lustre/mdt/mdt_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2010, 2016, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/mdt/mdt_handler.c
  33  *
  34  * Lustre Metadata Target (mdt) request handler
  35  *
  36  * Author: Peter Braam <braam@clusterfs.com>
  37  * Author: Andreas Dilger <adilger@clusterfs.com>
  38  * Author: Phil Schwan <phil@clusterfs.com>
  39  * Author: Mike Shaver <shaver@clusterfs.com>
  40  * Author: Nikita Danilov <nikita@clusterfs.com>
  41  * Author: Huang Hua <huanghua@clusterfs.com>
  42  * Author: Yury Umanets <umka@clusterfs.com>
  43  */
  44
  45 #define DEBUG_SUBSYSTEM S_MDS
  46
  47 #include <linux/module.h>
  48 #include <linux/pagemap.h>
  49
  50 #include <dt_object.h>
  51 #include <lustre_acl.h>
  52 #include <lustre_export.h>
  53 #include <uapi/linux/lustre/lustre_ioctl.h>
  54 #include <lustre_lfsck.h>
  55 #include <lustre_log.h>
  56 #include <lustre_nodemap.h>
  57 #include <lustre_mds.h>
  58 #include <uapi/linux/lustre/lustre_param.h>
  59 #include <lustre_quota.h>
  60 #include <lustre_swab.h>
  61 #include <obd.h>
  62 #include <obd_support.h>
  63 #include <lustre_barrier.h>
  64 #include <obd_cksum.h>
  65 #include <llog_swab.h>
  66
  67 #include "mdt_internal.h"
  68
  69
  70 static unsigned int max_mod_rpcs_per_client = 8;
  71 module_param(max_mod_rpcs_per_client, uint, 0644);
  72 MODULE_PARM_DESC(max_mod_rpcs_per_client, "maximum number of modify RPCs in flight allowed per client");
  73
  74 mdl_mode_t mdt_mdl_lock_modes[] = {
  75         [LCK_MINMODE] = MDL_MINMODE,
  76         [LCK_EX]      = MDL_EX,
  77         [LCK_PW]      = MDL_PW,
  78         [LCK_PR]      = MDL_PR,
  79         [LCK_CW]      = MDL_CW,
  80         [LCK_CR]      = MDL_CR,
  81         [LCK_NL]      = MDL_NL,
  82         [LCK_GROUP]   = MDL_GROUP
  83 };
  84
  85 enum ldlm_mode mdt_dlm_lock_modes[] = {
  86         [MDL_MINMODE]   = LCK_MINMODE,
  87         [MDL_EX]        = LCK_EX,
  88         [MDL_PW]        = LCK_PW,
  89         [MDL_PR]        = LCK_PR,
  90         [MDL_CW]        = LCK_CW,
  91         [MDL_CR]        = LCK_CR,
  92         [MDL_NL]        = LCK_NL,
  93         [MDL_GROUP]     = LCK_GROUP
  94 };
  95
  96 static struct mdt_device *mdt_dev(struct lu_device *d);
  97 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
  98
  99 static const struct lu_object_operations mdt_obj_ops;
 100
 101 /* Slab for MDT object allocation */
 102 static struct kmem_cache *mdt_object_kmem;
 103
 104 /* For HSM restore handles */
 105 struct kmem_cache *mdt_hsm_cdt_kmem;
 106
 107 /* For HSM request handles */
 108 struct kmem_cache *mdt_hsm_car_kmem;
 109
 110 static struct lu_kmem_descr mdt_caches[] = {
 111         {
 112                 .ckd_cache = &mdt_object_kmem,
 113                 .ckd_name  = "mdt_obj",
 114                 .ckd_size  = sizeof(struct mdt_object)
 115         },
 116         {
 117                 .ckd_cache      = &mdt_hsm_cdt_kmem,
 118                 .ckd_name       = "mdt_cdt_restore_handle",
 119                 .ckd_size       = sizeof(struct cdt_restore_handle)
 120         },
 121         {
 122                 .ckd_cache      = &mdt_hsm_car_kmem,
 123                 .ckd_name       = "mdt_cdt_agent_req",
 124                 .ckd_size       = sizeof(struct cdt_agent_req)
 125         },
 126         {
 127                 .ckd_cache = NULL
 128         }
 129 };
 130
 131 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag)
 132 {
 133         if (!rep)
 134                 return 0;
 135         return rep->lock_policy_res1 & op_flag;
 136 }
 137
 138 void mdt_clear_disposition(struct mdt_thread_info *info,
 139                            struct ldlm_reply *rep, __u64 op_flag)
 140 {
 141         if (info) {
 142                 info->mti_opdata &= ~op_flag;
 143                 tgt_opdata_clear(info->mti_env, op_flag);
 144         }
 145         if (rep)
 146                 rep->lock_policy_res1 &= ~op_flag;
 147 }
 148
 149 void mdt_set_disposition(struct mdt_thread_info *info,
 150                          struct ldlm_reply *rep, __u64 op_flag)
 151 {
 152         if (info) {
 153                 info->mti_opdata |= op_flag;
 154                 tgt_opdata_set(info->mti_env, op_flag);
 155         }
 156         if (rep)
 157                 rep->lock_policy_res1 |= op_flag;
 158 }
 159
 160 void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm)
 161 {
 162         lh->mlh_pdo_hash = 0;
 163         lh->mlh_reg_mode = lm;
 164         lh->mlh_rreg_mode = lm;
 165         lh->mlh_type = MDT_REG_LOCK;
 166 }
 167
 168 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode,
 169                        const struct lu_name *lname)
 170 {
 171         lh->mlh_reg_mode = lock_mode;
 172         lh->mlh_pdo_mode = LCK_MINMODE;
 173         lh->mlh_rreg_mode = lock_mode;
 174         lh->mlh_type = MDT_PDO_LOCK;
 175
 176         if (lu_name_is_valid(lname)) {
 177                 lh->mlh_pdo_hash = ll_full_name_hash(NULL, lname->ln_name,
 178                                                      lname->ln_namelen);
 179                 /* XXX Workaround for LU-2856
 180                  *
 181                  * Zero is a valid return value of full_name_hash, but
 182                  * several users of mlh_pdo_hash assume a non-zero
 183                  * hash value. We therefore map zero onto an
 184                  * arbitrary, but consistent value (1) to avoid
 185                  * problems further down the road. */
 186                 if (unlikely(lh->mlh_pdo_hash == 0))
 187                         lh->mlh_pdo_hash = 1;
 188         } else {
 189                 lh->mlh_pdo_hash = 0;
 190         }
 191 }
 192
 193 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 194                               struct mdt_lock_handle *lh)
 195 {
 196         mdl_mode_t mode;
 197         ENTRY;
 198
 199         /*
 200          * Any dir access needs couple of locks:
 201          *
 202          * 1) on part of dir we gonna take lookup/modify;
 203          *
 204          * 2) on whole dir to protect it from concurrent splitting and/or to
 205          * flush client's cache for readdir().
 206          *
 207          * so, for a given mode and object this routine decides what lock mode
 208          * to use for lock #2:
 209          *
 210          * 1) if caller's gonna lookup in dir then we need to protect dir from
 211          * being splitted only - LCK_CR
 212          *
 213          * 2) if caller's gonna modify dir then we need to protect dir from
 214          * being splitted and to flush cache - LCK_CW
 215          *
 216          * 3) if caller's gonna modify dir and that dir seems ready for
 217          * splitting then we need to protect it from any type of access
 218          * (lookup/modify/split) - LCK_EX --bzzz
 219          */
 220
 221         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 222         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 223
 224         /*
 225          * Ask underlaying level its opinion about preferable PDO lock mode
 226          * having access type passed as regular lock mode:
 227          *
 228          * - MDL_MINMODE means that lower layer does not want to specify lock
 229          * mode;
 230          *
 231          * - MDL_NL means that no PDO lock should be taken. This is used in some
 232          * cases. Say, for non-splittable directories no need to use PDO locks
 233          * at all.
 234          */
 235         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 236                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 237
 238         if (mode != MDL_MINMODE) {
 239                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 240         } else {
 241                 /*
 242                  * Lower layer does not want to specify locking mode. We do it
 243                  * our selves. No special protection is needed, just flush
 244                  * client's cache on modification and allow concurrent
 245                  * mondification.
 246                  */
 247                 switch (lh->mlh_reg_mode) {
 248                 case LCK_EX:
 249                         lh->mlh_pdo_mode = LCK_EX;
 250                         break;
 251                 case LCK_PR:
 252                         lh->mlh_pdo_mode = LCK_CR;
 253                         break;
 254                 case LCK_PW:
 255                         lh->mlh_pdo_mode = LCK_CW;
 256                         break;
 257                 default:
 258                         CERROR("Not expected lock type (0x%x)\n",
 259                                (int)lh->mlh_reg_mode);
 260                         LBUG();
 261                 }
 262         }
 263
 264         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 265         EXIT;
 266 }
 267
 268 static int mdt_lookup_fileset(struct mdt_thread_info *info, const char *fileset,
 269                               struct lu_fid *fid)
 270 {
 271         struct mdt_device *mdt = info->mti_mdt;
 272         struct lu_name *lname = &info->mti_name;
 273         char *name = NULL;
 274         struct mdt_object *parent;
 275         u32 mode;
 276         int rc = 0;
 277
 278         LASSERT(!info->mti_cross_ref);
 279
 280         OBD_ALLOC(name, NAME_MAX + 1);
 281         if (name == NULL)
 282                 return -ENOMEM;
 283         lname->ln_name = name;
 284
 285         /*
 286          * We may want to allow this to mount a completely separate
 287          * fileset from the MDT in the future, but keeping it to
 288          * ROOT/ only for now avoid potential security issues.
 289          */
 290         *fid = mdt->mdt_md_root_fid;
 291
 292         while (rc == 0 && fileset != NULL && *fileset != '\0') {
 293                 const char *s1 = fileset;
 294                 const char *s2;
 295
 296                 while (*++s1 == '/')
 297                         ;
 298                 s2 = s1;
 299                 while (*s2 != '/' && *s2 != '\0')
 300                         s2++;
 301
 302                 if (s2 == s1)
 303                         break;
 304
 305                 fileset = s2;
 306
 307                 lname->ln_namelen = s2 - s1;
 308                 if (lname->ln_namelen > NAME_MAX) {
 309                         rc = -EINVAL;
 310                         break;
 311                 }
 312
 313                 /* reject .. as a path component */
 314                 if (lname->ln_namelen == 2 &&
 315                     strncmp(s1, "..", 2) == 0) {
 316                         rc = -EINVAL;
 317                         break;
 318                 }
 319
 320                 strncpy(name, s1, lname->ln_namelen);
 321                 name[lname->ln_namelen] = '\0';
 322
 323                 parent = mdt_object_find(info->mti_env, mdt, fid);
 324                 if (IS_ERR(parent)) {
 325                         rc = PTR_ERR(parent);
 326                         break;
 327                 }
 328                 /* Only got the fid of this obj by name */
 329                 fid_zero(fid);
 330                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
 331                                 fid, &info->mti_spec);
 332                 mdt_object_put(info->mti_env, parent);
 333         }
 334         if (!rc) {
 335                 parent = mdt_object_find(info->mti_env, mdt, fid);
 336                 if (IS_ERR(parent))
 337                         rc = PTR_ERR(parent);
 338                 else {
 339                         mode = lu_object_attr(&parent->mot_obj);
 340                         mdt_object_put(info->mti_env, parent);
 341                         if (!S_ISDIR(mode))
 342                                 rc = -ENOTDIR;
 343                 }
 344         }
 345
 346         OBD_FREE(name, NAME_MAX + 1);
 347
 348         return rc;
 349 }
 350
 351 static int mdt_get_root(struct tgt_session_info *tsi)
 352 {
 353         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
 354         struct mdt_device       *mdt = info->mti_mdt;
 355         struct mdt_body         *repbody;
 356         char                    *fileset = NULL, *buffer = NULL;
 357         int                      rc;
 358         struct obd_export       *exp = info->mti_exp;
 359         char                    *nodemap_fileset;
 360
 361         ENTRY;
 362
 363         rc = mdt_check_ucred(info);
 364         if (rc)
 365                 GOTO(out, rc = err_serious(rc));
 366
 367         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GET_ROOT_PACK))
 368                 GOTO(out, rc = err_serious(-ENOMEM));
 369
 370         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 371         if (req_capsule_get_size(info->mti_pill, &RMF_NAME, RCL_CLIENT) > 0) {
 372                 fileset = req_capsule_client_get(info->mti_pill, &RMF_NAME);
 373                 if (fileset == NULL)
 374                         GOTO(out, rc = err_serious(-EFAULT));
 375         }
 376
 377         nodemap_fileset = nodemap_get_fileset(exp->exp_target_data.ted_nodemap);
 378         if (nodemap_fileset && nodemap_fileset[0]) {
 379                 CDEBUG(D_INFO, "nodemap fileset is %s\n", nodemap_fileset);
 380                 if (fileset) {
 381                         /* consider fileset from client as a sub-fileset
 382                          * of the nodemap one */
 383                         OBD_ALLOC(buffer, PATH_MAX + 1);
 384                         if (buffer == NULL)
 385                                 GOTO(out, rc = err_serious(-ENOMEM));
 386                         if (snprintf(buffer, PATH_MAX + 1, "%s/%s",
 387                                      nodemap_fileset, fileset) >= PATH_MAX + 1)
 388                                 GOTO(out, rc = err_serious(-EINVAL));
 389                         fileset = buffer;
 390                 } else {
 391                         /* enforce fileset as specified in the nodemap */
 392                         fileset = nodemap_fileset;
 393                 }
 394         }
 395
 396         if (fileset) {
 397                 CDEBUG(D_INFO, "Getting fileset %s\n", fileset);
 398                 rc = mdt_lookup_fileset(info, fileset, &repbody->mbo_fid1);
 399                 if (rc < 0)
 400                         GOTO(out, rc = err_serious(rc));
 401         } else {
 402                 repbody->mbo_fid1 = mdt->mdt_md_root_fid;
 403         }
 404         repbody->mbo_valid |= OBD_MD_FLID;
 405
 406         EXIT;
 407 out:
 408         mdt_thread_info_fini(info);
 409         if (buffer)
 410                 OBD_FREE(buffer, PATH_MAX+1);
 411         return rc;
 412 }
 413
 414 static int mdt_statfs(struct tgt_session_info *tsi)
 415 {
 416         struct ptlrpc_request           *req = tgt_ses_req(tsi);
 417         struct mdt_thread_info          *info = tsi2mdt_info(tsi);
 418         struct mdt_device               *mdt = info->mti_mdt;
 419         struct tg_grants_data           *tgd = &mdt->mdt_lut.lut_tgd;
 420         struct ptlrpc_service_part      *svcpt;
 421         struct obd_statfs               *osfs;
 422         int                             rc;
 423
 424         ENTRY;
 425
 426         svcpt = req->rq_rqbd->rqbd_svcpt;
 427
 428         /* This will trigger a watchdog timeout */
 429         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 430                          (MDT_SERVICE_WATCHDOG_FACTOR *
 431                           at_get(&svcpt->scp_at_estimate)) + 1);
 432
 433         rc = mdt_check_ucred(info);
 434         if (rc)
 435                 GOTO(out, rc = err_serious(rc));
 436
 437         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
 438                 GOTO(out, rc = err_serious(-ENOMEM));
 439
 440         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 441         if (!osfs)
 442                 GOTO(out, rc = -EPROTO);
 443
 444         rc = tgt_statfs_internal(tsi->tsi_env, &mdt->mdt_lut, osfs,
 445                                  cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 446                                  NULL);
 447         if (unlikely(rc))
 448                 GOTO(out, rc);
 449
 450         /* at least try to account for cached pages.  its still racy and
 451          * might be under-reporting if clients haven't announced their
 452          * caches with brw recently */
 453         CDEBUG(D_SUPER | D_CACHE, "blocks cached %llu granted %llu"
 454                " pending %llu free %llu avail %llu\n",
 455                tgd->tgd_tot_dirty, tgd->tgd_tot_granted,
 456                tgd->tgd_tot_pending,
 457                osfs->os_bfree << tgd->tgd_blockbits,
 458                osfs->os_bavail << tgd->tgd_blockbits);
 459
 460         osfs->os_bavail -= min_t(u64, osfs->os_bavail,
 461                                  ((tgd->tgd_tot_dirty + tgd->tgd_tot_pending +
 462                                    osfs->os_bsize - 1) >> tgd->tgd_blockbits));
 463
 464         tgt_grant_sanity_check(mdt->mdt_lu_dev.ld_obd, __func__);
 465         CDEBUG(D_CACHE, "%llu blocks: %llu free, %llu avail; "
 466                "%llu objects: %llu free; state %x\n",
 467                osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
 468                osfs->os_files, osfs->os_ffree, osfs->os_state);
 469
 470         if (!exp_grant_param_supp(tsi->tsi_exp) &&
 471             tgd->tgd_blockbits > COMPAT_BSIZE_SHIFT) {
 472                 /* clients which don't support OBD_CONNECT_GRANT_PARAM
 473                  * should not see a block size > page size, otherwise
 474                  * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
 475                  * block size which is the biggest block size known to work
 476                  * with all client's page size. */
 477                 osfs->os_blocks <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
 478                 osfs->os_bfree  <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
 479                 osfs->os_bavail <<= tgd->tgd_blockbits - COMPAT_BSIZE_SHIFT;
 480                 osfs->os_bsize = 1 << COMPAT_BSIZE_SHIFT;
 481         }
 482         if (rc == 0)
 483                 mdt_counter_incr(req, LPROC_MDT_STATFS);
 484 out:
 485         mdt_thread_info_fini(info);
 486         RETURN(rc);
 487 }
 488
 489 /**
 490  * Pack size attributes into the reply.
 491  */
 492 int mdt_pack_size2body(struct mdt_thread_info *info,
 493                         const struct lu_fid *fid, bool dom_lock)
 494 {
 495         struct mdt_body *b;
 496         struct md_attr *ma = &info->mti_attr;
 497         int dom_stripe;
 498
 499         ENTRY;
 500
 501         LASSERT(ma->ma_attr.la_valid & LA_MODE);
 502
 503         if (!S_ISREG(ma->ma_attr.la_mode) ||
 504             !(ma->ma_valid & MA_LOV && ma->ma_lmm != NULL))
 505                 RETURN(-ENODATA);
 506
 507         dom_stripe = mdt_lmm_dom_entry(ma->ma_lmm);
 508         /* no DoM stripe, no size in reply */
 509         if (dom_stripe == LMM_NO_DOM)
 510                 RETURN(-ENOENT);
 511
 512         /* no DoM lock, no size in reply */
 513         if (!dom_lock)
 514                 RETURN(0);
 515
 516         /* Either DoM lock exists or LMM has only DoM stripe then
 517          * return size on body. */
 518         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 519
 520         mdt_dom_object_size(info->mti_env, info->mti_mdt, fid, b, dom_lock);
 521         RETURN(0);
 522 }
 523
 524 #ifdef CONFIG_FS_POSIX_ACL
 525 /*
 526  * Pack ACL data into the reply. UIDs/GIDs are mapped and filtered by nodemap.
 527  *
 528  * \param       info    thread info object
 529  * \param       repbody reply to pack ACLs into
 530  * \param       o       mdt object of file to examine
 531  * \param       nodemap nodemap of client to reply to
 532  * \retval      0       success
 533  * \retval      -errno  error getting or parsing ACL from disk
 534  */
 535 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
 536                       struct mdt_object *o, struct lu_nodemap *nodemap)
 537 {
 538         const struct lu_env     *env = info->mti_env;
 539         struct md_object        *next = mdt_object_child(o);
 540         struct lu_buf           *buf = &info->mti_buf;
 541         struct mdt_device       *mdt = info->mti_mdt;
 542         int rc;
 543
 544         ENTRY;
 545
 546         buf->lb_buf = req_capsule_server_get(info->mti_pill, &RMF_ACL);
 547         buf->lb_len = req_capsule_get_size(info->mti_pill, &RMF_ACL,
 548                                            RCL_SERVER);
 549         if (buf->lb_len == 0)
 550                 RETURN(0);
 551
 552 again:
 553         rc = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_ACCESS);
 554         if (rc < 0) {
 555                 if (rc == -ENODATA) {
 556                         repbody->mbo_aclsize = 0;
 557                         repbody->mbo_valid |= OBD_MD_FLACL;
 558                         rc = 0;
 559                 } else if (rc == -EOPNOTSUPP) {
 560                         rc = 0;
 561                 } else {
 562                         if (rc == -ERANGE &&
 563                             exp_connect_large_acl(info->mti_exp) &&
 564                             buf->lb_buf != info->mti_big_acl) {
 565                                 if (info->mti_big_acl == NULL) {
 566                                         OBD_ALLOC_LARGE(info->mti_big_acl,
 567                                                         mdt->mdt_max_ea_size);
 568                                         if (info->mti_big_acl == NULL) {
 569                                                 CERROR("%s: unable to grow "
 570                                                        DFID" ACL buffer\n",
 571                                                        mdt_obd_name(mdt),
 572                                                        PFID(mdt_object_fid(o)));
 573                                                 RETURN(-ENOMEM);
 574                                         }
 575
 576                                         info->mti_big_aclsize =
 577                                                         mdt->mdt_max_ea_size;
 578                                 }
 579
 580                                 CDEBUG(D_INODE, "%s: grow the "DFID
 581                                        " ACL buffer to size %d\n",
 582                                        mdt_obd_name(mdt),
 583                                        PFID(mdt_object_fid(o)),
 584                                        mdt->mdt_max_ea_size);
 585
 586                                 buf->lb_buf = info->mti_big_acl;
 587                                 buf->lb_len = info->mti_big_aclsize;
 588
 589                                 goto again;
 590                         }
 591
 592                         CERROR("%s: unable to read "DFID" ACL: rc = %d\n",
 593                                mdt_obd_name(mdt), PFID(mdt_object_fid(o)), rc);
 594                 }
 595         } else {
 596                 if (buf->lb_buf == info->mti_big_acl)
 597                         info->mti_big_acl_used = 1;
 598
 599                 rc = nodemap_map_acl(nodemap, buf->lb_buf,
 600                                      rc, NODEMAP_FS_TO_CLIENT);
 601                 /* if all ACLs mapped out, rc is still >= 0 */
 602                 if (rc < 0) {
 603                         CERROR("%s: nodemap_map_acl unable to parse "DFID
 604                                " ACL: rc = %d\n", mdt_obd_name(mdt),
 605                                PFID(mdt_object_fid(o)), rc);
 606                 } else {
 607                         repbody->mbo_aclsize = rc;
 608                         repbody->mbo_valid |= OBD_MD_FLACL;
 609                         rc = 0;
 610                 }
 611         }
 612
 613         RETURN(rc);
 614 }
 615 #endif
 616
 617 /* XXX Look into layout in MDT layer. */
 618 static inline bool mdt_hsm_is_released(struct lov_mds_md *lmm)
 619 {
 620         struct lov_comp_md_v1   *comp_v1;
 621         struct lov_mds_md       *v1;
 622         int                      i;
 623
 624         if (lmm->lmm_magic == LOV_MAGIC_COMP_V1) {
 625                 comp_v1 = (struct lov_comp_md_v1 *)lmm;
 626
 627                 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
 628                         v1 = (struct lov_mds_md *)((char *)comp_v1 +
 629                                 comp_v1->lcm_entries[i].lcme_offset);
 630                         /* We don't support partial release for now */
 631                         if (!(v1->lmm_pattern & LOV_PATTERN_F_RELEASED))
 632                                 return false;
 633                 }
 634                 return true;
 635         } else {
 636                 return (lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) ?
 637                         true : false;
 638         }
 639 }
 640
 641 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 642                         const struct lu_attr *attr, const struct lu_fid *fid)
 643 {
 644         struct md_attr *ma = &info->mti_attr;
 645         struct obd_export *exp = info->mti_exp;
 646         struct lu_nodemap *nodemap = NULL;
 647
 648         LASSERT(ma->ma_valid & MA_INODE);
 649
 650         if (attr->la_valid & LA_ATIME) {
 651                 b->mbo_atime = attr->la_atime;
 652                 b->mbo_valid |= OBD_MD_FLATIME;
 653         }
 654         if (attr->la_valid & LA_MTIME) {
 655                 b->mbo_mtime = attr->la_mtime;
 656                 b->mbo_valid |= OBD_MD_FLMTIME;
 657         }
 658         if (attr->la_valid & LA_CTIME) {
 659                 b->mbo_ctime = attr->la_ctime;
 660                 b->mbo_valid |= OBD_MD_FLCTIME;
 661         }
 662         if (attr->la_valid & LA_FLAGS) {
 663                 b->mbo_flags = attr->la_flags;
 664                 b->mbo_valid |= OBD_MD_FLFLAGS;
 665         }
 666         if (attr->la_valid & LA_NLINK) {
 667                 b->mbo_nlink = attr->la_nlink;
 668                 b->mbo_valid |= OBD_MD_FLNLINK;
 669         }
 670         if (attr->la_valid & (LA_UID|LA_GID)) {
 671                 nodemap = nodemap_get_from_exp(exp);
 672                 if (IS_ERR(nodemap))
 673                         goto out;
 674         }
 675         if (attr->la_valid & LA_UID) {
 676                 b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 677                                             NODEMAP_FS_TO_CLIENT,
 678                                             attr->la_uid);
 679                 b->mbo_valid |= OBD_MD_FLUID;
 680         }
 681         if (attr->la_valid & LA_GID) {
 682                 b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 683                                             NODEMAP_FS_TO_CLIENT,
 684                                             attr->la_gid);
 685                 b->mbo_valid |= OBD_MD_FLGID;
 686         }
 687
 688         if (attr->la_valid & LA_PROJID) {
 689                 /* TODO, nodemap for project id */
 690                 b->mbo_projid = attr->la_projid;
 691                 b->mbo_valid |= OBD_MD_FLPROJID;
 692         }
 693
 694         b->mbo_mode = attr->la_mode;
 695         if (attr->la_valid & LA_MODE)
 696                 b->mbo_valid |= OBD_MD_FLMODE;
 697         if (attr->la_valid & LA_TYPE)
 698                 b->mbo_valid |= OBD_MD_FLTYPE;
 699
 700         if (fid != NULL) {
 701                 b->mbo_fid1 = *fid;
 702                 b->mbo_valid |= OBD_MD_FLID;
 703                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid=%#llx\n",
 704                        PFID(fid), b->mbo_nlink, b->mbo_mode, b->mbo_valid);
 705         }
 706
 707         if (!(attr->la_valid & LA_TYPE))
 708                 return;
 709
 710         b->mbo_rdev   = attr->la_rdev;
 711         b->mbo_size   = attr->la_size;
 712         b->mbo_blocks = attr->la_blocks;
 713
 714         if (!S_ISREG(attr->la_mode)) {
 715                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 716         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
 717                 /* means no objects are allocated on osts. */
 718                 LASSERT(!(ma->ma_valid & MA_LOV));
 719                 /* just ignore blocks occupied by extend attributes on MDS */
 720                 b->mbo_blocks = 0;
 721                 /* if no object is allocated on osts, the size on mds is valid.
