
5056         LASSERT(obd != NULL);
5057
5058         m->mdt_max_mdsize = MAX_MD_SIZE; /* 4 stripes */
5059         m->mdt_opts.mo_evict_tgt_nids = 1;
5060         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
5061
5062         lmi = server_get_mount(dev);
5063         if (lmi == NULL) {
5064                 CERROR("Cannot get mount info for %s!\n", dev);
5065                 RETURN(-EFAULT);
5066         } else {
5067                 lsi = s2lsi(lmi->lmi_sb);
5068                 /* CMD is supported only in IAM mode */
5069                 LASSERT(num);
5070                 node_id = simple_strtol(num, NULL, 10);
5071                 obd->u.obt.obt_magic = OBT_MAGIC;
5072                 if (lsi->lsi_lmd != NULL &&
5073                     lsi->lsi_lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
5074                         m->mdt_skip_lfsck = 1;
5075         }
5076
5077         /* DoM files get IO lock at open by default */
5078         m->mdt_opts.mo_dom_lock = 1;
5079
5080         m->mdt_squash.rsi_uid = 0;
5081         m->mdt_squash.rsi_gid = 0;
5082         INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
5083         init_rwsem(&m->mdt_squash.rsi_sem);
5084         spin_lock_init(&m->mdt_lock);
5085         m->mdt_enable_remote_dir = 0;
5086         m->mdt_enable_remote_dir_gid = 0;
5087
5088         atomic_set(&m->mdt_mds_mds_conns, 0);
5089         atomic_set(&m->mdt_async_commit_count, 0);
5090
5091         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
5092         m->mdt_lu_dev.ld_obd = obd;
5093         /* Set this lu_device to obd for error handling purposes. */
5094         obd->obd_lu_dev = &m->mdt_lu_dev;
5095
5096         /* init the stack */
5097         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
5098         if (rc) {
5099                 CERROR("%s: Can't init device stack, rc %d\n",
5100                        mdt_obd_name(m), rc);
5101                 GOTO(err_lmi, rc);
5102         }
5103
5104         s = mdt_lu_site(m);
5105         ss_site = mdt_seq_site(m);
5106         s->ld_seq_site = ss_site;
5107         ss_site->ss_lu = s;
5108
5109         /* set server index */
5110         ss_site->ss_node_id = node_id;
5111
5112         /* failover is the default
5113          * FIXME: we do not failout mds0/mgs, which may cause some problems.
5114          * assumed whose ss_node_id == 0 XXX
5115          * */
5116         obd->obd_replayable = 1;
5117         /* No connection accepted until configurations will finish */
5118         obd->obd_no_conn = 1;
5119
5120         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
5121                 char *str = lustre_cfg_string(cfg, 4);
5122                 if (strchr(str, 'n')) {
5123                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
5124                         obd->obd_replayable = 0;
5125                 }
5126         }
5127
5128         rc = mdt_fld_init(env, mdt_obd_name(m), m);
5129         if (rc)
5130                 GOTO(err_fini_stack, rc);
5131
5132         rc = mdt_seq_init(env, m);
5133         if (rc)
5134                 GOTO(err_fini_fld, rc);
5135
5136         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
5137                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
5138         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
5139                                               LDLM_NAMESPACE_SERVER,
5140                                               LDLM_NAMESPACE_GREEDY,
5141                                               LDLM_NS_TYPE_MDT);
5142         if (m->mdt_namespace == NULL)
5143                 GOTO(err_fini_seq, rc = -ENOMEM);
5144
5145         m->mdt_namespace->ns_lvbp = m;
5146         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
5147
5148         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
5149         /* set obd_namespace for compatibility with old code */
5150         obd->obd_namespace = m->mdt_namespace;
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_free_ns, rc);
5157         }
5158
5159         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
5160                       OBD_FAIL_MDS_ALL_REQUEST_NET,
5161                       OBD_FAIL_MDS_ALL_REPLY_NET);
5162         if (rc)
5163                 GOTO(err_free_hsm, rc);
5164
5165         /* Amount of available space excluded from granting and reserved
5166          * for metadata. It is in percentage and 50% is default value. */
5167         tgd->tgd_reserved_pcnt = 50;
5168
5169         if (ONE_MB_BRW_SIZE < (1U << tgd->tgd_blockbits))
5170                 m->mdt_brw_size = 1U << tgd->tgd_blockbits;
5171         else
5172                 m->mdt_brw_size = ONE_MB_BRW_SIZE;
5173
5174         rc = mdt_fs_setup(env, m, obd, lsi);
5175         if (rc)
5176                 GOTO(err_tgt, rc);
5177
5178         tgt_adapt_sptlrpc_conf(&m->mdt_lut);
5179
5180         next = m->mdt_child;
5181         dt_conf = next->md_ops->mdo_dtconf_get(env, next);
5182
5183         mntopts = dt_conf->ddp_mntopts;
5184
5185         if (mntopts & MNTOPT_USERXATTR)
5186                 m->mdt_opts.mo_user_xattr = 1;
5187         else
5188                 m->mdt_opts.mo_user_xattr = 0;
5189
5190         m->mdt_max_ea_size = dt_conf->ddp_max_ea_size;
5191
5192         if (mntopts & MNTOPT_ACL)
5193                 m->mdt_opts.mo_acl = 1;
5194         else
5195                 m->mdt_opts.mo_acl = 0;
5196
5197         /* XXX: to support suppgid for ACL, we enable identity_upcall
5198          * by default, otherwise, maybe got unexpected -EACCESS. */
5199         if (m->mdt_opts.mo_acl)
5200                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
5201
5202         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
5203                                                 identity_upcall,
5204                                                 &mdt_identity_upcall_cache_ops);
5205         if (IS_ERR(m->mdt_identity_cache)) {
5206                 rc = PTR_ERR(m->mdt_identity_cache);
5207                 m->mdt_identity_cache = NULL;
5208                 GOTO(err_fs_cleanup, rc);
5209         }
5210
5211         rc = mdt_procfs_init(m, dev);
5212         if (rc) {
5213                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
5214                 GOTO(err_recovery, rc);
5215         }
5216
5217         rc = mdt_quota_init(env, m, cfg);
5218         if (rc)
5219                 GOTO(err_procfs, rc);
5220
5221         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
5222         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
5223                            "mdt_ldlm_client", m->mdt_ldlm_client);
5224
5225         ping_evictor_start();
5226
5227         /* recovery will be started upon mdt_prepare()
5228          * when the whole stack is complete and ready
5229          * to serve the requests */
5230
5231         /* Reduce the initial timeout on an MDS because it doesn't need such
5232          * a long timeout as an OST does. Adaptive timeouts will adjust this
5233          * value appropriately. */
5234         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
5235                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
5236
5237         RETURN(0);
5238 err_procfs:
5239         mdt_procfs_fini(m);
5240 err_recovery:
5241         target_recovery_fini(obd);
5242         upcall_cache_cleanup(m->mdt_identity_cache);
5243         m->mdt_identity_cache = NULL;
5244 err_fs_cleanup:
5245         mdt_fs_cleanup(env, m);
5246 err_tgt:
5247         tgt_fini(env, &m->mdt_lut);
5248 err_free_hsm:
5249         mdt_hsm_cdt_fini(m);
5250 err_free_ns:
5251         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
5252         obd->obd_namespace = m->mdt_namespace = NULL;
5253 err_fini_seq:
5254         mdt_seq_fini(env, m);
5255 err_fini_fld:
5256         mdt_fld_fini(env, m);
5257 err_fini_stack:
5258         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
5259 err_lmi:
5260         if (lmi)
5261                 server_put_mount(dev, true);
5262         return(rc);
5263 }
5264
5265 /* For interoperability, the left element is old parameter, the right one
5266  * is the new version of the parameter, if some parameter is deprecated,
5267  * the new version should be set as NULL. */
5268 static struct cfg_interop_param mdt_interop_param[] = {
5269         { "mdt.group_upcall",   NULL },
5270         { "mdt.quota_type",     NULL },
5271         { "mdd.quota_type",     NULL },
5272         { "mdt.som",            NULL },
5273         { "mdt.rootsquash",     "mdt.root_squash" },
5274         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
5275         { NULL }
5276 };
5277
5278 /* used by MGS to process specific configurations */
5279 static int mdt_process_config(const struct lu_env *env,
5280                               struct lu_device *d, struct lustre_cfg *cfg)
5281 {
5282         struct mdt_device *m = mdt_dev(d);
5283         struct md_device *md_next = m->mdt_child;
5284         struct lu_device *next = md2lu_dev(md_next);
5285         int rc;
5286         ENTRY;
5287
5288         switch (cfg->lcfg_command) {
5289         case LCFG_PARAM: {
5290                 struct obd_device          *obd = d->ld_obd;
5291
5292                 /* For interoperability */
5293                 struct cfg_interop_param   *ptr = NULL;
5294                 struct lustre_cfg          *old_cfg = NULL;
5295                 char                       *param = NULL;
5296
5297                 param = lustre_cfg_string(cfg, 1);
5298                 if (param == NULL) {
5299                         CERROR("param is empty\n");
5300                         rc = -EINVAL;
5301                         break;
5302                 }
5303
5304                 ptr = class_find_old_param(param, mdt_interop_param);
5305                 if (ptr != NULL) {
5306                         if (ptr->new_param == NULL) {
5307                                 rc = 0;
