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