 722                  * b=22272 */
 723                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 724         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL) {
 725                 if (mdt_hsm_is_released(ma->ma_lmm)) {
 726                         /* A released file stores its size on MDS. */
 727                         /* But return 1 block for released file, unless tools
 728                          * like tar will consider it fully sparse. (LU-3864)
 729                          */
 730                         if (unlikely(b->mbo_size == 0))
 731                                 b->mbo_blocks = 0;
 732                         else
 733                                 b->mbo_blocks = 1;
 734                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 735                 } else if (info->mti_som_valid) { /* som is valid */
 736                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 737                 }
 738         }
 739
 740         if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE))
 741                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 742                        PFID(fid), (unsigned long long)b->mbo_size);
 743
 744 out:
 745         if (!IS_ERR_OR_NULL(nodemap))
 746                 nodemap_putref(nodemap);
 747 }
 748
 749 static inline int mdt_body_has_lov(const struct lu_attr *la,
 750                                    const struct mdt_body *body)
 751 {
 752         return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
 753                (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
 754 }
 755
 756 void mdt_client_compatibility(struct mdt_thread_info *info)
 757 {
 758         struct mdt_body       *body;
 759         struct ptlrpc_request *req = mdt_info_req(info);
 760         struct obd_export     *exp = req->rq_export;
 761         struct md_attr        *ma = &info->mti_attr;
 762         struct lu_attr        *la = &ma->ma_attr;
 763         ENTRY;
 764
 765         if (exp_connect_layout(exp))
 766                 /* the client can deal with 16-bit lmm_stripe_count */
 767                 RETURN_EXIT;
 768
 769         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 770
 771         if (!mdt_body_has_lov(la, body))
 772                 RETURN_EXIT;
 773
 774         /* now we have a reply with a lov for a client not compatible with the
 775          * layout lock so we have to clean the layout generation number */
 776         if (S_ISREG(la->la_mode))
 777                 ma->ma_lmm->lmm_layout_gen = 0;
 778         EXIT;
 779 }
 780
 781 static int mdt_attr_get_eabuf_size(struct mdt_thread_info *info,
 782                                    struct mdt_object *o)
 783 {
 784         const struct lu_env *env = info->mti_env;
 785         int rc, rc2;
 786
 787         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 788                           XATTR_NAME_LOV);
 789
 790         if (rc == -ENODATA)
 791                 rc = 0;
 792
 793         if (rc < 0)
 794                 goto out;
 795
 796         /* Is it a directory? Let's check for the LMV as well */
 797         if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
 798                 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 799                                    XATTR_NAME_LMV);
 800
 801                 if (rc2 == -ENODATA)
 802                         rc2 = mo_xattr_get(env, mdt_object_child(o),
 803                                            &LU_BUF_NULL,
 804                                            XATTR_NAME_DEFAULT_LMV);
 805
 806                 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
 807                         rc = rc2;
 808         }
 809
 810 out:
 811         return rc;
 812 }
 813
 814 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
 815                       const char *name)
 816 {
 817         const struct lu_env *env = info->mti_env;
 818         int rc;
 819         ENTRY;
 820
 821         LASSERT(info->mti_big_lmm_used == 0);
 822         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
 823         if (rc < 0)
 824                 RETURN(rc);
 825
 826         /* big_lmm may need to be grown */
 827         if (info->mti_big_lmmsize < rc) {
 828                 int size = size_roundup_power2(rc);
 829
 830                 if (info->mti_big_lmmsize > 0) {
 831                         /* free old buffer */
 832                         LASSERT(info->mti_big_lmm);
 833                         OBD_FREE_LARGE(info->mti_big_lmm,
 834                                        info->mti_big_lmmsize);
 835                         info->mti_big_lmm = NULL;
 836                         info->mti_big_lmmsize = 0;
 837                 }
 838
 839                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
 840                 if (info->mti_big_lmm == NULL)
 841                         RETURN(-ENOMEM);
 842                 info->mti_big_lmmsize = size;
 843         }
 844         LASSERT(info->mti_big_lmmsize >= rc);
 845
 846         info->mti_buf.lb_buf = info->mti_big_lmm;
 847         info->mti_buf.lb_len = info->mti_big_lmmsize;
 848         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
 849
 850         RETURN(rc);
 851 }
 852
 853 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
 854                    struct md_attr *ma, const char *name)
 855 {
 856         struct md_object *next = mdt_object_child(o);
 857         struct lu_buf    *buf = &info->mti_buf;
 858         int rc;
 859
 860         if (strcmp(name, XATTR_NAME_LOV) == 0) {
 861                 buf->lb_buf = ma->ma_lmm;
 862                 buf->lb_len = ma->ma_lmm_size;
 863                 LASSERT(!(ma->ma_valid & MA_LOV));
 864         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 865                 buf->lb_buf = ma->ma_lmv;
 866                 buf->lb_len = ma->ma_lmv_size;
 867                 LASSERT(!(ma->ma_valid & MA_LMV));
 868         } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 869                 buf->lb_buf = ma->ma_lmv;
 870                 buf->lb_len = ma->ma_lmv_size;
 871                 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
 872         } else {
 873                 return -EINVAL;
 874         }
 875
 876         rc = mo_xattr_get(info->mti_env, next, buf, name);
 877         if (rc > 0) {
 878
 879 got:
 880                 if (strcmp(name, XATTR_NAME_LOV) == 0) {
 881                         if (info->mti_big_lmm_used)
 882                                 ma->ma_lmm = info->mti_big_lmm;
 883
 884                         /* NOT return LOV EA with hole to old client. */
 885                         if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
 886                                      LOV_PATTERN_F_HOLE) &&
 887                             !(exp_connect_flags(info->mti_exp) &
 888                               OBD_CONNECT_LFSCK)) {
 889                                 return -EIO;
 890                         } else {
 891                                 ma->ma_lmm_size = rc;
 892                                 ma->ma_valid |= MA_LOV;
 893                         }
 894                 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 895                         if (info->mti_big_lmm_used)
 896                                 ma->ma_lmv = info->mti_big_lmm;
 897
 898                         ma->ma_lmv_size = rc;
 899                         ma->ma_valid |= MA_LMV;
 900                 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 901                         ma->ma_lmv_size = rc;
 902                         ma->ma_valid |= MA_LMV_DEF;
 903                 }
 904
 905                 /* Update mdt_max_mdsize so all clients will be aware that */
 906                 if (info->mti_mdt->mdt_max_mdsize < rc)
 907                         info->mti_mdt->mdt_max_mdsize = rc;
 908
 909                 rc = 0;
 910         } else if (rc == -ENODATA) {
 911                 /* no LOV EA */
 912                 rc = 0;
 913         } else if (rc == -ERANGE) {
 914                 /* Default LMV has fixed size, so it must be able to fit
 915                  * in the original buffer */
 916                 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
 917                         return rc;
 918                 rc = mdt_big_xattr_get(info, o, name);
 919                 if (rc > 0) {
 920                         info->mti_big_lmm_used = 1;
 921                         goto got;
 922                 }
 923         }
 924
 925         return rc;
 926 }
 927
 928 static int mdt_attr_get_pfid(struct mdt_thread_info *info,
 929                              struct mdt_object *o, struct lu_fid *pfid)
 930 {
 931         struct lu_buf           *buf = &info->mti_buf;
 932         struct link_ea_header   *leh;
 933         struct link_ea_entry    *lee;
 934         int                      rc;
 935         ENTRY;
 936
 937         buf->lb_buf = info->mti_big_lmm;
 938         buf->lb_len = info->mti_big_lmmsize;
 939         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
 940                           buf, XATTR_NAME_LINK);
 941         /* ignore errors, MA_PFID won't be set and it is
 942          * up to the caller to treat this as an error */
 943         if (rc == -ERANGE || buf->lb_len == 0) {
 944                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
 945                 buf->lb_buf = info->mti_big_lmm;
 946                 buf->lb_len = info->mti_big_lmmsize;
 947         }
 948
 949         if (rc < 0)
 950                 RETURN(rc);
 951         if (rc < sizeof(*leh)) {
 952                 CERROR("short LinkEA on "DFID": rc = %d\n",
 953                        PFID(mdt_object_fid(o)), rc);
 954                 RETURN(-ENODATA);
 955         }
 956
 957         leh = (struct link_ea_header *) buf->lb_buf;
 958         lee = (struct link_ea_entry *)(leh + 1);
 959         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
 960                 leh->leh_magic = LINK_EA_MAGIC;
 961                 leh->leh_reccount = __swab32(leh->leh_reccount);
 962                 leh->leh_len = __swab64(leh->leh_len);
 963         }
 964         if (leh->leh_magic != LINK_EA_MAGIC)
 965                 RETURN(-EINVAL);
 966         if (leh->leh_reccount == 0)
 967                 RETURN(-ENODATA);
 968
 969         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
 970         fid_be_to_cpu(pfid, pfid);
 971
 972         RETURN(0);
 973 }
 974
 975 int mdt_attr_get_complex(struct mdt_thread_info *info,
 976                          struct mdt_object *o, struct md_attr *ma)
 977 {
 978         const struct lu_env *env = info->mti_env;
 979         struct md_object    *next = mdt_object_child(o);
 980         struct lu_buf       *buf = &info->mti_buf;
 981         int                  need = ma->ma_need;
 982         int                  rc = 0, rc2;
 983         u32                  mode;
 984         ENTRY;
 985
 986         ma->ma_valid = 0;
 987
 988         if (mdt_object_exists(o) == 0)
 989                 GOTO(out, rc = -ENOENT);
 990         mode = lu_object_attr(&next->mo_lu);
 991
 992         if (need & MA_INODE) {
 993                 ma->ma_need = MA_INODE;
 994                 rc = mo_attr_get(env, next, ma);
 995                 if (rc)
 996                         GOTO(out, rc);
 997
 998                 if (S_ISREG(mode))
 999                         (void) mdt_get_som(info, o, &ma->ma_attr);
1000                 ma->ma_valid |= MA_INODE;
1001         }
1002
1003         if (need & MA_PFID) {
1004                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
1005                 if (rc == 0)
1006                         ma->ma_valid |= MA_PFID;
1007                 /* ignore this error, parent fid is not mandatory */
1008                 rc = 0;
1009         }
1010
1011         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
1012                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
1013                 if (rc)
1014                         GOTO(out, rc);
1015         }
1016
1017         if (need & MA_LMV && S_ISDIR(mode)) {
1018                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
1019                 if (rc != 0)
1020                         GOTO(out, rc);
1021         }
1022
1023         if (need & MA_LMV_DEF && S_ISDIR(mode)) {
1024                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
1025                 if (rc != 0)
1026                         GOTO(out, rc);
1027         }
1028
1029         if (need & MA_HSM && S_ISREG(mode)) {
1030                 buf->lb_buf = info->mti_xattr_buf;
1031                 buf->lb_len = sizeof(info->mti_xattr_buf);
1032                 CLASSERT(sizeof(struct hsm_attrs) <=
1033                          sizeof(info->mti_xattr_buf));
1034                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
1035                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
1036                 if (rc2 == 0)
1037                         ma->ma_valid |= MA_HSM;
1038                 else if (rc2 < 0 && rc2 != -ENODATA)
1039                         GOTO(out, rc = rc2);
1040         }
1041
1042 #ifdef CONFIG_FS_POSIX_ACL
1043         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
1044                 buf->lb_buf = ma->ma_acl;
1045                 buf->lb_len = ma->ma_acl_size;
1046                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
1047                 if (rc2 > 0) {
1048                         ma->ma_acl_size = rc2;
1049                         ma->ma_valid |= MA_ACL_DEF;
1050                 } else if (rc2 == -ENODATA) {
1051                         /* no ACLs */
1052                         ma->ma_acl_size = 0;
1053                 } else
1054                         GOTO(out, rc = rc2);
1055         }
1056 #endif
1057 out:
1058         ma->ma_need = need;
1059         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = %#llx ma_lmm=%p\n",
1060                rc, ma->ma_valid, ma->ma_lmm);
1061         RETURN(rc);
1062 }
1063
1064 static int mdt_getattr_internal(struct mdt_thread_info *info,
1065                                 struct mdt_object *o, int ma_need)
1066 {
1067         struct md_object        *next = mdt_object_child(o);
1068         const struct mdt_body   *reqbody = info->mti_body;
1069         struct ptlrpc_request   *req = mdt_info_req(info);
1070         struct md_attr          *ma = &info->mti_attr;
1071         struct lu_attr          *la = &ma->ma_attr;
1072         struct req_capsule      *pill = info->mti_pill;
1073         const struct lu_env     *env = info->mti_env;
1074         struct mdt_body         *repbody;
1075         struct lu_buf           *buffer = &info->mti_buf;
1076         struct obd_export       *exp = info->mti_exp;
1077         int                      rc;
1078         ENTRY;
1079
1080         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
1081                 RETURN(err_serious(-ENOMEM));
1082
1083         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1084
1085         ma->ma_valid = 0;
1086
1087         if (mdt_object_remote(o)) {
1088                 /* This object is located on remote node.*/
1089                 /* Return -ENOTSUPP for old client */
1090                 if (!mdt_is_dne_client(req->rq_export))
1091                         GOTO(out, rc = -ENOTSUPP);
1092
1093                 repbody->mbo_fid1 = *mdt_object_fid(o);
1094                 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
1095                 GOTO(out, rc = 0);
1096         }
1097
1098         if (reqbody->mbo_eadatasize > 0) {
1099                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
1100                 if (buffer->lb_buf == NULL)
1101                         GOTO(out, rc = -EPROTO);
1102                 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
1103                                                       RCL_SERVER);
1104         } else {
1105                 buffer->lb_buf = NULL;
1106                 buffer->lb_len = 0;
1107                 ma_need &= ~(MA_LOV | MA_LMV);
1108                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
1109                        mdt_obd_name(info->mti_mdt),
1110                        req->rq_export->exp_client_uuid.uuid);
1111         }
1112
1113         /* If it is dir object and client require MEA, then we got MEA */
1114         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1115             (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
1116                 /* Assumption: MDT_MD size is enough for lmv size. */
1117                 ma->ma_lmv = buffer->lb_buf;
1118                 ma->ma_lmv_size = buffer->lb_len;
1119                 ma->ma_need = MA_INODE;
1120                 if (ma->ma_lmv_size > 0) {
1121                         if (reqbody->mbo_valid & OBD_MD_MEA)
1122                                 ma->ma_need |= MA_LMV;
1123                         else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA)
1124                                 ma->ma_need |= MA_LMV_DEF;
1125                 }
1126         } else {
1127                 ma->ma_lmm = buffer->lb_buf;
1128                 ma->ma_lmm_size = buffer->lb_len;
1129                 ma->ma_need = MA_INODE | MA_HSM;
1130                 if (ma->ma_lmm_size > 0)
1131                         ma->ma_need |= MA_LOV;
1132         }
1133
1134         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1135             reqbody->mbo_valid & OBD_MD_FLDIREA  &&
1136             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
1137                 /* get default stripe info for this dir. */
1138                 ma->ma_need |= MA_LOV_DEF;
1139         }
1140         ma->ma_need |= ma_need;
1141
1142         rc = mdt_attr_get_complex(info, o, ma);
1143         if (unlikely(rc)) {
1144                 CDEBUG(rc == -ENOENT ? D_OTHER : D_ERROR,
1145                        "%s: getattr error for "DFID": rc = %d\n",
1146                        mdt_obd_name(info->mti_mdt),
1147                        PFID(mdt_object_fid(o)), rc);
1148                 RETURN(rc);
1149         }
1150
1151         /* if file is released, check if a restore is running */
1152         if (ma->ma_valid & MA_HSM) {
1153                 repbody->mbo_valid |= OBD_MD_TSTATE;
1154                 if ((ma->ma_hsm.mh_flags & HS_RELEASED) &&
1155                     mdt_hsm_restore_is_running(info, mdt_object_fid(o)))
1156                         repbody->mbo_t_state = MS_RESTORE;
1157         }
1158
1159         if (likely(ma->ma_valid & MA_INODE))
1160                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
1161         else
1162                 RETURN(-EFAULT);
1163
1164         if (mdt_body_has_lov(la, reqbody)) {
1165                 if (ma->ma_valid & MA_LOV) {
1166                         LASSERT(ma->ma_lmm_size);
1167                         repbody->mbo_eadatasize = ma->ma_lmm_size;
1168                         if (S_ISDIR(la->la_mode))
1169                                 repbody->mbo_valid |= OBD_MD_FLDIREA;
1170                         else
1171                                 repbody->mbo_valid |= OBD_MD_FLEASIZE;
1172                         mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
1173                 }
1174                 if (ma->ma_valid & MA_LMV) {
1175                         /* Return -ENOTSUPP for old client */
1176                         if (!mdt_is_striped_client(req->rq_export))
1177                                 RETURN(-ENOTSUPP);
1178
1179                         LASSERT(S_ISDIR(la->la_mode));
1180                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1181                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1182                         repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
1183                 }
1184                 if (ma->ma_valid & MA_LMV_DEF) {
1185                         /* Return -ENOTSUPP for old client */
1186                         if (!mdt_is_striped_client(req->rq_export))
1187                                 RETURN(-ENOTSUPP);
1188                         LASSERT(S_ISDIR(la->la_mode));
1189                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1190                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1191                         repbody->mbo_valid |= (OBD_MD_FLDIREA |
1192                                                OBD_MD_DEFAULT_MEA);
1193                 }
1194         } else if (S_ISLNK(la->la_mode) &&
1195                    reqbody->mbo_valid & OBD_MD_LINKNAME) {
1196                 buffer->lb_buf = ma->ma_lmm;
1197                 /* eadatasize from client includes NULL-terminator, so
1198                  * there is no need to read it */
1199                 buffer->lb_len = reqbody->mbo_eadatasize - 1;
1200                 rc = mo_readlink(env, next, buffer);
1201                 if (unlikely(rc <= 0)) {
1202                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
1203                                mdt_obd_name(info->mti_mdt),
1204                                PFID(mdt_object_fid(o)), rc);
1205                         rc = -EFAULT;
1206                 } else {
1207                         int print_limit = min_t(int, PAGE_SIZE - 128, rc);
1208
1209                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
1210                                 rc -= 2;
1211                         repbody->mbo_valid |= OBD_MD_LINKNAME;
1212                         /* we need to report back size with NULL-terminator
1213                          * because client expects that */
1214                         repbody->mbo_eadatasize = rc + 1;
1215                         if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
1216                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
1217                                        "on "DFID ", expected %d\n",
1218                                        mdt_obd_name(info->mti_mdt),
1219                                        rc, PFID(mdt_object_fid(o)),
1220                                        reqbody->mbo_eadatasize - 1);
1221                         /* NULL terminate */
1222                         ((char *)ma->ma_lmm)[rc] = 0;
1223
1224                         /* If the total CDEBUG() size is larger than a page, it
1225                          * will print a warning to the console, avoid this by
1226                          * printing just the last part of the symlink. */
1227                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
1228                                print_limit < rc ? "..." : "", print_limit,
1229                                (char *)ma->ma_lmm + rc - print_limit, rc);
1230                         rc = 0;
1231                 }
1232         }
1233
1234         if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
1235                 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1236                 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1237                 CDEBUG(D_INODE, "changing the max MD size to %u\n",
1238                        repbody->mbo_max_mdsize);
1239         }
1240
1241 #ifdef CONFIG_FS_POSIX_ACL
1242         if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1243                  (reqbody->mbo_valid & OBD_MD_FLACL)) {
1244                 struct lu_nodemap *nodemap = nodemap_get_from_exp(exp);
1245                 if (IS_ERR(nodemap))
1246                         RETURN(PTR_ERR(nodemap));
1247
1248                 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
1249                 nodemap_putref(nodemap);
1250         }
1251 #endif
1252
1253 out:
1254         if (rc == 0)
1255                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
1256
1257         RETURN(rc);
1258 }
1259
1260 static int mdt_getattr(struct tgt_session_info *tsi)
1261 {
1262         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1263         struct mdt_object       *obj = info->mti_object;
1264         struct req_capsule      *pill = info->mti_pill;
1265         struct mdt_body         *reqbody;
1266         struct mdt_body         *repbody;
1267         int rc, rc2;
1268         ENTRY;
1269
1270         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1271         LASSERT(reqbody);
1272         LASSERT(obj != NULL);
1273         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1274
1275         /* Unlike intent case where we need to pre-fill out buffers early on
1276          * in intent policy for ldlm reasons, here we can have a much better
1277          * guess at EA size by just reading it from disk.
1278          * Exceptions are readdir and (missing) directory striping */
1279         /* Readlink */
1280         if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1281                 /* No easy way to know how long is the symlink, but it cannot
1282                  * be more than PATH_MAX, so we allocate +1 */
1283                 rc = PATH_MAX + 1;
1284
1285         /* A special case for fs ROOT: getattr there might fetch
1286          * default EA for entire fs, not just for this dir!
1287          */
1288         } else if (lu_fid_eq(mdt_object_fid(obj),
1289                              &info->mti_mdt->mdt_md_root_fid) &&
1290                    (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1291                    (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1292                                                                  MDS_GETATTR)) {
1293                 /* Should the default strping be bigger, mdt_fix_reply
1294                  * will reallocate */
1295                 rc = DEF_REP_MD_SIZE;
1296         } else {
1297                 /* Read the actual EA size from disk */
1298                 rc = mdt_attr_get_eabuf_size(info, obj);
1299         }
1300
1301         if (rc < 0)
1302                 GOTO(out, rc = err_serious(rc));
1303
1304         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1305
1306         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
1307          * by default. If the target object has more ACL entries, then
1308          * enlarge the buffer when necessary. */
1309         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
1310                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
1311
1312         rc = req_capsule_server_pack(pill);
1313         if (unlikely(rc != 0))
1314                 GOTO(out, rc = err_serious(rc));
1315
1316         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1317         LASSERT(repbody != NULL);
1318         repbody->mbo_eadatasize = 0;
1319         repbody->mbo_aclsize = 0;
1320
1321         rc = mdt_check_ucred(info);
1322         if (unlikely(rc))
1323                 GOTO(out_shrink, rc);
1324
1325         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1326
1327         rc = mdt_getattr_internal(info, obj, 0);
1328         EXIT;
1329 out_shrink:
1330         mdt_client_compatibility(info);
1331         rc2 = mdt_fix_reply(info);
1332         if (rc == 0)
1333                 rc = rc2;
1334 out:
1335         mdt_thread_info_fini(info);
1336         return rc;
1337 }
1338
1339 /**
1340  * Handler of layout intent RPC requiring the layout modification
1341  *
1342  * \param[in] info      thread environment
1343  * \param[in] obj       object
1344  * \param[in] layout    layout change descriptor
1345  *
1346  * \retval 0    on success
1347  * \retval < 0  error code
1348  */
1349 static int mdt_layout_change(struct mdt_thread_info *info,
1350                              struct mdt_object *obj,
1351                              struct md_layout_change *layout)
1352 {
1353         struct mdt_lock_handle *lh = &info->mti_lh[MDT_LH_LOCAL];
1354         struct layout_intent *intent = layout->mlc_intent;
1355         int rc;
1356         ENTRY;
1357
1358         CDEBUG(D_INFO, "got layout change request from client: "
1359                "opc:%u flags:%#x extent "DEXT"\n",
1360                intent->li_opc, intent->li_flags, PEXT(&intent->li_extent));
1361
1362         if (intent->li_extent.e_start >= intent->li_extent.e_end) {
1363                 CERROR(DFID ":invalid range of layout change "DEXT"\n",
1364                        PFID(mdt_object_fid(obj)), PEXT(&intent->li_extent));
1365                 RETURN(-EINVAL);
1366         }
1367
1368         if (!mdt_object_exists(obj))
1369                 GOTO(out, rc = -ENOENT);
1370
1371         if (!S_ISREG(lu_object_attr(&obj->mot_obj)))
1372                 GOTO(out, rc = -EINVAL);
1373
1374         rc = mo_permission(info->mti_env, NULL, mdt_object_child(obj), NULL,
1375                            MAY_WRITE);
1376         if (rc)
1377                 GOTO(out, rc);
1378
1379         /* take layout lock to prepare layout change */
1380         mdt_lock_reg_init(lh, LCK_EX);
1381         rc = mdt_object_lock(info, obj, lh, MDS_INODELOCK_LAYOUT);
1382         if (rc)
1383                 GOTO(out, rc);
1384
1385         rc = mo_layout_change(info->mti_env, mdt_object_child(obj), layout);
1386
1387         mdt_object_unlock(info, obj, lh, 1);
1388 out:
1389         RETURN(rc);
1390 }
1391
1392 /**
1393  * Exchange MOF_LOV_CREATED flags between two objects after a
1394  * layout swap. No assumption is made on whether o1 or o2 have
1395  * created objects or not.
1396  *
1397  * \param[in,out] o1    First swap layout object
1398  * \param[in,out] o2    Second swap layout object
1399  */
1400 static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2)
1401 {
1402         unsigned int o1_lov_created = o1->mot_lov_created;
1403
1404         mutex_lock(&o1->mot_lov_mutex);
1405         mutex_lock(&o2->mot_lov_mutex);
1406
1407         o1->mot_lov_created = o2->mot_lov_created;
1408         o2->mot_lov_created = o1_lov_created;
1409
1410         mutex_unlock(&o2->mot_lov_mutex);
1411         mutex_unlock(&o1->mot_lov_mutex);
1412 }
1413
1414 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1415 {
1416         struct mdt_thread_info  *info;
1417         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1418         struct obd_export       *exp = req->rq_export;
1419         struct mdt_object       *o1, *o2, *o;
1420         struct mdt_lock_handle  *lh1, *lh2;
1421         struct mdc_swap_layouts *msl;
1422         int                      rc;
1423         ENTRY;
1424
1425         /* client does not support layout lock, so layout swaping
1426          * is disabled.
1427          * FIXME: there is a problem for old clients which don't support
1428          * layout lock yet. If those clients have already opened the file
1429          * they won't be notified at all so that old layout may still be
1430          * used to do IO. This can be fixed after file release is landed by
1431          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1432         if (!exp_connect_layout(exp))
1433                 RETURN(-EOPNOTSUPP);
1434
1435         info = tsi2mdt_info(tsi);
1436
1437         if (info->mti_dlm_req != NULL)
1438                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1439
1440         o1 = info->mti_object;
1441         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1442                                 &info->mti_body->mbo_fid2);
1443         if (IS_ERR(o))
1444                 GOTO(out, rc = PTR_ERR(o));
1445
1446         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1447                 GOTO(put, rc = -ENOENT);
1448
1449         rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1450         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1451                 GOTO(put, rc);
1452
1453         if (rc < 0)
1454                 swap(o1, o2);
1455
1456         /* permission check. Make sure the calling process having permission
1457          * to write both files. */
1458         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1459                                 MAY_WRITE);
1460         if (rc < 0)
1461                 GOTO(put, rc);
1462
1463         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1464                                 MAY_WRITE);
1465         if (rc < 0)
1466                 GOTO(put, rc);
1467
1468         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1469         if (msl == NULL)
1470                 GOTO(put, rc = -EPROTO);
1471
1472         lh1 = &info->mti_lh[MDT_LH_NEW];
1473         mdt_lock_reg_init(lh1, LCK_EX);
1474         lh2 = &info->mti_lh[MDT_LH_OLD];
1475         mdt_lock_reg_init(lh2, LCK_EX);
1476
1477         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1478                              MDS_INODELOCK_XATTR);
1479         if (rc < 0)
1480                 GOTO(put, rc);
1481
1482         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1483                              MDS_INODELOCK_XATTR);
1484         if (rc < 0)
1485                 GOTO(unlock1, rc);
1486
1487         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1488                              mdt_object_child(o2), msl->msl_flags);
1489         if (rc < 0)
1490                 GOTO(unlock2, rc);
1491
1492         mdt_swap_lov_flag(o1, o2);
1493
1494 unlock2:
1495         mdt_object_unlock(info, o2, lh2, rc);
1496 unlock1:
1497         mdt_object_unlock(info, o1, lh1, rc);
1498 put:
1499         mdt_object_put(info->mti_env, o);
1500 out:
1501         mdt_thread_info_fini(info);
1502         RETURN(rc);
1503 }
1504
1505 static int mdt_raw_lookup(struct mdt_thread_info *info,
1506                           struct mdt_object *parent,
1507                           const struct lu_name *lname,
1508                           struct ldlm_reply *ldlm_rep)
1509 {
1510         struct lu_fid   *child_fid = &info->mti_tmp_fid1;
1511         int              rc;
1512         ENTRY;
1513
1514         LASSERT(!info->mti_cross_ref);
1515
1516         /* Only got the fid of this obj by name */
1517         fid_zero(child_fid);
1518         rc = mdo_lookup(info->mti_env, mdt_object_child(info->mti_object),
1519                         lname, child_fid, &info->mti_spec);
1520         if (rc == 0) {
1521                 struct mdt_body *repbody;
1522
1523                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1524                 repbody->mbo_fid1 = *child_fid;
1525                 repbody->mbo_valid = OBD_MD_FLID;
1526                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1527         } else if (rc == -ENOENT) {
1528                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1529         }
1530
1531         RETURN(rc);
1532 }
1533
1534 /*
1535  * UPDATE lock should be taken against parent, and be released before exit;
1536  * child_bits lock should be taken against child, and be returned back:
1537  *            (1)normal request should release the child lock;
1538  *            (2)intent request will grant the lock to client.
1539  */
1540 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1541                                  struct mdt_lock_handle *lhc,
1542                                  __u64 child_bits,
1543                                  struct ldlm_reply *ldlm_rep)
1544 {
1545         struct ptlrpc_request  *req = mdt_info_req(info);
1546         struct mdt_body        *reqbody = NULL;
1547         struct mdt_object      *parent = info->mti_object;
1548         struct mdt_object      *child;
1549         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1550         struct lu_name         *lname = NULL;
1551         struct mdt_lock_handle *lhp = NULL;
1552         struct ldlm_lock       *lock;
1553         __u64 try_bits = 0;
1554         bool is_resent;
1555         int ma_need = 0;
1556         int rc;
1557
1558         ENTRY;
1559
1560         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1561         LASSERT(ergo(is_resent,
1562                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1563
1564         if (parent == NULL)
1565                 RETURN(-ENOENT);
1566
1567         if (info->mti_cross_ref) {
1568                 /* Only getattr on the child. Parent is on another node. */
1569                 mdt_set_disposition(info, ldlm_rep,
1570                                     DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1571                 child = parent;
1572                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1573                        "ldlm_rep = %p\n",
1574                        PFID(mdt_object_fid(child)), ldlm_rep);
1575
1576                 rc = mdt_check_resent_lock(info, child, lhc);
1577                 if (rc < 0) {
1578                         RETURN(rc);
1579                 } else if (rc > 0) {
1580                         mdt_lock_handle_init(lhc);
1581                         mdt_lock_reg_init(lhc, LCK_PR);
1582
1583                         /*
1584                          * Object's name is on another MDS, no lookup or layout
1585                          * lock is needed here but update lock is.