5308                                 CWARN("For interoperability, skip this %s."
5309                                       " It is obsolete.\n", ptr->old_param);
5310                                 break;
5311                         }
5312
5313                         CWARN("Found old param %s, changed it to %s.\n",
5314                               ptr->old_param, ptr->new_param);
5315
5316                         old_cfg = cfg;
5317                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
5318                         if (IS_ERR(cfg)) {
5319                                 rc = PTR_ERR(cfg);
5320                                 break;
5321                         }
5322                 }
5323
5324                 rc = class_process_proc_param(PARAM_MDT, obd->obd_vars,
5325                                               cfg, obd);
5326                 if (rc > 0 || rc == -ENOSYS) {
5327                         /* is it an HSM var ? */
5328                         rc = class_process_proc_param(PARAM_HSM,
5329                                                       hsm_cdt_get_proc_vars(),
5330                                                       cfg, obd);
5331                         if (rc > 0 || rc == -ENOSYS)
5332                                 /* we don't understand; pass it on */
5333                                 rc = next->ld_ops->ldo_process_config(env, next,
5334                                                                       cfg);
5335                 }
5336
5337                 if (old_cfg)
5338                         OBD_FREE(cfg, lustre_cfg_len(cfg->lcfg_bufcount,
5339                                                      cfg->lcfg_buflens));
5340                 break;
5341         }
5342         default:
5343                 /* others are passed further */
5344                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
5345                 break;
5346         }
5347         RETURN(rc);
5348 }
5349
5350 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
5351                                           const struct lu_object_header *hdr,
5352                                           struct lu_device *d)
5353 {
5354         struct mdt_object *mo;
5355
5356         ENTRY;
5357
5358         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
5359         if (mo != NULL) {
5360                 struct lu_object *o;
5361                 struct lu_object_header *h;
5362
5363                 o = &mo->mot_obj;
5364                 h = &mo->mot_header;
5365                 lu_object_header_init(h);
5366                 lu_object_init(o, h, d);
5367                 lu_object_add_top(h, o);
5368                 o->lo_ops = &mdt_obj_ops;
5369                 spin_lock_init(&mo->mot_write_lock);
5370                 mutex_init(&mo->mot_lov_mutex);
5371                 init_rwsem(&mo->mot_dom_sem);
5372                 init_rwsem(&mo->mot_open_sem);
5373                 atomic_set(&mo->mot_open_count, 0);
5374                 RETURN(o);
5375         }
5376         RETURN(NULL);
5377 }
5378
5379 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
5380                            const struct lu_object_conf *unused)
5381 {
5382         struct mdt_device *d = mdt_dev(o->lo_dev);
5383         struct lu_device  *under;
5384         struct lu_object  *below;
5385         int                rc = 0;
5386         ENTRY;
5387
5388         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
5389                PFID(lu_object_fid(o)));
5390
5391         under = &d->mdt_child->md_lu_dev;
5392         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
5393         if (below != NULL) {
5394                 lu_object_add(o, below);
5395         } else
5396                 rc = -ENOMEM;
5397
5398         RETURN(rc);
5399 }
5400
5401 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
5402 {
5403         struct mdt_object *mo = mdt_obj(o);
5404         struct lu_object_header *h;
5405         ENTRY;
5406
5407         h = o->lo_header;
5408         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
5409                PFID(lu_object_fid(o)));
5410
5411         LASSERT(atomic_read(&mo->mot_open_count) == 0);
5412         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
5413
5414         lu_object_fini(o);
5415         lu_object_header_fini(h);
5416         OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
5417
5418         EXIT;
5419 }
5420
5421 static int mdt_object_print(const struct lu_env *env, void *cookie,
5422                             lu_printer_t p, const struct lu_object *o)
5423 {
5424         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
5425
5426         return (*p)(env, cookie,
5427                     LUSTRE_MDT_NAME"-object@%p(%s %s, writecount=%d)",
5428                     mdto, mdto->mot_lov_created ? "lov_created" : "",
5429                     mdto->mot_cache_attr ? "cache_attr" : "",
5430                     mdto->mot_write_count);
5431 }
5432
5433 static int mdt_prepare(const struct lu_env *env,
5434                 struct lu_device *pdev,
5435                 struct lu_device *cdev)
5436 {
5437         struct mdt_device *mdt = mdt_dev(cdev);
5438         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
5439         struct obd_device *obd = cdev->ld_obd;
5440         int rc;
5441
5442         ENTRY;
5443
5444         LASSERT(obd);
5445
5446         rc = next->ld_ops->ldo_prepare(env, cdev, next);
5447         if (rc)
5448                 RETURN(rc);
5449
5450         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
5451         if (rc)
5452                 RETURN(rc);
5453
5454         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
5455         if (rc)
5456                 RETURN(rc);
5457
5458         rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
5459         /* The LFSCK instance is registered just now, so it must be there when
5460          * register the namespace to such instance. */
5461         LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
5462
5463         if (mdt->mdt_seq_site.ss_node_id == 0) {
5464                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
5465                                                          &mdt->mdt_md_root_fid);
5466                 if (rc)
5467                         RETURN(rc);
5468         }
5469
5470         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
5471
5472         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
5473         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
5474         LASSERT(obd->obd_no_conn);
5475         spin_lock(&obd->obd_dev_lock);
5476         obd->obd_no_conn = 0;
5477         spin_unlock(&obd->obd_dev_lock);
5478
5479         if (obd->obd_recovering == 0)
5480                 mdt_postrecov(env, mdt);
5481
5482         RETURN(rc);
5483 }
5484
5485 const struct lu_device_operations mdt_lu_ops = {
5486         .ldo_object_alloc   = mdt_object_alloc,
5487         .ldo_process_config = mdt_process_config,
5488         .ldo_prepare        = mdt_prepare,
5489 };
5490
5491 static const struct lu_object_operations mdt_obj_ops = {
5492         .loo_object_init    = mdt_object_init,
5493         .loo_object_free    = mdt_object_free,
5494         .loo_object_print   = mdt_object_print
5495 };
5496
5497 static int mdt_obd_set_info_async(const struct lu_env *env,
5498                                   struct obd_export *exp,
5499                                   __u32 keylen, void *key,
5500                                   __u32 vallen, void *val,
5501                                   struct ptlrpc_request_set *set)
5502 {
5503         int rc;
5504
5505         ENTRY;
5506
5507         if (KEY_IS(KEY_SPTLRPC_CONF)) {
5508                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp));
5509                 RETURN(rc);
5510         }
5511
5512         RETURN(0);
5513 }
5514
5515 /**
5516  * Match client and server connection feature flags.
