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 <lustre_ioctl.h>
  54 #include <lustre_lfsck.h>
  55 #include <lustre_log.h>
  56 #include <lustre_net.h>
  57 #include <lustre_nodemap.h>
  58 #include <lustre_mds.h>
  59 #include <lustre_param.h>
  60 #include <lustre_quota.h>
  61 #include <lustre_swab.h>
  62 #include <obd.h>
  63 #include <obd_support.h>
  64 #include <lustre_barrier.h>
  65
  66 #include <llog_swab.h>
  67
  68 #include "mdt_internal.h"
  69
  70
  71 static unsigned int max_mod_rpcs_per_client = 8;
  72 module_param(max_mod_rpcs_per_client, uint, 0644);
  73 MODULE_PARM_DESC(max_mod_rpcs_per_client, "maximum number of modify RPCs in flight allowed per client");
  74
  75 mdl_mode_t mdt_mdl_lock_modes[] = {
  76         [LCK_MINMODE] = MDL_MINMODE,
  77         [LCK_EX]      = MDL_EX,
  78         [LCK_PW]      = MDL_PW,
  79         [LCK_PR]      = MDL_PR,
  80         [LCK_CW]      = MDL_CW,
  81         [LCK_CR]      = MDL_CR,
  82         [LCK_NL]      = MDL_NL,
  83         [LCK_GROUP]   = MDL_GROUP
  84 };
  85
  86 enum ldlm_mode mdt_dlm_lock_modes[] = {
  87         [MDL_MINMODE]   = LCK_MINMODE,
  88         [MDL_EX]        = LCK_EX,
  89         [MDL_PW]        = LCK_PW,
  90         [MDL_PR]        = LCK_PR,
  91         [MDL_CW]        = LCK_CW,
  92         [MDL_CR]        = LCK_CR,
  93         [MDL_NL]        = LCK_NL,
  94         [MDL_GROUP]     = LCK_GROUP
  95 };
  96
  97 static struct mdt_device *mdt_dev(struct lu_device *d);
  98 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
  99
 100 static const struct lu_object_operations mdt_obj_ops;
 101
 102 /* Slab for MDT object allocation */
 103 static struct kmem_cache *mdt_object_kmem;
 104
 105 /* For HSM restore handles */
 106 struct kmem_cache *mdt_hsm_cdt_kmem;
 107
 108 /* For HSM request handles */
 109 struct kmem_cache *mdt_hsm_car_kmem;
 110
 111 static struct lu_kmem_descr mdt_caches[] = {
 112         {
 113                 .ckd_cache = &mdt_object_kmem,
 114                 .ckd_name  = "mdt_obj",
 115                 .ckd_size  = sizeof(struct mdt_object)
 116         },
 117         {
 118                 .ckd_cache      = &mdt_hsm_cdt_kmem,
 119                 .ckd_name       = "mdt_cdt_restore_handle",
 120                 .ckd_size       = sizeof(struct cdt_restore_handle)
 121         },
 122         {
 123                 .ckd_cache      = &mdt_hsm_car_kmem,
 124                 .ckd_name       = "mdt_cdt_agent_req",
 125                 .ckd_size       = sizeof(struct cdt_agent_req)
 126         },
 127         {
 128                 .ckd_cache = NULL
 129         }
 130 };
 131
 132 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag)
 133 {
 134         if (!rep)
 135                 return 0;
 136         return rep->lock_policy_res1 & op_flag;
 137 }
 138
 139 void mdt_clear_disposition(struct mdt_thread_info *info,
 140                            struct ldlm_reply *rep, __u64 op_flag)
 141 {
 142         if (info) {
 143                 info->mti_opdata &= ~op_flag;
 144                 tgt_opdata_clear(info->mti_env, op_flag);
 145         }
 146         if (rep)
 147                 rep->lock_policy_res1 &= ~op_flag;
 148 }
 149
 150 void mdt_set_disposition(struct mdt_thread_info *info,
 151                          struct ldlm_reply *rep, __u64 op_flag)
 152 {
 153         if (info) {
 154                 info->mti_opdata |= op_flag;
 155                 tgt_opdata_set(info->mti_env, op_flag);
 156         }
 157         if (rep)
 158                 rep->lock_policy_res1 |= op_flag;
 159 }
 160
 161 void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm)
 162 {
 163         lh->mlh_pdo_hash = 0;
 164         lh->mlh_reg_mode = lm;
 165         lh->mlh_rreg_mode = lm;
 166         lh->mlh_type = MDT_REG_LOCK;
 167 }
 168
 169 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode,
 170                        const struct lu_name *lname)
 171 {
 172         lh->mlh_reg_mode = lock_mode;
 173         lh->mlh_pdo_mode = LCK_MINMODE;
 174         lh->mlh_rreg_mode = lock_mode;
 175         lh->mlh_type = MDT_PDO_LOCK;
 176
 177         if (lu_name_is_valid(lname)) {
 178                 lh->mlh_pdo_hash = full_name_hash(lname->ln_name,
 179                                                   lname->ln_namelen);
 180                 /* XXX Workaround for LU-2856
 181                  *
 182                  * Zero is a valid return value of full_name_hash, but
 183                  * several users of mlh_pdo_hash assume a non-zero
 184                  * hash value. We therefore map zero onto an
 185                  * arbitrary, but consistent value (1) to avoid
 186                  * problems further down the road. */
 187                 if (unlikely(lh->mlh_pdo_hash == 0))
 188                         lh->mlh_pdo_hash = 1;
 189         } else {
 190                 lh->mlh_pdo_hash = 0;
 191         }
 192 }
 193
 194 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 195                               struct mdt_lock_handle *lh)
 196 {
 197         mdl_mode_t mode;
 198         ENTRY;
 199
 200         /*
 201          * Any dir access needs couple of locks:
 202          *
 203          * 1) on part of dir we gonna take lookup/modify;
 204          *
 205          * 2) on whole dir to protect it from concurrent splitting and/or to
 206          * flush client's cache for readdir().
 207          *
 208          * so, for a given mode and object this routine decides what lock mode
 209          * to use for lock #2:
 210          *
 211          * 1) if caller's gonna lookup in dir then we need to protect dir from
 212          * being splitted only - LCK_CR
 213          *
 214          * 2) if caller's gonna modify dir then we need to protect dir from
 215          * being splitted and to flush cache - LCK_CW
 216          *
 217          * 3) if caller's gonna modify dir and that dir seems ready for
 218          * splitting then we need to protect it from any type of access
 219          * (lookup/modify/split) - LCK_EX --bzzz
 220          */
 221
 222         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 223         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 224
 225         /*
 226          * Ask underlaying level its opinion about preferable PDO lock mode
 227          * having access type passed as regular lock mode:
 228          *
 229          * - MDL_MINMODE means that lower layer does not want to specify lock
 230          * mode;
 231          *
 232          * - MDL_NL means that no PDO lock should be taken. This is used in some
 233          * cases. Say, for non-splittable directories no need to use PDO locks
 234          * at all.
 235          */
 236         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 237                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 238
 239         if (mode != MDL_MINMODE) {
 240                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 241         } else {
 242                 /*
 243                  * Lower layer does not want to specify locking mode. We do it
 244                  * our selves. No special protection is needed, just flush
 245                  * client's cache on modification and allow concurrent
 246                  * mondification.
 247                  */
 248                 switch (lh->mlh_reg_mode) {
 249                 case LCK_EX:
 250                         lh->mlh_pdo_mode = LCK_EX;
 251                         break;
 252                 case LCK_PR:
 253                         lh->mlh_pdo_mode = LCK_CR;
 254                         break;
 255                 case LCK_PW:
 256                         lh->mlh_pdo_mode = LCK_CW;
 257                         break;
 258                 default:
 259                         CERROR("Not expected lock type (0x%x)\n",
 260                                (int)lh->mlh_reg_mode);
 261                         LBUG();
 262                 }
 263         }
 264
 265         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 266         EXIT;
 267 }
 268
 269 static int mdt_lookup_fileset(struct mdt_thread_info *info, const char *fileset,
 270                               struct lu_fid *fid)
 271 {
 272         struct mdt_device *mdt = info->mti_mdt;
 273         struct lu_name *lname = &info->mti_name;
 274         char *name = NULL;
 275         struct mdt_object *parent;
 276         u32 mode;
 277         int rc = 0;
 278
 279         LASSERT(!info->mti_cross_ref);
 280
 281         OBD_ALLOC(name, NAME_MAX + 1);
 282         if (name == NULL)
 283                 return -ENOMEM;
 284         lname->ln_name = name;
 285
 286         /*
 287          * We may want to allow this to mount a completely separate
 288          * fileset from the MDT in the future, but keeping it to
 289          * ROOT/ only for now avoid potential security issues.