1586                          */
1587                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
1588                                         MDS_INODELOCK_LAYOUT);
1589                         child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1590
1591                         rc = mdt_object_lock(info, child, lhc, child_bits);
1592                         if (rc < 0)
1593                                 RETURN(rc);
1594                 }
1595
1596                 /* Finally, we can get attr for child. */
1597                 if (!mdt_object_exists(child)) {
1598                         LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1599                                         &child->mot_obj,
1600                                         "remote object doesn't exist.");
1601                         mdt_object_unlock(info, child, lhc, 1);
1602                         RETURN(-ENOENT);
1603                 }
1604
1605                 rc = mdt_getattr_internal(info, child, 0);
1606                 if (unlikely(rc != 0))
1607                         mdt_object_unlock(info, child, lhc, 1);
1608
1609                 RETURN(rc);
1610         }
1611
1612         lname = &info->mti_name;
1613         mdt_name_unpack(info->mti_pill, &RMF_NAME, lname, MNF_FIX_ANON);
1614
1615         if (lu_name_is_valid(lname)) {
1616                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1617                        "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1618                        PNAME(lname), ldlm_rep);
1619         } else {
1620                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1621                 if (unlikely(reqbody == NULL))
1622                         RETURN(err_serious(-EPROTO));
1623
1624                 *child_fid = reqbody->mbo_fid2;
1625
1626                 if (unlikely(!fid_is_sane(child_fid)))
1627                         RETURN(err_serious(-EINVAL));
1628
1629                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1630                        "ldlm_rep = %p\n",
1631                        PFID(mdt_object_fid(parent)),
1632                        PFID(&reqbody->mbo_fid2), ldlm_rep);
1633         }
1634
1635         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1636
1637         if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1638                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1639                                 &parent->mot_obj,
1640                                 "Parent doesn't exist!");
1641                 RETURN(-ESTALE);
1642         }
1643
1644         if (mdt_object_remote(parent)) {
1645                 CERROR("%s: parent "DFID" is on remote target\n",
1646                        mdt_obd_name(info->mti_mdt),
1647                        PFID(mdt_object_fid(parent)));
1648                 RETURN(-EIO);
1649         }
1650
1651         if (lu_name_is_valid(lname)) {
1652                 /* Always allow to lookup ".." */
1653                 if (unlikely(lname->ln_namelen == 2 &&
1654                              lname->ln_name[0] == '.' &&
1655                              lname->ln_name[1] == '.'))
1656                         info->mti_spec.sp_permitted = 1;
1657
1658                 if (info->mti_body->mbo_valid == OBD_MD_FLID) {
1659                         rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1660
1661                         RETURN(rc);
1662                 }
1663
1664                 /* step 1: lock parent only if parent is a directory */
1665                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1666                         lhp = &info->mti_lh[MDT_LH_PARENT];
1667                         mdt_lock_pdo_init(lhp, LCK_PR, lname);
1668                         rc = mdt_object_lock(info, parent, lhp,
1669                                              MDS_INODELOCK_UPDATE);
1670                         if (unlikely(rc != 0))
1671                                 RETURN(rc);
1672                 }
1673
1674                 /* step 2: lookup child's fid by name */
1675                 fid_zero(child_fid);
1676                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
1677                                 child_fid, &info->mti_spec);
1678                 if (rc == -ENOENT)
1679                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1680
1681                 if (rc != 0)
1682                         GOTO(out_parent, rc);
1683         }
1684
1685         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1686
1687         /*
1688          *step 3: find the child object by fid & lock it.
1689          *        regardless if it is local or remote.
1690          *
1691          *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1692          *      set parent dir fid the same as child fid in getattr by fid case
1693          *      we should not lu_object_find() the object again, could lead
1694          *      to hung if there is a concurrent unlink destroyed the object.
1695          */
1696         if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1697                 mdt_object_get(info->mti_env, parent);
1698                 child = parent;
1699         } else {
1700                 child = mdt_object_find(info->mti_env, info->mti_mdt,
1701                                         child_fid);
1702         }
1703
1704         if (unlikely(IS_ERR(child)))
1705                 GOTO(out_parent, rc = PTR_ERR(child));
1706
1707         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
1708         if (!mdt_object_exists(child)) {
1709                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1710                                 &child->mot_obj,
1711                                 "Object doesn't exist!");
1712                 GOTO(out_child, rc = -ENOENT);
1713         }
1714
1715         rc = mdt_check_resent_lock(info, child, lhc);
1716         if (rc < 0) {
1717                 GOTO(out_child, rc);
1718         } else if (rc > 0) {
1719                 mdt_lock_handle_init(lhc);
1720                 mdt_lock_reg_init(lhc, LCK_PR);
1721
1722                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1723                       mdt_object_exists(child) && !mdt_object_remote(child)) {
1724                         struct md_attr *ma = &info->mti_attr;
1725
1726                         ma->ma_valid = 0;
1727                         ma->ma_need = MA_INODE;
1728                         rc = mdt_attr_get_complex(info, child, ma);
1729                         if (unlikely(rc != 0))
1730                                 GOTO(out_child, rc);
1731
1732                         /* If the file has not been changed for some time, we
1733                          * return not only a LOOKUP lock, but also an UPDATE
1734                          * lock and this might save us RPC on later STAT. For
1735                          * directories, it also let negative dentry cache start
1736                          * working for this dir. */
1737                         if (ma->ma_valid & MA_INODE &&
1738                             ma->ma_attr.la_valid & LA_CTIME &&
1739                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1740                                 ma->ma_attr.la_ctime < cfs_time_current_sec())
1741                                 child_bits |= MDS_INODELOCK_UPDATE;
1742                 }
1743
1744                 /* layout lock must be granted in a best-effort way
1745                  * for IT operations */
1746                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1747                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
1748                     !mdt_object_remote(child) && ldlm_rep != NULL) {
1749                         if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1750                             exp_connect_layout(info->mti_exp)) {
1751                                 /* try to grant layout lock for regular file. */
1752                                 try_bits = MDS_INODELOCK_LAYOUT;
1753                         }
1754                         /* Acquire DOM lock in advance for data-on-mdt file */
1755                         if (child != parent)
1756                                 try_bits |= MDS_INODELOCK_DOM;
1757                 }
1758
1759                 if (try_bits != 0) {
1760                         /* try layout lock, it may fail to be granted due to
1761                          * contention at LOOKUP or UPDATE */
1762                         rc = mdt_object_lock_try(info, child, lhc, &child_bits,
1763                                                  try_bits, false);
1764                         if (child_bits & MDS_INODELOCK_LAYOUT)
1765                                 ma_need |= MA_LOV;
1766                 } else {
1767                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1768                          * client will enqueue the lock to the remote MDT */
1769                         if (mdt_object_remote(child))
1770                                 child_bits &= ~MDS_INODELOCK_UPDATE;
1771                         rc = mdt_object_lock(info, child, lhc, child_bits);
1772                 }
1773                 if (unlikely(rc != 0))
1774                         GOTO(out_child, rc);
1775         }
1776
1777         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1778
1779         /* finally, we can get attr for child. */
1780         rc = mdt_getattr_internal(info, child, ma_need);
1781         if (unlikely(rc != 0)) {
1782                 mdt_object_unlock(info, child, lhc, 1);
1783         } else if (lock) {
1784                 /* Debugging code. */
1785                 LDLM_DEBUG(lock, "Returning lock to client");
1786                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1787                                          &lock->l_resource->lr_name),
1788                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1789                          PLDLMRES(lock->l_resource),
1790                          PFID(mdt_object_fid(child)));
1791
1792                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
1793                     mdt_object_exists(child) && !mdt_object_remote(child) &&
1794                     child != parent) {
1795                         LDLM_LOCK_PUT(lock);
1796                         mdt_object_put(info->mti_env, child);
1797                         /* NB: call the mdt_pack_size2body always after
1798                          * mdt_object_put(), that is why this speacial
1799                          * exit path is used. */
1800                         rc = mdt_pack_size2body(info, child_fid,
1801                                                 child_bits & MDS_INODELOCK_DOM);
1802                         if (rc != 0 && child_bits & MDS_INODELOCK_DOM) {
1803                                 /* DOM lock was taken in advance but this is
1804                                  * not DoM file. Drop the lock. */
1805                                 lock_res_and_lock(lock);
1806                                 ldlm_inodebits_drop(lock, MDS_INODELOCK_DOM);
1807                                 unlock_res_and_lock(lock);
1808                         }
1809
1810                         GOTO(out_parent, rc = 0);
1811                 }
1812         }
1813         if (lock)
1814                 LDLM_LOCK_PUT(lock);
1815
1816         EXIT;
1817 out_child:
1818         mdt_object_put(info->mti_env, child);
1819 out_parent:
1820         if (lhp)
1821                 mdt_object_unlock(info, parent, lhp, 1);
1822         return rc;
1823 }
1824
1825 /* normal handler: should release the child lock */
1826 static int mdt_getattr_name(struct tgt_session_info *tsi)
1827 {
1828         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1829         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1830         struct mdt_body        *reqbody;
1831         struct mdt_body        *repbody;
1832         int rc, rc2;
1833         ENTRY;
1834
1835         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1836         LASSERT(reqbody != NULL);
1837         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1838         LASSERT(repbody != NULL);
1839
1840         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1841         repbody->mbo_eadatasize = 0;
1842         repbody->mbo_aclsize = 0;
1843
1844         rc = mdt_init_ucred_intent_getattr(info, reqbody);
1845         if (unlikely(rc))
1846                 GOTO(out_shrink, rc);
1847
1848         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1849         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1850                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1851                 lhc->mlh_reg_lh.cookie = 0;
1852         }
1853         mdt_exit_ucred(info);
1854         EXIT;
1855 out_shrink:
1856         mdt_client_compatibility(info);
1857         rc2 = mdt_fix_reply(info);
1858         if (rc == 0)
1859                 rc = rc2;
1860         mdt_thread_info_fini(info);
1861         return rc;
1862 }
1863
1864 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1865                          void *karg, void __user *uarg);
1866
1867 static int mdt_set_info(struct tgt_session_info *tsi)
1868 {
1869         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1870         char                    *key;
1871         void                    *val;
1872         int                      keylen, vallen, rc = 0;
1873
1874         ENTRY;
1875
1876         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
1877         if (key == NULL) {
1878                 DEBUG_REQ(D_HA, req, "no set_info key");
1879                 RETURN(err_serious(-EFAULT));
1880         }
1881
1882         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
1883                                       RCL_CLIENT);
1884
1885         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
1886         if (val == NULL) {
1887                 DEBUG_REQ(D_HA, req, "no set_info val");
1888                 RETURN(err_serious(-EFAULT));
1889         }
1890
1891         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
1892                                       RCL_CLIENT);
1893
1894         /* Swab any part of val you need to here */
1895         if (KEY_IS(KEY_READ_ONLY)) {
1896                 spin_lock(&req->rq_export->exp_lock);
1897                 if (*(__u32 *)val)
1898                         *exp_connect_flags_ptr(req->rq_export) |=
1899                                 OBD_CONNECT_RDONLY;
1900                 else
1901                         *exp_connect_flags_ptr(req->rq_export) &=
1902                                 ~OBD_CONNECT_RDONLY;
1903                 spin_unlock(&req->rq_export->exp_lock);
1904         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1905                 struct changelog_setinfo *cs = val;
1906
1907                 if (vallen != sizeof(*cs)) {
1908                         CERROR("%s: bad changelog_clear setinfo size %d\n",
1909                                tgt_name(tsi->tsi_tgt), vallen);
1910                         RETURN(-EINVAL);
1911                 }
1912                 if (ptlrpc_req_need_swab(req)) {
1913                         __swab64s(&cs->cs_recno);
1914                         __swab32s(&cs->cs_id);
1915                 }
1916
1917                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
1918                                    vallen, val, NULL);
1919         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
1920                 if (vallen > 0)
1921                         obd_export_evict_by_nid(req->rq_export->exp_obd, val);
1922         } else {
1923                 RETURN(-EINVAL);
1924         }
1925         RETURN(rc);
1926 }
1927
1928 static int mdt_readpage(struct tgt_session_info *tsi)
1929 {
1930         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
1931         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
1932         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
1933         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
1934         struct mdt_body         *repbody;
1935         int                      rc;
1936         int                      i;
1937
1938         ENTRY;
1939
1940         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1941                 RETURN(err_serious(-ENOMEM));
1942
1943         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
1944         if (repbody == NULL || reqbody == NULL)
1945                 RETURN(err_serious(-EFAULT));
1946
1947         /*
1948          * prepare @rdpg before calling lower layers and transfer itself. Here
1949          * reqbody->size contains offset of where to start to read and
1950          * reqbody->nlink contains number bytes to read.
1951          */
1952         rdpg->rp_hash = reqbody->mbo_size;
1953         if (rdpg->rp_hash != reqbody->mbo_size) {
1954                 CERROR("Invalid hash: %#llx != %#llx\n",
1955                        rdpg->rp_hash, reqbody->mbo_size);
1956                 RETURN(-EFAULT);
1957         }
1958
1959         rdpg->rp_attrs = reqbody->mbo_mode;
1960         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
1961                 rdpg->rp_attrs |= LUDA_64BITHASH;
1962         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
1963                                 exp_max_brw_size(tsi->tsi_exp));
1964         rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >>
1965                           PAGE_SHIFT;
1966         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1967         if (rdpg->rp_pages == NULL)
1968                 RETURN(-ENOMEM);
1969
1970         for (i = 0; i < rdpg->rp_npages; ++i) {
1971                 rdpg->rp_pages[i] = alloc_page(GFP_NOFS);
1972                 if (rdpg->rp_pages[i] == NULL)
1973                         GOTO(free_rdpg, rc = -ENOMEM);
1974         }
1975
1976         /* call lower layers to fill allocated pages with directory data */
1977         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
1978         if (rc < 0)
1979                 GOTO(free_rdpg, rc);
1980
1981         /* send pages to client */
1982         rc = tgt_sendpage(tsi, rdpg, rc);
1983
1984         EXIT;
1985 free_rdpg:
1986
1987         for (i = 0; i < rdpg->rp_npages; i++)
1988                 if (rdpg->rp_pages[i] != NULL)
1989                         __free_page(rdpg->rp_pages[i]);
1990         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1991
1992         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1993                 RETURN(0);
1994
1995         return rc;
1996 }
1997
1998 static int mdt_fix_attr_ucred(struct mdt_thread_info *info, __u32 op)
1999 {
2000         struct lu_ucred *uc = mdt_ucred_check(info);
2001         struct lu_attr *attr = &info->mti_attr.ma_attr;
2002
2003         if (uc == NULL)
2004                 return -EINVAL;
2005
2006         if (op != REINT_SETATTR) {
2007                 if ((attr->la_valid & LA_UID) && (attr->la_uid != -1))
2008                         attr->la_uid = uc->uc_fsuid;
2009                 /* for S_ISGID, inherit gid from his parent, such work will be
2010                  * done in cmm/mdd layer, here set all cases as uc->uc_fsgid. */
2011                 if ((attr->la_valid & LA_GID) && (attr->la_gid != -1))
2012                         attr->la_gid = uc->uc_fsgid;
2013         }
2014
2015         return 0;
2016 }
2017
2018 static int mdt_reint_internal(struct mdt_thread_info *info,
2019                               struct mdt_lock_handle *lhc,
2020                               __u32 op)
2021 {
2022         struct req_capsule      *pill = info->mti_pill;
2023         struct mdt_body         *repbody;
2024         int                      rc = 0, rc2;
2025
2026         ENTRY;
2027
2028         rc = mdt_reint_unpack(info, op);
2029         if (rc != 0) {
2030                 CERROR("Can't unpack reint, rc %d\n", rc);
2031                 RETURN(err_serious(rc));
2032         }
2033
2034         /* for replay (no_create) lmm is not needed, client has it already */
2035         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2036                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
2037                                      DEF_REP_MD_SIZE);
2038
2039         /* llog cookies are always 0, the field is kept for compatibility */
2040         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
2041                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
2042
2043         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
2044          * by default. If the target object has more ACL entries, then
2045          * enlarge the buffer when necessary. */
2046         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
2047                 req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
2048                                      LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
2049
2050         rc = req_capsule_server_pack(pill);
2051         if (rc != 0) {
2052                 CERROR("Can't pack response, rc %d\n", rc);
2053                 RETURN(err_serious(rc));
2054         }
2055
2056         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
2057                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
2058                 LASSERT(repbody);
2059                 repbody->mbo_eadatasize = 0;
2060                 repbody->mbo_aclsize = 0;
2061         }
2062
2063         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
2064
2065         /* for replay no cookkie / lmm need, because client have this already */
2066         if (info->mti_spec.no_create)
2067                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
2068                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
2069
2070         rc = mdt_init_ucred_reint(info);
2071         if (rc)
2072                 GOTO(out_shrink, rc);
2073
2074         rc = mdt_fix_attr_ucred(info, op);
2075         if (rc != 0)
2076                 GOTO(out_ucred, rc = err_serious(rc));
2077
2078         rc = mdt_check_resent(info, mdt_reconstruct, lhc);
2079         if (rc < 0) {
2080                 GOTO(out_ucred, rc);
2081         } else if (rc == 1) {
2082                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
2083                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
2084                 GOTO(out_ucred, rc);
2085         }
2086         rc = mdt_reint_rec(info, lhc);
2087         EXIT;
2088 out_ucred:
2089         mdt_exit_ucred(info);
2090 out_shrink:
2091         mdt_client_compatibility(info);
2092         rc2 = mdt_fix_reply(info);
2093         if (rc == 0)
2094                 rc = rc2;
2095         return rc;
2096 }
2097
2098 static long mdt_reint_opcode(struct ptlrpc_request *req,
2099                              const struct req_format **fmt)
2100 {
2101         struct mdt_device       *mdt;
2102         struct mdt_rec_reint    *rec;
2103         long                     opc;
2104
2105         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
2106         if (rec != NULL) {
2107                 opc = rec->rr_opcode;
2108                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
2109                 if (opc < REINT_MAX && fmt[opc] != NULL)
2110                         req_capsule_extend(&req->rq_pill, fmt[opc]);
2111                 else {
2112                         mdt = mdt_exp2dev(req->rq_export);
2113                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
2114                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
2115                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
2116                         opc = err_serious(-EFAULT);
2117                 }
2118         } else {
2119                 opc = err_serious(-EFAULT);
2120         }
2121         return opc;
2122 }
2123
2124 static int mdt_reint(struct tgt_session_info *tsi)
2125 {
2126         long opc;
2127         int  rc;
2128         static const struct req_format *reint_fmts[REINT_MAX] = {
2129                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
2130                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
2131                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
2132                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
2133                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
2134                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
2135                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
2136                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
2137                 [REINT_MIGRATE]  = &RQF_MDS_REINT_RENAME
2138         };
2139
2140         ENTRY;
2141
2142         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
2143         if (opc >= 0) {
2144                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2145                 /*
2146                  * No lock possible here from client to pass it to reint code
2147                  * path.
2148                  */
2149                 rc = mdt_reint_internal(info, NULL, opc);
2150                 mdt_thread_info_fini(info);
2151         } else {
2152                 rc = opc;
2153         }
2154
2155         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
2156         RETURN(rc);
2157 }
2158
2159 /* this should sync the whole device */
2160 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
2161 {
2162         struct dt_device *dt = mdt->mdt_bottom;
2163         int rc;
2164         ENTRY;
2165
2166         rc = dt->dd_ops->dt_sync(env, dt);
2167         RETURN(rc);
2168 }
2169
2170 /* this should sync this object */
2171 static int mdt_object_sync(const struct lu_env *env, struct obd_export *exp,
2172                            struct mdt_object *mo)
2173 {
2174         int rc;
2175
2176         ENTRY;
2177
2178         if (!mdt_object_exists(mo)) {
2179                 CWARN("%s: non existing object "DFID": rc = %d\n",
2180                       exp->exp_obd->obd_name, PFID(mdt_object_fid(mo)),
2181                       -ESTALE);
2182                 RETURN(-ESTALE);
2183         }
2184
2185         rc = mo_object_sync(env, mdt_object_child(mo));
2186
2187         RETURN(rc);
2188 }
2189
2190 static int mdt_sync(struct tgt_session_info *tsi)
2191 {
2192         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2193         struct req_capsule      *pill = tsi->tsi_pill;
2194         struct mdt_body         *body;
2195         int                      rc;
2196
2197         ENTRY;
2198
2199         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
2200                 RETURN(err_serious(-ENOMEM));
2201
2202         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
2203                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
2204         } else {
2205                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2206
2207                 /* sync an object */
2208                 rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp,
2209                                      info->mti_object);
2210                 if (rc == 0) {
2211                         const struct lu_fid *fid;
2212                         struct lu_attr *la = &info->mti_attr.ma_attr;
2213
2214                         info->mti_attr.ma_need = MA_INODE;
2215                         info->mti_attr.ma_valid = 0;
2216                         rc = mdt_attr_get_complex(info, info->mti_object,
2217                                                   &info->mti_attr);
2218                         if (rc == 0) {
2219                                 body = req_capsule_server_get(pill,
2220                                                               &RMF_MDT_BODY);
2221                                 fid = mdt_object_fid(info->mti_object);
2222                                 mdt_pack_attr2body(info, body, la, fid);
2223                         }
2224                 }
2225                 mdt_thread_info_fini(info);
2226         }
2227         if (rc == 0)
2228                 mdt_counter_incr(req, LPROC_MDT_SYNC);
2229
2230         RETURN(rc);
2231 }
2232
2233 static int mdt_data_sync(struct tgt_session_info *tsi)
2234 {
2235         struct mdt_thread_info *info;
2236         struct mdt_device *mdt = mdt_exp2dev(tsi->tsi_exp);
2237         struct ost_body *body = tsi->tsi_ost_body;
2238         struct ost_body *repbody;
2239         struct mdt_object *mo = NULL;
2240         struct md_attr *ma;
2241         int rc = 0;
2242
2243         ENTRY;
2244
2245         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2246
2247         /* if no fid is specified then do nothing,
2248          * device sync is done via MDS_SYNC */
2249         if (fid_is_zero(&tsi->tsi_fid))
2250                 RETURN(0);
2251
2252         mo = mdt_object_find(tsi->tsi_env, mdt, &tsi->tsi_fid);
2253         if (IS_ERR(mo))
2254                 RETURN(PTR_ERR(mo));
2255
2256         rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp, mo);
2257         if (rc)
2258                 GOTO(put, rc);
2259
2260         repbody->oa.o_oi = body->oa.o_oi;
2261         repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
2262
2263         info = tsi2mdt_info(tsi);
2264         ma = &info->mti_attr;
2265         ma->ma_need = MA_INODE;
2266         ma->ma_valid = 0;
2267         rc = mdt_attr_get_complex(info, mo, ma);
2268         if (rc == 0)
2269                 obdo_from_la(&repbody->oa, &ma->ma_attr, VALID_FLAGS);
2270         else
2271                 rc = 0;
2272         mdt_thread_info_fini(info);
2273
2274         EXIT;
2275 put:
2276         if (mo != NULL)
2277                 mdt_object_put(tsi->tsi_env, mo);
2278         return rc;
2279 }
2280
2281 /*
2282  * Handle quota control requests to consult current usage/limit, but also
2283  * to configure quota enforcement
2284  */
2285 static int mdt_quotactl(struct tgt_session_info *tsi)
2286 {
2287         struct obd_export       *exp  = tsi->tsi_exp;
2288         struct req_capsule      *pill = tsi->tsi_pill;
2289         struct obd_quotactl     *oqctl, *repoqc;
2290         int                      id, rc;
2291         struct mdt_device       *mdt = mdt_exp2dev(exp);
2292         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2293         struct lu_nodemap       *nodemap;
2294         ENTRY;
2295
2296         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
2297         if (oqctl == NULL)
2298                 RETURN(err_serious(-EPROTO));
2299
2300         rc = req_capsule_server_pack(pill);
2301         if (rc)
2302                 RETURN(err_serious(rc));
2303
2304         nodemap = nodemap_get_from_exp(exp);
2305         if (IS_ERR(nodemap))
2306                 RETURN(PTR_ERR(nodemap));
2307
2308         switch (oqctl->qc_cmd) {
2309                 /* master quotactl */
2310         case Q_SETINFO:
2311         case Q_SETQUOTA:
2312                 if (!nodemap_can_setquota(nodemap))
2313                         GOTO(out_nodemap, rc = -EPERM);
2314         case Q_GETINFO:
2315         case Q_GETQUOTA:
2316                 if (qmt == NULL)
2317                         GOTO(out_nodemap, rc = -EOPNOTSUPP);
2318                 /* slave quotactl */
2319         case Q_GETOINFO:
2320         case Q_GETOQUOTA:
2321                 break;
2322         default:
2323                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2324                 GOTO(out_nodemap, rc = -EFAULT);
2325         }
2326
2327         id = oqctl->qc_id;
2328         switch (oqctl->qc_type) {
2329         case USRQUOTA:
2330                 id = nodemap_map_id(nodemap, NODEMAP_UID,
2331                                     NODEMAP_CLIENT_TO_FS, id);
2332                 break;
2333         case GRPQUOTA:
2334                 id = nodemap_map_id(nodemap, NODEMAP_GID,
2335                                     NODEMAP_CLIENT_TO_FS, id);
2336                 break;
2337         case PRJQUOTA:
2338                 /* todo: check/map project id */
2339                 id = oqctl->qc_id;
2340                 break;
2341         default:
2342                 GOTO(out_nodemap, rc = -EOPNOTSUPP);
2343         }
2344         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2345         if (repoqc == NULL)
2346                 GOTO(out_nodemap, rc = err_serious(-EFAULT));
2347
2348         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA)
2349                 barrier_exit(tsi->tsi_tgt->lut_bottom);
2350
2351         if (oqctl->qc_id != id)
2352                 swap(oqctl->qc_id, id);
2353
2354         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA) {
2355                 if (unlikely(!barrier_entry(tsi->tsi_tgt->lut_bottom)))
2356                         RETURN(-EINPROGRESS);
2357         }
2358
2359         switch (oqctl->qc_cmd) {
2360
2361         case Q_GETINFO:
2362         case Q_SETINFO:
2363         case Q_SETQUOTA:
2364         case Q_GETQUOTA:
2365                 /* forward quotactl request to QMT */
2366                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2367                 break;
2368
2369         case Q_GETOINFO:
2370         case Q_GETOQUOTA:
2371                 /* slave quotactl */
2372                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2373                                    oqctl);
2374                 break;
2375
2376         default:
2377                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2378                 GOTO(out_nodemap, rc = -EFAULT);
2379         }
2380
2381         if (oqctl->qc_id != id)
2382                 swap(oqctl->qc_id, id);
2383
2384         *repoqc = *oqctl;
2385
2386         EXIT;
2387
2388 out_nodemap:
2389         nodemap_putref(nodemap);
2390
2391         return rc;
2392 }
2393
2394 /** clone llog ctxt from child (mdd)
2395  * This allows remote llog (replicator) access.
2396  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2397  * context was originally set up, or we can handle them directly.
2398  * I choose the latter, but that means I need any llog
2399  * contexts set up by child to be accessable by the mdt.  So we clone the
2400  * context into our context list here.
2401  */
2402 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2403                                int idx)
2404 {
2405         struct md_device  *next = mdt->mdt_child;
2406         struct llog_ctxt *ctxt;
2407         int rc;
2408
2409         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2410                 return 0;
2411
2412         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2413         if (rc || ctxt == NULL) {
2414                 return 0;
2415         }
2416
2417         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2418         if (rc)
2419                 CERROR("Can't set mdt ctxt %d\n", rc);
2420
2421         return rc;
2422 }
2423
2424 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2425                                  struct mdt_device *mdt, int idx)
2426 {
2427         struct llog_ctxt *ctxt;
2428
2429         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2430         if (ctxt == NULL)
2431                 return 0;
2432         /* Put once for the get we just did, and once for the clone */
2433         llog_ctxt_put(ctxt);
2434         llog_ctxt_put(ctxt);
2435         return 0;
2436 }
2437
2438 /*
2439  * sec context handlers
2440  */
2441 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2442 {
2443         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2444
2445         return 0;
2446 }
2447
2448 /*
2449  * quota request handlers
2450  */
2451 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2452 {
2453         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
2454         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2455         int                      rc;
2456         ENTRY;
2457
2458         if (qmt == NULL)
2459                 RETURN(err_serious(-EOPNOTSUPP));
2460
2461         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2462         RETURN(rc);
2463 }
2464
2465 struct mdt_object *mdt_object_new(const struct lu_env *env,
2466                                   struct mdt_device *d,
2467                                   const struct lu_fid *f)
2468 {
2469         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2470         struct lu_object *o;
2471         struct mdt_object *m;
2472         ENTRY;
2473
2474         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2475         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2476         if (unlikely(IS_ERR(o)))
2477                 m = (struct mdt_object *)o;
2478         else
2479                 m = mdt_obj(o);
2480         RETURN(m);
2481 }
2482
2483 struct mdt_object *mdt_object_find(const struct lu_env *env,
2484                                    struct mdt_device *d,
2485                                    const struct lu_fid *f)
2486 {
2487         struct lu_object *o;
2488         struct mdt_object *m;
2489         ENTRY;
2490
2491         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2492         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2493         if (unlikely(IS_ERR(o)))
2494                 m = (struct mdt_object *)o;
2495         else
2496                 m = mdt_obj(o);
2497
2498         RETURN(m);
2499 }
2500
2501 /**
2502  * Asyncronous commit for mdt device.
2503  *
2504  * Pass asynchonous commit call down the MDS stack.
2505  *
2506  * \param env environment
2507  * \param mdt the mdt device
2508  */
2509 static void mdt_device_commit_async(const struct lu_env *env,
2510                                     struct mdt_device *mdt)
2511 {
2512         struct dt_device *dt = mdt->mdt_bottom;
2513         int rc;
2514         ENTRY;
2515
2516         rc = dt->dd_ops->dt_commit_async(env, dt);
2517         if (unlikely(rc != 0))
2518                 CWARN("%s: async commit start failed: rc = %d\n",
2519                       mdt_obd_name(mdt), rc);
2520         atomic_inc(&mdt->mdt_async_commit_count);
2521         EXIT;
2522 }
2523
2524 /**
2525  * Mark the lock as "synchonous".