5517  *
5518  * Compute the compatibility flags for a connection request based on
5519  * features mutually supported by client and server.
5520  *
5521  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
5522  * must not be updated here, otherwise a partially initialized value may
5523  * be exposed. After the connection request is successfully processed,
5524  * the top-level MDT connect request handler atomically updates the export
5525  * connect flags from the obd_connect_data::ocd_connect_flags field of the
5526  * reply. \see mdt_connect().
5527  *
5528  * Before 2.7.50 clients will send a struct obd_connect_data_v1 rather than a
5529  * full struct obd_connect_data. So care must be taken when accessing fields
5530  * that are not present in struct obd_connect_data_v1. See LU-16.
5531  *
5532  * \param exp   the obd_export associated with this client/target pair
5533  * \param mdt   the target device for the connection
5534  * \param data  stores data for this connect request
5535  *
5536  * \retval 0       success
5537  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
5538  * \retval -EBADE  client and server feature requirements are incompatible
5539  */
5540 static int mdt_connect_internal(const struct lu_env *env,
5541                                 struct obd_export *exp,
5542                                 struct mdt_device *mdt,
5543                                 struct obd_connect_data *data, bool reconnect)
5544 {
5545         LASSERT(data != NULL);
5546
5547         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
5548
5549         if (mdt->mdt_bottom->dd_rdonly &&
5550             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5551             !(data->ocd_connect_flags & OBD_CONNECT_RDONLY))
5552                 RETURN(-EACCES);
5553
5554         if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
5555                 data->ocd_connect_flags2 &= MDT_CONNECT_SUPPORTED2;
5556
5557         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
5558
5559         if (!(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5560             !(data->ocd_connect_flags & OBD_CONNECT_IBITS)) {
5561                 CWARN("%s: client %s does not support ibits lock, either "
5562                       "very old or an invalid client: flags %#llx\n",
5563                       mdt_obd_name(mdt), exp->exp_client_uuid.uuid,
5564                       data->ocd_connect_flags);
5565                 return -EBADE;
5566         }
5567
5568         if (!mdt->mdt_opts.mo_acl)
5569                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
5570
5571         if (!mdt->mdt_opts.mo_user_xattr)
5572                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
5573
5574         if (OCD_HAS_FLAG(data, BRW_SIZE)) {
5575                 data->ocd_brw_size = min(data->ocd_brw_size,
5576                                          mdt->mdt_brw_size);
5577                 if (data->ocd_brw_size == 0) {
5578                         CERROR("%s: cli %s/%p ocd_connect_flags: %#llx "
5579                                "ocd_version: %x ocd_grant: %d ocd_index: %u "
5580                                "ocd_brw_size unexpectedly zero, network data "
5581                                "corruption? Refusing to connect this client\n",
5582                                mdt_obd_name(mdt),
5583                                exp->exp_client_uuid.uuid,
5584                                exp, data->ocd_connect_flags, data->ocd_version,
5585                                data->ocd_grant, data->ocd_index);
5586                         return -EPROTO;
5587                 }
5588         }
5589
5590         if (OCD_HAS_FLAG(data, GRANT_PARAM)) {
5591                 struct dt_device_param *ddp = &mdt->mdt_lut.lut_dt_conf;
5592
5593                 /* client is reporting its page size, for future use */
5594                 exp->exp_target_data.ted_pagebits = data->ocd_grant_blkbits;
5595                 data->ocd_grant_blkbits  = mdt->mdt_lut.lut_tgd.tgd_blockbits;
5596                 /* ddp_inodespace may not be power-of-two value, eg. for ldiskfs
5597                  * it's LDISKFS_DIR_REC_LEN(20) = 28. */
5598                 data->ocd_grant_inobits = fls(ddp->ddp_inodespace - 1);
5599                 /* ocd_grant_tax_kb is in 1K byte blocks */
5600                 data->ocd_grant_tax_kb = ddp->ddp_extent_tax >> 10;
5601                 data->ocd_grant_max_blks = ddp->ddp_max_extent_blks;
5602         }
5603
5604         if (OCD_HAS_FLAG(data, GRANT)) {
5605                 /* Save connect_data we have so far because tgt_grant_connect()
5606                  * uses it to calculate grant. */
5607                 exp->exp_connect_data = *data;
5608                 tgt_grant_connect(env, exp, data, !reconnect);
5609         }
5610
5611         if (OCD_HAS_FLAG(data, MAXBYTES))
5612                 data->ocd_maxbytes = mdt->mdt_lut.lut_dt_conf.ddp_maxbytes;
5613
5614         /* NB: Disregard the rule against updating
5615          * exp_connect_data.ocd_connect_flags in this case, since
5616          * tgt_client_new() needs to know if this is a lightweight
5617          * connection, and it is safe to expose this flag before
5618          * connection processing completes. */
5619         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
5620                 spin_lock(&exp->exp_lock);
5621                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
5622                 spin_unlock(&exp->exp_lock);
5623         }
5624
5625         data->ocd_version = LUSTRE_VERSION_CODE;
5626
5627         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
5628                 CWARN("%s: MDS requires FID support, but client not\n",
5629                       mdt_obd_name(mdt));
5630                 return -EBADE;
5631         }
5632
5633         if (OCD_HAS_FLAG(data, PINGLESS)) {
5634                 if (ptlrpc_pinger_suppress_pings()) {
5635                         spin_lock(&exp->exp_obd->obd_dev_lock);
5636                         list_del_init(&exp->exp_obd_chain_timed);
5637                         spin_unlock(&exp->exp_obd->obd_dev_lock);
5638                 } else {
5639                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
5640                 }
5641         }
5642
5643         data->ocd_max_easize = mdt->mdt_max_ea_size;
5644
5645         /* NB: Disregard the rule against updating
5646          * exp_connect_data.ocd_connect_flags in this case, since
5647          * tgt_client_new() needs to know if this is client supports
5648          * multiple modify RPCs, and it is safe to expose this flag before
5649          * connection processing completes. */
5650         if (data->ocd_connect_flags & OBD_CONNECT_MULTIMODRPCS) {
5651                 data->ocd_maxmodrpcs = max_mod_rpcs_per_client;
5652                 spin_lock(&exp->exp_lock);
5653                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_MULTIMODRPCS;
5654                 spin_unlock(&exp->exp_lock);
5655         }
5656
5657         if (OCD_HAS_FLAG(data, CKSUM)) {
5658                 __u32 cksum_types = data->ocd_cksum_types;
5659
5660                 /* The client set in ocd_cksum_types the checksum types it
5661                  * supports. We have to mask off the algorithms that we don't
5662                  * support */
5663                 data->ocd_cksum_types &= cksum_types_supported_server();
5664
5665                 if (unlikely(data->ocd_cksum_types == 0)) {
5666                         CERROR("%s: Connect with checksum support but no "
5667                                "ocd_cksum_types is set\n",
5668                                exp->exp_obd->obd_name);
5669                         RETURN(-EPROTO);
5670                 }
5671
5672                 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
5673                        "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
5674                        cksum_types, data->ocd_cksum_types);
5675         } else {
5676                 /* This client does not support OBD_CONNECT_CKSUM
5677                  * fall back to CRC32 */
5678                 CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
5679                        "OBD_CONNECT_CKSUM, CRC32 will be used\n",
5680                        exp->exp_obd->obd_name, obd_export_nid2str(exp));
5681         }
5682
5683         return 0;
5684 }
5685
5686 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
5687                                    struct mdt_thread_info *info,
5688                                    struct mdt_file_data *mfd)
5689 {
5690         struct lu_context ses;
5691         int rc;
5692         ENTRY;
5693
5694         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
5695         if (rc)
5696                 RETURN(rc);
5697
5698         env->le_ses = &ses;
5699         lu_context_enter(&ses);
5700
5701         mdt_ucred(info)->uc_valid = UCRED_OLD;
5702         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
5703
5704         lu_context_exit(&ses);
5705         lu_context_fini(&ses);
5706         env->le_ses = NULL;
5707
5708         RETURN(rc);
5709 }
5710
5711 static int mdt_export_cleanup(struct obd_export *exp)
5712 {
5713         struct list_head         closing_list;
5714         struct mdt_export_data  *med = &exp->exp_mdt_data;
5715         struct obd_device       *obd = exp->exp_obd;
5716         struct mdt_device       *mdt;
5717         struct mdt_thread_info  *info;
5718         struct lu_env            env;
5719         struct mdt_file_data    *mfd, *n;
5720         int rc = 0;
5721         ENTRY;
5722
5723         INIT_LIST_HEAD(&closing_list);
5724         spin_lock(&med->med_open_lock);
5725         while (!list_empty(&med->med_open_head)) {
5726                 struct list_head *tmp = med->med_open_head.next;
5727                 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
5728
5729                 /* Remove mfd handle so it can't be found again.