 290          */
 291         *fid = mdt->mdt_md_root_fid;
 292
 293         while (rc == 0 && fileset != NULL && *fileset != '\0') {
 294                 const char *s1 = fileset;
 295                 const char *s2;
 296
 297                 while (*++s1 == '/')
 298                         ;
 299                 s2 = s1;
 300                 while (*s2 != '/' && *s2 != '\0')
 301                         s2++;
 302
 303                 if (s2 == s1)
 304                         break;
 305
 306                 fileset = s2;
 307
 308                 lname->ln_namelen = s2 - s1;
 309                 if (lname->ln_namelen > NAME_MAX) {
 310                         rc = -EINVAL;
 311                         break;
 312                 }
 313
 314                 /* reject .. as a path component */
 315                 if (lname->ln_namelen == 2 &&
 316                     strncmp(s1, "..", 2) == 0) {
 317                         rc = -EINVAL;
 318                         break;
 319                 }
 320
 321                 strncpy(name, s1, lname->ln_namelen);
 322                 name[lname->ln_namelen] = '\0';
 323
 324                 parent = mdt_object_find(info->mti_env, mdt, fid);
 325                 if (IS_ERR(parent)) {
 326                         rc = PTR_ERR(parent);
 327                         break;
 328                 }
 329                 /* Only got the fid of this obj by name */
 330                 fid_zero(fid);
 331                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
 332                                 fid, &info->mti_spec);
 333                 mdt_object_put(info->mti_env, parent);
 334         }
 335         if (!rc) {
 336                 parent = mdt_object_find(info->mti_env, mdt, fid);
 337                 if (IS_ERR(parent))
 338                         rc = PTR_ERR(parent);
 339                 else {
 340                         mode = lu_object_attr(&parent->mot_obj);
 341                         mdt_object_put(info->mti_env, parent);
 342                         if (!S_ISDIR(mode))
 343                                 rc = -ENOTDIR;
 344                 }
 345         }
 346
 347         OBD_FREE(name, NAME_MAX + 1);
 348
 349         return rc;
 350 }
 351
 352 static int mdt_get_root(struct tgt_session_info *tsi)
 353 {
 354         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
 355         struct mdt_device       *mdt = info->mti_mdt;
 356         struct mdt_body         *repbody;
 357         char                    *fileset = NULL, *buffer = NULL;
 358         int                      rc;
 359         struct obd_export       *exp = info->mti_exp;
 360         char                    *nodemap_fileset;
 361
 362         ENTRY;
 363
 364         rc = mdt_check_ucred(info);
 365         if (rc)
 366                 GOTO(out, rc = err_serious(rc));
 367
 368         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GET_ROOT_PACK))
 369                 GOTO(out, rc = err_serious(-ENOMEM));
 370
 371         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 372         if (req_capsule_get_size(info->mti_pill, &RMF_NAME, RCL_CLIENT) > 0) {
 373                 fileset = req_capsule_client_get(info->mti_pill, &RMF_NAME);
 374                 if (fileset == NULL)
 375                         GOTO(out, rc = err_serious(-EFAULT));
 376         }
 377
 378         nodemap_fileset = nodemap_get_fileset(exp->exp_target_data.ted_nodemap);
 379         if (nodemap_fileset && nodemap_fileset[0]) {
 380                 CDEBUG(D_INFO, "nodemap fileset is %s\n", nodemap_fileset);
 381                 if (fileset) {
 382                         /* consider fileset from client as a sub-fileset
 383                          * of the nodemap one */
 384                         OBD_ALLOC(buffer, PATH_MAX + 1);
 385                         if (buffer == NULL)
 386                                 GOTO(out, rc = err_serious(-ENOMEM));
 387                         if (snprintf(buffer, PATH_MAX + 1, "%s/%s",
 388                                      nodemap_fileset, fileset) >= PATH_MAX + 1)
 389                                 GOTO(out, rc = err_serious(-EINVAL));
 390                         fileset = buffer;
 391                 } else {
 392                         /* enforce fileset as specified in the nodemap */
 393                         fileset = nodemap_fileset;
 394                 }
 395         }
 396
 397         if (fileset) {
 398                 CDEBUG(D_INFO, "Getting fileset %s\n", fileset);
 399                 rc = mdt_lookup_fileset(info, fileset, &repbody->mbo_fid1);
 400                 if (rc < 0)
 401                         GOTO(out, rc = err_serious(rc));
 402         } else {
 403                 repbody->mbo_fid1 = mdt->mdt_md_root_fid;
 404         }
 405         repbody->mbo_valid |= OBD_MD_FLID;
 406
 407         EXIT;
 408 out:
 409         mdt_thread_info_fini(info);
 410         if (buffer)
 411                 OBD_FREE(buffer, PATH_MAX+1);
 412         return rc;
 413 }
 414
 415 static int mdt_statfs(struct tgt_session_info *tsi)
 416 {
 417         struct ptlrpc_request           *req = tgt_ses_req(tsi);
 418         struct mdt_thread_info          *info = tsi2mdt_info(tsi);
 419         struct md_device                *next = info->mti_mdt->mdt_child;
 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         /** statfs information are cached in the mdt_device */
 445         if (cfs_time_before_64(info->mti_mdt->mdt_osfs_age,
 446                                cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS))) {
 447                 /** statfs data is too old, get up-to-date one */
 448                 rc = next->md_ops->mdo_statfs(info->mti_env, next, osfs);
 449                 if (rc)
 450                         GOTO(out, rc);
 451                 spin_lock(&info->mti_mdt->mdt_lock);
 452                 info->mti_mdt->mdt_osfs = *osfs;
 453                 info->mti_mdt->mdt_osfs_age = cfs_time_current_64();
 454                 spin_unlock(&info->mti_mdt->mdt_lock);
 455         } else {
 456                 /** use cached statfs data */
 457                 spin_lock(&info->mti_mdt->mdt_lock);
 458                 *osfs = info->mti_mdt->mdt_osfs;
 459                 spin_unlock(&info->mti_mdt->mdt_lock);
 460         }
 461
 462         if (rc == 0)
 463                 mdt_counter_incr(req, LPROC_MDT_STATFS);
 464 out:
 465         mdt_thread_info_fini(info);
 466         RETURN(rc);
 467 }
 468
 469 #ifdef CONFIG_FS_POSIX_ACL
 470 /*
 471  * Pack ACL data into the reply. UIDs/GIDs are mapped and filtered by nodemap.
 472  *
 473  * \param       info    thread info object
 474  * \param       repbody reply to pack ACLs into
 475  * \param       o       mdt object of file to examine
 476  * \param       nodemap nodemap of client to reply to
 477  * \retval      0       success
 478  * \retval      -errno  error getting or parsing ACL from disk
 479  */
 480 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
 481                       struct mdt_object *o, struct lu_nodemap *nodemap)
 482 {
 483         const struct lu_env     *env = info->mti_env;
 484         struct md_object        *next = mdt_object_child(o);
 485         struct lu_buf           *buf = &info->mti_buf;
 486         int rc;
 487
 488         buf->lb_buf = req_capsule_server_get(info->mti_pill, &RMF_ACL);
 489         buf->lb_len = req_capsule_get_size(info->mti_pill, &RMF_ACL,
 490                                            RCL_SERVER);
 491         if (buf->lb_len == 0)
 492                 return 0;
 493
 494         rc = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_ACCESS);
 495         if (rc < 0) {
 496                 if (rc == -ENODATA) {
 497                         repbody->mbo_aclsize = 0;
 498                         repbody->mbo_valid |= OBD_MD_FLACL;
 499                         rc = 0;
 500                 } else if (rc == -EOPNOTSUPP) {
 501                         rc = 0;
 502                 } else {
 503                         CERROR("%s: unable to read "DFID" ACL: rc = %d\n",
 504                                mdt_obd_name(info->mti_mdt),
 505                                PFID(mdt_object_fid(o)), rc);
 506                 }
 507         } else {
 508                 rc = nodemap_map_acl(nodemap, buf->lb_buf,
 509                                      rc, NODEMAP_FS_TO_CLIENT);
 510                 /* if all ACLs mapped out, rc is still >= 0 */
 511                 if (rc < 0) {
 512                         CERROR("%s: nodemap_map_acl unable to parse "DFID
 513                                " ACL: rc = %d\n", mdt_obd_name(info->mti_mdt),
 514                                PFID(mdt_object_fid(o)), rc);
 515                 } else {
 516                         repbody->mbo_aclsize = rc;
 517                         repbody->mbo_valid |= OBD_MD_FLACL;
 518                         rc = 0;
 519                 }
 520         }
 521         return rc;
 522 }
 523 #endif
 524
 525 /* XXX Look into layout in MDT layer. */
 526 static inline bool mdt_hsm_is_released(struct lov_mds_md *lmm)
 527 {
 528         struct lov_comp_md_v1   *comp_v1;
 529         struct lov_mds_md       *v1;
 530         int                      i;
 531
 532         if (lmm->lmm_magic == LOV_MAGIC_COMP_V1) {
 533                 comp_v1 = (struct lov_comp_md_v1 *)lmm;
 534
 535                 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
 536                         v1 = (struct lov_mds_md *)((char *)comp_v1 +
 537                                 comp_v1->lcm_entries[i].lcme_offset);
 538                         /* We don't support partial release for now */
 539                         if (!(v1->lmm_pattern & LOV_PATTERN_F_RELEASED))
 540                                 return false;
 541                 }
 542                 return true;
 543         } else {
 544                 return (lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) ?
 545                         true : false;
 546         }
 547 }
 548
 549 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 550                         const struct lu_attr *attr, const struct lu_fid *fid)
 551 {
 552         struct md_attr *ma = &info->mti_attr;
 553         struct obd_export *exp = info->mti_exp;
 554         struct lu_nodemap *nodemap = NULL;
 555
 556         LASSERT(ma->ma_valid & MA_INODE);
 557
 558         if (attr->la_valid & LA_ATIME) {
 559                 b->mbo_atime = attr->la_atime;
 560                 b->mbo_valid |= OBD_MD_FLATIME;
 561         }
 562         if (attr->la_valid & LA_MTIME) {
 563                 b->mbo_mtime = attr->la_mtime;
 564                 b->mbo_valid |= OBD_MD_FLMTIME;
 565         }
 566         if (attr->la_valid & LA_CTIME) {
 567                 b->mbo_ctime = attr->la_ctime;
 568                 b->mbo_valid |= OBD_MD_FLCTIME;
 569         }
 570         if (attr->la_valid & LA_FLAGS) {
 571                 b->mbo_flags = attr->la_flags;
 572                 b->mbo_valid |= OBD_MD_FLFLAGS;
 573         }
 574         if (attr->la_valid & LA_NLINK) {
 575                 b->mbo_nlink = attr->la_nlink;
 576                 b->mbo_valid |= OBD_MD_FLNLINK;
 577         }
 578         if (attr->la_valid & (LA_UID|LA_GID)) {
 579                 nodemap = nodemap_get_from_exp(exp);
 580                 if (IS_ERR(nodemap))
 581                         goto out;
 582         }
 583         if (attr->la_valid & LA_UID) {
 584                 b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 585                                             NODEMAP_FS_TO_CLIENT,
 586                                             attr->la_uid);
 587                 b->mbo_valid |= OBD_MD_FLUID;
 588         }
 589         if (attr->la_valid & LA_GID) {
 590                 b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 591                                             NODEMAP_FS_TO_CLIENT,
 592                                             attr->la_gid);
 593                 b->mbo_valid |= OBD_MD_FLGID;
 594         }
 595
 596         if (attr->la_valid & LA_PROJID) {
 597                 /* TODO, nodemap for project id */
 598                 b->mbo_projid = attr->la_projid;
 599                 b->mbo_valid |= OBD_MD_FLPROJID;
 600         }
 601
 602         b->mbo_mode = attr->la_mode;
 603         if (attr->la_valid & LA_MODE)
 604                 b->mbo_valid |= OBD_MD_FLMODE;
 605         if (attr->la_valid & LA_TYPE)
 606                 b->mbo_valid |= OBD_MD_FLTYPE;
 607
 608         if (fid != NULL) {
 609                 b->mbo_fid1 = *fid;
 610                 b->mbo_valid |= OBD_MD_FLID;
 611                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid=%#llx\n",
 612                        PFID(fid), b->mbo_nlink, b->mbo_mode, b->mbo_valid);
 613         }
 614
 615         if (!(attr->la_valid & LA_TYPE))
 616                 return;
 617
 618         b->mbo_rdev   = attr->la_rdev;
 619         b->mbo_size   = attr->la_size;
 620         b->mbo_blocks = attr->la_blocks;
 621
 622         if (!S_ISREG(attr->la_mode)) {
 623                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 624         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
 625                 /* means no objects are allocated on osts. */
 626                 LASSERT(!(ma->ma_valid & MA_LOV));
 627                 /* just ignore blocks occupied by extend attributes on MDS */
 628                 b->mbo_blocks = 0;
 629                 /* if no object is allocated on osts, the size on mds is valid.