2526  *
2527  * Mark the lock to deffer transaction commit to the unlock time.
2528  *
2529  * \param lock the lock to mark as "synchonous"
2530  *
2531  * \see mdt_is_lock_sync
2532  * \see mdt_save_lock
2533  */
2534 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2535 {
2536         lock->l_ast_data = (void*)1;
2537 }
2538
2539 /**
2540  * Check whehter the lock "synchonous" or not.
2541  *
2542  * \param lock the lock to check
2543  * \retval 1 the lock is "synchonous"
2544  * \retval 0 the lock isn't "synchronous"
2545  *
2546  * \see mdt_set_lock_sync
2547  * \see mdt_save_lock
2548  */
2549 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2550 {
2551         return lock->l_ast_data != NULL;
2552 }
2553
2554 /**
2555  * Blocking AST for mdt locks.
2556  *
2557  * Starts transaction commit if in case of COS lock conflict or
2558  * deffers such a commit to the mdt_save_lock.
2559  *
2560  * \param lock the lock which blocks a request or cancelling lock
2561  * \param desc unused
2562  * \param data unused
2563  * \param flag indicates whether this cancelling or blocking callback
2564  * \retval 0
2565  * \see ldlm_blocking_ast_nocheck
2566  */
2567 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2568                      void *data, int flag)
2569 {
2570         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2571         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2572         bool commit_async = false;
2573         int rc;
2574         ENTRY;
2575
2576         if (flag == LDLM_CB_CANCELING)
2577                 RETURN(0);
2578
2579         lock_res_and_lock(lock);
2580         if (lock->l_blocking_ast != mdt_blocking_ast) {
2581                 unlock_res_and_lock(lock);
2582                 RETURN(0);
2583         }
2584         /* There is no lock conflict if l_blocking_lock == NULL,
2585          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2586          * when the last reference to a local lock was released */
2587         if (lock->l_req_mode & (LCK_PW | LCK_EX) &&
2588             lock->l_blocking_lock != NULL) {
2589                 if (mdt_cos_is_enabled(mdt)) {
2590                         if (lock->l_client_cookie !=
2591                             lock->l_blocking_lock->l_client_cookie)
2592                                 mdt_set_lock_sync(lock);
2593                 } else if (mdt_slc_is_enabled(mdt) &&
2594                            ldlm_is_cos_incompat(lock->l_blocking_lock)) {
2595                         mdt_set_lock_sync(lock);
2596                         /*
2597                          * we may do extra commit here, but there is a small
2598                          * window to miss a commit: lock was unlocked (saved),
2599                          * then a conflict lock queued and we come here, but
2600                          * REP-ACK not received, so lock was not converted to
2601                          * COS mode yet.
2602                          * Fortunately this window is quite small, so the
2603                          * extra commit should be rare (not to say distributed
2604                          * operation is rare too).
2605                          */
2606                         commit_async = true;
2607                 }
2608         } else if (lock->l_req_mode == LCK_COS &&
2609                    lock->l_blocking_lock != NULL) {
2610                 commit_async = true;
2611         }
2612
2613         rc = ldlm_blocking_ast_nocheck(lock);
2614
2615         if (commit_async) {
2616                 struct lu_env env;
2617
2618                 rc = lu_env_init(&env, LCT_LOCAL);
2619                 if (unlikely(rc != 0))
2620                         CWARN("%s: lu_env initialization failed, cannot "
2621                               "start asynchronous commit: rc = %d\n",
2622                               obd->obd_name, rc);
2623                 else
2624                         mdt_device_commit_async(&env, mdt);
2625                 lu_env_fini(&env);
2626         }
2627         RETURN(rc);
2628 }
2629
2630 /*
2631  * Blocking AST for cross-MDT lock
2632  *
2633  * Discard lock from uncommitted_slc_locks and cancel it.
2634  *
2635  * \param lock  the lock which blocks a request or cancelling lock
2636  * \param desc  unused
2637  * \param data  unused
2638  * \param flag  indicates whether this cancelling or blocking callback
2639  * \retval      0 on success
2640  * \retval      negative number on error
2641  */
2642 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2643                             void *data, int flag)
2644 {
2645         int rc = 0;
2646         ENTRY;
2647
2648         switch (flag) {
2649         case LDLM_CB_BLOCKING: {
2650                 struct lustre_handle lockh;
2651
2652                 ldlm_lock2handle(lock, &lockh);
2653                 rc = ldlm_cli_cancel(&lockh,
2654                         ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
2655                 if (rc < 0) {
2656                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2657                         RETURN(rc);
2658                 }
2659                 break;
2660         }
2661         case LDLM_CB_CANCELING: {
2662                 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2663                 struct mdt_device *mdt =
2664                                 mdt_dev(obd->obd_lu_dev->ld_site->ls_top_dev);
2665
2666                 LDLM_DEBUG(lock, "Revoke remote lock\n");
2667
2668                 /* discard slc lock here so that it can be cleaned anytime,
2669                  * especially for cleanup_resource() */
2670                 tgt_discard_slc_lock(&mdt->mdt_lut, lock);
2671
2672                 /* once we cache lock, l_ast_data is set to mdt_object */
2673                 if (lock->l_ast_data != NULL) {
2674                         struct mdt_object *mo = lock->l_ast_data;
2675                         struct lu_env env;
2676
2677                         rc = lu_env_init(&env, LCT_MD_THREAD);
2678                         if (unlikely(rc != 0)) {
2679                                 CWARN("%s: lu_env initialization failed, object"
2680                                       "%p "DFID" is leaked!\n",
2681                                       obd->obd_name, mo,
2682                                       PFID(mdt_object_fid(mo)));
2683                                 RETURN(rc);
2684                         }
2685
2686                         if (lock->l_policy_data.l_inodebits.bits &
2687                             (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)) {
2688                                 rc = mo_invalidate(&env, mdt_object_child(mo));
2689                                 mo->mot_cache_attr = 0;
2690                         }
2691                         mdt_object_put(&env, mo);
2692                         lu_env_fini(&env);
2693                 }
2694                 break;
2695         }
2696         default:
2697                 LBUG();
2698         }
2699
2700         RETURN(rc);
2701 }
2702
2703 int mdt_check_resent_lock(struct mdt_thread_info *info,
2704                           struct mdt_object *mo,
2705                           struct mdt_lock_handle *lhc)
2706 {
2707         /* the lock might already be gotten in ldlm_handle_enqueue() */
2708         if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
2709                 struct ptlrpc_request *req = mdt_info_req(info);
2710                 struct ldlm_lock      *lock;
2711
2712                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2713                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
2714                 if (lock == NULL) {
2715                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
2716                          * a resend case, however, it could be already destroyed
2717                          * due to client eviction or a raced cancel RPC. */
2718                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx",
2719                                           lhc->mlh_reg_lh.cookie);
2720                         RETURN(-ESTALE);
2721                 }
2722
2723                 if (!fid_res_name_eq(mdt_object_fid(mo),
2724                                      &lock->l_resource->lr_name)) {
2725                         CWARN("%s: Although resent, but still not "
2726                               "get child lock:"DFID"\n",
2727                               info->mti_exp->exp_obd->obd_name,
2728                               PFID(mdt_object_fid(mo)));
2729                         LDLM_LOCK_PUT(lock);
2730                         RETURN(-EPROTO);
2731                 }
2732                 LDLM_LOCK_PUT(lock);
2733                 return 0;
2734         }
2735         return 1;
2736 }
2737
2738 int mdt_remote_object_lock_try(struct mdt_thread_info *mti,
2739                                struct mdt_object *o, const struct lu_fid *fid,
2740                                struct lustre_handle *lh, enum ldlm_mode mode,
2741                                __u64 *ibits, __u64 trybits, bool cache)
2742 {
2743         struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
2744         union ldlm_policy_data *policy = &mti->mti_policy;
2745         struct ldlm_res_id *res_id = &mti->mti_res_id;
2746         int rc = 0;
2747         ENTRY;
2748
2749         LASSERT(mdt_object_remote(o));
2750
2751         fid_build_reg_res_name(fid, res_id);
2752
2753         memset(einfo, 0, sizeof(*einfo));
2754         einfo->ei_type = LDLM_IBITS;
2755         einfo->ei_mode = mode;
2756         einfo->ei_cb_bl = mdt_remote_blocking_ast;
2757         einfo->ei_cb_cp = ldlm_completion_ast;
2758         einfo->ei_enq_slave = 0;
2759         einfo->ei_res_id = res_id;
2760
2761         if (cache) {
2762                 /*
2763                  * if we cache lock, couple lock with mdt_object, so that object
2764                  * can be easily found in lock ASTs.
2765                  */
2766                 mdt_object_get(mti->mti_env, o);
2767                 einfo->ei_cbdata = o;
2768         }
2769
2770
2771         memset(policy, 0, sizeof(*policy));
2772         policy->l_inodebits.bits = *ibits;
2773         policy->l_inodebits.try_bits = trybits;
2774
2775         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2776                             policy);
2777         if (rc < 0 && cache) {
2778                 mdt_object_put(mti->mti_env, o);
2779                 einfo->ei_cbdata = NULL;
2780         }
2781
2782         /* Return successfully acquired bits to a caller */
2783         if (rc == 0) {
2784                 struct ldlm_lock *lock = ldlm_handle2lock(lh);
2785
2786                 LASSERT(lock);
2787                 *ibits = lock->l_policy_data.l_inodebits.bits;
2788                 LDLM_LOCK_PUT(lock);
2789         }
2790         RETURN(rc);
2791 }
2792
2793 int mdt_remote_object_lock(struct mdt_thread_info *mti, struct mdt_object *o,
2794                            const struct lu_fid *fid, struct lustre_handle *lh,
2795                            enum ldlm_mode mode, __u64 ibits, bool cache)
2796 {
2797         return mdt_remote_object_lock_try(mti, o, fid, lh, mode, &ibits, 0,
2798                                           cache);
2799 }
2800
2801 static int mdt_object_local_lock(struct mdt_thread_info *info,
2802                                  struct mdt_object *o,
2803                                  struct mdt_lock_handle *lh, __u64 *ibits,
2804                                  __u64 trybits, bool cos_incompat)
2805 {
2806         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2807         union ldlm_policy_data *policy = &info->mti_policy;
2808         struct ldlm_res_id *res_id = &info->mti_res_id;
2809         __u64 dlmflags = 0;
2810         int rc;
2811         ENTRY;
2812
2813         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2814         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2815         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2816         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2817
2818         if (cos_incompat) {
2819                 LASSERT(lh->mlh_reg_mode == LCK_PW ||
2820                         lh->mlh_reg_mode == LCK_EX);
2821                 dlmflags |= LDLM_FL_COS_INCOMPAT;
2822         } else if (mdt_cos_is_enabled(info->mti_mdt)) {
2823                 dlmflags |= LDLM_FL_COS_ENABLED;
2824         }
2825
2826         /* Only enqueue LOOKUP lock for remote object */
2827         if (mdt_object_remote(o)) {
2828                 LASSERT(*ibits == MDS_INODELOCK_LOOKUP);
2829         }
2830
2831         if (lh->mlh_type == MDT_PDO_LOCK) {
2832                 /* check for exists after object is locked */
2833                 if (mdt_object_exists(o) == 0) {
2834                         /* Non-existent object shouldn't have PDO lock */
2835                         RETURN(-ESTALE);
2836                 } else {
2837                         /* Non-dir object shouldn't have PDO lock */
2838                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2839                                 RETURN(-ENOTDIR);
2840                 }
2841         }
2842
2843
2844         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2845         dlmflags |= LDLM_FL_ATOMIC_CB;
2846
2847         /*
2848          * Take PDO lock on whole directory and build correct @res_id for lock
2849          * on part of directory.
2850          */
2851         if (lh->mlh_pdo_hash != 0) {
2852                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2853                 mdt_lock_pdo_mode(info, o, lh);
2854                 if (lh->mlh_pdo_mode != LCK_NL) {
2855                         /*
2856                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2857                          * is never going to be sent to client and we do not
2858                          * want it slowed down due to possible cancels.
2859                          */
2860                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2861                         policy->l_inodebits.try_bits = 0;
2862                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2863                                           policy, res_id, dlmflags,
2864                                           info->mti_exp == NULL ? NULL :
2865                                           &info->mti_exp->exp_handle.h_cookie);
2866                         if (unlikely(rc != 0))
2867                                 GOTO(out_unlock, rc);
2868                 }
2869
2870                 /*
2871                  * Finish res_id initializing by name hash marking part of
2872                  * directory which is taking modification.
2873                  */
2874                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2875         }
2876
2877         policy->l_inodebits.bits = *ibits;
2878         policy->l_inodebits.try_bits = trybits;
2879
2880         /*
2881          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2882          * going to be sent to client. If it is - mdt_intent_policy() path will
2883          * fix it up and turn FL_LOCAL flag off.
2884          */
2885         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2886                           res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2887                           info->mti_exp == NULL ? NULL :
2888                           &info->mti_exp->exp_handle.h_cookie);
2889 out_unlock:
2890         if (rc != 0)
2891                 mdt_object_unlock(info, o, lh, 1);
2892         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2893                    lh->mlh_pdo_hash != 0 &&
2894                    (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
2895                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2896
2897         /* Return successfully acquired bits to a caller */
2898         if (rc == 0) {
2899                 struct ldlm_lock *lock = ldlm_handle2lock(&lh->mlh_reg_lh);
2900
2901                 LASSERT(lock);
2902                 *ibits = lock->l_policy_data.l_inodebits.bits;
2903                 LDLM_LOCK_PUT(lock);
2904         }
2905         RETURN(rc);
2906 }
2907
2908 static int
2909 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
2910                          struct mdt_lock_handle *lh, __u64 *ibits,
2911                          __u64 trybits, bool cos_incompat)
2912 {
2913         struct mdt_lock_handle *local_lh = NULL;
2914         int rc;
2915         ENTRY;
2916
2917         if (!mdt_object_remote(o)) {
2918                 rc = mdt_object_local_lock(info, o, lh, ibits, trybits,
2919                                            cos_incompat);
2920                 RETURN(rc);
2921         }
2922
2923         /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
2924         *ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
2925                     MDS_INODELOCK_XATTR);
2926
2927         /* Only enqueue LOOKUP lock for remote object */
2928         if (*ibits & MDS_INODELOCK_LOOKUP) {
2929                 __u64 local = MDS_INODELOCK_LOOKUP;
2930
2931                 rc = mdt_object_local_lock(info, o, lh, &local, 0,
2932                                            cos_incompat);
2933                 if (rc != ELDLM_OK)
2934                         RETURN(rc);
2935
2936                 local_lh = lh;
2937         }
2938
2939         if ((*ibits | trybits) & MDS_INODELOCK_UPDATE) {
2940                 /* Sigh, PDO needs to enqueue 2 locks right now, but
2941                  * enqueue RPC can only request 1 lock, to avoid extra
2942                  * RPC, so it will instead enqueue EX lock for remote
2943                  * object anyway XXX*/
2944                 if (lh->mlh_type == MDT_PDO_LOCK &&
2945                     lh->mlh_pdo_hash != 0) {
2946                         CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
2947                                "EX lock.\n", mdt_obd_name(info->mti_mdt),
2948                                PFID(mdt_object_fid(o)));
2949                         lh->mlh_pdo_hash = 0;
2950                         lh->mlh_rreg_mode = LCK_EX;
2951                         lh->mlh_type = MDT_REG_LOCK;
2952                 }
2953
2954                 rc = mdt_remote_object_lock_try(info, o, mdt_object_fid(o),
2955                                                 &lh->mlh_rreg_lh,
2956                                                 lh->mlh_rreg_mode,
2957                                                 ibits, trybits, false);
2958                 if (rc != ELDLM_OK) {
2959                         if (local_lh != NULL)
2960                                 mdt_object_unlock(info, o, local_lh, rc);
2961                         RETURN(rc);
2962                 }
2963         }
2964
2965         RETURN(0);
2966 }
2967
2968 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2969                     struct mdt_lock_handle *lh, __u64 ibits)
2970 {
2971         return mdt_object_lock_internal(info, o, lh, &ibits, 0, false);
2972 }
2973
2974 int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2975                           struct mdt_lock_handle *lh, __u64 ibits,
2976                           bool cos_incompat)
2977 {
2978         LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
2979         return mdt_object_lock_internal(info, o, lh, &ibits, 0,
2980                                         cos_incompat);
2981 }
2982
2983 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
2984                         struct mdt_lock_handle *lh, __u64 *ibits,
2985                         __u64 trybits, bool cos_incompat)
2986 {
2987         return mdt_object_lock_internal(info, o, lh, ibits, trybits,
2988                                         cos_incompat);
2989 }
2990
2991 /**
2992  * Save a lock within request object.
2993  *
2994  * Keep the lock referenced until whether client ACK or transaction
2995  * commit happens or release the lock immediately depending on input
2996  * parameters. If COS is ON, a write lock is converted to COS lock
2997  * before saving.
2998  *
2999  * \param info thead info object
3000  * \param h lock handle
3001  * \param mode lock mode
3002  * \param decref force immediate lock releasing
3003  */
3004 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
3005                    enum ldlm_mode mode, int decref)
3006 {
3007         ENTRY;
3008
3009         if (lustre_handle_is_used(h)) {
3010                 if (decref || !info->mti_has_trans ||
3011                     !(mode & (LCK_PW | LCK_EX))) {
3012                         mdt_fid_unlock(h, mode);
3013                 } else {
3014                         struct mdt_device *mdt = info->mti_mdt;
3015                         struct ldlm_lock *lock = ldlm_handle2lock(h);
3016                         struct ptlrpc_request *req = mdt_info_req(info);
3017                         bool cos = mdt_cos_is_enabled(mdt);
3018                         bool convert_lock = !cos && mdt_slc_is_enabled(mdt);
3019
3020                         LASSERTF(lock != NULL, "no lock for cookie %#llx\n",
3021                                  h->cookie);
3022
3023                         /* there is no request if mdt_object_unlock() is called
3024                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
3025                         if (likely(req != NULL)) {
3026                                 LDLM_DEBUG(lock, "save lock request %p reply "
3027                                         "state %p transno %lld\n", req,
3028                                         req->rq_reply_state, req->rq_transno);
3029                                 if (cos) {
3030                                         ldlm_lock_downgrade(lock, LCK_COS);
3031                                         mode = LCK_COS;
3032                                 }
3033                                 if (req->rq_export->exp_disconnected)
3034                                         mdt_fid_unlock(h, mode);
3035                                 else
3036                                         ptlrpc_save_lock(req, h, mode, cos,
3037                                                          convert_lock);
3038                         } else {
3039                                 mdt_fid_unlock(h, mode);
3040                         }
3041                         if (mdt_is_lock_sync(lock)) {
3042                                 CDEBUG(D_HA, "found sync-lock,"
3043                                        " async commit started\n");
3044                                 mdt_device_commit_async(info->mti_env,
3045                                                         mdt);
3046                         }
3047                         LDLM_LOCK_PUT(lock);
3048                 }
3049                 h->cookie = 0ull;
3050         }
3051
3052         EXIT;
3053 }
3054
3055 /**
3056  * Save cross-MDT lock in uncommitted_slc_locks
3057  *
3058  * Keep the lock referenced until transaction commit happens or release the lock
3059  * immediately depending on input parameters.
3060  *
3061  * \param info thead info object
3062  * \param h lock handle
3063  * \param mode lock mode
3064  * \param decref force immediate lock releasing
3065  */
3066 static void mdt_save_remote_lock(struct mdt_thread_info *info,
3067                                  struct mdt_object *o, struct lustre_handle *h,
3068                                  enum ldlm_mode mode, int decref)
3069 {
3070         ENTRY;
3071
3072         if (lustre_handle_is_used(h)) {
3073                 struct ldlm_lock *lock = ldlm_handle2lock(h);
3074
3075                 if (o != NULL &&
3076                     (lock->l_policy_data.l_inodebits.bits &
3077                      (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)))
3078                         mo_invalidate(info->mti_env, mdt_object_child(o));
3079
3080                 if (decref || !info->mti_has_trans ||
3081                     !(mode & (LCK_PW | LCK_EX))) {
3082                         ldlm_lock_decref_and_cancel(h, mode);
3083                         LDLM_LOCK_PUT(lock);
3084                 } else {
3085                         struct ptlrpc_request *req = mdt_info_req(info);
3086
3087                         LASSERT(req != NULL);
3088                         tgt_save_slc_lock(&info->mti_mdt->mdt_lut, lock,
3089                                           req->rq_transno);
3090                         ldlm_lock_decref(h, mode);
3091                 }
3092                 h->cookie = 0ull;
3093         }
3094
3095         EXIT;
3096 }
3097
3098 /**
3099  * Unlock mdt object.
3100  *
3101  * Immeditely release the regular lock and the PDO lock or save the
3102  * lock in request and keep them referenced until client ACK or
3103  * transaction commit.
3104  *
3105  * \param info thread info object
3106  * \param o mdt object
3107  * \param lh mdt lock handle referencing regular and PDO locks
3108  * \param decref force immediate lock releasing
3109  *
3110  * XXX o is not used and may be NULL, see hsm_cdt_request_completed().
3111  */
3112 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
3113                        struct mdt_lock_handle *lh, int decref)
3114 {
3115         ENTRY;
3116
3117         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
3118         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
3119         mdt_save_remote_lock(info, o, &lh->mlh_rreg_lh, lh->mlh_rreg_mode,
3120                              decref);
3121
3122         EXIT;
3123 }
3124
3125 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
3126                                         const struct lu_fid *f,
3127                                         struct mdt_lock_handle *lh,
3128                                         __u64 ibits)
3129 {
3130         struct mdt_object *o;
3131
3132         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
3133         if (!IS_ERR(o)) {
3134                 int rc;
3135
3136                 rc = mdt_object_lock(info, o, lh, ibits);
3137                 if (rc != 0) {
3138                         mdt_object_put(info->mti_env, o);
3139                         o = ERR_PTR(rc);
3140                 }
3141         }
3142         return o;
3143 }
3144
3145 void mdt_object_unlock_put(struct mdt_thread_info * info,
3146                            struct mdt_object * o,
3147                            struct mdt_lock_handle *lh,
3148                            int decref)
3149 {
3150         mdt_object_unlock(info, o, lh, decref);
3151         mdt_object_put(info->mti_env, o);
3152 }
3153
3154 /*
3155  * Generic code handling requests that have struct mdt_body passed in:
3156  *
3157  *  - extract mdt_body from request and save it in @info, if present;
3158  *
3159  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
3160  *  @info;
3161  *
3162  *  - if HABEO_CORPUS flag is set for this request type check whether object
3163  *  actually exists on storage (lu_object_exists()).
3164  *
3165  */
3166 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
3167 {
3168         const struct mdt_body    *body;
3169         struct mdt_object        *obj;
3170         const struct lu_env      *env;
3171         struct req_capsule       *pill;
3172         int                       rc;
3173         ENTRY;
3174
3175         env = info->mti_env;
3176         pill = info->mti_pill;
3177
3178         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
3179         if (body == NULL)
3180                 RETURN(-EFAULT);
3181
3182         if (!(body->mbo_valid & OBD_MD_FLID))
3183                 RETURN(0);
3184
3185         if (!fid_is_sane(&body->mbo_fid1)) {
3186                 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
3187                 RETURN(-EINVAL);
3188         }
3189
3190         obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
3191         if (!IS_ERR(obj)) {
3192                 if ((flags & HABEO_CORPUS) && !mdt_object_exists(obj)) {
3193                         mdt_object_put(env, obj);
3194                         rc = -ENOENT;
3195                 } else {
3196                         info->mti_object = obj;
3197                         rc = 0;
3198                 }
3199         } else
3200                 rc = PTR_ERR(obj);
3201
3202         RETURN(rc);
3203 }
3204
3205 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
3206 {
3207         struct req_capsule *pill = info->mti_pill;
3208         int rc;
3209         ENTRY;
3210
3211         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
3212                 rc = mdt_body_unpack(info, flags);
3213         else
3214                 rc = 0;
3215
3216         if (rc == 0 && (flags & HABEO_REFERO)) {
3217                 /* Pack reply. */
3218                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
3219                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
3220                                              DEF_REP_MD_SIZE);
3221                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
3222                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
3223                                              RCL_SERVER, 0);
3224
3225                 /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
3226                  * by default. If the target object has more ACL entries, then
3227                  * enlarge the buffer when necessary. */
3228                 if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
3229                         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
3230                                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
3231
3232                 rc = req_capsule_server_pack(pill);
3233         }
3234         RETURN(rc);
3235 }
3236
3237 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
3238 {
3239         lh->mlh_type = MDT_NUL_LOCK;
3240         lh->mlh_reg_lh.cookie = 0ull;
3241         lh->mlh_reg_mode = LCK_MINMODE;
3242         lh->mlh_pdo_lh.cookie = 0ull;
3243         lh->mlh_pdo_mode = LCK_MINMODE;
3244         lh->mlh_rreg_lh.cookie = 0ull;
3245         lh->mlh_rreg_mode = LCK_MINMODE;
3246 }
3247
3248 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
3249 {
3250         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
3251         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
3252 }
3253
3254 /*
3255  * Initialize fields of struct mdt_thread_info. Other fields are left in
3256  * uninitialized state, because it's too expensive to zero out whole
3257  * mdt_thread_info (> 1K) on each request arrival.
3258  */
3259 void mdt_thread_info_init(struct ptlrpc_request *req,
3260                           struct mdt_thread_info *info)
3261 {
3262         int i;
3263
3264         info->mti_pill = &req->rq_pill;
3265
3266         /* lock handle */
3267         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3268                 mdt_lock_handle_init(&info->mti_lh[i]);
3269
3270         /* mdt device: it can be NULL while CONNECT */
3271         if (req->rq_export) {
3272                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
3273                 info->mti_exp = req->rq_export;
3274         } else
3275                 info->mti_mdt = NULL;
3276         info->mti_env = req->rq_svc_thread->t_env;
3277         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
3278
3279         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
3280         info->mti_big_buf = LU_BUF_NULL;
3281         info->mti_body = NULL;
3282         info->mti_object = NULL;
3283         info->mti_dlm_req = NULL;
3284         info->mti_has_trans = 0;
3285         info->mti_cross_ref = 0;
3286         info->mti_opdata = 0;
3287         info->mti_big_lmm_used = 0;
3288         info->mti_big_acl_used = 0;
3289         info->mti_som_valid = 0;
3290
3291         info->mti_spec.no_create = 0;
3292         info->mti_spec.sp_rm_entry = 0;
3293         info->mti_spec.sp_permitted = 0;
3294         info->mti_spec.sp_migrate_close = 0;
3295
3296         info->mti_spec.u.sp_ea.eadata = NULL;
3297         info->mti_spec.u.sp_ea.eadatalen = 0;
3298 }
3299
3300 void mdt_thread_info_fini(struct mdt_thread_info *info)
3301 {
3302         int i;
3303
3304         if (info->mti_object != NULL) {
3305                 mdt_object_put(info->mti_env, info->mti_object);
3306                 info->mti_object = NULL;
3307         }
3308
3309         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3310                 mdt_lock_handle_fini(&info->mti_lh[i]);
3311         info->mti_env = NULL;
3312         info->mti_pill = NULL;
3313         info->mti_exp = NULL;
3314
3315         if (unlikely(info->mti_big_buf.lb_buf != NULL))
3316                 lu_buf_free(&info->mti_big_buf);
3317 }
3318
3319 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
3320 {
3321         struct mdt_thread_info  *mti;
3322
3323         mti = mdt_th_info(tsi->tsi_env);
3324         LASSERT(mti != NULL);
3325
3326         mdt_thread_info_init(tgt_ses_req(tsi), mti);
3327         if (tsi->tsi_corpus != NULL) {
3328                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
3329                 lu_object_get(tsi->tsi_corpus);
3330         }
3331         mti->mti_body = tsi->tsi_mdt_body;
3332         mti->mti_dlm_req = tsi->tsi_dlm_req;
3333
3334         return mti;
3335 }
3336
3337 static int mdt_tgt_connect(struct tgt_session_info *tsi)
3338 {
3339         if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
3340             cfs_fail_val ==
3341             tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id) {
3342                 set_current_state(TASK_UNINTERRUPTIBLE);
3343                 schedule_timeout(msecs_to_jiffies(3 * MSEC_PER_SEC));
3344         }
3345
3346         return tgt_connect(tsi);
3347 }
3348
3349 enum mdt_it_code {
3350         MDT_IT_OPEN,
3351         MDT_IT_OCREAT,
3352         MDT_IT_CREATE,
3353         MDT_IT_GETATTR,
3354         MDT_IT_READDIR,
3355         MDT_IT_LOOKUP,
3356         MDT_IT_UNLINK,
3357         MDT_IT_TRUNC,
3358         MDT_IT_GETXATTR,
3359         MDT_IT_LAYOUT,
3360         MDT_IT_QUOTA,
3361         MDT_IT_GLIMPSE,
3362         MDT_IT_BRW,
3363         MDT_IT_NR
3364 };
3365
3366 static int mdt_intent_getattr(enum mdt_it_code opcode,
3367                               struct mdt_thread_info *info,
3368                               struct ldlm_lock **, __u64);
3369
3370 static int mdt_intent_getxattr(enum mdt_it_code opcode,
3371                                 struct mdt_thread_info *info,
3372                                 struct ldlm_lock **lockp,
3373                                 __u64 flags);
3374
3375 static int mdt_intent_layout(enum mdt_it_code opcode,
3376                              struct mdt_thread_info *info,
3377                              struct ldlm_lock **,
3378                              __u64);
3379 static int mdt_intent_reint(enum mdt_it_code opcode,
3380                             struct mdt_thread_info *info,
3381                             struct ldlm_lock **,
3382                             __u64);
3383 static int mdt_intent_glimpse(enum mdt_it_code opcode,
3384                               struct mdt_thread_info *info,
3385                               struct ldlm_lock **lockp, __u64 flags)
3386 {
3387         return mdt_glimpse_enqueue(info, info->mti_mdt->mdt_namespace,
3388                                    lockp, flags);
3389 }
3390 static int mdt_intent_brw(enum mdt_it_code opcode,
3391                           struct mdt_thread_info *info,
3392                           struct ldlm_lock **lockp, __u64 flags)
3393 {
3394         return mdt_brw_enqueue(info, info->mti_mdt->mdt_namespace,
3395                                lockp, flags);
3396 }
3397
3398 static struct mdt_it_flavor {
3399         const struct req_format *it_fmt;
3400         __u32                    it_flags;
3401         int                    (*it_act)(enum mdt_it_code ,
3402                                          struct mdt_thread_info *,
3403                                          struct ldlm_lock **,
3404                                          __u64);
3405         long                     it_reint;
3406 } mdt_it_flavor[] = {
3407         [MDT_IT_OPEN]     = {
3408                 .it_fmt   = &RQF_LDLM_INTENT,
3409                 /*.it_flags = HABEO_REFERO,*/
3410                 .it_flags = 0,
3411                 .it_act   = mdt_intent_reint,
3412                 .it_reint = REINT_OPEN
3413         },
3414         [MDT_IT_OCREAT]   = {
3415                 .it_fmt   = &RQF_LDLM_INTENT,
3416                 /*
3417                  * OCREAT is not a MUTABOR request as if the file
3418                  * already exists.