5730                  * We are consuming the mfd_list reference here. */
5731                 class_handle_unhash(&mfd->mfd_handle);
5732                 list_move_tail(&mfd->mfd_list, &closing_list);
5733         }
5734         spin_unlock(&med->med_open_lock);
5735         mdt = mdt_dev(obd->obd_lu_dev);
5736         LASSERT(mdt != NULL);
5737
5738         rc = lu_env_init(&env, LCT_MD_THREAD);
5739         if (rc)
5740                 RETURN(rc);
5741
5742         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5743         LASSERT(info != NULL);
5744         memset(info, 0, sizeof *info);
5745         info->mti_env = &env;
5746         info->mti_mdt = mdt;
5747         info->mti_exp = exp;
5748
5749         if (!list_empty(&closing_list)) {
5750                 struct md_attr *ma = &info->mti_attr;
5751
5752                 /* Close any open files (which may also cause orphan
5753                  * unlinking). */
5754                 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5755                         list_del_init(&mfd->mfd_list);
5756                         ma->ma_need = ma->ma_valid = 0;
5757
5758                         /* This file is being closed due to an eviction, it
5759                          * could have been modified and now dirty regarding to
5760                          * HSM archive, check this!
5761                          * The logic here is to mark a file dirty if there's a
5762                          * chance it was dirtied before the client was evicted,
5763                          * so that we don't have to wait for a release attempt
5764                          * before finding out the file was actually dirty and
5765                          * fail the release. Aggressively marking it dirty here
5766                          * will cause the policy engine to attempt to
5767                          * re-archive it; when rearchiving, we can compare the
5768                          * current version to the HSM data_version and make the
5769                          * archive request into a noop if it's not actually
5770                          * dirty.
5771                          */
5772                         if (mfd->mfd_mode & FMODE_WRITE)
5773                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
5774
5775                         /* Don't unlink orphan on failover umount, LU-184 */
5776                         if (exp->exp_flags & OBD_OPT_FAILOVER) {
5777                                 ma->ma_valid = MA_FLAGS;
5778                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5779                         }
5780                         mdt_mfd_close(info, mfd);
5781                 }
5782         }
5783         info->mti_mdt = NULL;
5784         /* cleanup client slot early */
5785         /* Do not erase record for recoverable client. */
5786         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5787                 tgt_client_del(&env, exp);
5788         lu_env_fini(&env);
5789
5790         RETURN(rc);
5791 }
5792
5793 static inline void mdt_enable_slc(struct mdt_device *mdt)
5794 {
5795         if (mdt->mdt_lut.lut_sync_lock_cancel == NEVER_SYNC_ON_CANCEL)
5796                 mdt->mdt_lut.lut_sync_lock_cancel = BLOCKING_SYNC_ON_CANCEL;
5797 }
5798
5799 static inline void mdt_disable_slc(struct mdt_device *mdt)
5800 {
5801         if (mdt->mdt_lut.lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL)
5802                 mdt->mdt_lut.lut_sync_lock_cancel = NEVER_SYNC_ON_CANCEL;
5803 }
5804
5805 static int mdt_obd_disconnect(struct obd_export *exp)
5806 {
5807         int rc;
5808
5809         ENTRY;
5810
5811         LASSERT(exp);
5812         class_export_get(exp);
5813
5814         if (!(exp->exp_flags & OBD_OPT_FORCE))
5815                 tgt_grant_sanity_check(exp->exp_obd, __func__);
5816
5817         if ((exp_connect_flags(exp) & OBD_CONNECT_MDS_MDS) &&
5818             !(exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT)) {
5819                 struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
5820
5821                 if (atomic_dec_and_test(&mdt->mdt_mds_mds_conns))
5822                         mdt_disable_slc(mdt);
5823         }
5824
5825         rc = server_disconnect_export(exp);
5826         if (rc != 0)
5827                 CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
5828
5829         tgt_grant_discard(exp);
5830
5831         rc = mdt_export_cleanup(exp);
5832         nodemap_del_member(exp);
5833         class_export_put(exp);
5834         RETURN(rc);
5835 }
5836
5837 /* mds_connect copy */
5838 static int mdt_obd_connect(const struct lu_env *env,
5839                            struct obd_export **exp, struct obd_device *obd,
5840                            struct obd_uuid *cluuid,
5841                            struct obd_connect_data *data,
5842                            void *localdata)
5843 {
5844         struct obd_export       *lexp;
5845         struct lustre_handle    conn = { 0 };
5846         struct mdt_device       *mdt;
5847         int                      rc;
5848         lnet_nid_t              *client_nid = localdata;
5849         ENTRY;
5850
5851         LASSERT(env != NULL);
5852         LASSERT(data != NULL);
5853
5854         if (!exp || !obd || !cluuid)
5855                 RETURN(-EINVAL);
5856
5857         mdt = mdt_dev(obd->obd_lu_dev);
5858
5859         if ((data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5860             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
5861                 atomic_inc(&mdt->mdt_mds_mds_conns);
5862                 mdt_enable_slc(mdt);
5863         }
5864
5865         /*
5866          * first, check whether the stack is ready to handle requests
5867          * XXX: probably not very appropriate method is used now
5868          *      at some point we should find a better one
5869          */
5870         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) &&
5871             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) &&
5872             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
5873                 rc = obd_get_info(env, mdt->mdt_child_exp,
5874                                   sizeof(KEY_OSP_CONNECTED),
5875                                   KEY_OSP_CONNECTED, NULL, NULL);
5876                 if (rc)
5877                         RETURN(-EAGAIN);