 630                  * b=22272 */
 631                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 632         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL &&
 633                    mdt_hsm_is_released(ma->ma_lmm)) {
 634                 /* A released file stores its size on MDS. */
 635                 /* But return 1 block for released file, unless tools like tar
 636                  * will consider it fully sparse. (LU-3864)
 637                  */
 638                 if (unlikely(b->mbo_size == 0))
 639                         b->mbo_blocks = 0;
 640                 else
 641                         b->mbo_blocks = 1;
 642                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 643         }
 644
 645         if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE))
 646                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 647                        PFID(fid), (unsigned long long)b->mbo_size);
 648
 649 out:
 650         if (!IS_ERR_OR_NULL(nodemap))
 651                 nodemap_putref(nodemap);
 652 }
 653
 654 static inline int mdt_body_has_lov(const struct lu_attr *la,
 655                                    const struct mdt_body *body)
 656 {
 657         return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
 658                (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
 659 }
 660
 661 void mdt_client_compatibility(struct mdt_thread_info *info)
 662 {
 663         struct mdt_body       *body;
 664         struct ptlrpc_request *req = mdt_info_req(info);
 665         struct obd_export     *exp = req->rq_export;
 666         struct md_attr        *ma = &info->mti_attr;
 667         struct lu_attr        *la = &ma->ma_attr;
 668         ENTRY;
 669
 670         if (exp_connect_layout(exp))
 671                 /* the client can deal with 16-bit lmm_stripe_count */
 672                 RETURN_EXIT;
 673
 674         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 675
 676         if (!mdt_body_has_lov(la, body))
 677                 RETURN_EXIT;
 678
 679         /* now we have a reply with a lov for a client not compatible with the
 680          * layout lock so we have to clean the layout generation number */
 681         if (S_ISREG(la->la_mode))
 682                 ma->ma_lmm->lmm_layout_gen = 0;
 683         EXIT;
 684 }
 685
 686 static int mdt_attr_get_eabuf_size(struct mdt_thread_info *info,
 687                                    struct mdt_object *o)
 688 {
 689         const struct lu_env *env = info->mti_env;
 690         int rc, rc2;
 691
 692         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 693                           XATTR_NAME_LOV);
 694
 695         if (rc == -ENODATA)
 696                 rc = 0;
 697
 698         if (rc < 0)
 699                 goto out;
 700
 701         /* Is it a directory? Let's check for the LMV as well */
 702         if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
 703                 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 704                                    XATTR_NAME_LMV);
 705
 706                 if (rc2 == -ENODATA)
 707                         rc2 = mo_xattr_get(env, mdt_object_child(o),
 708                                            &LU_BUF_NULL,
 709                                            XATTR_NAME_DEFAULT_LMV);
 710
 711                 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
 712                         rc = rc2;
 713         }
 714
 715 out:
 716         return rc;
 717 }
 718
 719 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
 720                       const char *name)
 721 {
 722         const struct lu_env *env = info->mti_env;
 723         int rc;
 724         ENTRY;
 725
 726         LASSERT(info->mti_big_lmm_used == 0);
 727         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
 728         if (rc < 0)
 729                 RETURN(rc);
 730
 731         /* big_lmm may need to be grown */
 732         if (info->mti_big_lmmsize < rc) {
 733                 int size = size_roundup_power2(rc);
 734
 735                 if (info->mti_big_lmmsize > 0) {
 736                         /* free old buffer */
 737                         LASSERT(info->mti_big_lmm);
 738                         OBD_FREE_LARGE(info->mti_big_lmm,
 739                                        info->mti_big_lmmsize);
 740                         info->mti_big_lmm = NULL;
 741                         info->mti_big_lmmsize = 0;
 742                 }
 743
 744                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
 745                 if (info->mti_big_lmm == NULL)
 746                         RETURN(-ENOMEM);
 747                 info->mti_big_lmmsize = size;
 748         }
 749         LASSERT(info->mti_big_lmmsize >= rc);
 750
 751         info->mti_buf.lb_buf = info->mti_big_lmm;
 752         info->mti_buf.lb_len = info->mti_big_lmmsize;
 753         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
 754
 755         RETURN(rc);
 756 }
 757
 758 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
 759                    struct md_attr *ma, const char *name)
 760 {
 761         struct md_object *next = mdt_object_child(o);
 762         struct lu_buf    *buf = &info->mti_buf;
 763         int rc;
 764
 765         if (strcmp(name, XATTR_NAME_LOV) == 0) {
 766                 buf->lb_buf = ma->ma_lmm;
 767                 buf->lb_len = ma->ma_lmm_size;
 768                 LASSERT(!(ma->ma_valid & MA_LOV));
 769         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 770                 buf->lb_buf = ma->ma_lmv;
 771                 buf->lb_len = ma->ma_lmv_size;
 772                 LASSERT(!(ma->ma_valid & MA_LMV));
 773         } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 774                 buf->lb_buf = ma->ma_lmv;
 775                 buf->lb_len = ma->ma_lmv_size;
 776                 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
 777         } else {
 778                 return -EINVAL;
 779         }
 780
 781         rc = mo_xattr_get(info->mti_env, next, buf, name);
 782         if (rc > 0) {
 783
 784 got:
 785                 if (strcmp(name, XATTR_NAME_LOV) == 0) {
 786                         if (info->mti_big_lmm_used)
 787                                 ma->ma_lmm = info->mti_big_lmm;
 788
 789                         /* NOT return LOV EA with hole to old client. */
 790                         if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
 791                                      LOV_PATTERN_F_HOLE) &&
 792                             !(exp_connect_flags(info->mti_exp) &
 793                               OBD_CONNECT_LFSCK)) {
 794                                 return -EIO;
 795                         } else {
 796                                 ma->ma_lmm_size = rc;
 797                                 ma->ma_valid |= MA_LOV;
 798                         }
 799                 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 800                         if (info->mti_big_lmm_used)
 801                                 ma->ma_lmv = info->mti_big_lmm;
 802
 803                         ma->ma_lmv_size = rc;
 804                         ma->ma_valid |= MA_LMV;
 805                 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 806                         ma->ma_lmv_size = rc;
 807                         ma->ma_valid |= MA_LMV_DEF;
 808                 }
 809
 810                 /* Update mdt_max_mdsize so all clients will be aware that */
 811                 if (info->mti_mdt->mdt_max_mdsize < rc)
 812                         info->mti_mdt->mdt_max_mdsize = rc;
 813
 814                 rc = 0;
 815         } else if (rc == -ENODATA) {
 816                 /* no LOV EA */
 817                 rc = 0;
 818         } else if (rc == -ERANGE) {
 819                 /* Default LMV has fixed size, so it must be able to fit
 820                  * in the original buffer */
 821                 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
 822                         return rc;
 823                 rc = mdt_big_xattr_get(info, o, name);
 824                 if (rc > 0) {
 825                         info->mti_big_lmm_used = 1;
 826                         goto got;
 827                 }
 828         }
 829
 830         return rc;
 831 }
 832
 833 static int mdt_attr_get_pfid(struct mdt_thread_info *info,
 834                              struct mdt_object *o, struct lu_fid *pfid)
 835 {
 836         struct lu_buf           *buf = &info->mti_buf;
 837         struct link_ea_header   *leh;
 838         struct link_ea_entry    *lee;
 839         int                      rc;
 840         ENTRY;
 841
 842         buf->lb_buf = info->mti_big_lmm;
 843         buf->lb_len = info->mti_big_lmmsize;
 844         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
 845                           buf, XATTR_NAME_LINK);
 846         /* ignore errors, MA_PFID won't be set and it is
 847          * up to the caller to treat this as an error */
 848         if (rc == -ERANGE || buf->lb_len == 0) {
 849                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
 850                 buf->lb_buf = info->mti_big_lmm;
 851                 buf->lb_len = info->mti_big_lmmsize;
 852         }
 853
 854         if (rc < 0)
 855                 RETURN(rc);
 856         if (rc < sizeof(*leh)) {
 857                 CERROR("short LinkEA on "DFID": rc = %d\n",
 858                        PFID(mdt_object_fid(o)), rc);
 859                 RETURN(-ENODATA);
 860         }
 861
 862         leh = (struct link_ea_header *) buf->lb_buf;
 863         lee = (struct link_ea_entry *)(leh + 1);
 864         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
 865                 leh->leh_magic = LINK_EA_MAGIC;
 866                 leh->leh_reccount = __swab32(leh->leh_reccount);
 867                 leh->leh_len = __swab64(leh->leh_len);
 868         }
 869         if (leh->leh_magic != LINK_EA_MAGIC)
 870                 RETURN(-EINVAL);
 871         if (leh->leh_reccount == 0)
 872                 RETURN(-ENODATA);
 873
 874         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
 875         fid_be_to_cpu(pfid, pfid);
 876
 877         RETURN(0);
 878 }
 879
 880 int mdt_attr_get_complex(struct mdt_thread_info *info,
 881                          struct mdt_object *o, struct md_attr *ma)
 882 {
 883         const struct lu_env *env = info->mti_env;
 884         struct md_object    *next = mdt_object_child(o);
 885         struct lu_buf       *buf = &info->mti_buf;
 886         int                  need = ma->ma_need;
 887         int                  rc = 0, rc2;
 888         u32                  mode;
 889         ENTRY;
 890
 891         ma->ma_valid = 0;
 892
 893         if (mdt_object_exists(o) == 0)
 894                 GOTO(out, rc = -ENOENT);
 895         mode = lu_object_attr(&next->mo_lu);
 896
 897         if (need & MA_INODE) {
 898                 ma->ma_need = MA_INODE;
 899                 rc = mo_attr_get(env, next, ma);
 900                 if (rc)
 901                         GOTO(out, rc);
 902                 ma->ma_valid |= MA_INODE;
 903         }
 904
 905         if (need & MA_PFID) {
 906                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
 907                 if (rc == 0)
 908                         ma->ma_valid |= MA_PFID;
 909                 /* ignore this error, parent fid is not mandatory */
 910                 rc = 0;
 911         }
 912
 913         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
 914                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
 915                 if (rc)
 916                         GOTO(out, rc);
 917         }
 918
 919         if (need & MA_LMV && S_ISDIR(mode)) {
 920                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
 921                 if (rc != 0)
 922                         GOTO(out, rc);
 923         }
 924
 925         if (need & MA_LMV_DEF && S_ISDIR(mode)) {
 926                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
 927                 if (rc != 0)
 928                         GOTO(out, rc);
 929         }
 930
 931         if (need & MA_HSM && S_ISREG(mode)) {
 932                 buf->lb_buf = info->mti_xattr_buf;
 933                 buf->lb_len = sizeof(info->mti_xattr_buf);
 934                 CLASSERT(sizeof(struct hsm_attrs) <=
 935                          sizeof(info->mti_xattr_buf));
 936                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
 937                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
 938                 if (rc2 == 0)
 939                         ma->ma_valid |= MA_HSM;
 940                 else if (rc2 < 0 && rc2 != -ENODATA)
 941                         GOTO(out, rc = rc2);
 942         }
 943
 944 #ifdef CONFIG_FS_POSIX_ACL
 945         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
 946                 buf->lb_buf = ma->ma_acl;
 947                 buf->lb_len = ma->ma_acl_size;
 948                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
 949                 if (rc2 > 0) {
 950                         ma->ma_acl_size = rc2;
 951                         ma->ma_valid |= MA_ACL_DEF;
 952                 } else if (rc2 == -ENODATA) {
 953                         /* no ACLs */
 954                         ma->ma_acl_size = 0;
 955                 } else
 956                         GOTO(out, rc = rc2);
 957         }
 958 #endif
 959 out:
 960         ma->ma_need = need;
 961         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = %#llx ma_lmm=%p\n",
 962                rc, ma->ma_valid, ma->ma_lmm);
 963         RETURN(rc);
 964 }
 965
 966 static int mdt_getattr_internal(struct mdt_thread_info *info,
 967                                 struct mdt_object *o, int ma_need)
 968 {
 969         struct md_object        *next = mdt_object_child(o);
 970         const struct mdt_body   *reqbody = info->mti_body;
 971         struct ptlrpc_request   *req = mdt_info_req(info);
 972         struct md_attr          *ma = &info->mti_attr;
 973         struct lu_attr          *la = &ma->ma_attr;
 974         struct req_capsule      *pill = info->mti_pill;
 975         const struct lu_env     *env = info->mti_env;