3419                  * We do the extra check of OBD_CONNECT_RDONLY in
3420                  * mdt_reint_open() when we really need to create
3421                  * the object.
3422                  */
3423                 .it_flags = 0,
3424                 .it_act   = mdt_intent_reint,
3425                 .it_reint = REINT_OPEN
3426         },
3427         [MDT_IT_CREATE]   = {
3428                 .it_fmt   = &RQF_LDLM_INTENT,
3429                 .it_flags = MUTABOR,
3430                 .it_act   = mdt_intent_reint,
3431                 .it_reint = REINT_CREATE
3432         },
3433         [MDT_IT_GETATTR]  = {
3434                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3435                 .it_flags = HABEO_REFERO,
3436                 .it_act   = mdt_intent_getattr
3437         },
3438         [MDT_IT_READDIR]  = {
3439                 .it_fmt   = NULL,
3440                 .it_flags = 0,
3441                 .it_act   = NULL
3442         },
3443         [MDT_IT_LOOKUP]   = {
3444                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3445                 .it_flags = HABEO_REFERO,
3446                 .it_act   = mdt_intent_getattr
3447         },
3448         [MDT_IT_UNLINK]   = {
3449                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
3450                 .it_flags = MUTABOR,
3451                 .it_act   = NULL,
3452                 .it_reint = REINT_UNLINK
3453         },
3454         [MDT_IT_TRUNC]    = {
3455                 .it_fmt   = NULL,
3456                 .it_flags = MUTABOR,
3457                 .it_act   = NULL
3458         },
3459         [MDT_IT_GETXATTR] = {
3460                 .it_fmt   = &RQF_LDLM_INTENT_GETXATTR,
3461                 .it_flags = HABEO_CORPUS,
3462                 .it_act   = mdt_intent_getxattr
3463         },
3464         [MDT_IT_LAYOUT] = {
3465                 .it_fmt   = &RQF_LDLM_INTENT_LAYOUT,
3466                 .it_flags = 0,
3467                 .it_act   = mdt_intent_layout
3468         },
3469         [MDT_IT_GLIMPSE] = {
3470                 .it_fmt = &RQF_LDLM_INTENT,
3471                 .it_flags = 0,
3472                 .it_act = mdt_intent_glimpse,
3473         },
3474         [MDT_IT_BRW] = {
3475                 .it_fmt = &RQF_LDLM_INTENT,
3476                 .it_flags = 0,
3477                 .it_act = mdt_intent_brw,
3478         },
3479
3480 };
3481
3482 int mdt_intent_lock_replace(struct mdt_thread_info *info,
3483                             struct ldlm_lock **lockp,
3484                             struct mdt_lock_handle *lh,
3485                             __u64 flags, int result)
3486 {
3487         struct ptlrpc_request  *req = mdt_info_req(info);
3488         struct ldlm_lock       *lock = *lockp;
3489         struct ldlm_lock       *new_lock;
3490
3491         /* If possible resent found a lock, @lh is set to its handle */
3492         new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3493
3494         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3495                 lh->mlh_reg_lh.cookie = 0;
3496                 RETURN(0);
3497         }
3498
3499         if (new_lock == NULL && (flags & LDLM_FL_RESENT)) {
3500                 /* Lock is pinned by ldlm_handle_enqueue0() as it is
3501                  * a resend case, however, it could be already destroyed
3502                  * due to client eviction or a raced cancel RPC. */
3503                 LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx\n",
3504                                   lh->mlh_reg_lh.cookie);
3505                 lh->mlh_reg_lh.cookie = 0;
3506                 RETURN(-ESTALE);
3507         }
3508
3509         LASSERTF(new_lock != NULL,
3510                  "lockh %#llx flags %#llx : rc = %d\n",
3511                  lh->mlh_reg_lh.cookie, flags, result);
3512
3513         /*
3514          * If we've already given this lock to a client once, then we should
3515          * have no readers or writers.  Otherwise, we should have one reader
3516          * _or_ writer ref (which will be zeroed below) before returning the
3517          * lock to a client.
3518          */
3519         if (new_lock->l_export == req->rq_export) {
3520                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3521         } else {
3522                 LASSERT(new_lock->l_export == NULL);
3523                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3524         }
3525
3526         *lockp = new_lock;
3527
3528         if (new_lock->l_export == req->rq_export) {
3529                 /*
3530                  * Already gave this to the client, which means that we
3531                  * reconstructed a reply.
3532                  */
3533                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3534                         MSG_RESENT);
3535
3536                 LDLM_LOCK_RELEASE(new_lock);
3537                 lh->mlh_reg_lh.cookie = 0;
3538                 RETURN(ELDLM_LOCK_REPLACED);
3539         }
3540
3541         /*
3542          * Fixup the lock to be given to the client.
3543          */
3544         lock_res_and_lock(new_lock);
3545         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3546          * possible blocking AST. */
3547         while (new_lock->l_readers > 0) {
3548                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3549                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3550                 new_lock->l_readers--;
3551         }
3552         while (new_lock->l_writers > 0) {
3553                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3554                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3555                 new_lock->l_writers--;
3556         }
3557
3558         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3559         new_lock->l_blocking_ast = lock->l_blocking_ast;
3560         new_lock->l_completion_ast = lock->l_completion_ast;
3561         if (ldlm_has_dom(new_lock))
3562                 new_lock->l_glimpse_ast = ldlm_server_glimpse_ast;
3563         new_lock->l_remote_handle = lock->l_remote_handle;
3564         new_lock->l_flags &= ~LDLM_FL_LOCAL;
3565
3566         unlock_res_and_lock(new_lock);
3567
3568         cfs_hash_add(new_lock->l_export->exp_lock_hash,
3569                      &new_lock->l_remote_handle,
3570                      &new_lock->l_exp_hash);
3571
3572         LDLM_LOCK_RELEASE(new_lock);
3573         lh->mlh_reg_lh.cookie = 0;
3574
3575         RETURN(ELDLM_LOCK_REPLACED);
3576 }
3577
3578 void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3579                              struct ldlm_lock *new_lock,
3580                              struct mdt_lock_handle *lh, __u64 flags)
3581 {
3582         struct ptlrpc_request  *req = mdt_info_req(info);
3583         struct ldlm_request    *dlmreq;
3584
3585         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3586                 return;
3587
3588         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3589
3590         /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
3591          * lock was found by ldlm_handle_enqueue(); if so @lh must be
3592          * initialized. */
3593         if (flags & LDLM_FL_RESENT) {
3594                 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
3595                 lh->mlh_reg_mode = new_lock->l_granted_mode;
3596
3597                 LDLM_DEBUG(new_lock, "Restoring lock cookie");
3598                 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie %#llx",
3599                           lh->mlh_reg_lh.cookie);
3600                 return;
3601         }
3602
3603         /*
3604          * If the xid matches, then we know this is a resent request, and allow
3605          * it. (It's probably an OPEN, for which we don't send a lock.
3606          */
3607         if (req_can_reconstruct(req, NULL))
3608                 return;
3609
3610         /*
3611          * This remote handle isn't enqueued, so we never received or processed
3612          * this request.  Clear MSG_RESENT, because it can be handled like any
3613          * normal request now.
3614          */
3615         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3616
3617         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle %#llx",
3618                   dlmreq->lock_handle[0].cookie);
3619 }
3620
3621 static int mdt_intent_getxattr(enum mdt_it_code opcode,
3622                                 struct mdt_thread_info *info,
3623                                 struct ldlm_lock **lockp,
3624                                 __u64 flags)
3625 {
3626         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3627         struct ldlm_reply      *ldlm_rep = NULL;
3628         int rc;
3629         ENTRY;
3630
3631         /*
3632          * Initialize lhc->mlh_reg_lh either from a previously granted lock
3633          * (for the resend case) or a new lock. Below we will use it to
3634          * replace the original lock.
3635          */
3636         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3637         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3638                 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3639                 rc = mdt_object_lock(info, info->mti_object, lhc,
3640                                      MDS_INODELOCK_XATTR);
3641                 if (rc)
3642                         return rc;
3643         }
3644
3645         rc = mdt_getxattr(info);
3646
3647         if (mdt_info_req(info)->rq_repmsg != NULL)
3648                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3649
3650         if (ldlm_rep == NULL ||
3651             OBD_FAIL_CHECK(OBD_FAIL_MDS_XATTR_REP)) {
3652                 mdt_object_unlock(info,  info->mti_object, lhc, 1);
3653                 RETURN(err_serious(-EFAULT));
3654         }
3655
3656         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3657
3658         /* This is left for interop instead of adding a new interop flag.
3659          * LU-7433 */
3660 #if LUSTRE_VERSION_CODE > OBD_OCD_VERSION(3, 0, 0, 0)
3661         if (ldlm_rep->lock_policy_res2) {
3662                 mdt_object_unlock(info, info->mti_object, lhc, 1);
3663                 RETURN(ELDLM_LOCK_ABORTED);
3664         }
3665 #endif
3666
3667         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3668         RETURN(rc);
3669 }
3670
3671 static int mdt_intent_getattr(enum mdt_it_code opcode,
3672                               struct mdt_thread_info *info,
3673                               struct ldlm_lock **lockp,
3674                               __u64 flags)
3675 {
3676         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3677         __u64                   child_bits;
3678         struct ldlm_reply      *ldlm_rep;
3679         struct mdt_body        *reqbody;
3680         struct mdt_body        *repbody;
3681         int                     rc, rc2;
3682         ENTRY;
3683
3684         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3685         LASSERT(reqbody);
3686
3687         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3688         LASSERT(repbody);
3689
3690         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
3691         repbody->mbo_eadatasize = 0;
3692         repbody->mbo_aclsize = 0;
3693
3694         switch (opcode) {
3695         case MDT_IT_LOOKUP:
3696                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3697                 break;
3698         case MDT_IT_GETATTR:
3699                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3700                              MDS_INODELOCK_PERM;
3701                 break;
3702         default:
3703                 CERROR("Unsupported intent (%d)\n", opcode);
3704                 GOTO(out_shrink, rc = -EINVAL);
3705         }
3706
3707         rc = mdt_init_ucred_intent_getattr(info, reqbody);
3708         if (rc)
3709                 GOTO(out_shrink, rc);
3710
3711         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3712         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3713
3714         /* Get lock from request for possible resent case. */
3715         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3716
3717         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3718         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3719
3720         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3721                 ldlm_rep->lock_policy_res2 = 0;
3722         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3723             ldlm_rep->lock_policy_res2) {
3724                 lhc->mlh_reg_lh.cookie = 0ull;
3725                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3726         }
3727
3728         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3729         EXIT;
3730 out_ucred:
3731         mdt_exit_ucred(info);
3732 out_shrink:
3733         mdt_client_compatibility(info);
3734         rc2 = mdt_fix_reply(info);
3735         if (rc == 0)
3736                 rc = rc2;
3737         return rc;
3738 }
3739
3740 static int mdt_intent_layout(enum mdt_it_code opcode,
3741                              struct mdt_thread_info *info,
3742                              struct ldlm_lock **lockp,
3743                              __u64 flags)
3744 {
3745         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_LAYOUT];
3746         struct md_layout_change layout = { .mlc_opc = MD_LAYOUT_NOP };
3747         struct layout_intent *intent;
3748         struct lu_fid *fid;
3749         struct mdt_object *obj = NULL;
3750         int layout_size = 0;
3751         int rc = 0;
3752         ENTRY;
3753
3754         if (opcode != MDT_IT_LAYOUT) {
3755                 CERROR("%s: Unknown intent (%d)\n", mdt_obd_name(info->mti_mdt),
3756                         opcode);
3757                 RETURN(-EINVAL);
3758         }
3759
3760         intent = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3761         if (intent == NULL)
3762                 RETURN(-EPROTO);
3763
3764         switch (intent->li_opc) {
3765         case LAYOUT_INTENT_TRUNC:
3766         case LAYOUT_INTENT_WRITE:
3767                 layout.mlc_opc = MD_LAYOUT_WRITE;
3768                 layout.mlc_intent = intent;
3769                 break;
3770         case LAYOUT_INTENT_ACCESS:
3771                 break;
3772         case LAYOUT_INTENT_READ:
3773         case LAYOUT_INTENT_GLIMPSE:
3774         case LAYOUT_INTENT_RELEASE:
3775         case LAYOUT_INTENT_RESTORE:
3776                 CERROR("%s: Unsupported layout intent opc %d\n",
3777                        mdt_obd_name(info->mti_mdt), intent->li_opc);
3778                 rc = -ENOTSUPP;
3779                 break;
3780         default:
3781                 CERROR("%s: Unknown layout intent opc %d\n",
3782                        mdt_obd_name(info->mti_mdt), intent->li_opc);
3783                 rc = -EINVAL;
3784                 break;
3785         }
3786         if (rc < 0)
3787                 RETURN(rc);
3788
3789         fid = &info->mti_tmp_fid2;
3790         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3791
3792         /* Get lock from request for possible resent case. */
3793         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3794
3795         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3796         if (IS_ERR(obj))
3797                 GOTO(out, rc = PTR_ERR(obj));
3798
3799         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3800                 layout_size = mdt_attr_get_eabuf_size(info, obj);
3801                 if (layout_size < 0)
3802                         GOTO(out_obj, rc = layout_size);
3803
3804                 if (layout_size > info->mti_mdt->mdt_max_mdsize)
3805                         info->mti_mdt->mdt_max_mdsize = layout_size;
3806         }
3807
3808         /*
3809          * set reply buffer size, so that ldlm_handle_enqueue0()->
3810          * ldlm_lvbo_fill() will fill the reply buffer with lovea.
3811          */
3812         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3813         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
3814                              layout_size);
3815         rc = req_capsule_server_pack(info->mti_pill);
3816         if (rc)
3817                 GOTO(out_obj, rc);
3818
3819
3820         if (layout.mlc_opc != MD_LAYOUT_NOP) {
3821                 struct lu_buf *buf = &layout.mlc_buf;
3822
3823                 /**
3824                  * mdt_layout_change is a reint operation, when the request
3825                  * is resent, layout write shouldn't reprocess it again.
3826                  */
3827                 rc = mdt_check_resent(info, mdt_reconstruct_generic, lhc);
3828                 if (rc)
3829                         GOTO(out_obj, rc = rc < 0 ? rc : 0);
3830
3831                 /**
3832                  * There is another resent case: the client's job has been
3833                  * done by another client, referring lod_declare_layout_change
3834                  * -EALREADY case, and it became a operation w/o transaction,
3835                  * so we should not do the layout change, otherwise
3836                  * mdt_layout_change() will try to cancel the granted server
3837                  * CR lock whose remote counterpart is still in hold on the
3838                  * client, and a deadlock ensues.
3839                  */
3840                 rc = mdt_check_resent_lock(info, obj, lhc);
3841                 if (rc <= 0)
3842                         GOTO(out_obj, rc);
3843
3844                 buf->lb_buf = NULL;
3845                 buf->lb_len = 0;
3846                 if (unlikely(req_is_replay(mdt_info_req(info)))) {
3847                         buf->lb_buf = req_capsule_client_get(info->mti_pill,
3848                                         &RMF_EADATA);
3849                         buf->lb_len = req_capsule_get_size(info->mti_pill,
3850                                         &RMF_EADATA, RCL_CLIENT);
3851                         /*
3852                          * If it's a replay of layout write intent RPC, the
3853                          * client has saved the extended lovea when
3854                          * it get reply then.
3855                          */
3856                         if (buf->lb_len > 0)
3857                                 mdt_fix_lov_magic(info, buf->lb_buf);
3858                 }
3859
3860                 /*
3861                  * Instantiate some layout components, if @buf contains
3862                  * lovea, then it's a replay of the layout intent write
3863                  * RPC.
3864                  */
3865                 rc = mdt_layout_change(info, obj, &layout);
3866                 if (rc)
3867                         GOTO(out_obj, rc);
3868         }
3869 out_obj:
3870         mdt_object_put(info->mti_env, obj);
3871
3872         if (rc == 0 && lustre_handle_is_used(&lhc->mlh_reg_lh))
3873                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3874
3875 out:
3876         lhc->mlh_reg_lh.cookie = 0;
3877
3878         return rc;
3879 }
3880
3881 static int mdt_intent_reint(enum mdt_it_code opcode,
3882                             struct mdt_thread_info *info,
3883                             struct ldlm_lock **lockp,
3884                             __u64 flags)
3885 {
3886         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3887         struct ldlm_reply      *rep = NULL;
3888         long                    opc;
3889         int                     rc;
3890
3891         static const struct req_format *intent_fmts[REINT_MAX] = {
3892                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3893                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3894         };
3895
3896         ENTRY;
3897
3898         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
3899         if (opc < 0)
3900                 RETURN(opc);
3901
3902         if (mdt_it_flavor[opcode].it_reint != opc) {
3903                 CERROR("Reint code %ld doesn't match intent: %d\n",
3904                        opc, opcode);
3905                 RETURN(err_serious(-EPROTO));
3906         }
3907
3908         /* Get lock from request for possible resent case. */
3909         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3910
3911         rc = mdt_reint_internal(info, lhc, opc);
3912
3913         /* Check whether the reply has been packed successfully. */
3914         if (mdt_info_req(info)->rq_repmsg != NULL)
3915                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3916         if (rep == NULL)
3917                 RETURN(err_serious(-EFAULT));
3918
3919         /* MDC expects this in any case */
3920         if (rc != 0)
3921                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3922
3923         /* the open lock or the lock for cross-ref object should be
3924          * returned to the client */
3925         if (lustre_handle_is_used(&lhc->mlh_reg_lh) &&
3926             (rc == 0 || rc == -MDT_EREMOTE_OPEN)) {
3927                 rep->lock_policy_res2 = 0;
3928                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3929                 RETURN(rc);
3930         }
3931
3932         rep->lock_policy_res2 = clear_serious(rc);
3933
3934         if (rep->lock_policy_res2 == -ENOENT &&
3935             mdt_get_disposition(rep, DISP_LOOKUP_NEG) &&
3936             !mdt_get_disposition(rep, DISP_OPEN_CREATE))
3937                 rep->lock_policy_res2 = 0;
3938
3939         lhc->mlh_reg_lh.cookie = 0ull;
3940         if (rc == -ENOTCONN || rc == -ENODEV ||
3941             rc == -EOVERFLOW) { /**< if VBR failure then return error */
3942                 /*
3943                  * If it is the disconnect error (ENODEV & ENOCONN), the error
3944                  * will be returned by rq_status, and client at ptlrpc layer
3945                  * will detect this, then disconnect, reconnect the import
3946                  * immediately, instead of impacting the following the rpc.
3947                  */
3948                 RETURN(rc);
3949         }
3950         /*
3951          * For other cases, the error will be returned by intent, and client
3952          * will retrieve the result from intent.
3953          */
3954         RETURN(ELDLM_LOCK_ABORTED);
3955 }
3956
3957 static int mdt_intent_code(enum ldlm_intent_flags itcode)
3958 {
3959         int rc;
3960
3961         switch (itcode) {
3962         case IT_OPEN:
3963                 rc = MDT_IT_OPEN;
3964                 break;
3965         case IT_OPEN|IT_CREAT:
3966                 rc = MDT_IT_OCREAT;
3967                 break;
3968         case IT_CREAT:
3969                 rc = MDT_IT_CREATE;
3970                 break;
3971         case IT_READDIR:
3972                 rc = MDT_IT_READDIR;
3973                 break;
3974         case IT_GETATTR:
3975                 rc = MDT_IT_GETATTR;
3976                 break;
3977         case IT_LOOKUP:
3978                 rc = MDT_IT_LOOKUP;
3979                 break;
3980         case IT_UNLINK:
3981                 rc = MDT_IT_UNLINK;
3982                 break;
3983         case IT_TRUNC:
3984                 rc = MDT_IT_TRUNC;
3985                 break;
3986         case IT_GETXATTR:
3987                 rc = MDT_IT_GETXATTR;
3988                 break;
3989         case IT_LAYOUT:
3990                 rc = MDT_IT_LAYOUT;
3991                 break;
3992         case IT_QUOTA_DQACQ:
3993         case IT_QUOTA_CONN:
3994                 rc = MDT_IT_QUOTA;
3995                 break;
3996         case IT_GLIMPSE:
3997                 rc = MDT_IT_GLIMPSE;
3998                 break;
3999         case IT_BRW:
4000                 rc = MDT_IT_BRW;
4001                 break;
4002         default:
4003                 CERROR("Unknown intent opcode: 0x%08x\n", itcode);
4004                 rc = -EINVAL;
4005                 break;
4006         }
4007         return rc;
4008 }
4009
4010 static int mdt_intent_opc(enum ldlm_intent_flags itopc,
4011                           struct mdt_thread_info *info,
4012                           struct ldlm_lock **lockp, __u64 flags)
4013 {
4014         struct req_capsule      *pill = info->mti_pill;
4015         struct ptlrpc_request   *req = mdt_info_req(info);
4016         struct mdt_it_flavor    *flv;
4017         int opc;
4018         int rc;
4019         ENTRY;
4020
4021         opc = mdt_intent_code(itopc);
4022         if (opc < 0)
4023                 RETURN(-EINVAL);
4024
4025         if (opc == MDT_IT_QUOTA) {
4026                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
4027
4028                 if (qmt == NULL)
4029                         RETURN(-EOPNOTSUPP);
4030
4031                 if (mdt_rdonly(req->rq_export))
4032                         RETURN(-EROFS);
4033
4034                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
4035                 /* pass the request to quota master */
4036                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
4037                                                  mdt_info_req(info), lockp,
4038                                                  flags);
4039                 RETURN(rc);
4040         }
4041
4042         flv = &mdt_it_flavor[opc];
4043         if (flv->it_fmt != NULL)
4044                 req_capsule_extend(pill, flv->it_fmt);
4045
4046         rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
4047         if (rc < 0)
4048                 RETURN(rc);
4049
4050         if (flv->it_flags & MUTABOR && mdt_rdonly(req->rq_export))
4051                 RETURN(-EROFS);
4052
4053         if (flv->it_act != NULL) {
4054                 struct ldlm_reply *rep;
4055
4056                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_INTENT_DELAY, 10);
4057
4058                 /* execute policy */
4059                 rc = flv->it_act(opc, info, lockp, flags);
4060
4061                 /* Check whether the reply has been packed successfully. */
4062                 if (req->rq_repmsg != NULL) {
4063                         rep = req_capsule_server_get(info->mti_pill,
4064                                                      &RMF_DLM_REP);
4065                         rep->lock_policy_res2 =
4066                                 ptlrpc_status_hton(rep->lock_policy_res2);
4067                 }
4068         }
4069
4070         RETURN(rc);
4071 }
4072
4073 static void mdt_ptlrpc_stats_update(struct ptlrpc_request *req,
4074                                     enum ldlm_intent_flags it_opc)
4075 {
4076         struct lprocfs_stats *srv_stats = ptlrpc_req2svc(req)->srv_stats;
4077
4078         /* update stats when IT code is known */
4079         if (srv_stats != NULL)
4080                 lprocfs_counter_incr(srv_stats,
4081                                 PTLRPC_LAST_CNTR + (it_opc == IT_GLIMPSE ?
4082                                 LDLM_GLIMPSE_ENQUEUE : LDLM_IBITS_ENQUEUE));
4083 }
4084
4085 static int mdt_intent_policy(struct ldlm_namespace *ns,
4086                              struct ldlm_lock **lockp, void *req_cookie,
4087                              enum ldlm_mode mode, __u64 flags, void *data)
4088 {
4089         struct tgt_session_info *tsi;
4090         struct mdt_thread_info  *info;
4091         struct ptlrpc_request   *req  =  req_cookie;
4092         struct ldlm_intent      *it;
4093         struct req_capsule      *pill;
4094         const struct ldlm_lock_desc *ldesc;
4095         int rc;
4096
4097         ENTRY;
4098
4099         LASSERT(req != NULL);
4100
4101         tsi = tgt_ses_info(req->rq_svc_thread->t_env);
4102
4103         info = tsi2mdt_info(tsi);
4104         LASSERT(info != NULL);
4105         pill = info->mti_pill;
4106         LASSERT(pill->rc_req == req);
4107         ldesc = &info->mti_dlm_req->lock_desc;
4108
4109         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
4110                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
4111                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
4112                 if (it != NULL) {
4113                         mdt_ptlrpc_stats_update(req, it->opc);
4114                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
4115                         if (rc == 0)
4116                                 rc = ELDLM_OK;
4117
4118                         /* Lock without inodebits makes no sense and will oops
4119                          * later in ldlm. Let's check it now to see if we have
4120                          * ibits corrupted somewhere in mdt_intent_opc().
4121                          * The case for client miss to set ibits has been
4122                          * processed by others. */
4123                         LASSERT(ergo(ldesc->l_resource.lr_type == LDLM_IBITS,
4124                                 ldesc->l_policy_data.l_inodebits.bits != 0));
4125                 } else {
4126                         rc = err_serious(-EFAULT);
4127                 }
4128         } else {
4129                 /* No intent was provided */
4130                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
4131                 rc = req_capsule_server_pack(pill);
4132                 if (rc)
4133                         rc = err_serious(rc);
4134         }
4135         mdt_thread_info_fini(info);
4136         RETURN(rc);
4137 }
4138
4139 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
4140 {
4141         struct seq_server_site  *ss = mdt_seq_site(mdt);
4142
4143         if (ss->ss_node_id == 0)
4144                 return;
4145
4146         if (ss->ss_client_seq != NULL) {
4147                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
4148                 ss->ss_client_seq->lcs_exp = NULL;
4149         }
4150
4151         if (ss->ss_server_fld != NULL) {
4152                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
4153                 ss->ss_server_fld->lsf_control_exp = NULL;
4154         }
4155 }
4156
4157 static void mdt_seq_fini_cli(struct mdt_device *mdt)
4158 {
4159         struct seq_server_site *ss = mdt_seq_site(mdt);
4160
4161         if (ss == NULL)
4162                 return;
4163
4164         if (ss->ss_server_seq != NULL)
4165                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
4166 }
4167
4168 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
4169 {
4170         mdt_seq_fini_cli(mdt);
4171         mdt_deregister_seq_exp(mdt);
4172
4173         return seq_site_fini(env, mdt_seq_site(mdt));
4174 }
4175
4176 /**
4177  * It will retrieve its FLDB entries from MDT0, and it only happens
4178  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
4179  * and it needs to refresh FLDB from the MDT0.
4180  **/
4181 static int mdt_register_lwp_callback(void *data)
4182 {
4183         struct lu_env           env;
4184         struct mdt_device       *mdt = data;
4185         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
4186         int                     rc;
4187         ENTRY;
4188
4189         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
4190
4191         rc = lu_env_init(&env, LCT_MD_THREAD);
4192         if (rc < 0) {
4193                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
4194                 RETURN(rc);
4195         }
4196
4197         /* Allocate new sequence now to avoid creating local transaction
4198          * in the normal transaction process */
4199         rc = seq_server_check_and_alloc_super(&env,
4200                                               mdt_seq_site(mdt)->ss_server_seq);
4201         if (rc < 0)
4202                 GOTO(out, rc);
4203
4204         if (fld->lsf_new) {
4205                 rc = fld_update_from_controller(&env, fld);
4206                 if (rc != 0) {
4207                         CERROR("%s: cannot update controller: rc = %d\n",
4208                                mdt_obd_name(mdt), rc);
4209                         GOTO(out, rc);
4210                 }
4211         }
4212 out:
4213         lu_env_fini(&env);
4214         RETURN(rc);
4215 }
4216
4217 static int mdt_register_seq_exp(struct mdt_device *mdt)
4218 {
4219         struct seq_server_site  *ss = mdt_seq_site(mdt);
4220         char                    *lwp_name = NULL;
4221         int                     rc;
4222
4223         if (ss->ss_node_id == 0)
4224                 return 0;
4225
4226         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
4227         if (lwp_name == NULL)
4228                 GOTO(out_free, rc = -ENOMEM);
4229
4230         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
4231         if (rc != 0)
4232                 GOTO(out_free, rc);
4233
4234         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
4235                                       NULL, NULL);
4236         if (rc != 0)
4237                 GOTO(out_free, rc);
4238
4239         rc = lustre_register_lwp_item(lwp_name,
4240                                       &ss->ss_server_fld->lsf_control_exp,
4241                                       mdt_register_lwp_callback, mdt);
4242         if (rc != 0) {
4243                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
4244                 ss->ss_client_seq->lcs_exp = NULL;
4245                 GOTO(out_free, rc);
4246         }
4247 out_free:
4248         if (lwp_name != NULL)
4249                 OBD_FREE(lwp_name, MAX_OBD_NAME);
4250
4251         return rc;
4252 }
4253
4254 /*
4255  * Init client sequence manager which is used by local MDS to talk to sequence
4256  * controller on remote node.