5878                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
5879         }
5880
5881         rc = class_connect(&conn, obd, cluuid);
5882         if (rc)
5883                 RETURN(rc);
5884
5885         lexp = class_conn2export(&conn);
5886         LASSERT(lexp != NULL);
5887
5888         rc = nodemap_add_member(*client_nid, lexp);
5889         if (rc != 0 && rc != -EEXIST)
5890                 GOTO(out, rc);
5891
5892         rc = mdt_connect_internal(env, lexp, mdt, data, false);
5893         if (rc == 0) {
5894                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5895
5896                 LASSERT(lcd);
5897                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5898                 rc = tgt_client_new(env, lexp);
5899                 if (rc == 0)
5900                         mdt_export_stats_init(obd, lexp, localdata);
5901         }
5902 out:
5903         if (rc != 0) {
5904                 class_disconnect(lexp);
5905                 nodemap_del_member(lexp);
5906                 *exp = NULL;
5907         } else {
5908                 *exp = lexp;
5909                 /* Because we do not want this export to be evicted by pinger,
5910                  * let's not add this export to the timed chain list. */
5911                 if (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) {
5912                         spin_lock(&lexp->exp_obd->obd_dev_lock);
5913                         list_del_init(&lexp->exp_obd_chain_timed);
5914                         spin_unlock(&lexp->exp_obd->obd_dev_lock);
5915                 }
5916         }
5917
5918         RETURN(rc);
5919 }
5920
5921 static int mdt_obd_reconnect(const struct lu_env *env,
5922                              struct obd_export *exp, struct obd_device *obd,
5923                              struct obd_uuid *cluuid,
5924                              struct obd_connect_data *data,
5925                              void *localdata)
5926 {
5927         lnet_nid_t             *client_nid = localdata;
5928         int                     rc;
5929         ENTRY;
5930
5931         if (exp == NULL || obd == NULL || cluuid == NULL)
5932                 RETURN(-EINVAL);
5933
5934         rc = nodemap_add_member(*client_nid, exp);
5935         if (rc != 0 && rc != -EEXIST)
5936                 RETURN(rc);
5937
5938         rc = mdt_connect_internal(env, exp, mdt_dev(obd->obd_lu_dev), data,
5939                                   true);
5940         if (rc == 0)
5941                 mdt_export_stats_init(obd, exp, localdata);
5942         else
5943                 nodemap_del_member(exp);
5944
5945         RETURN(rc);
5946 }
5947
5948 /* FIXME: Can we avoid using these two interfaces? */
5949 static int mdt_init_export(struct obd_export *exp)
5950 {
5951         struct mdt_export_data *med = &exp->exp_mdt_data;
5952         int                     rc;
5953         ENTRY;
5954
5955         INIT_LIST_HEAD(&med->med_open_head);
5956         spin_lock_init(&med->med_open_lock);
5957         spin_lock(&exp->exp_lock);
5958         exp->exp_connecting = 1;
5959         spin_unlock(&exp->exp_lock);
5960
5961         /* self-export doesn't need client data and ldlm initialization */
5962         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5963                                      &exp->exp_client_uuid)))
5964                 RETURN(0);
5965
5966         rc = tgt_client_alloc(exp);
5967         if (rc)
5968                 GOTO(err, rc);
5969
5970         rc = ldlm_init_export(exp);
5971         if (rc)
5972                 GOTO(err_free, rc);
5973
5974         RETURN(rc);
5975
5976 err_free:
5977         tgt_client_free(exp);
5978 err:
5979         CERROR("%s: Failed to initialize export: rc = %d\n",
5980                exp->exp_obd->obd_name, rc);
5981         return rc;
5982 }
5983
5984 static int mdt_destroy_export(struct obd_export *exp)
5985 {
5986         ENTRY;
5987
5988         target_destroy_export(exp);
5989         /* destroy can be called from failed obd_setup, so
5990          * checking uuid is safer than obd_self_export */
5991         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5992                                      &exp->exp_client_uuid)))
5993                 RETURN(0);
5994
5995         ldlm_destroy_export(exp);
5996         tgt_client_free(exp);
5997
5998         LASSERT(list_empty(&exp->exp_outstanding_replies));
5999         LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
6000
6001         /*
6002          * discard grants once we're sure no more
6003          * interaction with the client is possible
6004          */
6005         tgt_grant_discard(exp);
6006         if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
6007                 exp->exp_obd->u.obt.obt_lut->lut_tgd.tgd_tot_granted_clients--;
6008
6009         if (!(exp->exp_flags & OBD_OPT_FORCE))
6010                 tgt_grant_sanity_check(exp->exp_obd, __func__);
6011
6012         RETURN(0);
6013 }
6014
6015 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
6016                    struct linkea_data *ldata)
6017 {
6018         int rc;
6019
6020         LASSERT(ldata->ld_buf->lb_buf != NULL);
6021
6022         if (!mdt_object_exists(mdt_obj))
6023                 return -ENODATA;
6024
6025         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6026                           ldata->ld_buf, XATTR_NAME_LINK);
6027         if (rc == -ERANGE) {
6028                 /* Buf was too small, figure out what we need. */
6029                 lu_buf_free(ldata->ld_buf);
6030                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6031                                   ldata->ld_buf, XATTR_NAME_LINK);
6032                 if (rc < 0)
6033                         return rc;
6034                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
6035                 if (ldata->ld_buf->lb_buf == NULL)
6036                         return -ENOMEM;
6037                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6038                                   ldata->ld_buf, XATTR_NAME_LINK);
6039         }
6040         if (rc < 0)
6041                 return rc;
6042
6043         return linkea_init_with_rec(ldata);
6044 }
6045
6046 /**
6047  * Given an MDT object, try to look up the full path to the object.
6048  * Part of the MDT layer implementation of lfs fid2path.
6049  *
6050  * \param[in]     info  Per-thread common data shared by MDT level handlers.