 976         struct mdt_body         *repbody;
 977         struct lu_buf           *buffer = &info->mti_buf;
 978         struct obd_export       *exp = info->mti_exp;
 979         int                      rc;
 980         ENTRY;
 981
 982         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
 983                 RETURN(err_serious(-ENOMEM));
 984
 985         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 986
 987         ma->ma_valid = 0;
 988
 989         if (mdt_object_remote(o)) {
 990                 /* This object is located on remote node.*/
 991                 /* Return -ENOTSUPP for old client */
 992                 if (!mdt_is_dne_client(req->rq_export))
 993                         GOTO(out, rc = -ENOTSUPP);
 994
 995                 repbody->mbo_fid1 = *mdt_object_fid(o);
 996                 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
 997                 GOTO(out, rc = 0);
 998         }
 999
1000         if (reqbody->mbo_eadatasize > 0) {
1001                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
1002                 if (buffer->lb_buf == NULL)
1003                         GOTO(out, rc = -EPROTO);
1004                 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
1005                                                       RCL_SERVER);
1006         } else {
1007                 buffer->lb_buf = NULL;
1008                 buffer->lb_len = 0;
1009                 ma_need &= ~(MA_LOV | MA_LMV);
1010                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
1011                        mdt_obd_name(info->mti_mdt),
1012                        req->rq_export->exp_client_uuid.uuid);
1013         }
1014
1015         /* If it is dir object and client require MEA, then we got MEA */
1016         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1017             (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
1018                 /* Assumption: MDT_MD size is enough for lmv size. */
1019                 ma->ma_lmv = buffer->lb_buf;
1020                 ma->ma_lmv_size = buffer->lb_len;
1021                 ma->ma_need = MA_INODE;
1022                 if (ma->ma_lmv_size > 0) {
1023                         if (reqbody->mbo_valid & OBD_MD_MEA)
1024                                 ma->ma_need |= MA_LMV;
1025                         else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA)
1026                                 ma->ma_need |= MA_LMV_DEF;
1027                 }
1028         } else {
1029                 ma->ma_lmm = buffer->lb_buf;
1030                 ma->ma_lmm_size = buffer->lb_len;
1031                 ma->ma_need = MA_INODE | MA_HSM;
1032                 if (ma->ma_lmm_size > 0)
1033                         ma->ma_need |= MA_LOV;
1034         }
1035
1036         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1037             reqbody->mbo_valid & OBD_MD_FLDIREA  &&
1038             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
1039                 /* get default stripe info for this dir. */
1040                 ma->ma_need |= MA_LOV_DEF;
1041         }
1042         ma->ma_need |= ma_need;
1043
1044         rc = mdt_attr_get_complex(info, o, ma);
1045         if (unlikely(rc)) {
1046                 CDEBUG(rc == -ENOENT ? D_OTHER : D_ERROR,
1047                        "%s: getattr error for "DFID": rc = %d\n",
1048                        mdt_obd_name(info->mti_mdt),
1049                        PFID(mdt_object_fid(o)), rc);
1050                 RETURN(rc);
1051         }
1052
1053         /* if file is released, check if a restore is running */
1054         if (ma->ma_valid & MA_HSM) {
1055                 repbody->mbo_valid |= OBD_MD_TSTATE;
1056                 if ((ma->ma_hsm.mh_flags & HS_RELEASED) &&
1057                     mdt_hsm_restore_is_running(info, mdt_object_fid(o)))
1058                         repbody->mbo_t_state = MS_RESTORE;
1059         }
1060
1061         if (likely(ma->ma_valid & MA_INODE))
1062                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
1063         else
1064                 RETURN(-EFAULT);
1065
1066         if (mdt_body_has_lov(la, reqbody)) {
1067                 if (ma->ma_valid & MA_LOV) {
1068                         LASSERT(ma->ma_lmm_size);
1069                         repbody->mbo_eadatasize = ma->ma_lmm_size;
1070                         if (S_ISDIR(la->la_mode))
1071                                 repbody->mbo_valid |= OBD_MD_FLDIREA;
1072                         else
1073                                 repbody->mbo_valid |= OBD_MD_FLEASIZE;
1074                         mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
1075                 }
1076                 if (ma->ma_valid & MA_LMV) {
1077                         /* Return -ENOTSUPP for old client */
1078                         if (!mdt_is_striped_client(req->rq_export))
1079                                 RETURN(-ENOTSUPP);
1080
1081                         LASSERT(S_ISDIR(la->la_mode));
1082                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1083                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1084                         repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
1085                 }
1086                 if (ma->ma_valid & MA_LMV_DEF) {
1087                         /* Return -ENOTSUPP for old client */
1088                         if (!mdt_is_striped_client(req->rq_export))
1089                                 RETURN(-ENOTSUPP);
1090                         LASSERT(S_ISDIR(la->la_mode));
1091                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1092                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1093                         repbody->mbo_valid |= (OBD_MD_FLDIREA |
1094                                                OBD_MD_DEFAULT_MEA);
1095                 }
1096         } else if (S_ISLNK(la->la_mode) &&
1097                    reqbody->mbo_valid & OBD_MD_LINKNAME) {
1098                 buffer->lb_buf = ma->ma_lmm;
1099                 /* eadatasize from client includes NULL-terminator, so
1100                  * there is no need to read it */
1101                 buffer->lb_len = reqbody->mbo_eadatasize - 1;
1102                 rc = mo_readlink(env, next, buffer);
1103                 if (unlikely(rc <= 0)) {
1104                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
1105                                mdt_obd_name(info->mti_mdt),
1106                                PFID(mdt_object_fid(o)), rc);
1107                         rc = -EFAULT;
1108                 } else {
1109                         int print_limit = min_t(int, PAGE_SIZE - 128, rc);
1110
1111                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
1112                                 rc -= 2;
1113                         repbody->mbo_valid |= OBD_MD_LINKNAME;
1114                         /* we need to report back size with NULL-terminator
1115                          * because client expects that */
1116                         repbody->mbo_eadatasize = rc + 1;
1117                         if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
1118                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
1119                                        "on "DFID ", expected %d\n",
1120                                        mdt_obd_name(info->mti_mdt),
1121                                        rc, PFID(mdt_object_fid(o)),
1122                                        reqbody->mbo_eadatasize - 1);
1123                         /* NULL terminate */
1124                         ((char *)ma->ma_lmm)[rc] = 0;
1125
1126                         /* If the total CDEBUG() size is larger than a page, it
1127                          * will print a warning to the console, avoid this by
1128                          * printing just the last part of the symlink. */
1129                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
1130                                print_limit < rc ? "..." : "", print_limit,
1131                                (char *)ma->ma_lmm + rc - print_limit, rc);
1132                         rc = 0;
1133                 }
1134         }
1135
1136         if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
1137                 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1138                 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1139                 CDEBUG(D_INODE, "changing the max MD size to %u\n",
1140                        repbody->mbo_max_mdsize);
1141         }
1142
1143 #ifdef CONFIG_FS_POSIX_ACL
1144         if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1145                  (reqbody->mbo_valid & OBD_MD_FLACL)) {
1146                 struct lu_nodemap *nodemap = nodemap_get_from_exp(exp);
1147                 if (IS_ERR(nodemap))
1148                         RETURN(PTR_ERR(nodemap));
1149
1150                 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
1151                 nodemap_putref(nodemap);
1152         }
1153 #endif
1154
1155 out:
1156         if (rc == 0)
1157                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
1158
1159         RETURN(rc);
1160 }
1161
1162 static int mdt_getattr(struct tgt_session_info *tsi)
1163 {
1164         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1165         struct mdt_object       *obj = info->mti_object;
1166         struct req_capsule      *pill = info->mti_pill;
1167         struct mdt_body         *reqbody;
1168         struct mdt_body         *repbody;
1169         int rc, rc2;
1170         ENTRY;
1171
1172         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1173         LASSERT(reqbody);
1174         LASSERT(obj != NULL);
1175         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1176
1177         /* Unlike intent case where we need to pre-fill out buffers early on
1178          * in intent policy for ldlm reasons, here we can have a much better
1179          * guess at EA size by just reading it from disk.
1180          * Exceptions are readdir and (missing) directory striping */
1181         /* Readlink */
1182         if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1183                 /* No easy way to know how long is the symlink, but it cannot
1184                  * be more than PATH_MAX, so we allocate +1 */
1185                 rc = PATH_MAX + 1;
1186
1187         /* A special case for fs ROOT: getattr there might fetch
1188          * default EA for entire fs, not just for this dir!
1189          */
1190         } else if (lu_fid_eq(mdt_object_fid(obj),
1191                              &info->mti_mdt->mdt_md_root_fid) &&
1192                    (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1193                    (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1194                                                                  MDS_GETATTR)) {
1195                 /* Should the default strping be bigger, mdt_fix_reply
1196                  * will reallocate */
1197                 rc = DEF_REP_MD_SIZE;
1198         } else {
1199                 /* Read the actual EA size from disk */
1200                 rc = mdt_attr_get_eabuf_size(info, obj);
1201         }
1202
1203         if (rc < 0)
1204                 GOTO(out, rc = err_serious(rc));
1205
1206         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1207
1208         rc = req_capsule_server_pack(pill);
1209         if (unlikely(rc != 0))
1210                 GOTO(out, rc = err_serious(rc));
1211
1212         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1213         LASSERT(repbody != NULL);
1214         repbody->mbo_eadatasize = 0;
1215         repbody->mbo_aclsize = 0;
1216
1217         rc = mdt_check_ucred(info);
1218         if (unlikely(rc))
1219                 GOTO(out_shrink, rc);
1220
1221         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1222
1223         rc = mdt_getattr_internal(info, obj, 0);
1224         EXIT;
1225 out_shrink:
1226         mdt_client_compatibility(info);
1227         rc2 = mdt_fix_reply(info);
1228         if (rc == 0)
1229                 rc = rc2;
1230 out:
1231         mdt_thread_info_fini(info);
1232         return rc;
1233 }
1234
1235 /**
1236  * Handler of layout intent RPC requiring the layout modification
1237  *
1238  * \param[in] info      thread environment
1239  * \param[in] obj       object
1240  * \param[in] layout    layout intent
1241  * \param[in] buf       buffer containing client's lovea, could be empty
1242  *
1243  * \retval 0    on success
1244  * \retval < 0  error code
1245  */
1246 static int mdt_layout_change(struct mdt_thread_info *info,
1247                              struct mdt_object *obj,
1248                              struct layout_intent *layout,
1249                              const struct lu_buf *buf)
1250 {
1251         struct mdt_lock_handle *lh = &info->mti_lh[MDT_LH_LOCAL];
1252         int rc;
1253         ENTRY;
1254
1255         if (layout->li_start >= layout->li_end) {
1256                 CERROR("Recieved an invalid layout change range [%llu, %llu) "
1257                        "for "DFID"\n", layout->li_start, layout->li_end,
1258                        PFID(mdt_object_fid(obj)));
1259                 RETURN(-EINVAL);
1260         }
1261
1262         if (!S_ISREG(lu_object_attr(&obj->mot_obj)))
1263                 GOTO(out, rc = -EINVAL);
1264
1265         rc = mo_permission(info->mti_env, NULL, mdt_object_child(obj), NULL,
1266                            MAY_WRITE);
1267         if (rc)
1268                 GOTO(out, rc);
1269
1270         /* take layout lock to prepare layout change */
1271         mdt_lock_reg_init(lh, LCK_EX);
1272         rc = mdt_object_lock(info, obj, lh,
1273                              MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR);
1274         if (rc)
1275                 GOTO(out, rc);
1276
1277         rc = mo_layout_change(info->mti_env, mdt_object_child(obj), layout,
1278                               buf);
1279
1280         mdt_object_unlock(info, obj, lh, 1);
1281 out:
1282         RETURN(rc);
1283 }
1284
1285 /**
1286  * Exchange MOF_LOV_CREATED flags between two objects after a
1287  * layout swap. No assumption is made on whether o1 or o2 have
1288  * created objects or not.