4257  */
4258 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
4259 {
4260         struct seq_server_site  *ss = mdt_seq_site(mdt);
4261         int                     rc;
4262         char                    *prefix;
4263         ENTRY;
4264
4265         /* check if this is adding the first MDC and controller is not yet
4266          * initialized. */
4267         OBD_ALLOC_PTR(ss->ss_client_seq);
4268         if (ss->ss_client_seq == NULL)
4269                 RETURN(-ENOMEM);
4270
4271         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
4272         if (prefix == NULL) {
4273                 OBD_FREE_PTR(ss->ss_client_seq);
4274                 ss->ss_client_seq = NULL;
4275                 RETURN(-ENOMEM);
4276         }
4277
4278         /* Note: seq_client_fini will be called in seq_site_fini */
4279         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
4280         rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
4281                              prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
4282                                                             NULL);
4283         OBD_FREE(prefix, MAX_OBD_NAME + 5);
4284         if (rc != 0) {
4285                 OBD_FREE_PTR(ss->ss_client_seq);
4286                 ss->ss_client_seq = NULL;
4287                 RETURN(rc);
4288         }
4289
4290         rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
4291
4292         RETURN(rc);
4293 }
4294
4295 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
4296 {
4297         struct seq_server_site  *ss;
4298         int                     rc;
4299         ENTRY;
4300
4301         ss = mdt_seq_site(mdt);
4302         /* init sequence controller server(MDT0) */
4303         if (ss->ss_node_id == 0) {
4304                 OBD_ALLOC_PTR(ss->ss_control_seq);
4305                 if (ss->ss_control_seq == NULL)
4306                         RETURN(-ENOMEM);
4307
4308                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
4309                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
4310                                      ss);
4311                 if (rc)
4312                         GOTO(out_seq_fini, rc);
4313         }
4314
4315         /* Init normal sequence server */
4316         OBD_ALLOC_PTR(ss->ss_server_seq);
4317         if (ss->ss_server_seq == NULL)
4318                 GOTO(out_seq_fini, rc = -ENOMEM);
4319
4320         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
4321                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
4322         if (rc)
4323                 GOTO(out_seq_fini, rc);
4324
4325         /* init seq client for seq server to talk to seq controller(MDT0) */
4326         rc = mdt_seq_init_cli(env, mdt);
4327         if (rc != 0)
4328                 GOTO(out_seq_fini, rc);
4329
4330         if (ss->ss_node_id != 0)
4331                 /* register controller export through lwp */
4332                 rc = mdt_register_seq_exp(mdt);
4333
4334         EXIT;
4335 out_seq_fini:
4336         if (rc)
4337                 mdt_seq_fini(env, mdt);
4338
4339         return rc;
4340 }
4341
4342 /*
4343  * FLD wrappers
4344  */
4345 static int mdt_fld_fini(const struct lu_env *env,
4346                         struct mdt_device *m)
4347 {
4348         struct seq_server_site *ss = mdt_seq_site(m);
4349         ENTRY;
4350
4351         if (ss && ss->ss_server_fld) {
4352                 fld_server_fini(env, ss->ss_server_fld);
4353                 OBD_FREE_PTR(ss->ss_server_fld);
4354                 ss->ss_server_fld = NULL;
4355         }
4356
4357         RETURN(0);
4358 }
4359
4360 static int mdt_fld_init(const struct lu_env *env,
4361                         const char *uuid,
4362                         struct mdt_device *m)
4363 {
4364         struct seq_server_site *ss;
4365         int rc;
4366         ENTRY;
4367
4368         ss = mdt_seq_site(m);
4369
4370         OBD_ALLOC_PTR(ss->ss_server_fld);
4371         if (ss->ss_server_fld == NULL)
4372                 RETURN(rc = -ENOMEM);
4373
4374         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
4375                              LU_SEQ_RANGE_MDT);
4376         if (rc) {
4377                 OBD_FREE_PTR(ss->ss_server_fld);
4378                 ss->ss_server_fld = NULL;
4379                 RETURN(rc);
4380         }
4381
4382         RETURN(0);
4383 }
4384
4385 static void mdt_stack_pre_fini(const struct lu_env *env,
4386                            struct mdt_device *m, struct lu_device *top)
4387 {
4388         struct lustre_cfg_bufs  *bufs;
4389         struct lustre_cfg       *lcfg;
4390         struct mdt_thread_info  *info;
4391         ENTRY;
4392
4393         LASSERT(top);
4394
4395         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4396         LASSERT(info != NULL);
4397
4398         bufs = &info->mti_u.bufs;
4399
4400         LASSERT(m->mdt_child_exp);
4401         LASSERT(m->mdt_child_exp->exp_obd);
4402
4403         /* process cleanup, pass mdt obd name to get obd umount flags */
4404         /* XXX: this is needed because all layers are referenced by
4405          * objects (some of them are pinned by osd, for example *
4406          * the proper solution should be a model where object used
4407          * by osd only doesn't have mdt/mdd slices -bzzz */
4408         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4409         lustre_cfg_bufs_set_string(bufs, 1, NULL);
4410         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4411         if (!lcfg)
4412                 RETURN_EXIT;
4413         lustre_cfg_init(lcfg, LCFG_PRE_CLEANUP, bufs);
4414
4415         top->ld_ops->ldo_process_config(env, top, lcfg);
4416         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4417         EXIT;
4418 }
4419
4420 static void mdt_stack_fini(const struct lu_env *env,
4421                            struct mdt_device *m, struct lu_device *top)
4422 {
4423         struct obd_device       *obd = mdt2obd_dev(m);
4424         struct lustre_cfg_bufs  *bufs;
4425         struct lustre_cfg       *lcfg;
4426         struct mdt_thread_info  *info;
4427         char                     flags[3] = "";
4428         ENTRY;
4429
4430         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4431         LASSERT(info != NULL);
4432
4433         lu_dev_del_linkage(top->ld_site, top);
4434
4435         lu_site_purge(env, top->ld_site, -1);
4436
4437         bufs = &info->mti_u.bufs;
4438         /* process cleanup, pass mdt obd name to get obd umount flags */
4439         /* another purpose is to let all layers to release their objects */
4440         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4441         if (obd->obd_force)
4442                 strcat(flags, "F");
4443         if (obd->obd_fail)
4444                 strcat(flags, "A");
4445         lustre_cfg_bufs_set_string(bufs, 1, flags);
4446         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4447         if (!lcfg)
4448                 RETURN_EXIT;
4449         lustre_cfg_init(lcfg, LCFG_CLEANUP, bufs);
4450
4451         LASSERT(top);
4452         top->ld_ops->ldo_process_config(env, top, lcfg);
4453         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4454
4455         lu_site_purge(env, top->ld_site, -1);
4456
4457         m->mdt_child = NULL;
4458         m->mdt_bottom = NULL;
4459
4460         obd_disconnect(m->mdt_child_exp);
4461         m->mdt_child_exp = NULL;
4462
4463         obd_disconnect(m->mdt_bottom_exp);
4464         m->mdt_child_exp = NULL;
4465 }
4466
4467 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
4468                                const char *next, struct obd_export **exp)
4469 {
4470         struct obd_connect_data *data = NULL;
4471         struct obd_device       *obd;
4472         int                      rc;
4473         ENTRY;
4474
4475         OBD_ALLOC_PTR(data);
4476         if (data == NULL)
4477                 GOTO(out, rc = -ENOMEM);
4478
4479         obd = class_name2obd(next);
4480         if (obd == NULL) {
4481                 CERROR("%s: can't locate next device: %s\n",
4482                        mdt_obd_name(m), next);
4483                 GOTO(out, rc = -ENOTCONN);
4484         }
4485
4486         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4487         data->ocd_version = LUSTRE_VERSION_CODE;
4488
4489         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
4490         if (rc) {
4491                 CERROR("%s: cannot connect to next dev %s (%d)\n",
4492                        mdt_obd_name(m), next, rc);
4493                 GOTO(out, rc);
4494         }
4495
4496 out:
4497         if (data)
4498                 OBD_FREE_PTR(data);
4499         RETURN(rc);
4500 }
4501
4502 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
4503                           struct lustre_cfg *cfg)
4504 {
4505         char                   *dev = lustre_cfg_string(cfg, 0);
4506         int                     rc, name_size, uuid_size;
4507         char                   *name, *uuid, *p;
4508         struct lustre_cfg_bufs *bufs;
4509         struct lustre_cfg      *lcfg;
4510         struct obd_device      *obd;
4511         struct lustre_profile  *lprof;
4512         struct lu_site         *site;
4513         ENTRY;
4514
4515         /* in 1.8 we had the only device in the stack - MDS.
4516          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
4517          * in 2.3 OSD is instantiated by obd_mount.c, so we need
4518          * to generate names and setup MDT, MDD. MDT will be using
4519          * generated name to connect to MDD. for MDD the next device
4520          * will be LOD with name taken from so called "profile" which
4521          * is generated by mount_option line
4522          *
4523          * 1.8 MGS generates config. commands like this:
4524          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
4525          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
4526          * 2.0 MGS generates config. commands like this:
4527          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
4528          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4529          *                    3:lustre-MDT0000-mdtlov  4:f
4530          *
4531          * we generate MDD name from MDT one, just replacing T with D
4532          *
4533          * after all the preparations, the logical equivalent will be
4534          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
4535          *                    3:lustre-MDT0000-mdtlov  4:f
4536          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4537          *                    3:lustre-MDD0000  4:f
4538          *
4539          *  notice we build the stack from down to top: MDD first, then MDT */
4540
4541         name_size = MAX_OBD_NAME;
4542         uuid_size = MAX_OBD_NAME;
4543
4544         OBD_ALLOC(name, name_size);
4545         OBD_ALLOC(uuid, uuid_size);
4546         if (name == NULL || uuid == NULL)
4547                 GOTO(cleanup_mem, rc = -ENOMEM);
4548
4549         OBD_ALLOC_PTR(bufs);
4550         if (!bufs)
4551                 GOTO(cleanup_mem, rc = -ENOMEM);
4552
4553         strcpy(name, dev);
4554         p = strstr(name, "-MDT");
4555         if (p == NULL)
4556                 GOTO(free_bufs, rc = -ENOMEM);
4557         p[3] = 'D';
4558
4559         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
4560
4561         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4562         if (lprof == NULL || lprof->lp_dt == NULL) {
4563                 CERROR("can't find the profile: %s\n",
4564                        lustre_cfg_string(cfg, 0));
4565                 GOTO(free_bufs, rc = -EINVAL);
4566         }
4567
4568         lustre_cfg_bufs_reset(bufs, name);
4569         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
4570         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4571         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4572
4573         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4574         if (!lcfg)
4575                 GOTO(put_profile, rc = -ENOMEM);
4576         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
4577
4578         rc = class_attach(lcfg);
4579         if (rc)
4580                 GOTO(lcfg_cleanup, rc);
4581
4582         obd = class_name2obd(name);
4583         if (!obd) {
4584                 CERROR("Can not find obd %s (%s in config)\n",
4585                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
4586                 GOTO(lcfg_cleanup, rc = -EINVAL);
4587         }
4588
4589         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4590
4591         lustre_cfg_bufs_reset(bufs, name);
4592         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4593         lustre_cfg_bufs_set_string(bufs, 2, dev);
4594         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4595
4596         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4597         if (!lcfg)
4598                 GOTO(class_detach, rc = -ENOMEM);
4599         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
4600
4601         rc = class_setup(obd, lcfg);
4602         if (rc)
4603                 GOTO(class_detach, rc);
4604
4605         /* connect to MDD we just setup */
4606         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
4607         if (rc)
4608                 GOTO(class_detach, rc);
4609
4610         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
4611         LASSERT(site);
4612         LASSERT(mdt_lu_site(mdt) == NULL);
4613         mdt->mdt_lu_dev.ld_site = site;
4614         site->ls_top_dev = &mdt->mdt_lu_dev;
4615         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
4616
4617         /* now connect to bottom OSD */
4618         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
4619         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
4620         if (rc)
4621                 GOTO(class_detach, rc);
4622         mdt->mdt_bottom =
4623                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
4624
4625         rc = lu_env_refill((struct lu_env *)env);
4626         if (rc != 0)
4627                 CERROR("Failure to refill session: '%d'\n", rc);
4628
4629         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
4630
4631         EXIT;
4632 class_detach:
4633         if (rc)
4634                 class_detach(obd, lcfg);
4635 lcfg_cleanup:
4636         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4637 put_profile:
4638         class_put_profile(lprof);
4639 free_bufs:
4640         OBD_FREE_PTR(bufs);
4641 cleanup_mem:
4642         if (name)
4643                 OBD_FREE(name, name_size);
4644         if (uuid)
4645                 OBD_FREE(uuid, uuid_size);
4646         RETURN(rc);
4647 }
4648
4649 /* setup quota master target on MDT0 */
4650 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
4651                           struct lustre_cfg *cfg)
4652 {
4653         struct obd_device       *obd;
4654         char                    *dev = lustre_cfg_string(cfg, 0);
4655         char                    *qmtname, *uuid, *p;
4656         struct lustre_cfg_bufs  *bufs;
4657         struct lustre_cfg       *lcfg;
4658         struct lustre_profile   *lprof;
4659         struct obd_connect_data *data;
4660         int                      rc;
4661         ENTRY;
4662
4663         LASSERT(mdt->mdt_qmt_exp == NULL);
4664         LASSERT(mdt->mdt_qmt_dev == NULL);
4665
4666         /* quota master is on MDT0 only for now */
4667         if (mdt->mdt_seq_site.ss_node_id != 0)
4668                 RETURN(0);
4669
4670         /* MGS generates config commands which look as follows:
4671          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4672          *                    3:lustre-MDT0000-mdtlov  4:f
4673          *
4674          * We generate the QMT name from the MDT one, just replacing MD with QM
4675          * after all the preparations, the logical equivalent will be:
4676          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
4677          *                    3:lustre-MDT0000-osd  4:f */
4678         OBD_ALLOC(qmtname, MAX_OBD_NAME);
4679         OBD_ALLOC(uuid, UUID_MAX);
4680         OBD_ALLOC_PTR(bufs);
4681         OBD_ALLOC_PTR(data);
4682         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
4683                 GOTO(cleanup_mem, rc = -ENOMEM);
4684
4685         strcpy(qmtname, dev);
4686         p = strstr(qmtname, "-MDT");
4687         if (p == NULL)
4688                 GOTO(cleanup_mem, rc = -ENOMEM);
4689         /* replace MD with QM */
4690         p[1] = 'Q';
4691         p[2] = 'M';
4692
4693         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
4694
4695         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4696         if (lprof == NULL || lprof->lp_dt == NULL) {
4697                 CERROR("can't find profile for %s\n",
4698                        lustre_cfg_string(cfg, 0));
4699                 GOTO(cleanup_mem, rc = -EINVAL);
4700         }
4701
4702         lustre_cfg_bufs_reset(bufs, qmtname);
4703         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
4704         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4705         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4706
4707         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4708         if (!lcfg)
4709                 GOTO(put_profile, rc = -ENOMEM);
4710         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
4711
4712         rc = class_attach(lcfg);
4713         if (rc)
4714                 GOTO(lcfg_cleanup, rc);
4715
4716         obd = class_name2obd(qmtname);
4717         if (!obd) {
4718                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
4719                        lustre_cfg_string(cfg, 0));
4720                 GOTO(lcfg_cleanup, rc = -EINVAL);
4721         }
4722
4723         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4724
4725         lustre_cfg_bufs_reset(bufs, qmtname);
4726         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4727         lustre_cfg_bufs_set_string(bufs, 2, dev);
4728
4729         /* for quota, the next device should be the OSD device */
4730         lustre_cfg_bufs_set_string(bufs, 3,
4731                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
4732
4733         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
4734         if (!lcfg)
4735                 GOTO(class_detach, rc = -ENOMEM);
4736         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
4737
4738         rc = class_setup(obd, lcfg);
4739         if (rc)
4740                 GOTO(class_detach, rc);
4741
4742         mdt->mdt_qmt_dev = obd->obd_lu_dev;
4743
4744         /* configure local quota objects */
4745         rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
4746                                                    &mdt->mdt_lu_dev,
4747                                                    mdt->mdt_qmt_dev);
4748         if (rc)
4749                 GOTO(class_cleanup, rc);
4750
4751         /* connect to quota master target */
4752         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4753         data->ocd_version = LUSTRE_VERSION_CODE;
4754         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
4755                          data, NULL);
4756         if (rc) {
4757                 CERROR("cannot connect to quota master device %s (%d)\n",
4758                        qmtname, rc);
4759                 GOTO(class_cleanup, rc);
4760         }
4761
4762         EXIT;
4763 class_cleanup:
4764         if (rc) {
4765                 class_manual_cleanup(obd);
4766                 mdt->mdt_qmt_dev = NULL;
4767         }
4768 class_detach:
4769         if (rc)
4770                 class_detach(obd, lcfg);
4771 lcfg_cleanup:
4772         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
4773 put_profile:
4774         class_put_profile(lprof);
4775 cleanup_mem:
4776         if (bufs)
4777                 OBD_FREE_PTR(bufs);
4778         if (qmtname)
4779                 OBD_FREE(qmtname, MAX_OBD_NAME);
4780         if (uuid)
4781                 OBD_FREE(uuid, UUID_MAX);
4782         if (data)
4783                 OBD_FREE_PTR(data);
4784         return rc;
4785 }
4786
4787 /* Shutdown quota master target associated with mdt */
4788 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
4789 {
4790         ENTRY;
4791
4792         if (mdt->mdt_qmt_exp == NULL)
4793                 RETURN_EXIT;
4794         LASSERT(mdt->mdt_qmt_dev != NULL);
4795
4796         /* the qmt automatically shuts down when the mdt disconnects */
4797         obd_disconnect(mdt->mdt_qmt_exp);
4798         mdt->mdt_qmt_exp = NULL;
4799         mdt->mdt_qmt_dev = NULL;
4800         EXIT;
4801 }
4802
4803 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
4804  * for now. This will be removed along with converting rest of MDT code
4805  * to use tgt_session_info */
4806 static int mdt_tgt_getxattr(struct tgt_session_info *tsi)
4807 {
4808         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
4809         int                      rc;
4810
4811         rc = mdt_getxattr(info);
4812
4813         mdt_thread_info_fini(info);
4814         return rc;
4815 }
4816
4817 #define OBD_FAIL_OST_READ_NET   OBD_FAIL_OST_BRW_NET
4818 #define OBD_FAIL_OST_WRITE_NET  OBD_FAIL_OST_BRW_NET
4819 #define OST_BRW_READ    OST_READ
4820 #define OST_BRW_WRITE   OST_WRITE
4821
4822 static struct tgt_handler mdt_tgt_handlers[] = {
4823 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4824                 0,                      MDS_CONNECT,    mdt_tgt_connect,
4825                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
4826 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4827                 0,                      MDS_DISCONNECT, tgt_disconnect,
4828                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
4829 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4830                 HABEO_REFERO,           MDS_SET_INFO,   mdt_set_info,
4831                 &RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION),
4832 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
4833 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_GET_ROOT,   mdt_get_root),
4834 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETATTR,    mdt_getattr),
4835 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_GETATTR_NAME,
4836                                                         mdt_getattr_name),
4837 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETXATTR,   mdt_tgt_getxattr),
4838 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_STATFS,     mdt_statfs),
4839 TGT_MDT_HDL(0           | MUTABOR,      MDS_REINT,      mdt_reint),
4840 TGT_MDT_HDL(HABEO_CORPUS,               MDS_CLOSE,      mdt_close),
4841 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_READPAGE,   mdt_readpage),
4842 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_SYNC,       mdt_sync),
4843 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
4844 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_PROGRESS,
4845                                                         mdt_hsm_progress),
4846 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_REGISTER,
4847                                                         mdt_hsm_ct_register),
4848 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_UNREGISTER,
4849                                                         mdt_hsm_ct_unregister),
4850 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_STATE_GET,
4851                                                         mdt_hsm_state_get),
4852 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_STATE_SET,
4853                                                         mdt_hsm_state_set),
4854 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_ACTION, mdt_hsm_action),
4855 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_REQUEST,
4856                                                         mdt_hsm_request),
4857 TGT_MDT_HDL(HABEO_CLAVIS | HABEO_CORPUS | HABEO_REFERO | MUTABOR,
4858             MDS_SWAP_LAYOUTS,
4859             mdt_swap_layouts),
4860 };
4861
4862 static struct tgt_handler mdt_io_ops[] = {
4863 TGT_OST_HDL(HABEO_CORPUS | HABEO_REFERO, OST_BRW_READ,  tgt_brw_read),
4864 TGT_OST_HDL(HABEO_CORPUS | MUTABOR,      OST_BRW_WRITE, tgt_brw_write),
4865 TGT_OST_HDL(HABEO_CORPUS | HABEO_REFERO | MUTABOR,
4866                                          OST_PUNCH,     mdt_punch_hdl),
4867 TGT_OST_HDL(HABEO_CORPUS | HABEO_REFERO, OST_SYNC,      mdt_data_sync),
4868 };
4869
4870 static struct tgt_handler mdt_sec_ctx_ops[] = {
4871 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
4872 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
4873 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
4874 };
4875
4876 static struct tgt_handler mdt_quota_ops[] = {
4877 TGT_QUOTA_HDL(HABEO_REFERO,             QUOTA_DQACQ,      mdt_quota_dqacq),
4878 };
4879
4880 static struct tgt_opc_slice mdt_common_slice[] = {
4881         {
4882                 .tos_opc_start  = MDS_FIRST_OPC,
4883                 .tos_opc_end    = MDS_LAST_OPC,
4884                 .tos_hs         = mdt_tgt_handlers
4885         },
4886         {
4887                 .tos_opc_start  = OBD_FIRST_OPC,
4888                 .tos_opc_end    = OBD_LAST_OPC,
4889                 .tos_hs         = tgt_obd_handlers
4890         },
4891         {
4892                 .tos_opc_start  = LDLM_FIRST_OPC,
4893                 .tos_opc_end    = LDLM_LAST_OPC,
4894                 .tos_hs         = tgt_dlm_handlers
4895         },
4896         {
4897                 .tos_opc_start  = SEC_FIRST_OPC,
4898                 .tos_opc_end    = SEC_LAST_OPC,
4899                 .tos_hs         = mdt_sec_ctx_ops
4900         },
4901         {
4902                 .tos_opc_start  = OUT_UPDATE_FIRST_OPC,
4903                 .tos_opc_end    = OUT_UPDATE_LAST_OPC,
4904                 .tos_hs         = tgt_out_handlers
4905         },
4906         {
4907                 .tos_opc_start  = FLD_FIRST_OPC,
4908                 .tos_opc_end    = FLD_LAST_OPC,
4909                 .tos_hs         = fld_handlers
4910         },
4911         {
4912                 .tos_opc_start  = SEQ_FIRST_OPC,
4913                 .tos_opc_end    = SEQ_LAST_OPC,
4914                 .tos_hs         = seq_handlers
4915         },
4916         {
4917                 .tos_opc_start  = QUOTA_DQACQ,
4918                 .tos_opc_end    = QUOTA_LAST_OPC,
4919                 .tos_hs         = mdt_quota_ops
4920         },
4921         {
4922                 .tos_opc_start  = LLOG_FIRST_OPC,
4923                 .tos_opc_end    = LLOG_LAST_OPC,
4924                 .tos_hs         = tgt_llog_handlers
4925         },
4926         {
4927                 .tos_opc_start  = LFSCK_FIRST_OPC,
4928                 .tos_opc_end    = LFSCK_LAST_OPC,
4929                 .tos_hs         = tgt_lfsck_handlers
4930         },
4931         {
4932                 .tos_opc_start  = OST_FIRST_OPC,
4933                 .tos_opc_end    = OST_LAST_OPC,
4934                 .tos_hs         = mdt_io_ops
4935         },
4936         {
4937                 .tos_hs         = NULL
4938         }
4939 };
4940
4941 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4942 {
4943         struct md_device *next = m->mdt_child;
4944         struct lu_device *d = &m->mdt_lu_dev;
4945         struct obd_device *obd = mdt2obd_dev(m);
4946         struct lfsck_stop stop;
4947
4948         ENTRY;
4949         stop.ls_status = LS_PAUSED;
4950         stop.ls_flags = 0;
4951         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
4952
4953         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
4954         ping_evictor_stop();
4955
4956         /* Remove the HSM /proc entry so the coordinator cannot be
4957          * restarted by a user while it's shutting down. */
4958         hsm_cdt_procfs_fini(m);
4959         mdt_hsm_cdt_stop(m);
4960
4961         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
4962         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4963
4964         if (m->mdt_namespace != NULL)
4965                 ldlm_namespace_free_prior(m->mdt_namespace, NULL,
4966                                           d->ld_obd->obd_force);
4967
4968         obd_exports_barrier(obd);
4969         obd_zombie_barrier();
4970
4971         mdt_quota_fini(env, m);
4972
4973         cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
4974
4975         /* Calling the cleanup functions in the same order as in the mdt_init0
4976          * error path
4977          */
4978         mdt_procfs_fini(m);
4979
4980         target_recovery_fini(obd);
4981         upcall_cache_cleanup(m->mdt_identity_cache);
4982         m->mdt_identity_cache = NULL;
4983
4984         mdt_fs_cleanup(env, m);
4985
4986         tgt_fini(env, &m->mdt_lut);
4987
4988         mdt_hsm_cdt_fini(m);
4989
4990         if (m->mdt_namespace != NULL) {
4991                 ldlm_namespace_free_post(m->mdt_namespace);
4992                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4993         }
4994
4995         if (m->mdt_md_root != NULL) {
4996                 mdt_object_put(env, m->mdt_md_root);
4997                 m->mdt_md_root = NULL;
4998         }
4999
5000         mdt_seq_fini(env, m);
5001
5002         mdt_fld_fini(env, m);
5003
5004         /*
5005          * Finish the stack
5006          */
5007         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
5008
5009         LASSERT(atomic_read(&d->ld_ref) == 0);
5010
5011         server_put_mount(mdt_obd_name(m), true);
5012
5013         EXIT;
5014 }
5015
5016 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
5017
5018 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
5019                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
5020 {
5021         const struct dt_device_param *dt_conf;
5022         struct mdt_thread_info *info;
5023         struct obd_device *obd;
5024         const char *dev = lustre_cfg_string(cfg, 0);
5025         const char *num = lustre_cfg_string(cfg, 2);
5026         struct tg_grants_data *tgd = &m->mdt_lut.lut_tgd;
5027         struct lustre_mount_info *lmi = NULL;
5028         struct lustre_sb_info *lsi;
5029         struct lu_site *s;
5030         struct seq_server_site *ss_site;
5031         const char *identity_upcall = "NONE";
5032         struct md_device *next;
5033         int rc;
5034         long node_id;
5035         mntopt_t mntopts;
5036         ENTRY;
5037
5038         lu_device_init(&m->mdt_lu_dev, ldt);
5039         /*
5040          * Environment (env) might be missing mdt_thread_key values at that
5041          * point, if device is allocated when mdt_thread_key is in QUIESCENT
5042          * mode.
5043          *
5044          * Usually device allocation path doesn't use module key values, but
5045          * mdt has to do a lot of work here, so allocate key value.