6051  * \param[in]     obj   Object to do path lookup of
6052  * \param[in,out] fp    User-provided struct to store path information
6053  * \param[in]     root_fid Root FID of current path should reach
6054  *
6055  * \retval 0 Lookup successful, path information stored in fp
6056  * \retval -EAGAIN Lookup failed, usually because object is being moved
6057  * \retval negative errno if there was a problem
6058  */
6059 static int mdt_path_current(struct mdt_thread_info *info,
6060                             struct mdt_object *obj,
6061                             struct getinfo_fid2path *fp,
6062                             struct lu_fid *root_fid)
6063 {
6064         struct mdt_device       *mdt = info->mti_mdt;
6065         struct mdt_object       *mdt_obj;
6066         struct link_ea_header   *leh;
6067         struct link_ea_entry    *lee;
6068         struct lu_name          *tmpname = &info->mti_name;
6069         struct lu_fid           *tmpfid = &info->mti_tmp_fid1;
6070         struct lu_buf           *buf = &info->mti_big_buf;
6071         char                    *ptr;
6072         int                     reclen;
6073         struct linkea_data      ldata = { NULL };
6074         int                     rc = 0;
6075         bool                    first = true;
6076         ENTRY;
6077
6078         /* temp buffer for path element, the buffer will be finally freed
6079          * in mdt_thread_info_fini */
6080         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
6081         if (buf->lb_buf == NULL)
6082                 RETURN(-ENOMEM);
6083
6084         ldata.ld_buf = buf;
6085         ptr = fp->gf_u.gf_path + fp->gf_pathlen - 1;
6086         *ptr = 0;
6087         --ptr;
6088         *tmpfid = fp->gf_fid = *mdt_object_fid(obj);
6089
6090         while (!lu_fid_eq(root_fid, &fp->gf_fid)) {
6091                 struct lu_buf           lmv_buf;
6092
6093                 if (!lu_fid_eq(root_fid, &mdt->mdt_md_root_fid) &&
6094                     lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid))
6095                         GOTO(out, rc = -ENOENT);
6096
6097                 if (lu_fid_eq(mdt_object_fid(obj), tmpfid)) {
6098                         mdt_obj = obj;
6099                         mdt_object_get(info->mti_env, mdt_obj);
6100                 } else {
6101                         mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
6102                         if (IS_ERR(mdt_obj))
6103                                 GOTO(out, rc = PTR_ERR(mdt_obj));
6104                 }
6105
6106                 if (!mdt_object_exists(mdt_obj)) {
6107                         mdt_object_put(info->mti_env, mdt_obj);
6108                         GOTO(out, rc = -ENOENT);
6109                 }
6110
6111                 if (mdt_object_remote(mdt_obj)) {
6112                         mdt_object_put(info->mti_env, mdt_obj);
6113                         GOTO(remote_out, rc = -EREMOTE);
6114                 }
6115
6116                 rc = mdt_links_read(info, mdt_obj, &ldata);
6117                 if (rc != 0) {
6118                         mdt_object_put(info->mti_env, mdt_obj);
6119                         GOTO(out, rc);
6120                 }
6121
6122                 leh = buf->lb_buf;
6123                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
6124                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
6125                 /* If set, use link #linkno for path lookup, otherwise use
6126                    link #0.  Only do this for the final path element. */
6127                 if (first && fp->gf_linkno < leh->leh_reccount) {
6128                         int count;
6129                         for (count = 0; count < fp->gf_linkno; count++) {
6130                                 lee = (struct link_ea_entry *)
6131                                      ((char *)lee + reclen);
6132                                 linkea_entry_unpack(lee, &reclen, tmpname,
6133                                                     tmpfid);
6134                         }
6135                         if (fp->gf_linkno < leh->leh_reccount - 1)
6136                                 /* indicate to user there are more links */
6137                                 fp->gf_linkno++;
6138                 }
6139
6140                 lmv_buf.lb_buf = info->mti_xattr_buf;
6141                 lmv_buf.lb_len = sizeof(info->mti_xattr_buf);
6142                 /* Check if it is slave stripes */
6143                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6144                                   &lmv_buf, XATTR_NAME_LMV);
6145                 mdt_object_put(info->mti_env, mdt_obj);
6146                 if (rc > 0) {
6147                         union lmv_mds_md *lmm = lmv_buf.lb_buf;
6148
6149                         /* For slave stripes, get its master */
6150                         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE) {
6151                                 fp->gf_fid = *tmpfid;
6152                                 continue;
6153                         }
6154                 } else if (rc < 0 && rc != -ENODATA) {
6155                         GOTO(out, rc);
6156                 }
6157
6158                 rc = 0;
6159
6160                 /* Pack the name in the end of the buffer */
6161                 ptr -= tmpname->ln_namelen;
6162                 if (ptr - 1 <= fp->gf_u.gf_path)
6163                         GOTO(out, rc = -EOVERFLOW);
6164                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
6165                 *(--ptr) = '/';
6166
6167                 /* keep the last resolved fid to the client, so the
6168                  * client will build the left path on another MDT for
6169                  * remote object */
6170                 fp->gf_fid = *tmpfid;
6171
6172                 first = false;
6173         }
6174
6175 remote_out:
6176         ptr++; /* skip leading / */
6177         memmove(fp->gf_u.gf_path, ptr,
6178                 fp->gf_u.gf_path + fp->gf_pathlen - ptr);
6179
6180 out:
6181         RETURN(rc);
6182 }
6183
6184 /**
6185  * Given an MDT object, use mdt_path_current to get the path.
6186  * Essentially a wrapper to retry mdt_path_current a set number of times
6187  * if -EAGAIN is returned (usually because an object is being moved).
6188  *
6189  * Part of the MDT layer implementation of lfs fid2path.
6190  *
6191  * \param[in]     info  Per-thread common data shared by mdt level handlers.
6192  * \param[in]     obj   Object to do path lookup of
6193  * \param[in,out] fp    User-provided struct for arguments and to store path
6194  *                      information
6195  *
6196  * \retval 0 Lookup successful, path information stored in fp
6197  * \retval negative errno if there was a problem
6198  */
6199 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
6200                     struct getinfo_fid2path *fp, struct lu_fid *root_fid)
6201 {
6202         struct mdt_device       *mdt = info->mti_mdt;
6203         int                     tries = 3;
6204         int                     rc = -EAGAIN;
6205         ENTRY;
6206
6207         if (fp->gf_pathlen < 3)
6208                 RETURN(-EOVERFLOW);
6209
6210         if (root_fid == NULL)
6211                 root_fid = &mdt->mdt_md_root_fid;
6212
6213         if (lu_fid_eq(root_fid, mdt_object_fid(obj))) {
6214                 fp->gf_u.gf_path[0] = '\0';
6215                 RETURN(0);
6216         }
6217
6218         /* Retry multiple times in case file is being moved */
6219         while (tries-- && rc == -EAGAIN)
6220                 rc = mdt_path_current(info, obj, fp, root_fid);
6221
6222         RETURN(rc);
6223 }
6224
6225 /**
6226  * Get the full path of the provided FID, as of changelog record recno.
6227  *
6228  * This checks sanity and looks up object for user provided FID
6229  * before calling the actual path lookup code.
6230  *
6231  * Part of the MDT layer implementation of lfs fid2path.
6232  *
6233  * \param[in]     info  Per-thread common data shared by mdt level handlers.