1289  *
1290  * \param[in,out] o1    First swap layout object
1291  * \param[in,out] o2    Second swap layout object
1292  */
1293 static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2)
1294 {
1295         unsigned int o1_lov_created = o1->mot_lov_created;
1296
1297         mutex_lock(&o1->mot_lov_mutex);
1298         mutex_lock(&o2->mot_lov_mutex);
1299
1300         o1->mot_lov_created = o2->mot_lov_created;
1301         o2->mot_lov_created = o1_lov_created;
1302
1303         mutex_unlock(&o2->mot_lov_mutex);
1304         mutex_unlock(&o1->mot_lov_mutex);
1305 }
1306
1307 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1308 {
1309         struct mdt_thread_info  *info;
1310         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1311         struct obd_export       *exp = req->rq_export;
1312         struct mdt_object       *o1, *o2, *o;
1313         struct mdt_lock_handle  *lh1, *lh2;
1314         struct mdc_swap_layouts *msl;
1315         int                      rc;
1316         ENTRY;
1317
1318         /* client does not support layout lock, so layout swaping
1319          * is disabled.
1320          * FIXME: there is a problem for old clients which don't support
1321          * layout lock yet. If those clients have already opened the file
1322          * they won't be notified at all so that old layout may still be
1323          * used to do IO. This can be fixed after file release is landed by
1324          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1325         if (!exp_connect_layout(exp))
1326                 RETURN(-EOPNOTSUPP);
1327
1328         info = tsi2mdt_info(tsi);
1329
1330         if (info->mti_dlm_req != NULL)
1331                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1332
1333         o1 = info->mti_object;
1334         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1335                                 &info->mti_body->mbo_fid2);
1336         if (IS_ERR(o))
1337                 GOTO(out, rc = PTR_ERR(o));
1338
1339         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1340                 GOTO(put, rc = -ENOENT);
1341
1342         rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1343         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1344                 GOTO(put, rc);
1345
1346         if (rc < 0)
1347                 swap(o1, o2);
1348
1349         /* permission check. Make sure the calling process having permission
1350          * to write both files. */
1351         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1352                                 MAY_WRITE);
1353         if (rc < 0)
1354                 GOTO(put, rc);
1355
1356         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1357                                 MAY_WRITE);
1358         if (rc < 0)
1359                 GOTO(put, rc);
1360
1361         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1362         if (msl == NULL)
1363                 GOTO(put, rc = -EPROTO);
1364
1365         lh1 = &info->mti_lh[MDT_LH_NEW];
1366         mdt_lock_reg_init(lh1, LCK_EX);
1367         lh2 = &info->mti_lh[MDT_LH_OLD];
1368         mdt_lock_reg_init(lh2, LCK_EX);
1369
1370         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1371                              MDS_INODELOCK_XATTR);
1372         if (rc < 0)
1373                 GOTO(put, rc);
1374
1375         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1376                              MDS_INODELOCK_XATTR);
1377         if (rc < 0)
1378                 GOTO(unlock1, rc);
1379
1380         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1381                              mdt_object_child(o2), msl->msl_flags);
1382         if (rc < 0)
1383                 GOTO(unlock2, rc);
1384
1385         mdt_swap_lov_flag(o1, o2);
1386
1387 unlock2:
1388         mdt_object_unlock(info, o2, lh2, rc);
1389 unlock1:
1390         mdt_object_unlock(info, o1, lh1, rc);
1391 put:
1392         mdt_object_put(info->mti_env, o);
1393 out:
1394         mdt_thread_info_fini(info);
1395         RETURN(rc);
1396 }
1397
1398 static int mdt_raw_lookup(struct mdt_thread_info *info,
1399                           struct mdt_object *parent,
1400                           const struct lu_name *lname,
1401                           struct ldlm_reply *ldlm_rep)
1402 {
1403         struct lu_fid   *child_fid = &info->mti_tmp_fid1;
1404         int              rc;
1405         ENTRY;
1406
1407         LASSERT(!info->mti_cross_ref);
1408
1409         /* Only got the fid of this obj by name */
1410         fid_zero(child_fid);
1411         rc = mdo_lookup(info->mti_env, mdt_object_child(info->mti_object),
1412                         lname, child_fid, &info->mti_spec);
1413         if (rc == 0) {
1414                 struct mdt_body *repbody;
1415
1416                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1417                 repbody->mbo_fid1 = *child_fid;
1418                 repbody->mbo_valid = OBD_MD_FLID;
1419                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1420         } else if (rc == -ENOENT) {
1421                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1422         }
1423
1424         RETURN(rc);
1425 }
1426
1427 /*
1428  * UPDATE lock should be taken against parent, and be released before exit;
1429  * child_bits lock should be taken against child, and be returned back:
1430  *            (1)normal request should release the child lock;
1431  *            (2)intent request will grant the lock to client.
1432  */
1433 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1434                                  struct mdt_lock_handle *lhc,
1435                                  __u64 child_bits,
1436                                  struct ldlm_reply *ldlm_rep)
1437 {
1438         struct ptlrpc_request  *req       = mdt_info_req(info);
1439         struct mdt_body        *reqbody   = NULL;
1440         struct mdt_object      *parent    = info->mti_object;
1441         struct mdt_object      *child;
1442         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1443         struct lu_name         *lname     = NULL;
1444         struct mdt_lock_handle *lhp       = NULL;
1445         struct ldlm_lock       *lock;
1446         bool                    is_resent;
1447         bool                    try_layout;
1448         int                     ma_need = 0;
1449         int                     rc;
1450         ENTRY;
1451
1452         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1453         LASSERT(ergo(is_resent,
1454                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1455
1456         if (parent == NULL)
1457                 RETURN(-ENOENT);
1458
1459         if (info->mti_cross_ref) {
1460                 /* Only getattr on the child. Parent is on another node. */
1461                 mdt_set_disposition(info, ldlm_rep,
1462                                     DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1463                 child = parent;
1464                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1465                        "ldlm_rep = %p\n",
1466                        PFID(mdt_object_fid(child)), ldlm_rep);
1467
1468                 rc = mdt_check_resent_lock(info, child, lhc);
1469                 if (rc < 0) {
1470                         RETURN(rc);
1471                 } else if (rc > 0) {
1472                         mdt_lock_handle_init(lhc);
1473                         mdt_lock_reg_init(lhc, LCK_PR);
1474
1475                         /*
1476                          * Object's name is on another MDS, no lookup or layout
1477                          * lock is needed here but update lock is.
1478                          */
1479                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
1480                                         MDS_INODELOCK_LAYOUT);
1481                         child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1482
1483                         rc = mdt_object_lock(info, child, lhc, child_bits);
1484                         if (rc < 0)
1485                                 RETURN(rc);
1486                 }
1487
1488                 /* Finally, we can get attr for child. */
1489                 if (!mdt_object_exists(child)) {
1490                         LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1491                                         &child->mot_obj,
1492                                         "remote object doesn't exist.");
1493                         mdt_object_unlock(info, child, lhc, 1);
1494                         RETURN(-ENOENT);
1495                 }
1496
1497                 rc = mdt_getattr_internal(info, child, 0);
1498                 if (unlikely(rc != 0))
1499                         mdt_object_unlock(info, child, lhc, 1);
1500
1501                 RETURN(rc);
1502         }
1503
1504         lname = &info->mti_name;
1505         mdt_name_unpack(info->mti_pill, &RMF_NAME, lname, MNF_FIX_ANON);
1506
1507         if (lu_name_is_valid(lname)) {
1508                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1509                        "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1510                        PNAME(lname), ldlm_rep);
1511         } else {
1512                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1513                 if (unlikely(reqbody == NULL))
1514                         RETURN(err_serious(-EPROTO));
1515
1516                 *child_fid = reqbody->mbo_fid2;
1517
1518                 if (unlikely(!fid_is_sane(child_fid)))
1519                         RETURN(err_serious(-EINVAL));
1520
1521                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1522                        "ldlm_rep = %p\n",
1523                        PFID(mdt_object_fid(parent)),
1524                        PFID(&reqbody->mbo_fid2), ldlm_rep);
1525         }
1526
1527         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1528
1529         if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1530                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1531                                 &parent->mot_obj,
1532                                 "Parent doesn't exist!");
1533                 RETURN(-ESTALE);
1534         }
1535
1536         if (mdt_object_remote(parent)) {
1537                 CERROR("%s: parent "DFID" is on remote target\n",
1538                        mdt_obd_name(info->mti_mdt),
1539                        PFID(mdt_object_fid(parent)));
1540                 RETURN(-EIO);
1541         }
1542
1543         if (lu_name_is_valid(lname)) {
1544                 /* Always allow to lookup ".." */
1545                 if (unlikely(lname->ln_namelen == 2 &&
1546                              lname->ln_name[0] == '.' &&
1547                              lname->ln_name[1] == '.'))
1548                         info->mti_spec.sp_permitted = 1;
1549
1550                 if (info->mti_body->mbo_valid == OBD_MD_FLID) {
1551                         rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1552
1553                         RETURN(rc);
1554                 }
1555
1556                 /* step 1: lock parent only if parent is a directory */
1557                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1558                         lhp = &info->mti_lh[MDT_LH_PARENT];
1559                         mdt_lock_pdo_init(lhp, LCK_PR, lname);
1560                         rc = mdt_object_lock(info, parent, lhp,
1561                                              MDS_INODELOCK_UPDATE);
1562                         if (unlikely(rc != 0))
1563                                 RETURN(rc);
1564                 }
1565
1566                 /* step 2: lookup child's fid by name */
1567                 fid_zero(child_fid);
1568                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
1569                                 child_fid, &info->mti_spec);
1570                 if (rc == -ENOENT)
1571                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1572
1573                 if (rc != 0)
1574                         GOTO(out_parent, rc);
1575         }
1576
1577         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1578
1579         /*
1580          *step 3: find the child object by fid & lock it.
1581          *        regardless if it is local or remote.
1582          *
1583          *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1584          *      set parent dir fid the same as child fid in getattr by fid case
1585          *      we should not lu_object_find() the object again, could lead
1586          *      to hung if there is a concurrent unlink destroyed the object.