5046          */
5047         rc = lu_env_refill((struct lu_env *)env);
5048         if (rc != 0)
5049                 RETURN(rc);
5050
5051         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5052         LASSERT(info != NULL);
5053
5054         obd = class_name2obd(dev);
5055         LASSERT(obd != NULL);
5056
5057         m->mdt_max_mdsize = MAX_MD_SIZE; /* 4 stripes */
5058         m->mdt_opts.mo_evict_tgt_nids = 1;
5059         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
5060
5061         lmi = server_get_mount(dev);
5062         if (lmi == NULL) {
5063                 CERROR("Cannot get mount info for %s!\n", dev);
5064                 RETURN(-EFAULT);
5065         } else {
5066                 lsi = s2lsi(lmi->lmi_sb);
5067                 /* CMD is supported only in IAM mode */
5068                 LASSERT(num);
5069                 node_id = simple_strtol(num, NULL, 10);
5070                 obd->u.obt.obt_magic = OBT_MAGIC;
5071                 if (lsi->lsi_lmd != NULL &&
5072                     lsi->lsi_lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
5073                         m->mdt_skip_lfsck = 1;
5074         }
5075
5076         /* DoM files get IO lock at open by default */
5077         m->mdt_opts.mo_dom_lock = 1;
5078
5079         m->mdt_squash.rsi_uid = 0;
5080         m->mdt_squash.rsi_gid = 0;
5081         INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
5082         init_rwsem(&m->mdt_squash.rsi_sem);
5083         spin_lock_init(&m->mdt_lock);
5084         m->mdt_enable_remote_dir = 0;
5085         m->mdt_enable_remote_dir_gid = 0;
5086
5087         atomic_set(&m->mdt_mds_mds_conns, 0);
5088         atomic_set(&m->mdt_async_commit_count, 0);
5089
5090         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
5091         m->mdt_lu_dev.ld_obd = obd;
5092         /* Set this lu_device to obd for error handling purposes. */
5093         obd->obd_lu_dev = &m->mdt_lu_dev;
5094
5095         /* init the stack */
5096         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
5097         if (rc) {
5098                 CERROR("%s: Can't init device stack, rc %d\n",
5099                        mdt_obd_name(m), rc);
5100                 GOTO(err_lmi, rc);
5101         }
5102
5103         s = mdt_lu_site(m);
5104         ss_site = mdt_seq_site(m);
5105         s->ld_seq_site = ss_site;
5106         ss_site->ss_lu = s;
5107
5108         /* set server index */
5109         ss_site->ss_node_id = node_id;
5110
5111         /* failover is the default
5112          * FIXME: we do not failout mds0/mgs, which may cause some problems.
5113          * assumed whose ss_node_id == 0 XXX
5114          * */
5115         obd->obd_replayable = 1;
5116         /* No connection accepted until configurations will finish */
5117         obd->obd_no_conn = 1;
5118
5119         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
5120                 char *str = lustre_cfg_string(cfg, 4);
5121                 if (strchr(str, 'n')) {
5122                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
5123                         obd->obd_replayable = 0;
5124                 }
5125         }
5126
5127         rc = mdt_fld_init(env, mdt_obd_name(m), m);
5128         if (rc)
5129                 GOTO(err_fini_stack, rc);
5130
5131         rc = mdt_seq_init(env, m);
5132         if (rc)
5133                 GOTO(err_fini_fld, rc);
5134
5135         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
5136                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
5137         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
5138                                               LDLM_NAMESPACE_SERVER,
5139                                               LDLM_NAMESPACE_GREEDY,
5140                                               LDLM_NS_TYPE_MDT);
5141         if (m->mdt_namespace == NULL)
5142                 GOTO(err_fini_seq, rc = -ENOMEM);
5143
5144         m->mdt_namespace->ns_lvbp = m;
5145         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
5146
5147         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
5148         /* set obd_namespace for compatibility with old code */
5149         obd->obd_namespace = m->mdt_namespace;
5150
5151         rc = mdt_hsm_cdt_init(m);
5152         if (rc != 0) {
5153                 CERROR("%s: error initializing coordinator, rc %d\n",
5154                        mdt_obd_name(m), rc);
5155                 GOTO(err_free_ns, rc);
5156         }
5157
5158         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
5159                       OBD_FAIL_MDS_ALL_REQUEST_NET,
5160                       OBD_FAIL_MDS_ALL_REPLY_NET);
5161         if (rc)
5162                 GOTO(err_free_hsm, rc);
5163
5164         /* Amount of available space excluded from granting and reserved
5165          * for metadata. It is in percentage and 50% is default value. */
5166         tgd->tgd_reserved_pcnt = 50;
5167
5168         if (ONE_MB_BRW_SIZE < (1U << tgd->tgd_blockbits))
5169                 m->mdt_brw_size = 1U << tgd->tgd_blockbits;
5170         else
5171                 m->mdt_brw_size = ONE_MB_BRW_SIZE;
5172
5173         rc = mdt_fs_setup(env, m, obd, lsi);
5174         if (rc)
5175                 GOTO(err_tgt, rc);
5176
5177         tgt_adapt_sptlrpc_conf(&m->mdt_lut);
5178
5179         next = m->mdt_child;
5180         dt_conf = next->md_ops->mdo_dtconf_get(env, next);
5181
5182         mntopts = dt_conf->ddp_mntopts;
5183
5184         if (mntopts & MNTOPT_USERXATTR)
5185                 m->mdt_opts.mo_user_xattr = 1;
5186         else
5187                 m->mdt_opts.mo_user_xattr = 0;
5188
5189         m->mdt_max_ea_size = dt_conf->ddp_max_ea_size;
5190
5191         if (mntopts & MNTOPT_ACL)
5192                 m->mdt_opts.mo_acl = 1;
5193         else
5194                 m->mdt_opts.mo_acl = 0;
5195
5196         /* XXX: to support suppgid for ACL, we enable identity_upcall
5197          * by default, otherwise, maybe got unexpected -EACCESS. */
5198         if (m->mdt_opts.mo_acl)
5199                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
5200
5201         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
5202                                                 identity_upcall,
5203                                                 &mdt_identity_upcall_cache_ops);
5204         if (IS_ERR(m->mdt_identity_cache)) {
5205                 rc = PTR_ERR(m->mdt_identity_cache);
5206                 m->mdt_identity_cache = NULL;
5207                 GOTO(err_fs_cleanup, rc);
5208         }
5209
5210         rc = mdt_procfs_init(m, dev);
5211         if (rc) {
5212                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
5213                 GOTO(err_recovery, rc);
5214         }
5215
5216         rc = mdt_quota_init(env, m, cfg);
5217         if (rc)
5218                 GOTO(err_procfs, rc);
5219
5220         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
5221         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
5222                            "mdt_ldlm_client", m->mdt_ldlm_client);
5223
5224         ping_evictor_start();
5225
5226         /* recovery will be started upon mdt_prepare()
5227          * when the whole stack is complete and ready
5228          * to serve the requests */
5229
5230         /* Reduce the initial timeout on an MDS because it doesn't need such
5231          * a long timeout as an OST does. Adaptive timeouts will adjust this
5232          * value appropriately. */
5233         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
5234                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
5235
5236         RETURN(0);
5237 err_procfs:
5238         mdt_procfs_fini(m);
5239 err_recovery:
5240         target_recovery_fini(obd);
5241         upcall_cache_cleanup(m->mdt_identity_cache);
5242         m->mdt_identity_cache = NULL;
5243 err_fs_cleanup:
5244         mdt_fs_cleanup(env, m);
5245 err_tgt:
5246         tgt_fini(env, &m->mdt_lut);
5247 err_free_hsm:
5248         mdt_hsm_cdt_fini(m);
5249 err_free_ns:
5250         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
5251         obd->obd_namespace = m->mdt_namespace = NULL;
5252 err_fini_seq:
5253         mdt_seq_fini(env, m);
5254 err_fini_fld:
5255         mdt_fld_fini(env, m);
5256 err_fini_stack:
5257         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
5258 err_lmi:
5259         if (lmi)
5260                 server_put_mount(dev, true);
5261         return(rc);
5262 }
5263
5264 /* For interoperability, the left element is old parameter, the right one
5265  * is the new version of the parameter, if some parameter is deprecated,
5266  * the new version should be set as NULL. */
5267 static struct cfg_interop_param mdt_interop_param[] = {
5268         { "mdt.group_upcall",   NULL },
5269         { "mdt.quota_type",     NULL },
5270         { "mdd.quota_type",     NULL },
5271         { "mdt.som",            NULL },
5272         { "mdt.rootsquash",     "mdt.root_squash" },
5273         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
5274         { NULL }
5275 };
5276
5277 /* used by MGS to process specific configurations */
5278 static int mdt_process_config(const struct lu_env *env,
5279                               struct lu_device *d, struct lustre_cfg *cfg)
5280 {
5281         struct mdt_device *m = mdt_dev(d);
5282         struct md_device *md_next = m->mdt_child;
5283         struct lu_device *next = md2lu_dev(md_next);
5284         int rc;
5285         ENTRY;
5286
5287         switch (cfg->lcfg_command) {
5288         case LCFG_PARAM: {
5289                 struct obd_device          *obd = d->ld_obd;
5290
5291                 /* For interoperability */
5292                 struct cfg_interop_param   *ptr = NULL;
5293                 struct lustre_cfg          *old_cfg = NULL;
5294                 char                       *param = NULL;
5295
5296                 param = lustre_cfg_string(cfg, 1);
5297                 if (param == NULL) {
5298                         CERROR("param is empty\n");
5299                         rc = -EINVAL;
5300                         break;
5301                 }
5302
5303                 ptr = class_find_old_param(param, mdt_interop_param);
5304                 if (ptr != NULL) {
5305                         if (ptr->new_param == NULL) {
5306                                 rc = 0;
5307                                 CWARN("For interoperability, skip this %s."
5308                                       " It is obsolete.\n", ptr->old_param);
5309                                 break;
5310                         }
5311
5312                         CWARN("Found old param %s, changed it to %s.\n",
5313                               ptr->old_param, ptr->new_param);
5314
5315                         old_cfg = cfg;
5316                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
5317                         if (IS_ERR(cfg)) {
5318                                 rc = PTR_ERR(cfg);
5319                                 break;
5320                         }
5321                 }
5322
5323                 rc = class_process_proc_param(PARAM_MDT, obd->obd_vars,
5324                                               cfg, obd);
5325                 if (rc > 0 || rc == -ENOSYS) {
5326                         /* is it an HSM var ? */
5327                         rc = class_process_proc_param(PARAM_HSM,
5328                                                       hsm_cdt_get_proc_vars(),
5329                                                       cfg, obd);
5330                         if (rc > 0 || rc == -ENOSYS)
5331                                 /* we don't understand; pass it on */
5332                                 rc = next->ld_ops->ldo_process_config(env, next,
5333                                                                       cfg);
5334                 }
5335
5336                 if (old_cfg)
5337                         OBD_FREE(cfg, lustre_cfg_len(cfg->lcfg_bufcount,
5338                                                      cfg->lcfg_buflens));
5339                 break;
5340         }
5341         default:
5342                 /* others are passed further */
5343                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
5344                 break;
5345         }
5346         RETURN(rc);
5347 }
5348
5349 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
5350                                           const struct lu_object_header *hdr,
5351                                           struct lu_device *d)
5352 {
5353         struct mdt_object *mo;
5354
5355         ENTRY;
5356
5357         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
5358         if (mo != NULL) {
5359                 struct lu_object *o;
5360                 struct lu_object_header *h;
5361
5362                 o = &mo->mot_obj;
5363                 h = &mo->mot_header;
5364                 lu_object_header_init(h);
5365                 lu_object_init(o, h, d);
5366                 lu_object_add_top(h, o);
5367                 o->lo_ops = &mdt_obj_ops;
5368                 spin_lock_init(&mo->mot_write_lock);
5369                 mutex_init(&mo->mot_lov_mutex);
5370                 init_rwsem(&mo->mot_dom_sem);
5371                 init_rwsem(&mo->mot_open_sem);
5372                 atomic_set(&mo->mot_open_count, 0);
5373                 RETURN(o);
5374         }
5375         RETURN(NULL);
5376 }
5377
5378 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
5379                            const struct lu_object_conf *unused)
5380 {
5381         struct mdt_device *d = mdt_dev(o->lo_dev);
5382         struct lu_device  *under;
5383         struct lu_object  *below;
5384         int                rc = 0;
5385         ENTRY;
5386
5387         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
5388                PFID(lu_object_fid(o)));
5389
5390         under = &d->mdt_child->md_lu_dev;
5391         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
5392         if (below != NULL) {
5393                 lu_object_add(o, below);
5394         } else
5395                 rc = -ENOMEM;
5396
5397         RETURN(rc);
5398 }
5399
5400 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
5401 {
5402         struct mdt_object *mo = mdt_obj(o);
5403         struct lu_object_header *h;
5404         ENTRY;
5405
5406         h = o->lo_header;
5407         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
5408                PFID(lu_object_fid(o)));
5409
5410         LASSERT(atomic_read(&mo->mot_open_count) == 0);
5411         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
5412
5413         lu_object_fini(o);
5414         lu_object_header_fini(h);
5415         OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
5416
5417         EXIT;
5418 }
5419
5420 static int mdt_object_print(const struct lu_env *env, void *cookie,
5421                             lu_printer_t p, const struct lu_object *o)
5422 {
5423         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
5424
5425         return (*p)(env, cookie,
5426                     LUSTRE_MDT_NAME"-object@%p(%s %s, writecount=%d)",
5427                     mdto, mdto->mot_lov_created ? "lov_created" : "",
5428                     mdto->mot_cache_attr ? "cache_attr" : "",
5429                     mdto->mot_write_count);
5430 }
5431
5432 static int mdt_prepare(const struct lu_env *env,
5433                 struct lu_device *pdev,
5434                 struct lu_device *cdev)
5435 {
5436         struct mdt_device *mdt = mdt_dev(cdev);
5437         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
5438         struct obd_device *obd = cdev->ld_obd;
5439         int rc;
5440
5441         ENTRY;
5442
5443         LASSERT(obd);
5444
5445         rc = next->ld_ops->ldo_prepare(env, cdev, next);
5446         if (rc)
5447                 RETURN(rc);
5448
5449         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
5450         if (rc)
5451                 RETURN(rc);
5452
5453         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
5454         if (rc)
5455                 RETURN(rc);
5456
5457         rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
5458         /* The LFSCK instance is registered just now, so it must be there when
5459          * register the namespace to such instance. */
5460         LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
5461
5462         if (mdt->mdt_seq_site.ss_node_id == 0) {
5463                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
5464                                                          &mdt->mdt_md_root_fid);
5465                 if (rc)
5466                         RETURN(rc);
5467         }
5468
5469         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
5470
5471         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
5472         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
5473         LASSERT(obd->obd_no_conn);
5474         spin_lock(&obd->obd_dev_lock);
5475         obd->obd_no_conn = 0;
5476         spin_unlock(&obd->obd_dev_lock);
5477
5478         if (obd->obd_recovering == 0)
5479                 mdt_postrecov(env, mdt);
5480
5481         RETURN(rc);
5482 }
5483
5484 const struct lu_device_operations mdt_lu_ops = {
5485         .ldo_object_alloc   = mdt_object_alloc,
5486         .ldo_process_config = mdt_process_config,
5487         .ldo_prepare        = mdt_prepare,
5488 };
5489
5490 static const struct lu_object_operations mdt_obj_ops = {
5491         .loo_object_init    = mdt_object_init,
5492         .loo_object_free    = mdt_object_free,
5493         .loo_object_print   = mdt_object_print
5494 };
5495
5496 static int mdt_obd_set_info_async(const struct lu_env *env,
5497                                   struct obd_export *exp,
5498                                   __u32 keylen, void *key,
5499                                   __u32 vallen, void *val,
5500                                   struct ptlrpc_request_set *set)
5501 {
5502         int rc;
5503
5504         ENTRY;
5505
5506         if (KEY_IS(KEY_SPTLRPC_CONF)) {
5507                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp));
5508                 RETURN(rc);
5509         }
5510
5511         RETURN(0);
5512 }
5513
5514 /**
5515  * Match client and server connection feature flags.
5516  *
5517  * Compute the compatibility flags for a connection request based on
5518  * features mutually supported by client and server.
5519  *
5520  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
5521  * must not be updated here, otherwise a partially initialized value may
5522  * be exposed. After the connection request is successfully processed,
5523  * the top-level MDT connect request handler atomically updates the export
5524  * connect flags from the obd_connect_data::ocd_connect_flags field of the
5525  * reply. \see mdt_connect().
5526  *
5527  * Before 2.7.50 clients will send a struct obd_connect_data_v1 rather than a
5528  * full struct obd_connect_data. So care must be taken when accessing fields
5529  * that are not present in struct obd_connect_data_v1. See LU-16.
5530  *
5531  * \param exp   the obd_export associated with this client/target pair
5532  * \param mdt   the target device for the connection
5533  * \param data  stores data for this connect request
5534  *
5535  * \retval 0       success
5536  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
5537  * \retval -EBADE  client and server feature requirements are incompatible
5538  */
5539 static int mdt_connect_internal(const struct lu_env *env,
5540                                 struct obd_export *exp,
5541                                 struct mdt_device *mdt,
5542                                 struct obd_connect_data *data, bool reconnect)
5543 {
5544         LASSERT(data != NULL);
5545
5546         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
5547
5548         if (mdt->mdt_bottom->dd_rdonly &&
5549             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5550             !(data->ocd_connect_flags & OBD_CONNECT_RDONLY))
5551                 RETURN(-EACCES);
5552
5553         if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
5554                 data->ocd_connect_flags2 &= MDT_CONNECT_SUPPORTED2;
5555
5556         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
5557
5558         if (!(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5559             !(data->ocd_connect_flags & OBD_CONNECT_IBITS)) {
5560                 CWARN("%s: client %s does not support ibits lock, either "
5561                       "very old or an invalid client: flags %#llx\n",
5562                       mdt_obd_name(mdt), exp->exp_client_uuid.uuid,
5563                       data->ocd_connect_flags);
5564                 return -EBADE;
5565         }
5566
5567         if (!mdt->mdt_opts.mo_acl)
5568                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
5569
5570         if (!mdt->mdt_opts.mo_user_xattr)
5571                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
5572
5573         if (OCD_HAS_FLAG(data, BRW_SIZE)) {
5574                 data->ocd_brw_size = min(data->ocd_brw_size,
5575                                          mdt->mdt_brw_size);
5576                 if (data->ocd_brw_size == 0) {
5577                         CERROR("%s: cli %s/%p ocd_connect_flags: %#llx "
5578                                "ocd_version: %x ocd_grant: %d ocd_index: %u "
5579                                "ocd_brw_size unexpectedly zero, network data "
5580                                "corruption? Refusing to connect this client\n",
5581                                mdt_obd_name(mdt),
5582                                exp->exp_client_uuid.uuid,
5583                                exp, data->ocd_connect_flags, data->ocd_version,
5584                                data->ocd_grant, data->ocd_index);
5585                         return -EPROTO;
5586                 }
5587         }
5588
5589         if (OCD_HAS_FLAG(data, GRANT_PARAM)) {
5590                 struct dt_device_param *ddp = &mdt->mdt_lut.lut_dt_conf;
5591
5592                 /* client is reporting its page size, for future use */
5593                 exp->exp_target_data.ted_pagebits = data->ocd_grant_blkbits;
5594                 data->ocd_grant_blkbits  = mdt->mdt_lut.lut_tgd.tgd_blockbits;
5595                 /* ddp_inodespace may not be power-of-two value, eg. for ldiskfs
5596                  * it's LDISKFS_DIR_REC_LEN(20) = 28. */
5597                 data->ocd_grant_inobits = fls(ddp->ddp_inodespace - 1);
5598                 /* ocd_grant_tax_kb is in 1K byte blocks */
5599                 data->ocd_grant_tax_kb = ddp->ddp_extent_tax >> 10;
5600                 data->ocd_grant_max_blks = ddp->ddp_max_extent_blks;
5601         }
5602
5603         if (OCD_HAS_FLAG(data, GRANT)) {
5604                 /* Save connect_data we have so far because tgt_grant_connect()
5605                  * uses it to calculate grant. */
5606                 exp->exp_connect_data = *data;
5607                 tgt_grant_connect(env, exp, data, !reconnect);
5608         }
5609
5610         if (OCD_HAS_FLAG(data, MAXBYTES))
5611                 data->ocd_maxbytes = mdt->mdt_lut.lut_dt_conf.ddp_maxbytes;
5612
5613         /* NB: Disregard the rule against updating
5614          * exp_connect_data.ocd_connect_flags in this case, since
5615          * tgt_client_new() needs to know if this is a lightweight
5616          * connection, and it is safe to expose this flag before
5617          * connection processing completes. */
5618         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
5619                 spin_lock(&exp->exp_lock);
5620                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
5621                 spin_unlock(&exp->exp_lock);
5622         }
5623
5624         data->ocd_version = LUSTRE_VERSION_CODE;
5625
5626         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
5627                 CWARN("%s: MDS requires FID support, but client not\n",
5628                       mdt_obd_name(mdt));
5629                 return -EBADE;
5630         }
5631
5632         if (OCD_HAS_FLAG(data, PINGLESS)) {
5633                 if (ptlrpc_pinger_suppress_pings()) {
5634                         spin_lock(&exp->exp_obd->obd_dev_lock);
5635                         list_del_init(&exp->exp_obd_chain_timed);
5636                         spin_unlock(&exp->exp_obd->obd_dev_lock);
5637                 } else {
5638                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
5639                 }
5640         }
5641
5642         data->ocd_max_easize = mdt->mdt_max_ea_size;
5643
5644         /* NB: Disregard the rule against updating
5645          * exp_connect_data.ocd_connect_flags in this case, since
5646          * tgt_client_new() needs to know if this is client supports
5647          * multiple modify RPCs, and it is safe to expose this flag before
5648          * connection processing completes. */
5649         if (data->ocd_connect_flags & OBD_CONNECT_MULTIMODRPCS) {
5650                 data->ocd_maxmodrpcs = max_mod_rpcs_per_client;
5651                 spin_lock(&exp->exp_lock);
5652                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_MULTIMODRPCS;
5653                 spin_unlock(&exp->exp_lock);
5654         }
5655
5656         if (OCD_HAS_FLAG(data, CKSUM)) {
5657                 __u32 cksum_types = data->ocd_cksum_types;
5658
5659                 /* The client set in ocd_cksum_types the checksum types it
5660                  * supports. We have to mask off the algorithms that we don't
5661                  * support */
5662                 data->ocd_cksum_types &= cksum_types_supported_server();
5663
5664                 if (unlikely(data->ocd_cksum_types == 0)) {
5665                         CERROR("%s: Connect with checksum support but no "
5666                                "ocd_cksum_types is set\n",
5667                                exp->exp_obd->obd_name);
5668                         RETURN(-EPROTO);
5669                 }
5670
5671                 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return "
5672                        "%x\n", exp->exp_obd->obd_name, obd_export_nid2str(exp),
5673                        cksum_types, data->ocd_cksum_types);
5674         } else {
5675                 /* This client does not support OBD_CONNECT_CKSUM
5676                  * fall back to CRC32 */
5677                 CDEBUG(D_RPCTRACE, "%s: cli %s does not support "
5678                        "OBD_CONNECT_CKSUM, CRC32 will be used\n",
5679                        exp->exp_obd->obd_name, obd_export_nid2str(exp));
5680         }
5681
5682         return 0;
5683 }
5684
5685 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
5686                                    struct mdt_thread_info *info,
5687                                    struct mdt_file_data *mfd)
5688 {
5689         struct lu_context ses;
5690         int rc;
5691         ENTRY;
5692
5693         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
5694         if (rc)
5695                 RETURN(rc);
5696
5697         env->le_ses = &ses;
5698         lu_context_enter(&ses);
5699
5700         mdt_ucred(info)->uc_valid = UCRED_OLD;
5701         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
5702
5703         lu_context_exit(&ses);
5704         lu_context_fini(&ses);
5705         env->le_ses = NULL;
5706
5707         RETURN(rc);
5708 }
5709
5710 static int mdt_export_cleanup(struct obd_export *exp)
5711 {
5712         struct list_head         closing_list;
5713         struct mdt_export_data  *med = &exp->exp_mdt_data;
5714         struct obd_device       *obd = exp->exp_obd;
5715         struct mdt_device       *mdt;
5716         struct mdt_thread_info  *info;
5717         struct lu_env            env;
5718         struct mdt_file_data    *mfd, *n;
5719         int rc = 0;
5720         ENTRY;
5721
5722         INIT_LIST_HEAD(&closing_list);
5723         spin_lock(&med->med_open_lock);
5724         while (!list_empty(&med->med_open_head)) {
5725                 struct list_head *tmp = med->med_open_head.next;
5726                 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
5727
5728                 /* Remove mfd handle so it can't be found again.
5729                  * We are consuming the mfd_list reference here. */
5730                 class_handle_unhash(&mfd->mfd_handle);
5731                 list_move_tail(&mfd->mfd_list, &closing_list);
5732         }
5733         spin_unlock(&med->med_open_lock);
5734         mdt = mdt_dev(obd->obd_lu_dev);
5735         LASSERT(mdt != NULL);
5736
5737         rc = lu_env_init(&env, LCT_MD_THREAD);
5738         if (rc)
5739                 RETURN(rc);
5740
5741         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5742         LASSERT(info != NULL);
5743         memset(info, 0, sizeof *info);
5744         info->mti_env = &env;
5745         info->mti_mdt = mdt;
5746         info->mti_exp = exp;
5747
5748         if (!list_empty(&closing_list)) {
5749                 struct md_attr *ma = &info->mti_attr;
5750
5751                 /* Close any open files (which may also cause orphan
5752                  * unlinking). */
5753                 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5754                         list_del_init(&mfd->mfd_list);
5755                         ma->ma_need = ma->ma_valid = 0;
5756
5757                         /* This file is being closed due to an eviction, it
5758                          * could have been modified and now dirty regarding to
5759                          * HSM archive, check this!
5760                          * The logic here is to mark a file dirty if there's a
5761                          * chance it was dirtied before the client was evicted,
5762                          * so that we don't have to wait for a release attempt
5763                          * before finding out the file was actually dirty and
5764                          * fail the release. Aggressively marking it dirty here
5765                          * will cause the policy engine to attempt to
5766                          * re-archive it; when rearchiving, we can compare the
5767                          * current version to the HSM data_version and make the
5768                          * archive request into a noop if it's not actually
5769                          * dirty.