6234  * \param[in,out] fp    User-provided struct for arguments and to store path
6235  *                      information
6236  *
6237  * \retval 0 Lookup successful, path information and recno stored in fp
6238  * \retval -ENOENT, object does not exist
6239  * \retval negative errno if there was a problem
6240  */
6241 static int mdt_fid2path(struct mdt_thread_info *info,
6242                         struct lu_fid *root_fid,
6243                         struct getinfo_fid2path *fp)
6244 {
6245         struct mdt_device *mdt = info->mti_mdt;
6246         struct mdt_object *obj;
6247         int    rc;
6248         ENTRY;
6249
6250         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
6251                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
6252
6253         if (!fid_is_sane(&fp->gf_fid))
6254                 RETURN(-EINVAL);
6255
6256         if (!fid_is_namespace_visible(&fp->gf_fid)) {
6257                 CDEBUG(D_INFO, "%s: "DFID" is invalid, f_seq should be >= %#llx"
6258                        ", or f_oid != 0, or f_ver == 0\n", mdt_obd_name(mdt),
6259                        PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
6260                 RETURN(-EINVAL);
6261         }
6262
6263         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
6264         if (IS_ERR(obj)) {
6265                 rc = PTR_ERR(obj);
6266                 CDEBUG(D_IOCTL, "cannot find "DFID": rc = %d\n",
6267                        PFID(&fp->gf_fid), rc);
6268                 RETURN(rc);
6269         }
6270
6271         if (mdt_object_remote(obj))
6272                 rc = -EREMOTE;
6273         else if (!mdt_object_exists(obj))
6274                 rc = -ENOENT;
6275         else
6276                 rc = 0;
6277
6278         if (rc < 0) {
6279                 mdt_object_put(info->mti_env, obj);
6280                 CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n",
6281                        PFID(&fp->gf_fid), rc);
6282                 RETURN(rc);
6283         }
6284
6285         rc = mdt_path(info, obj, fp, root_fid);
6286
6287         CDEBUG(D_INFO, "fid "DFID", path %s recno %#llx linkno %u\n",
6288                PFID(&fp->gf_fid), fp->gf_u.gf_path,
6289                fp->gf_recno, fp->gf_linkno);
6290
6291         mdt_object_put(info->mti_env, obj);
6292
6293         RETURN(rc);
6294 }
6295
6296 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, int keylen,
6297                             void *val, int vallen)
6298 {
6299         struct getinfo_fid2path *fpout, *fpin;
6300         struct lu_fid *root_fid = NULL;
6301         int rc = 0;
6302
6303         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
6304         fpout = val;
6305
6306         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
6307                 lustre_swab_fid2path(fpin);
6308
6309         memcpy(fpout, fpin, sizeof(*fpin));
6310         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
6311                 RETURN(-EINVAL);
6312
6313         if (keylen >= cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*fpin) +
6314                       sizeof(struct lu_fid)) {
6315                 /* client sent its root FID, which is normally fileset FID */
6316                 root_fid = fpin->gf_u.gf_root_fid;
6317                 if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
6318                         lustre_swab_lu_fid(root_fid);
6319
6320                 if (root_fid != NULL && !fid_is_sane(root_fid))
6321                         RETURN(-EINVAL);
6322         }
6323
6324         rc = mdt_fid2path(info, root_fid, fpout);
6325         RETURN(rc);
6326 }
6327
6328 int mdt_get_info(struct tgt_session_info *tsi)
6329 {
6330         char    *key;
6331         int      keylen;
6332         __u32   *vallen;
6333         void    *valout;
6334         int      rc;
6335
6336         ENTRY;
6337
6338         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
6339         if (key == NULL) {
6340                 CDEBUG(D_IOCTL, "No GETINFO key\n");
6341                 RETURN(err_serious(-EFAULT));
6342         }
6343         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
6344                                       RCL_CLIENT);
6345
6346         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
6347         if (vallen == NULL) {
6348                 CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n",
6349                                 tgt_name(tsi->tsi_tgt));
6350                 RETURN(err_serious(-EFAULT));
6351         }
6352
6353         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
6354                              *vallen);
6355         rc = req_capsule_server_pack(tsi->tsi_pill);
6356         if (rc)
6357                 RETURN(err_serious(rc));
6358
6359         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
6360         if (valout == NULL) {
6361                 CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n",
6362                                 tgt_name(tsi->tsi_tgt));
6363                 RETURN(err_serious(-EFAULT));
6364         }
6365
6366         if (KEY_IS(KEY_FID2PATH)) {
6367                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
6368
6369                 rc = mdt_rpc_fid2path(info, key, keylen, valout, *vallen);
6370                 mdt_thread_info_fini(info);
6371         } else {
6372                 rc = -EINVAL;
6373         }
6374         RETURN(rc);
6375 }
6376
6377 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
6378 {
6379         struct obd_ioctl_data *data = karg;
6380         struct lu_fid *fid;
6381         __u64 version;
6382         struct mdt_object *obj;
6383         struct mdt_lock_handle  *lh;
6384         int rc;
6385         ENTRY;
6386
6387         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
6388             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
6389                 RETURN(-EINVAL);
6390
6391         fid = (struct lu_fid *)data->ioc_inlbuf1;
6392
6393         if (!fid_is_sane(fid))
6394                 RETURN(-EINVAL);
6395
6396         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
6397
6398         lh = &mti->mti_lh[MDT_LH_PARENT];
6399         mdt_lock_reg_init(lh, LCK_CR);
6400
6401         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
6402         if (IS_ERR(obj))
6403                 RETURN(PTR_ERR(obj));
6404
6405         if (mdt_object_remote(obj)) {
6406                 rc = -EREMOTE;
6407                 /**
6408                  * before calling version get the correct MDS should be
6409                  * fid, this is error to find remote object here
6410                  */
6411                 CERROR("nonlocal object "DFID"\n", PFID(fid));
6412         } else if (!mdt_object_exists(obj)) {
6413                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
6414                 rc = -ENOENT;
6415         } else {
6416                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
6417                *(__u64 *)data->ioc_inlbuf2 = version;
6418                 rc = 0;
6419         }
6420         mdt_object_unlock_put(mti, obj, lh, 1);
6421         RETURN(rc);
6422 }
6423
6424 /* ioctls on obd dev */
6425 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
6426                          void *karg, void __user *uarg)
6427 {
6428         struct lu_env      env;
6429         struct obd_device *obd = exp->exp_obd;
6430         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
6431         struct dt_device  *dt = mdt->mdt_bottom;
6432         int rc;
6433
6434         ENTRY;
6435         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
6436         rc = lu_env_init(&env, LCT_MD_THREAD);
6437         if (rc)
6438                 RETURN(rc);
6439
6440         switch (cmd) {
6441         case OBD_IOC_SYNC:
6442                 rc = mdt_device_sync(&env, mdt);
6443                 break;
6444         case OBD_IOC_SET_READONLY:
6445                 rc = dt_sync(&env, dt);
6446                 if (rc == 0)
6447                         rc = dt_ro(&env, dt);
6448                 break;
6449         case OBD_IOC_ABORT_RECOVERY:
6450                 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
6451                 obd->obd_abort_recovery = 1;
6452                 target_stop_recovery_thread(obd);
6453                 rc = 0;
6454                 break;
6455         case OBD_IOC_CHANGELOG_REG:
6456         case OBD_IOC_CHANGELOG_DEREG:
6457         case OBD_IOC_CHANGELOG_CLEAR:
6458                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(&env,
6459                                                            mdt->mdt_child,
6460                                                            cmd, len, karg);
6461                 break;
6462         case OBD_IOC_START_LFSCK: {
6463                 struct md_device *next = mdt->mdt_child;
6464                 struct obd_ioctl_data *data = karg;
6465                 struct lfsck_start_param lsp;
6466
6467                 if (unlikely(data == NULL)) {
6468                         rc = -EINVAL;
6469                         break;
6470                 }
6471
6472                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
6473                 lsp.lsp_index_valid = 0;
6474                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
6475                 break;
6476         }
6477         case OBD_IOC_STOP_LFSCK: {
6478                 struct md_device        *next = mdt->mdt_child;