1587          */
1588         if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1589                 mdt_object_get(info->mti_env, parent);
1590                 child = parent;
1591         } else {
1592                 child = mdt_object_find(info->mti_env, info->mti_mdt,
1593                                         child_fid);
1594         }
1595
1596         if (unlikely(IS_ERR(child)))
1597                 GOTO(out_parent, rc = PTR_ERR(child));
1598
1599         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
1600         if (!mdt_object_exists(child)) {
1601                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1602                                 &child->mot_obj,
1603                                 "Object doesn't exist!");
1604                 GOTO(out_child, rc = -ENOENT);
1605         }
1606
1607         rc = mdt_check_resent_lock(info, child, lhc);
1608         if (rc < 0) {
1609                 GOTO(out_child, rc);
1610         } else if (rc > 0) {
1611                 mdt_lock_handle_init(lhc);
1612                 mdt_lock_reg_init(lhc, LCK_PR);
1613                 try_layout = false;
1614
1615                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1616                       mdt_object_exists(child) && !mdt_object_remote(child)) {
1617                         struct md_attr *ma = &info->mti_attr;
1618
1619                         ma->ma_valid = 0;
1620                         ma->ma_need = MA_INODE;
1621                         rc = mdt_attr_get_complex(info, child, ma);
1622                         if (unlikely(rc != 0))
1623                                 GOTO(out_child, rc);
1624
1625                         /* If the file has not been changed for some time, we
1626                          * return not only a LOOKUP lock, but also an UPDATE
1627                          * lock and this might save us RPC on later STAT. For
1628                          * directories, it also let negative dentry cache start
1629                          * working for this dir. */
1630                         if (ma->ma_valid & MA_INODE &&
1631                             ma->ma_attr.la_valid & LA_CTIME &&
1632                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1633                                 ma->ma_attr.la_ctime < cfs_time_current_sec())
1634                                 child_bits |= MDS_INODELOCK_UPDATE;
1635                 }
1636
1637                 /* layout lock must be granted in a best-effort way
1638                  * for IT operations */
1639                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1640                 if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1641                     exp_connect_layout(info->mti_exp) &&
1642                     S_ISREG(lu_object_attr(&child->mot_obj)) &&
1643                     !mdt_object_remote(child) && ldlm_rep != NULL) {
1644                         /* try to grant layout lock for regular file. */
1645                         try_layout = true;
1646                 }
1647
1648                 rc = 0;
1649                 if (try_layout) {
1650                         child_bits |= MDS_INODELOCK_LAYOUT;
1651                         /* try layout lock, it may fail to be granted due to
1652                          * contention at LOOKUP or UPDATE */
1653                         if (!mdt_object_lock_try(info, child, lhc,
1654                                                  child_bits)) {
1655                                 child_bits &= ~MDS_INODELOCK_LAYOUT;
1656                                 LASSERT(child_bits != 0);
1657                                 rc = mdt_object_lock(info, child, lhc,
1658                                                      child_bits);
1659                         } else {
1660                                 ma_need |= MA_LOV;
1661                         }
1662                 } else {
1663                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1664                          * client will enqueue the lock to the remote MDT */
1665                         if (mdt_object_remote(child))
1666                                 child_bits &= ~MDS_INODELOCK_UPDATE;
1667                         rc = mdt_object_lock(info, child, lhc, child_bits);
1668                 }
1669                 if (unlikely(rc != 0))
1670                         GOTO(out_child, rc);
1671         }
1672
1673         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1674
1675         /* finally, we can get attr for child. */
1676         rc = mdt_getattr_internal(info, child, ma_need);
1677         if (unlikely(rc != 0)) {
1678                 mdt_object_unlock(info, child, lhc, 1);
1679         } else if (lock) {
1680                 /* Debugging code. */
1681                 LDLM_DEBUG(lock, "Returning lock to client");
1682                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1683                                          &lock->l_resource->lr_name),
1684                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1685                          PLDLMRES(lock->l_resource),
1686                          PFID(mdt_object_fid(child)));
1687         }
1688         if (lock)
1689                 LDLM_LOCK_PUT(lock);
1690
1691         EXIT;
1692 out_child:
1693         mdt_object_put(info->mti_env, child);
1694 out_parent:
1695         if (lhp)
1696                 mdt_object_unlock(info, parent, lhp, 1);
1697         return rc;
1698 }
1699
1700 /* normal handler: should release the child lock */
1701 static int mdt_getattr_name(struct tgt_session_info *tsi)
1702 {
1703         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1704         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1705         struct mdt_body        *reqbody;
1706         struct mdt_body        *repbody;
1707         int rc, rc2;
1708         ENTRY;
1709
1710         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1711         LASSERT(reqbody != NULL);
1712         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1713         LASSERT(repbody != NULL);
1714
1715         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1716         repbody->mbo_eadatasize = 0;
1717         repbody->mbo_aclsize = 0;
1718
1719         rc = mdt_init_ucred_intent_getattr(info, reqbody);
1720         if (unlikely(rc))
1721                 GOTO(out_shrink, rc);
1722
1723         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1724         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1725                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1726                 lhc->mlh_reg_lh.cookie = 0;
1727         }
1728         mdt_exit_ucred(info);
1729         EXIT;
1730 out_shrink:
1731         mdt_client_compatibility(info);
1732         rc2 = mdt_fix_reply(info);
1733         if (rc == 0)
1734                 rc = rc2;
1735         mdt_thread_info_fini(info);
1736         return rc;
1737 }
1738
1739 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1740                          void *karg, void __user *uarg);
1741
1742 static int mdt_set_info(struct tgt_session_info *tsi)
1743 {
1744         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1745         char                    *key;
1746         void                    *val;
1747         int                      keylen, vallen, rc = 0;
1748
1749         ENTRY;
1750
1751         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
1752         if (key == NULL) {
1753                 DEBUG_REQ(D_HA, req, "no set_info key");
1754                 RETURN(err_serious(-EFAULT));
1755         }
1756
1757         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
1758                                       RCL_CLIENT);
1759
1760         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
1761         if (val == NULL) {
1762                 DEBUG_REQ(D_HA, req, "no set_info val");
1763                 RETURN(err_serious(-EFAULT));
1764         }
1765
1766         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
1767                                       RCL_CLIENT);
1768
1769         /* Swab any part of val you need to here */
1770         if (KEY_IS(KEY_READ_ONLY)) {
1771                 spin_lock(&req->rq_export->exp_lock);
1772                 if (*(__u32 *)val)
1773                         *exp_connect_flags_ptr(req->rq_export) |=
1774                                 OBD_CONNECT_RDONLY;
1775                 else
1776                         *exp_connect_flags_ptr(req->rq_export) &=
1777                                 ~OBD_CONNECT_RDONLY;
1778                 spin_unlock(&req->rq_export->exp_lock);
1779         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1780                 struct changelog_setinfo *cs = val;
1781
1782                 if (vallen != sizeof(*cs)) {
1783                         CERROR("%s: bad changelog_clear setinfo size %d\n",
1784                                tgt_name(tsi->tsi_tgt), vallen);
1785                         RETURN(-EINVAL);
1786                 }
1787                 if (ptlrpc_req_need_swab(req)) {
1788                         __swab64s(&cs->cs_recno);
1789                         __swab32s(&cs->cs_id);
1790                 }
1791
1792                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
1793                                    vallen, val, NULL);
1794         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
1795                 if (vallen > 0)
1796                         obd_export_evict_by_nid(req->rq_export->exp_obd, val);
1797         } else {
1798                 RETURN(-EINVAL);
1799         }
1800         RETURN(rc);
1801 }
1802
1803 static int mdt_readpage(struct tgt_session_info *tsi)
1804 {
1805         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
1806         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
1807         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
1808         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
1809         struct mdt_body         *repbody;
1810         int                      rc;
1811         int                      i;
1812
1813         ENTRY;
1814
1815         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1816                 RETURN(err_serious(-ENOMEM));
1817
1818         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
1819         if (repbody == NULL || reqbody == NULL)
1820                 RETURN(err_serious(-EFAULT));
1821
1822         /*
1823          * prepare @rdpg before calling lower layers and transfer itself. Here
1824          * reqbody->size contains offset of where to start to read and
1825          * reqbody->nlink contains number bytes to read.
1826          */
1827         rdpg->rp_hash = reqbody->mbo_size;
1828         if (rdpg->rp_hash != reqbody->mbo_size) {
1829                 CERROR("Invalid hash: %#llx != %#llx\n",
1830                        rdpg->rp_hash, reqbody->mbo_size);
1831                 RETURN(-EFAULT);
1832         }
1833
1834         rdpg->rp_attrs = reqbody->mbo_mode;
1835         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
1836                 rdpg->rp_attrs |= LUDA_64BITHASH;
1837         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
1838                                 exp_max_brw_size(tsi->tsi_exp));
1839         rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >>
1840                           PAGE_SHIFT;
1841         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1842         if (rdpg->rp_pages == NULL)
1843                 RETURN(-ENOMEM);
1844
1845         for (i = 0; i < rdpg->rp_npages; ++i) {
1846                 rdpg->rp_pages[i] = alloc_page(GFP_NOFS);
1847                 if (rdpg->rp_pages[i] == NULL)
1848                         GOTO(free_rdpg, rc = -ENOMEM);
1849         }
1850
1851         /* call lower layers to fill allocated pages with directory data */
1852         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
1853         if (rc < 0)
1854                 GOTO(free_rdpg, rc);
1855
1856         /* send pages to client */
1857         rc = tgt_sendpage(tsi, rdpg, rc);
1858
1859         EXIT;
1860 free_rdpg:
1861
1862         for (i = 0; i < rdpg->rp_npages; i++)
1863                 if (rdpg->rp_pages[i] != NULL)
1864                         __free_page(rdpg->rp_pages[i]);
1865         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1866
1867         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1868                 RETURN(0);
1869
1870         return rc;
1871 }
1872
1873 static int mdt_fix_attr_ucred(struct mdt_thread_info *info, __u32 op)
1874 {
1875         struct lu_ucred *uc = mdt_ucred_check(info);
1876         struct lu_attr *attr = &info->mti_attr.ma_attr;
1877
1878         if (uc == NULL)
1879                 return -EINVAL;
1880
1881         if (op != REINT_SETATTR) {
1882                 if ((attr->la_valid & LA_UID) && (attr->la_uid != -1))
1883                         attr->la_uid = uc->uc_fsuid;
1884                 /* for S_ISGID, inherit gid from his parent, such work will be
1885                  * done in cmm/mdd layer, here set all cases as uc->uc_fsgid. */
1886                 if ((attr->la_valid & LA_GID) && (attr->la_gid != -1))
1887                         attr->la_gid = uc->uc_fsgid;
1888         }
1889
1890         return 0;
1891 }
1892
1893 static int mdt_reint_internal(struct mdt_thread_info *info,
1894                               struct mdt_lock_handle *lhc,
1895                               __u32 op)
1896 {
1897         struct req_capsule      *pill = info->mti_pill;
1898         struct mdt_body         *repbody;
1899         int                      rc = 0, rc2;
1900
1901         ENTRY;
1902
1903         rc = mdt_reint_unpack(info, op);
1904         if (rc != 0) {
1905                 CERROR("Can't unpack reint, rc %d\n", rc);
1906                 RETURN(err_serious(rc));
1907         }
1908
1909         /* for replay (no_create) lmm is not needed, client has it already */
1910         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1911                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1912                                      DEF_REP_MD_SIZE);
1913
1914         /* llog cookies are always 0, the field is kept for compatibility */
1915         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1916                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
1917
1918         rc = req_capsule_server_pack(pill);
1919         if (rc != 0) {
1920                 CERROR("Can't pack response, rc %d\n", rc);
1921                 RETURN(err_serious(rc));
1922         }
1923
1924         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1925                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1926                 LASSERT(repbody);
1927                 repbody->mbo_eadatasize = 0;
1928                 repbody->mbo_aclsize = 0;
1929         }
1930
1931         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1932
1933         /* for replay no cookkie / lmm need, because client have this already */
1934         if (info->mti_spec.no_create)
1935                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1936                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1937
1938         rc = mdt_init_ucred_reint(info);
1939         if (rc)
1940                 GOTO(out_shrink, rc);
1941
1942         rc = mdt_fix_attr_ucred(info, op);
1943         if (rc != 0)
1944                 GOTO(out_ucred, rc = err_serious(rc));
1945
1946         rc = mdt_check_resent(info, mdt_reconstruct, lhc);
1947         if (rc < 0) {
1948                 GOTO(out_ucred, rc);
1949         } else if (rc == 1) {
1950                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
1951                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1952                 GOTO(out_ucred, rc);
1953         }
1954         rc = mdt_reint_rec(info, lhc);
1955         EXIT;
1956 out_ucred:
1957         mdt_exit_ucred(info);
1958 out_shrink:
1959         mdt_client_compatibility(info);
1960         rc2 = mdt_fix_reply(info);
1961         if (rc == 0)
1962                 rc = rc2;
1963         return rc;
1964 }
1965
1966 static long mdt_reint_opcode(struct ptlrpc_request *req,
1967                              const struct req_format **fmt)
1968 {
1969         struct mdt_device       *mdt;
1970         struct mdt_rec_reint    *rec;
1971         long                     opc;
1972
1973         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
1974         if (rec != NULL) {
1975                 opc = rec->rr_opcode;
1976                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
1977                 if (opc < REINT_MAX && fmt[opc] != NULL)
1978                         req_capsule_extend(&req->rq_pill, fmt[opc]);
1979                 else {
1980                         mdt = mdt_exp2dev(req->rq_export);
1981                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
1982                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
1983                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
1984                         opc = err_serious(-EFAULT);
1985                 }
1986         } else {
1987                 opc = err_serious(-EFAULT);
1988         }
1989         return opc;
1990 }
1991
1992 static int mdt_reint(struct tgt_session_info *tsi)
1993 {
1994         long opc;
1995         int  rc;
1996         static const struct req_format *reint_fmts[REINT_MAX] = {
1997                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
1998                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
1999                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
2000                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
2001                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
2002                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
2003                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
2004                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
2005                 [REINT_MIGRATE]  = &RQF_MDS_REINT_RENAME
2006         };
2007
2008         ENTRY;
2009
2010         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
2011         if (opc >= 0) {
2012                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2013                 /*
2014                  * No lock possible here from client to pass it to reint code
2015                  * path.