5770                          */
5771                         if (mfd->mfd_mode & FMODE_WRITE)
5772                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
5773
5774                         /* Don't unlink orphan on failover umount, LU-184 */
5775                         if (exp->exp_flags & OBD_OPT_FAILOVER) {
5776                                 ma->ma_valid = MA_FLAGS;
5777                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5778                         }
5779                         mdt_mfd_close(info, mfd);
5780                 }
5781         }
5782         info->mti_mdt = NULL;
5783         /* cleanup client slot early */
5784         /* Do not erase record for recoverable client. */
5785         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5786                 tgt_client_del(&env, exp);
5787         lu_env_fini(&env);
5788
5789         RETURN(rc);
5790 }
5791
5792 static inline void mdt_enable_slc(struct mdt_device *mdt)
5793 {
5794         if (mdt->mdt_lut.lut_sync_lock_cancel == NEVER_SYNC_ON_CANCEL)
5795                 mdt->mdt_lut.lut_sync_lock_cancel = BLOCKING_SYNC_ON_CANCEL;
5796 }
5797
5798 static inline void mdt_disable_slc(struct mdt_device *mdt)
5799 {
5800         if (mdt->mdt_lut.lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL)
5801                 mdt->mdt_lut.lut_sync_lock_cancel = NEVER_SYNC_ON_CANCEL;
5802 }
5803
5804 static int mdt_obd_disconnect(struct obd_export *exp)
5805 {
5806         int rc;
5807
5808         ENTRY;
5809
5810         LASSERT(exp);
5811         class_export_get(exp);
5812
5813         if (!(exp->exp_flags & OBD_OPT_FORCE))
5814                 tgt_grant_sanity_check(exp->exp_obd, __func__);
5815
5816         if ((exp_connect_flags(exp) & OBD_CONNECT_MDS_MDS) &&
5817             !(exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT)) {
5818                 struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
5819
5820                 if (atomic_dec_and_test(&mdt->mdt_mds_mds_conns))
5821                         mdt_disable_slc(mdt);
5822         }
5823
5824         rc = server_disconnect_export(exp);
5825         if (rc != 0)
5826                 CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
5827
5828         tgt_grant_discard(exp);
5829
5830         rc = mdt_export_cleanup(exp);
5831         nodemap_del_member(exp);
5832         class_export_put(exp);
5833         RETURN(rc);
5834 }
5835
5836 /* mds_connect copy */
5837 static int mdt_obd_connect(const struct lu_env *env,
5838                            struct obd_export **exp, struct obd_device *obd,
5839                            struct obd_uuid *cluuid,
5840                            struct obd_connect_data *data,
5841                            void *localdata)
5842 {
5843         struct obd_export       *lexp;
5844         struct lustre_handle    conn = { 0 };
5845         struct mdt_device       *mdt;
5846         int                      rc;
5847         lnet_nid_t              *client_nid = localdata;
5848         ENTRY;
5849
5850         LASSERT(env != NULL);
5851         LASSERT(data != NULL);
5852
5853         if (!exp || !obd || !cluuid)
5854                 RETURN(-EINVAL);
5855
5856         mdt = mdt_dev(obd->obd_lu_dev);
5857
5858         if ((data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5859             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
5860                 atomic_inc(&mdt->mdt_mds_mds_conns);
5861                 mdt_enable_slc(mdt);
5862         }
5863
5864         /*
5865          * first, check whether the stack is ready to handle requests
5866          * XXX: probably not very appropriate method is used now
5867          *      at some point we should find a better one
5868          */
5869         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) &&
5870             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) &&
5871             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
5872                 rc = obd_get_info(env, mdt->mdt_child_exp,
5873                                   sizeof(KEY_OSP_CONNECTED),
5874                                   KEY_OSP_CONNECTED, NULL, NULL);
5875                 if (rc)
5876                         RETURN(-EAGAIN);
5877                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
5878         }
5879
5880         rc = class_connect(&conn, obd, cluuid);
5881         if (rc)
5882                 RETURN(rc);
5883
5884         lexp = class_conn2export(&conn);
5885         LASSERT(lexp != NULL);
5886
5887         rc = nodemap_add_member(*client_nid, lexp);
5888         if (rc != 0 && rc != -EEXIST)
5889                 GOTO(out, rc);
5890
5891         rc = mdt_connect_internal(env, lexp, mdt, data, false);
5892         if (rc == 0) {
5893                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5894
5895                 LASSERT(lcd);
5896                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5897                 rc = tgt_client_new(env, lexp);
5898                 if (rc == 0)
5899                         mdt_export_stats_init(obd, lexp, localdata);
5900         }
5901 out:
5902         if (rc != 0) {
5903                 class_disconnect(lexp);
5904                 nodemap_del_member(lexp);
5905                 *exp = NULL;
5906         } else {
5907                 *exp = lexp;
5908                 /* Because we do not want this export to be evicted by pinger,
5909                  * let's not add this export to the timed chain list. */
5910                 if (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) {
5911                         spin_lock(&lexp->exp_obd->obd_dev_lock);
5912                         list_del_init(&lexp->exp_obd_chain_timed);
5913                         spin_unlock(&lexp->exp_obd->obd_dev_lock);
5914                 }
5915         }
5916
5917         RETURN(rc);
5918 }
5919
5920 static int mdt_obd_reconnect(const struct lu_env *env,
5921                              struct obd_export *exp, struct obd_device *obd,
5922                              struct obd_uuid *cluuid,
5923                              struct obd_connect_data *data,
5924                              void *localdata)
5925 {
5926         lnet_nid_t             *client_nid = localdata;
5927         int                     rc;
5928         ENTRY;
5929
5930         if (exp == NULL || obd == NULL || cluuid == NULL)
5931                 RETURN(-EINVAL);
5932
5933         rc = nodemap_add_member(*client_nid, exp);
5934         if (rc != 0 && rc != -EEXIST)
5935                 RETURN(rc);
5936
5937         rc = mdt_connect_internal(env, exp, mdt_dev(obd->obd_lu_dev), data,
5938                                   true);
5939         if (rc == 0)
5940                 mdt_export_stats_init(obd, exp, localdata);
5941         else
5942                 nodemap_del_member(exp);
5943
5944         RETURN(rc);
5945 }
5946
5947 /* FIXME: Can we avoid using these two interfaces? */
5948 static int mdt_init_export(struct obd_export *exp)
5949 {
5950         struct mdt_export_data *med = &exp->exp_mdt_data;
5951         int                     rc;
5952         ENTRY;
5953
5954         INIT_LIST_HEAD(&med->med_open_head);
5955         spin_lock_init(&med->med_open_lock);
5956         spin_lock(&exp->exp_lock);
5957         exp->exp_connecting = 1;
5958         spin_unlock(&exp->exp_lock);
5959
5960         /* self-export doesn't need client data and ldlm initialization */
5961         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5962                                      &exp->exp_client_uuid)))
5963                 RETURN(0);
5964
5965         rc = tgt_client_alloc(exp);
5966         if (rc)
5967                 GOTO(err, rc);
5968
5969         rc = ldlm_init_export(exp);
5970         if (rc)
5971                 GOTO(err_free, rc);
5972
5973         RETURN(rc);
5974
5975 err_free:
5976         tgt_client_free(exp);
5977 err:
5978         CERROR("%s: Failed to initialize export: rc = %d\n",
5979                exp->exp_obd->obd_name, rc);
5980         return rc;
5981 }
5982
5983 static int mdt_destroy_export(struct obd_export *exp)
5984 {
5985         ENTRY;
5986
5987         target_destroy_export(exp);
5988         /* destroy can be called from failed obd_setup, so
5989          * checking uuid is safer than obd_self_export */
5990         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5991                                      &exp->exp_client_uuid)))
5992                 RETURN(0);
5993
5994         ldlm_destroy_export(exp);
5995         tgt_client_free(exp);
5996
5997         LASSERT(list_empty(&exp->exp_outstanding_replies));
5998         LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
5999
6000         /*
6001          * discard grants once we're sure no more
6002          * interaction with the client is possible
6003          */
6004         tgt_grant_discard(exp);
6005         if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
6006                 exp->exp_obd->u.obt.obt_lut->lut_tgd.tgd_tot_granted_clients--;
6007
6008         if (!(exp->exp_flags & OBD_OPT_FORCE))
6009                 tgt_grant_sanity_check(exp->exp_obd, __func__);
6010
6011         RETURN(0);
6012 }
6013
6014 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
6015                    struct linkea_data *ldata)
6016 {
6017         int rc;
6018
6019         LASSERT(ldata->ld_buf->lb_buf != NULL);
6020
6021         if (!mdt_object_exists(mdt_obj))
6022                 return -ENODATA;
6023
6024         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6025                           ldata->ld_buf, XATTR_NAME_LINK);
6026         if (rc == -ERANGE) {
6027                 /* Buf was too small, figure out what we need. */
6028                 lu_buf_free(ldata->ld_buf);
6029                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6030                                   ldata->ld_buf, XATTR_NAME_LINK);
6031                 if (rc < 0)
6032                         return rc;
6033                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
6034                 if (ldata->ld_buf->lb_buf == NULL)
6035                         return -ENOMEM;
6036                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6037                                   ldata->ld_buf, XATTR_NAME_LINK);
6038         }
6039         if (rc < 0)
6040                 return rc;
6041
6042         return linkea_init_with_rec(ldata);
6043 }
6044
6045 /**
6046  * Given an MDT object, try to look up the full path to the object.
6047  * Part of the MDT layer implementation of lfs fid2path.
6048  *
6049  * \param[in]     info  Per-thread common data shared by MDT level handlers.
6050  * \param[in]     obj   Object to do path lookup of
6051  * \param[in,out] fp    User-provided struct to store path information
6052  * \param[in]     root_fid Root FID of current path should reach
6053  *
6054  * \retval 0 Lookup successful, path information stored in fp
6055  * \retval -EAGAIN Lookup failed, usually because object is being moved
6056  * \retval negative errno if there was a problem
6057  */
6058 static int mdt_path_current(struct mdt_thread_info *info,
6059                             struct mdt_object *obj,
6060                             struct getinfo_fid2path *fp,
6061                             struct lu_fid *root_fid)
6062 {
6063         struct mdt_device       *mdt = info->mti_mdt;
6064         struct mdt_object       *mdt_obj;
6065         struct link_ea_header   *leh;
6066         struct link_ea_entry    *lee;
6067         struct lu_name          *tmpname = &info->mti_name;
6068         struct lu_fid           *tmpfid = &info->mti_tmp_fid1;
6069         struct lu_buf           *buf = &info->mti_big_buf;
6070         char                    *ptr;
6071         int                     reclen;
6072         struct linkea_data      ldata = { NULL };
6073         int                     rc = 0;
6074         bool                    first = true;
6075         ENTRY;
6076
6077         /* temp buffer for path element, the buffer will be finally freed
6078          * in mdt_thread_info_fini */
6079         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
6080         if (buf->lb_buf == NULL)
6081                 RETURN(-ENOMEM);
6082
6083         ldata.ld_buf = buf;
6084         ptr = fp->gf_u.gf_path + fp->gf_pathlen - 1;
6085         *ptr = 0;
6086         --ptr;
6087         *tmpfid = fp->gf_fid = *mdt_object_fid(obj);
6088
6089         while (!lu_fid_eq(root_fid, &fp->gf_fid)) {
6090                 struct lu_buf           lmv_buf;
6091
6092                 if (!lu_fid_eq(root_fid, &mdt->mdt_md_root_fid) &&
6093                     lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid))
6094                         GOTO(out, rc = -ENOENT);
6095
6096                 if (lu_fid_eq(mdt_object_fid(obj), tmpfid)) {
6097                         mdt_obj = obj;
6098                         mdt_object_get(info->mti_env, mdt_obj);
6099                 } else {
6100                         mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
6101                         if (IS_ERR(mdt_obj))
6102                                 GOTO(out, rc = PTR_ERR(mdt_obj));
6103                 }
6104
6105                 if (!mdt_object_exists(mdt_obj)) {
6106                         mdt_object_put(info->mti_env, mdt_obj);
6107                         GOTO(out, rc = -ENOENT);
6108                 }
6109
6110                 if (mdt_object_remote(mdt_obj)) {
6111                         mdt_object_put(info->mti_env, mdt_obj);
6112                         GOTO(remote_out, rc = -EREMOTE);
6113                 }
6114
6115                 rc = mdt_links_read(info, mdt_obj, &ldata);
6116                 if (rc != 0) {
6117                         mdt_object_put(info->mti_env, mdt_obj);
6118                         GOTO(out, rc);
6119                 }
6120
6121                 leh = buf->lb_buf;
6122                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
6123                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
6124                 /* If set, use link #linkno for path lookup, otherwise use
6125                    link #0.  Only do this for the final path element. */
6126                 if (first && fp->gf_linkno < leh->leh_reccount) {
6127                         int count;
6128                         for (count = 0; count < fp->gf_linkno; count++) {
6129                                 lee = (struct link_ea_entry *)
6130                                      ((char *)lee + reclen);
6131                                 linkea_entry_unpack(lee, &reclen, tmpname,
6132                                                     tmpfid);
6133                         }
6134                         if (fp->gf_linkno < leh->leh_reccount - 1)
6135                                 /* indicate to user there are more links */
6136                                 fp->gf_linkno++;
6137                 }
6138
6139                 lmv_buf.lb_buf = info->mti_xattr_buf;
6140                 lmv_buf.lb_len = sizeof(info->mti_xattr_buf);
6141                 /* Check if it is slave stripes */
6142                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
6143                                   &lmv_buf, XATTR_NAME_LMV);
6144                 mdt_object_put(info->mti_env, mdt_obj);
6145                 if (rc > 0) {
6146                         union lmv_mds_md *lmm = lmv_buf.lb_buf;
6147
6148                         /* For slave stripes, get its master */
6149                         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE) {
6150                                 fp->gf_fid = *tmpfid;
6151                                 continue;
6152                         }
6153                 } else if (rc < 0 && rc != -ENODATA) {
6154                         GOTO(out, rc);
6155                 }
6156
6157                 rc = 0;
6158
6159                 /* Pack the name in the end of the buffer */
6160                 ptr -= tmpname->ln_namelen;
6161                 if (ptr - 1 <= fp->gf_u.gf_path)
6162                         GOTO(out, rc = -EOVERFLOW);
6163                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
6164                 *(--ptr) = '/';
6165
6166                 /* keep the last resolved fid to the client, so the
6167                  * client will build the left path on another MDT for
6168                  * remote object */
6169                 fp->gf_fid = *tmpfid;
6170
6171                 first = false;
6172         }
6173
6174 remote_out:
6175         ptr++; /* skip leading / */
6176         memmove(fp->gf_u.gf_path, ptr,
6177                 fp->gf_u.gf_path + fp->gf_pathlen - ptr);
6178
6179 out:
6180         RETURN(rc);
6181 }
6182
6183 /**
6184  * Given an MDT object, use mdt_path_current to get the path.
6185  * Essentially a wrapper to retry mdt_path_current a set number of times
6186  * if -EAGAIN is returned (usually because an object is being moved).
6187  *
6188  * Part of the MDT layer implementation of lfs fid2path.
6189  *
6190  * \param[in]     info  Per-thread common data shared by mdt level handlers.
6191  * \param[in]     obj   Object to do path lookup of
6192  * \param[in,out] fp    User-provided struct for arguments and to store path
6193  *                      information
6194  *
6195  * \retval 0 Lookup successful, path information stored in fp
6196  * \retval negative errno if there was a problem
6197  */
6198 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
6199                     struct getinfo_fid2path *fp, struct lu_fid *root_fid)
6200 {
6201         struct mdt_device       *mdt = info->mti_mdt;
6202         int                     tries = 3;
6203         int                     rc = -EAGAIN;
6204         ENTRY;
6205
6206         if (fp->gf_pathlen < 3)
6207                 RETURN(-EOVERFLOW);
6208
6209         if (root_fid == NULL)
6210                 root_fid = &mdt->mdt_md_root_fid;
6211
6212         if (lu_fid_eq(root_fid, mdt_object_fid(obj))) {
6213                 fp->gf_u.gf_path[0] = '\0';
6214                 RETURN(0);
6215         }
6216
6217         /* Retry multiple times in case file is being moved */
6218         while (tries-- && rc == -EAGAIN)
6219                 rc = mdt_path_current(info, obj, fp, root_fid);
6220
6221         RETURN(rc);
6222 }
6223
6224 /**
6225  * Get the full path of the provided FID, as of changelog record recno.
6226  *
6227  * This checks sanity and looks up object for user provided FID
6228  * before calling the actual path lookup code.
6229  *
6230  * Part of the MDT layer implementation of lfs fid2path.
6231  *
6232  * \param[in]     info  Per-thread common data shared by mdt level handlers.
6233  * \param[in,out] fp    User-provided struct for arguments and to store path
6234  *                      information
6235  *
6236  * \retval 0 Lookup successful, path information and recno stored in fp
6237  * \retval -ENOENT, object does not exist
6238  * \retval negative errno if there was a problem
6239  */
6240 static int mdt_fid2path(struct mdt_thread_info *info,
6241                         struct lu_fid *root_fid,
6242                         struct getinfo_fid2path *fp)
6243 {
6244         struct mdt_device *mdt = info->mti_mdt;
6245         struct mdt_object *obj;
6246         int    rc;
6247         ENTRY;
6248
6249         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
6250                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
6251
6252         if (!fid_is_sane(&fp->gf_fid))
6253                 RETURN(-EINVAL);
6254
6255         if (!fid_is_namespace_visible(&fp->gf_fid)) {
6256                 CDEBUG(D_INFO, "%s: "DFID" is invalid, f_seq should be >= %#llx"
6257                        ", or f_oid != 0, or f_ver == 0\n", mdt_obd_name(mdt),
6258                        PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
6259                 RETURN(-EINVAL);
6260         }
6261
6262         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
6263         if (IS_ERR(obj)) {
6264                 rc = PTR_ERR(obj);
6265                 CDEBUG(D_IOCTL, "cannot find "DFID": rc = %d\n",
6266                        PFID(&fp->gf_fid), rc);
6267                 RETURN(rc);
6268         }
6269
6270         if (mdt_object_remote(obj))
6271                 rc = -EREMOTE;
6272         else if (!mdt_object_exists(obj))
6273                 rc = -ENOENT;
6274         else
6275                 rc = 0;
6276
6277         if (rc < 0) {
6278                 mdt_object_put(info->mti_env, obj);
6279                 CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n",
6280                        PFID(&fp->gf_fid), rc);
6281                 RETURN(rc);
6282         }
6283
6284         rc = mdt_path(info, obj, fp, root_fid);
6285
6286         CDEBUG(D_INFO, "fid "DFID", path %s recno %#llx linkno %u\n",
6287                PFID(&fp->gf_fid), fp->gf_u.gf_path,
6288                fp->gf_recno, fp->gf_linkno);
6289
6290         mdt_object_put(info->mti_env, obj);
6291
6292         RETURN(rc);
6293 }
6294
6295 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, int keylen,
6296                             void *val, int vallen)
6297 {
6298         struct getinfo_fid2path *fpout, *fpin;
6299         struct lu_fid *root_fid = NULL;
6300         int rc = 0;
6301
6302         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
6303         fpout = val;
6304
6305         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
6306                 lustre_swab_fid2path(fpin);
6307
6308         memcpy(fpout, fpin, sizeof(*fpin));
6309         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
6310                 RETURN(-EINVAL);
6311
6312         if (keylen >= cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*fpin) +
6313                       sizeof(struct lu_fid)) {
6314                 /* client sent its root FID, which is normally fileset FID */
6315                 root_fid = fpin->gf_u.gf_root_fid;
6316                 if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
6317                         lustre_swab_lu_fid(root_fid);
6318
6319                 if (root_fid != NULL && !fid_is_sane(root_fid))
6320                         RETURN(-EINVAL);
6321         }
6322
6323         rc = mdt_fid2path(info, root_fid, fpout);
6324         RETURN(rc);
6325 }
6326
6327 int mdt_get_info(struct tgt_session_info *tsi)
6328 {
6329         char    *key;
6330         int      keylen;
6331         __u32   *vallen;
6332         void    *valout;
6333         int      rc;
6334
6335         ENTRY;
6336
6337         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
6338         if (key == NULL) {
6339                 CDEBUG(D_IOCTL, "No GETINFO key\n");
6340                 RETURN(err_serious(-EFAULT));
6341         }
6342         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
6343                                       RCL_CLIENT);
6344
6345         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
6346         if (vallen == NULL) {
6347                 CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n",
6348                                 tgt_name(tsi->tsi_tgt));
6349                 RETURN(err_serious(-EFAULT));
6350         }
6351
6352         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
6353                              *vallen);
6354         rc = req_capsule_server_pack(tsi->tsi_pill);
6355         if (rc)
6356                 RETURN(err_serious(rc));
6357
6358         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
6359         if (valout == NULL) {
6360                 CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n",
6361                                 tgt_name(tsi->tsi_tgt));
6362                 RETURN(err_serious(-EFAULT));
6363         }
6364
6365         if (KEY_IS(KEY_FID2PATH)) {
6366                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
6367
6368                 rc = mdt_rpc_fid2path(info, key, keylen, valout, *vallen);
6369                 mdt_thread_info_fini(info);
6370         } else {
6371                 rc = -EINVAL;
6372         }
6373         RETURN(rc);
6374 }
6375
6376 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
6377 {
6378         struct obd_ioctl_data *data = karg;
6379         struct lu_fid *fid;
6380         __u64 version;
6381         struct mdt_object *obj;
6382         struct mdt_lock_handle  *lh;
6383         int rc;
6384         ENTRY;
6385
6386         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
6387             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
6388                 RETURN(-EINVAL);
6389
6390         fid = (struct lu_fid *)data->ioc_inlbuf1;
6391
6392         if (!fid_is_sane(fid))
6393                 RETURN(-EINVAL);
6394
6395         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
6396
6397         lh = &mti->mti_lh[MDT_LH_PARENT];
6398         mdt_lock_reg_init(lh, LCK_CR);
6399
6400         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
6401         if (IS_ERR(obj))
6402                 RETURN(PTR_ERR(obj));
6403
6404         if (mdt_object_remote(obj)) {
6405                 rc = -EREMOTE;
6406                 /**
6407                  * before calling version get the correct MDS should be
6408                  * fid, this is error to find remote object here
6409                  */
6410                 CERROR("nonlocal object "DFID"\n", PFID(fid));
6411         } else if (!mdt_object_exists(obj)) {
6412                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
6413                 rc = -ENOENT;
6414         } else {
6415                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
6416                *(__u64 *)data->ioc_inlbuf2 = version;
6417                 rc = 0;
6418         }
6419         mdt_object_unlock_put(mti, obj, lh, 1);
6420         RETURN(rc);
6421 }
6422
6423 /* ioctls on obd dev */
6424 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
6425                          void *karg, void __user *uarg)
6426 {
6427         struct lu_env      env;
6428         struct obd_device *obd = exp->exp_obd;
6429         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
6430         struct dt_device  *dt = mdt->mdt_bottom;
6431         int rc;
6432
6433         ENTRY;
6434         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
6435         rc = lu_env_init(&env, LCT_MD_THREAD);
6436         if (rc)
6437                 RETURN(rc);
6438
6439         switch (cmd) {
6440         case OBD_IOC_SYNC:
6441                 rc = mdt_device_sync(&env, mdt);
6442                 break;
6443         case OBD_IOC_SET_READONLY:
6444                 rc = dt_sync(&env, dt);
6445                 if (rc == 0)
6446                         rc = dt_ro(&env, dt);
6447                 break;
6448         case OBD_IOC_ABORT_RECOVERY:
6449                 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
6450                 obd->obd_abort_recovery = 1;
6451                 target_stop_recovery_thread(obd);
6452                 rc = 0;
6453                 break;
6454         case OBD_IOC_CHANGELOG_REG:
6455         case OBD_IOC_CHANGELOG_DEREG:
6456         case OBD_IOC_CHANGELOG_CLEAR:
6457                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(&env,
6458                                                            mdt->mdt_child,
6459                                                            cmd, len, karg);
6460                 break;
6461         case OBD_IOC_START_LFSCK: {
6462                 struct md_device *next = mdt->mdt_child;
6463                 struct obd_ioctl_data *data = karg;
6464                 struct lfsck_start_param lsp;
6465
6466                 if (unlikely(data == NULL)) {
6467                         rc = -EINVAL;
6468                         break;
6469                 }
6470
6471                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
6472                 lsp.lsp_index_valid = 0;
6473                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
6474                 break;
6475         }
6476         case OBD_IOC_STOP_LFSCK: {
6477                 struct md_device        *next = mdt->mdt_child;
6478                 struct obd_ioctl_data   *data = karg;
6479                 struct lfsck_stop        stop;
6480
6481                 stop.ls_status = LS_STOPPED;
6482                 /* Old lfsck utils may pass NULL @stop. */
6483                 if (data->ioc_inlbuf1 == NULL)
6484                         stop.ls_flags = 0;
6485                 else
6486                         stop.ls_flags =
6487                         ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
6488
6489                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
6490                 break;
6491         }
6492         case OBD_IOC_QUERY_LFSCK: {
6493                 struct md_device        *next = mdt->mdt_child;
6494                 struct obd_ioctl_data   *data = karg;
6495
6496                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0,
6497                                                  data->ioc_inlbuf1);
6498                 break;
6499         }
6500         case OBD_IOC_GET_OBJ_VERSION: {
6501                 struct mdt_thread_info *mti;
6502                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6503                 memset(mti, 0, sizeof *mti);
6504                 mti->mti_env = &env;
6505                 mti->mti_mdt = mdt;
6506                 mti->mti_exp = exp;
6507
6508                 rc = mdt_ioc_version_get(mti, karg);
6509                 break;
6510         }
6511         case OBD_IOC_CATLOGLIST: {
6512                 struct mdt_thread_info *mti;
6513
6514                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6515                 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
6516                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
6517                                        &mti->mti_tmp_fid1);
6518                 break;
6519          }
6520         default:
6521                 rc = -EOPNOTSUPP;
6522                 CERROR("%s: Not supported cmd = %d, rc = %d\n",
6523                         mdt_obd_name(mdt), cmd, rc);
6524         }
6525
6526         lu_env_fini(&env);
6527         RETURN(rc);
6528 }
6529
6530 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
6531 {
6532         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
6533         int rc;
6534         ENTRY;
6535
6536         if (!mdt->mdt_skip_lfsck && !mdt->mdt_bottom->dd_rdonly) {
6537                 struct lfsck_start_param lsp;
6538
6539                 lsp.lsp_start = NULL;
6540                 lsp.lsp_index_valid = 0;
6541                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
6542                                                            OBD_IOC_START_LFSCK,
6543                                                            0, &lsp);
6544                 if (rc != 0 && rc != -EALREADY)
6545                         CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
6546                               mdt_obd_name(mdt), rc);
6547         }
6548
6549         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
6550         RETURN(rc);
6551 }
6552
6553 static int mdt_obd_postrecov(struct obd_device *obd)
6554 {
6555         struct lu_env env;
6556         int rc;
6557
6558         rc = lu_env_init(&env, LCT_MD_THREAD);
6559         if (rc)
6560                 RETURN(rc);
6561         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
6562         lu_env_fini(&env);
6563         return rc;
6564 }
6565
6566 static struct obd_ops mdt_obd_device_ops = {
6567         .o_owner          = THIS_MODULE,
6568         .o_set_info_async = mdt_obd_set_info_async,
6569         .o_connect        = mdt_obd_connect,
6570         .o_reconnect      = mdt_obd_reconnect,
6571         .o_disconnect     = mdt_obd_disconnect,
6572         .o_init_export    = mdt_init_export,
6573         .o_destroy_export = mdt_destroy_export,
6574         .o_iocontrol      = mdt_iocontrol,
6575         .o_postrecov      = mdt_obd_postrecov,
6576         /* Data-on-MDT IO methods */
6577         .o_preprw         = mdt_obd_preprw,
6578         .o_commitrw       = mdt_obd_commitrw,
6579 };
6580
6581 static struct lu_device* mdt_device_fini(const struct lu_env *env,
6582                                          struct lu_device *d)
6583 {
6584         struct mdt_device *m = mdt_dev(d);
6585         ENTRY;
6586
6587         mdt_fini(env, m);
6588         RETURN(NULL);
6589 }
6590
6591 static struct lu_device *mdt_device_free(const struct lu_env *env,
6592                                          struct lu_device *d)
6593 {
6594         struct mdt_device *m = mdt_dev(d);
6595         ENTRY;
6596
6597         lu_device_fini(&m->mdt_lu_dev);
6598         OBD_FREE_PTR(m);
6599
6600         RETURN(NULL);
6601 }
6602
6603 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
6604                                           struct lu_device_type *t,
6605                                           struct lustre_cfg *cfg)
6606 {
6607         struct lu_device  *l;
6608         struct mdt_device *m;
6609
6610         OBD_ALLOC_PTR(m);
6611         if (m != NULL) {
6612                 int rc;
6613
6614                 l = &m->mdt_lu_dev;
6615                 rc = mdt_init0(env, m, t, cfg);
6616                 if (rc != 0) {
6617                         mdt_device_free(env, l);
6618                         l = ERR_PTR(rc);
6619                         return l;
6620                 }
6621         } else
6622                 l = ERR_PTR(-ENOMEM);
6623         return l;
6624 }
6625
6626 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
6627 LU_KEY_INIT(mdt, struct mdt_thread_info);
6628
6629 static void mdt_key_fini(const struct lu_context *ctx,
6630                          struct lu_context_key *key, void* data)
6631 {
6632         struct mdt_thread_info *info = data;
6633
6634         if (info->mti_big_lmm) {
6635                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
6636                 info->mti_big_lmm = NULL;
6637                 info->mti_big_lmmsize = 0;
6638         }
6639
6640         if (info->mti_big_acl) {
6641                 OBD_FREE_LARGE(info->mti_big_acl, info->mti_big_aclsize);
6642                 info->mti_big_acl = NULL;
6643                 info->mti_big_aclsize = 0;
6644         }
6645
6646         OBD_FREE_PTR(info);
6647 }
6648
6649 /* context key: mdt_thread_key */
6650 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
6651
6652 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
6653 {
6654         return lu_ucred(info->mti_env);
6655 }
6656
6657 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
6658 {
6659         return lu_ucred_check(info->mti_env);
6660 }
6661
6662 /**
6663  * Enable/disable COS (Commit On Sharing).
6664  *
6665  * Set/Clear the COS flag in mdt options.
6666  *
6667  * \param mdt mdt device
6668  * \param val 0 disables COS, other values enable COS
6669  */
6670 void mdt_enable_cos(struct mdt_device *mdt, int val)
6671 {
6672         struct lu_env env;
6673         int rc;
6674
6675         mdt->mdt_opts.mo_cos = !!val;
6676         rc = lu_env_init(&env, LCT_LOCAL);
6677         if (unlikely(rc != 0)) {
6678                 CWARN("%s: lu_env initialization failed, cannot "
6679                       "sync: rc = %d\n", mdt_obd_name(mdt), rc);
6680                 return;
6681         }
6682         mdt_device_sync(&env, mdt);
6683         lu_env_fini(&env);
6684 }
6685
6686 /**
6687  * Check COS (Commit On Sharing) status.
6688  *
6689  * Return COS flag status.
6690  *
6691  * \param mdt mdt device
6692  */
6693 int mdt_cos_is_enabled(struct mdt_device *mdt)
6694 {
6695         return mdt->mdt_opts.mo_cos != 0;
6696 }
6697
6698 static struct lu_device_type_operations mdt_device_type_ops = {
6699         .ldto_device_alloc = mdt_device_alloc,
6700         .ldto_device_free  = mdt_device_free,
6701         .ldto_device_fini  = mdt_device_fini
6702 };
6703
6704 static struct lu_device_type mdt_device_type = {
6705         .ldt_tags     = LU_DEVICE_MD,
6706         .ldt_name     = LUSTRE_MDT_NAME,
6707         .ldt_ops      = &mdt_device_type_ops,
6708         .ldt_ctx_tags = LCT_MD_THREAD
6709 };
6710
6711 static int __init mdt_init(void)
6712 {
6713         int rc;
6714
6715         CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") ==
6716                  FID_NOBRACE_LEN + 1);
6717         CLASSERT(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") ==
6718                  FID_LEN + 1);
6719         rc = lu_kmem_init(mdt_caches);
6720         if (rc)
6721                 return rc;
6722
6723         rc = mds_mod_init();
6724         if (rc)
6725                 GOTO(lu_fini, rc);
6726
6727         rc = class_register_type(&mdt_obd_device_ops, NULL, true, NULL,
6728                                  LUSTRE_MDT_NAME, &mdt_device_type);
6729         if (rc)
6730                 GOTO(mds_fini, rc);
6731 lu_fini:
6732         if (rc)
6733                 lu_kmem_fini(mdt_caches);
6734 mds_fini:
6735         if (rc)
6736                 mds_mod_exit();
6737         return rc;
6738 }
6739
6740 static void __exit mdt_exit(void)
6741 {
6742         class_unregister_type(LUSTRE_MDT_NAME);
6743         mds_mod_exit();
6744         lu_kmem_fini(mdt_caches);
6745 }
6746
6747 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
6748 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
6749 MODULE_VERSION(LUSTRE_VERSION_STRING);
6750 MODULE_LICENSE("GPL");
6751
6752 module_init(mdt_init);
6753 module_exit(mdt_exit);