6479                 struct obd_ioctl_data   *data = karg;
6480                 struct lfsck_stop        stop;
6481
6482                 stop.ls_status = LS_STOPPED;
6483                 /* Old lfsck utils may pass NULL @stop. */
6484                 if (data->ioc_inlbuf1 == NULL)
6485                         stop.ls_flags = 0;
6486                 else
6487                         stop.ls_flags =
6488                         ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
6489
6490                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
6491                 break;
6492         }
6493         case OBD_IOC_QUERY_LFSCK: {
6494                 struct md_device        *next = mdt->mdt_child;
6495                 struct obd_ioctl_data   *data = karg;
6496
6497                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0,
6498                                                  data->ioc_inlbuf1);
6499                 break;
6500         }
6501         case OBD_IOC_GET_OBJ_VERSION: {
6502                 struct mdt_thread_info *mti;
6503                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6504                 memset(mti, 0, sizeof *mti);
6505                 mti->mti_env = &env;
6506                 mti->mti_mdt = mdt;
6507                 mti->mti_exp = exp;
6508
6509                 rc = mdt_ioc_version_get(mti, karg);
6510                 break;
6511         }
6512         case OBD_IOC_CATLOGLIST: {
6513                 struct mdt_thread_info *mti;
6514
6515                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6516                 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
6517                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
6518                                        &mti->mti_tmp_fid1);
6519                 break;
6520          }
6521         default:
6522                 rc = -EOPNOTSUPP;
6523                 CERROR("%s: Not supported cmd = %d, rc = %d\n",
6524                         mdt_obd_name(mdt), cmd, rc);
6525         }
6526
6527         lu_env_fini(&env);
6528         RETURN(rc);
6529 }
6530
6531 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
6532 {
6533         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
6534         int rc;
6535         ENTRY;
6536
6537         if (!mdt->mdt_skip_lfsck && !mdt->mdt_bottom->dd_rdonly) {
6538                 struct lfsck_start_param lsp;
6539
6540                 lsp.lsp_start = NULL;
6541                 lsp.lsp_index_valid = 0;
6542                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
6543                                                            OBD_IOC_START_LFSCK,
6544                                                            0, &lsp);
6545                 if (rc != 0 && rc != -EALREADY)
6546                         CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
6547                               mdt_obd_name(mdt), rc);
6548         }
6549
6550         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
6551         RETURN(rc);
6552 }
6553
6554 static int mdt_obd_postrecov(struct obd_device *obd)
6555 {
6556         struct lu_env env;
6557         int rc;
6558
6559         rc = lu_env_init(&env, LCT_MD_THREAD);
6560         if (rc)
6561                 RETURN(rc);
6562         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
6563         lu_env_fini(&env);
6564         return rc;
6565 }
6566
6567 static struct obd_ops mdt_obd_device_ops = {
6568         .o_owner          = THIS_MODULE,
6569         .o_set_info_async = mdt_obd_set_info_async,
6570         .o_connect        = mdt_obd_connect,
6571         .o_reconnect      = mdt_obd_reconnect,
6572         .o_disconnect     = mdt_obd_disconnect,
6573         .o_init_export    = mdt_init_export,
6574         .o_destroy_export = mdt_destroy_export,
6575         .o_iocontrol      = mdt_iocontrol,
6576         .o_postrecov      = mdt_obd_postrecov,
6577         /* Data-on-MDT IO methods */
6578         .o_preprw         = mdt_obd_preprw,
6579         .o_commitrw       = mdt_obd_commitrw,
6580 };
6581
6582 static struct lu_device* mdt_device_fini(const struct lu_env *env,
6583                                          struct lu_device *d)
6584 {
6585         struct mdt_device *m = mdt_dev(d);
6586         ENTRY;
6587
6588         mdt_fini(env, m);
6589         RETURN(NULL);
6590 }
6591
6592 static struct lu_device *mdt_device_free(const struct lu_env *env,
6593                                          struct lu_device *d)
6594 {
6595         struct mdt_device *m = mdt_dev(d);
6596         ENTRY;
6597
6598         lu_device_fini(&m->mdt_lu_dev);
6599         OBD_FREE_PTR(m);
6600
6601         RETURN(NULL);
6602 }
6603
6604 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
6605                                           struct lu_device_type *t,
6606                                           struct lustre_cfg *cfg)
6607 {
6608         struct lu_device  *l;
6609         struct mdt_device *m;
6610
6611         OBD_ALLOC_PTR(m);
6612         if (m != NULL) {
6613                 int rc;
6614
6615                 l = &m->mdt_lu_dev;
6616                 rc = mdt_init0(env, m, t, cfg);
6617                 if (rc != 0) {
6618                         mdt_device_free(env, l);
6619                         l = ERR_PTR(rc);
6620                         return l;
6621                 }
6622         } else
6623                 l = ERR_PTR(-ENOMEM);
6624         return l;
6625 }
6626
6627 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
6628 LU_KEY_INIT(mdt, struct mdt_thread_info);
6629
6630 static void mdt_key_fini(const struct lu_context *ctx,
6631                          struct lu_context_key *key, void* data)
6632 {
6633         struct mdt_thread_info *info = data;
6634
6635         if (info->mti_big_lmm) {
6636                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
6637                 info->mti_big_lmm = NULL;
6638                 info->mti_big_lmmsize = 0;
6639         }
6640
6641         if (info->mti_big_acl) {
6642                 OBD_FREE_LARGE(info->mti_big_acl, info->mti_big_aclsize);
6643                 info->mti_big_acl = NULL;
6644                 info->mti_big_aclsize = 0;
6645         }
6646
6647         OBD_FREE_PTR(info);
6648 }
6649
6650 /* context key: mdt_thread_key */
6651 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
6652
6653 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
6654 {
6655         return lu_ucred(info->mti_env);
6656 }
6657
6658 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
6659 {
6660         return lu_ucred_check(info->mti_env);
6661 }
6662
6663 /**
6664  * Enable/disable COS (Commit On Sharing).
6665  *
6666  * Set/Clear the COS flag in mdt options.
6667  *
6668  * \param mdt mdt device
6669  * \param val 0 disables COS, other values enable COS
6670  */
6671 void mdt_enable_cos(struct mdt_device *mdt, int val)
6672 {
6673         struct lu_env env;
6674         int rc;
6675
6676         mdt->mdt_opts.mo_cos = !!val;
6677         rc = lu_env_init(&env, LCT_LOCAL);
6678         if (unlikely(rc != 0)) {
6679                 CWARN("%s: lu_env initialization failed, cannot "
6680                       "sync: rc = %d\n", mdt_obd_name(mdt), rc);
6681                 return;
6682         }
6683         mdt_device_sync(&env, mdt);
6684         lu_env_fini(&env);
6685 }
6686
6687 /**
6688  * Check COS (Commit On Sharing) status.
6689  *
6690  * Return COS flag status.
6691  *
6692  * \param mdt mdt device
6693  */
6694 int mdt_cos_is_enabled(struct mdt_device *mdt)
6695 {
6696         return mdt->mdt_opts.mo_cos != 0;
6697 }
6698
6699 static struct lu_device_type_operations mdt_device_type_ops = {
6700         .ldto_device_alloc = mdt_device_alloc,
6701         .ldto_device_free  = mdt_device_free,
6702         .ldto_device_fini  = mdt_device_fini
6703 };
6704
6705 static struct lu_device_type mdt_device_type = {
6706         .ldt_tags     = LU_DEVICE_MD,
6707         .ldt_name     = LUSTRE_MDT_NAME,
6708         .ldt_ops      = &mdt_device_type_ops,
6709         .ldt_ctx_tags = LCT_MD_THREAD
6710 };
6711
6712 static int __init mdt_init(void)
6713 {
6714         int rc;
6715
6716         CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") ==
6717                  FID_NOBRACE_LEN + 1);
6718         CLASSERT(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") ==
6719                  FID_LEN + 1);
6720         rc = lu_kmem_init(mdt_caches);
6721         if (rc)
6722                 return rc;
6723
6724         rc = mds_mod_init();
6725         if (rc)
6726                 GOTO(lu_fini, rc);
6727
6728         rc = class_register_type(&mdt_obd_device_ops, NULL, true, NULL,
6729                                  LUSTRE_MDT_NAME, &mdt_device_type);
6730         if (rc)
6731                 GOTO(mds_fini, rc);
6732 lu_fini:
6733         if (rc)
6734                 lu_kmem_fini(mdt_caches);
6735 mds_fini:
6736         if (rc)
6737                 mds_mod_exit();
6738         return rc;
6739 }
6740
6741 static void __exit mdt_exit(void)
6742 {
6743         class_unregister_type(LUSTRE_MDT_NAME);
6744         mds_mod_exit();
6745         lu_kmem_fini(mdt_caches);
6746 }
6747
6748 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
6749 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
6750 MODULE_VERSION(LUSTRE_VERSION_STRING);
6751 MODULE_LICENSE("GPL");
6752
6753 module_init(mdt_init);
6754 module_exit(mdt_exit);