2016                  */
2017                 rc = mdt_reint_internal(info, NULL, opc);
2018                 mdt_thread_info_fini(info);
2019         } else {
2020                 rc = opc;
2021         }
2022
2023         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
2024         RETURN(rc);
2025 }
2026
2027 /* this should sync the whole device */
2028 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
2029 {
2030         struct dt_device *dt = mdt->mdt_bottom;
2031         int rc;
2032         ENTRY;
2033
2034         rc = dt->dd_ops->dt_sync(env, dt);
2035         RETURN(rc);
2036 }
2037
2038 /* this should sync this object */
2039 static int mdt_object_sync(struct mdt_thread_info *info)
2040 {
2041         struct md_object *next;
2042         int rc;
2043         ENTRY;
2044
2045         if (!mdt_object_exists(info->mti_object)) {
2046                 CWARN("%s: non existing object "DFID": rc = %d\n",
2047                       mdt_obd_name(info->mti_mdt),
2048                       PFID(mdt_object_fid(info->mti_object)), -ESTALE);
2049                 RETURN(-ESTALE);
2050         }
2051         next = mdt_object_child(info->mti_object);
2052         rc = mo_object_sync(info->mti_env, next);
2053
2054         RETURN(rc);
2055 }
2056
2057 static int mdt_sync(struct tgt_session_info *tsi)
2058 {
2059         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2060         struct req_capsule      *pill = tsi->tsi_pill;
2061         struct mdt_body         *body;
2062         int                      rc;
2063
2064         ENTRY;
2065
2066         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
2067                 RETURN(err_serious(-ENOMEM));
2068
2069         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
2070                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
2071         } else {
2072                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2073
2074                 /* sync an object */
2075                 rc = mdt_object_sync(info);
2076                 if (rc == 0) {
2077                         const struct lu_fid *fid;
2078                         struct lu_attr *la = &info->mti_attr.ma_attr;
2079
2080                         info->mti_attr.ma_need = MA_INODE;
2081                         info->mti_attr.ma_valid = 0;
2082                         rc = mdt_attr_get_complex(info, info->mti_object,
2083                                                   &info->mti_attr);
2084                         if (rc == 0) {
2085                                 body = req_capsule_server_get(pill,
2086                                                               &RMF_MDT_BODY);
2087                                 fid = mdt_object_fid(info->mti_object);
2088                                 mdt_pack_attr2body(info, body, la, fid);
2089                         }
2090                 }
2091                 mdt_thread_info_fini(info);
2092         }
2093         if (rc == 0)
2094                 mdt_counter_incr(req, LPROC_MDT_SYNC);
2095
2096         RETURN(rc);
2097 }
2098
2099 /*
2100  * Handle quota control requests to consult current usage/limit, but also
2101  * to configure quota enforcement
2102  */
2103 static int mdt_quotactl(struct tgt_session_info *tsi)
2104 {
2105         struct obd_export       *exp  = tsi->tsi_exp;
2106         struct req_capsule      *pill = tsi->tsi_pill;
2107         struct obd_quotactl     *oqctl, *repoqc;
2108         int                      id, rc;
2109         struct mdt_device       *mdt = mdt_exp2dev(exp);
2110         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2111         struct lu_nodemap       *nodemap;
2112         ENTRY;
2113
2114         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
2115         if (oqctl == NULL)
2116                 RETURN(err_serious(-EPROTO));
2117
2118         rc = req_capsule_server_pack(pill);
2119         if (rc)
2120                 RETURN(err_serious(rc));
2121
2122         nodemap = nodemap_get_from_exp(exp);
2123         if (IS_ERR(nodemap))
2124                 RETURN(PTR_ERR(nodemap));
2125
2126         switch (oqctl->qc_cmd) {
2127                 /* master quotactl */
2128         case Q_SETINFO:
2129         case Q_SETQUOTA:
2130                 if (!nodemap_can_setquota(nodemap))
2131                         GOTO(out_nodemap, rc = -EPERM);
2132         case Q_GETINFO:
2133         case Q_GETQUOTA:
2134                 if (qmt == NULL)
2135                         GOTO(out_nodemap, rc = -EOPNOTSUPP);
2136                 /* slave quotactl */
2137         case Q_GETOINFO:
2138         case Q_GETOQUOTA:
2139                 break;
2140         default:
2141                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2142                 GOTO(out_nodemap, rc = -EFAULT);
2143         }
2144
2145         id = oqctl->qc_id;
2146         switch (oqctl->qc_type) {
2147         case USRQUOTA:
2148                 id = nodemap_map_id(nodemap, NODEMAP_UID,
2149                                     NODEMAP_CLIENT_TO_FS, id);
2150                 break;
2151         case GRPQUOTA:
2152                 id = nodemap_map_id(nodemap, NODEMAP_UID,
2153                                     NODEMAP_CLIENT_TO_FS, id);
2154                 break;
2155         case PRJQUOTA:
2156                 /* todo: check/map project id */
2157                 id = oqctl->qc_id;
2158                 break;
2159         default:
2160                 GOTO(out_nodemap, rc = -EOPNOTSUPP);
2161         }
2162         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2163         if (repoqc == NULL)
2164                 GOTO(out_nodemap, rc = err_serious(-EFAULT));
2165
2166         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA)
2167                 barrier_exit(tsi->tsi_tgt->lut_bottom);
2168
2169         if (oqctl->qc_id != id)
2170                 swap(oqctl->qc_id, id);
2171
2172         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA) {
2173                 if (unlikely(!barrier_entry(tsi->tsi_tgt->lut_bottom)))
2174                         RETURN(-EINPROGRESS);
2175         }
2176
2177         switch (oqctl->qc_cmd) {
2178
2179         case Q_GETINFO:
2180         case Q_SETINFO:
2181         case Q_SETQUOTA:
2182         case Q_GETQUOTA:
2183                 /* forward quotactl request to QMT */
2184                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2185                 break;
2186
2187         case Q_GETOINFO:
2188         case Q_GETOQUOTA:
2189                 /* slave quotactl */
2190                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2191                                    oqctl);
2192                 break;
2193
2194         default:
2195                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2196                 GOTO(out_nodemap, rc = -EFAULT);
2197         }
2198
2199         if (oqctl->qc_id != id)
2200                 swap(oqctl->qc_id, id);
2201
2202         *repoqc = *oqctl;
2203
2204         EXIT;
2205
2206 out_nodemap:
2207         nodemap_putref(nodemap);
2208
2209         return rc;
2210 }
2211
2212 /** clone llog ctxt from child (mdd)
2213  * This allows remote llog (replicator) access.
2214  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2215  * context was originally set up, or we can handle them directly.
2216  * I choose the latter, but that means I need any llog
2217  * contexts set up by child to be accessable by the mdt.  So we clone the
2218  * context into our context list here.
2219  */
2220 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2221                                int idx)
2222 {
2223         struct md_device  *next = mdt->mdt_child;
2224         struct llog_ctxt *ctxt;
2225         int rc;
2226
2227         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2228                 return 0;
2229
2230         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2231         if (rc || ctxt == NULL) {
2232                 return 0;
2233         }
2234
2235         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2236         if (rc)
2237                 CERROR("Can't set mdt ctxt %d\n", rc);
2238
2239         return rc;
2240 }
2241
2242 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2243                                  struct mdt_device *mdt, int idx)
2244 {
2245         struct llog_ctxt *ctxt;
2246
2247         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2248         if (ctxt == NULL)
2249                 return 0;
2250         /* Put once for the get we just did, and once for the clone */
2251         llog_ctxt_put(ctxt);
2252         llog_ctxt_put(ctxt);
2253         return 0;
2254 }
2255
2256 /*
2257  * sec context handlers
2258  */
2259 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2260 {
2261         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2262
2263         return 0;
2264 }
2265
2266 /*
2267  * quota request handlers
2268  */
2269 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2270 {
2271         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
2272         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2273         int                      rc;
2274         ENTRY;
2275
2276         if (qmt == NULL)
2277                 RETURN(err_serious(-EOPNOTSUPP));
2278
2279         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2280         RETURN(rc);
2281 }
2282
2283 struct mdt_object *mdt_object_new(const struct lu_env *env,
2284                                   struct mdt_device *d,
2285                                   const struct lu_fid *f)
2286 {
2287         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2288         struct lu_object *o;
2289         struct mdt_object *m;
2290         ENTRY;
2291
2292         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2293         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2294         if (unlikely(IS_ERR(o)))
2295                 m = (struct mdt_object *)o;
2296         else
2297                 m = mdt_obj(o);
2298         RETURN(m);
2299 }
2300
2301 struct mdt_object *mdt_object_find(const struct lu_env *env,
2302                                    struct mdt_device *d,
2303                                    const struct lu_fid *f)
2304 {
2305         struct lu_object *o;
2306         struct mdt_object *m;
2307         ENTRY;
2308
2309         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2310         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2311         if (unlikely(IS_ERR(o)))
2312                 m = (struct mdt_object *)o;
2313         else
2314                 m = mdt_obj(o);
2315
2316         RETURN(m);
2317 }
2318
2319 /**
2320  * Asyncronous commit for mdt device.
2321  *
2322  * Pass asynchonous commit call down the MDS stack.
2323  *
2324  * \param env environment
2325  * \param mdt the mdt device
2326  */
2327 static void mdt_device_commit_async(const struct lu_env *env,
2328                                     struct mdt_device *mdt)
2329 {
2330         struct dt_device *dt = mdt->mdt_bottom;
2331         int rc;
2332         ENTRY;
2333
2334         rc = dt->dd_ops->dt_commit_async(env, dt);
2335         if (unlikely(rc != 0))
2336                 CWARN("%s: async commit start failed: rc = %d\n",
2337                       mdt_obd_name(mdt), rc);
2338         atomic_inc(&mdt->mdt_async_commit_count);
2339         EXIT;
2340 }
2341
2342 /**
2343  * Mark the lock as "synchonous".
2344  *
2345  * Mark the lock to deffer transaction commit to the unlock time.
2346  *
2347  * \param lock the lock to mark as "synchonous"
2348  *
2349  * \see mdt_is_lock_sync
2350  * \see mdt_save_lock
2351  */
2352 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2353 {
2354         lock->l_ast_data = (void*)1;
2355 }
2356
2357 /**
2358  * Check whehter the lock "synchonous" or not.
2359  *
2360  * \param lock the lock to check
2361  * \retval 1 the lock is "synchonous"
2362  * \retval 0 the lock isn't "synchronous"
2363  *
2364  * \see mdt_set_lock_sync
2365  * \see mdt_save_lock
2366  */
2367 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2368 {
2369         return lock->l_ast_data != NULL;
2370 }
2371
2372 /**
2373  * Blocking AST for mdt locks.
2374  *
2375  * Starts transaction commit if in case of COS lock conflict or
2376  * deffers such a commit to the mdt_save_lock.
2377  *
2378  * \param lock the lock which blocks a request or cancelling lock
2379  * \param desc unused
2380  * \param data unused
2381  * \param flag indicates whether this cancelling or blocking callback
2382  * \retval 0
2383  * \see ldlm_blocking_ast_nocheck
2384  */
2385 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2386                      void *data, int flag)
2387 {
2388         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2389         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2390         int rc;
2391         ENTRY;
2392
2393         if (flag == LDLM_CB_CANCELING)
2394                 RETURN(0);
2395
2396         lock_res_and_lock(lock);
2397         if (lock->l_blocking_ast != mdt_blocking_ast) {
2398                 unlock_res_and_lock(lock);
2399                 RETURN(0);
2400         }
2401         if (lock->l_req_mode & (LCK_PW | LCK_EX) &&
2402             lock->l_blocking_lock != NULL) {
2403                 if (mdt_cos_is_enabled(mdt) &&
2404                     lock->l_client_cookie !=
2405                     lock->l_blocking_lock->l_client_cookie)
2406                         mdt_set_lock_sync(lock);
2407                 else if (mdt_slc_is_enabled(mdt) &&
2408                          ldlm_is_cos_incompat(lock->l_blocking_lock))
2409                         mdt_set_lock_sync(lock);
2410         }
2411         rc = ldlm_blocking_ast_nocheck(lock);
2412
2413         /* There is no lock conflict if l_blocking_lock == NULL,
2414          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2415          * when the last reference to a local lock was released */
2416         if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2417                 struct lu_env env;
2418
2419                 rc = lu_env_init(&env, LCT_LOCAL);
2420                 if (unlikely(rc != 0))
2421                         CWARN("%s: lu_env initialization failed, cannot "
2422                               "start asynchronous commit: rc = %d\n",
2423                               obd->obd_name, rc);
2424                 else
2425                         mdt_device_commit_async(&env, mdt);
2426                 lu_env_fini(&env);
2427         }
2428         RETURN(rc);
2429 }
2430
2431 /*
2432  * Blocking AST for cross-MDT lock
2433  *
2434  * Discard lock from uncommitted_slc_locks and cancel it.
2435  *
2436  * \param lock  the lock which blocks a request or cancelling lock
2437  * \param desc  unused
2438  * \param data  unused
2439  * \param flag  indicates whether this cancelling or blocking callback
2440  * \retval      0 on success
2441  * \retval      negative number on error
2442  */
2443 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2444                             void *data, int flag)
2445 {
2446         int rc = 0;
2447         ENTRY;
2448
2449         switch (flag) {
2450         case LDLM_CB_BLOCKING: {
2451                 struct lustre_handle lockh;
2452
2453                 ldlm_lock2handle(lock, &lockh);
2454                 rc = ldlm_cli_cancel(&lockh,
2455                         ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
2456                 if (rc < 0) {
2457                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2458                         RETURN(rc);
2459                 }
2460                 break;
2461         }
2462         case LDLM_CB_CANCELING: {
2463                 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2464                 struct mdt_device *mdt =
2465                                 mdt_dev(obd->obd_lu_dev->ld_site->ls_top_dev);
2466
2467                 LDLM_DEBUG(lock, "Revoke remote lock\n");
2468
2469                 /* discard slc lock here so that it can be cleaned anytime,
2470                  * especially for cleanup_resource() */
2471                 tgt_discard_slc_lock(&mdt->mdt_lut, lock);
2472
2473                 /* once we cache lock, l_ast_data is set to mdt_object */
2474                 if (lock->l_ast_data != NULL) {
2475                         struct mdt_object *mo = lock->l_ast_data;
2476                         struct lu_env env;
2477
2478                         rc = lu_env_init(&env, LCT_MD_THREAD);
2479                         if (unlikely(rc != 0)) {
2480                                 CWARN("%s: lu_env initialization failed, object"
2481                                       "%p "DFID" is leaked!\n",
2482                                       obd->obd_name, mo,
2483                                       PFID(mdt_object_fid(mo)));
2484                                 RETURN(rc);
2485                         }
2486
2487                         if (lock->l_policy_data.l_inodebits.bits &
2488                             (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)) {
2489                                 rc = mo_invalidate(&env, mdt_object_child(mo));
2490                                 mo->mot_cache_attr = 0;
2491                         }
2492                         mdt_object_put(&env, mo);
2493                         lu_env_fini(&env);
2494                 }
2495                 break;
2496         }
2497         default:
2498                 LBUG();
2499         }
2500
2501         RETURN(rc);
2502 }
2503
2504 int mdt_check_resent_lock(struct mdt_thread_info *info,
2505                           struct mdt_object *mo,
2506                           struct mdt_lock_handle *lhc)
2507 {
2508         /* the lock might already be gotten in ldlm_handle_enqueue() */
2509         if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
2510                 struct ptlrpc_request *req = mdt_info_req(info);
2511                 struct ldlm_lock      *lock;
2512
2513                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2514                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
2515                 if (lock == NULL) {
2516                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
2517                          * a resend case, however, it could be already destroyed
2518                          * due to client eviction or a raced cancel RPC. */
2519                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx",
2520                                           lhc->mlh_reg_lh.cookie);
2521                         RETURN(-ESTALE);
2522                 }
2523
2524                 if (!fid_res_name_eq(mdt_object_fid(mo),
2525                                      &lock->l_resource->lr_name)) {
2526                         CWARN("%s: Although resent, but still not "
2527                               "get child lock:"DFID"\n",
2528                               info->mti_exp->exp_obd->obd_name,
2529                               PFID(mdt_object_fid(mo)));
2530                         LDLM_LOCK_PUT(lock);
2531                         RETURN(-EPROTO);
2532                 }
2533                 LDLM_LOCK_PUT(lock);
2534                 return 0;
2535         }
2536         return 1;
2537 }
2538
2539 int mdt_remote_object_lock(struct mdt_thread_info *mti, struct mdt_object *o,
2540                            const struct lu_fid *fid, struct lustre_handle *lh,
2541                            enum ldlm_mode mode, __u64 ibits, bool nonblock,
2542                            bool cache)
2543 {
2544         struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
2545         union ldlm_policy_data *policy = &mti->mti_policy;
2546         struct ldlm_res_id *res_id = &mti->mti_res_id;
2547         int rc = 0;
2548         ENTRY;
2549
2550         LASSERT(mdt_object_remote(o));
2551
2552         fid_build_reg_res_name(fid, res_id);
2553
2554         memset(einfo, 0, sizeof(*einfo));
2555         einfo->ei_type = LDLM_IBITS;
2556         einfo->ei_mode = mode;
2557         einfo->ei_cb_bl = mdt_remote_blocking_ast;
2558         einfo->ei_cb_cp = ldlm_completion_ast;
2559         einfo->ei_enq_slave = 0;
2560         einfo->ei_res_id = res_id;
2561         if (nonblock)
2562                 einfo->ei_nonblock = 1;
2563         if (cache) {
2564                 /*
2565                  * if we cache lock, couple lock with mdt_object, so that object
2566                  * can be easily found in lock ASTs.
2567                  */
2568                 mdt_object_get(mti->mti_env, o);
2569                 einfo->ei_cbdata = o;
2570         }
2571
2572         memset(policy, 0, sizeof(*policy));
2573         policy->l_inodebits.bits = ibits;
2574
2575         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2576                             policy);
2577         if (rc < 0 && cache) {
2578                 mdt_object_put(mti->mti_env, o);
2579                 einfo->ei_cbdata = NULL;
2580         }
2581         RETURN(rc);
2582 }
2583
2584 static int mdt_object_local_lock(struct mdt_thread_info *info,
2585                                  struct mdt_object *o,
2586                                  struct mdt_lock_handle *lh, __u64 ibits,
2587                                  bool nonblock, bool cos_incompat)
2588 {
2589         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2590         union ldlm_policy_data *policy = &info->mti_policy;
2591         struct ldlm_res_id *res_id = &info->mti_res_id;
2592         __u64 dlmflags = 0;
2593         int rc;
2594         ENTRY;
2595
2596         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2597         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2598         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2599         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2600
2601         if (cos_incompat) {
2602                 LASSERT(lh->mlh_reg_mode == LCK_PW ||
2603                         lh->mlh_reg_mode == LCK_EX);
2604                 dlmflags |= LDLM_FL_COS_INCOMPAT;
2605         } else if (mdt_cos_is_enabled(info->mti_mdt)) {
2606                 dlmflags |= LDLM_FL_COS_ENABLED;
2607         }
2608
2609         /* Only enqueue LOOKUP lock for remote object */
2610         if (mdt_object_remote(o))
2611                 LASSERT(ibits == MDS_INODELOCK_LOOKUP);
2612
2613         if (lh->mlh_type == MDT_PDO_LOCK) {
2614                 /* check for exists after object is locked */
2615                 if (mdt_object_exists(o) == 0) {
2616                         /* Non-existent object shouldn't have PDO lock */
2617                         RETURN(-ESTALE);
2618                 } else {
2619                         /* Non-dir object shouldn't have PDO lock */
2620                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2621                                 RETURN(-ENOTDIR);
2622                 }
2623         }
2624
2625         memset(policy, 0, sizeof(*policy));
2626         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2627
2628         dlmflags |= LDLM_FL_ATOMIC_CB;
2629         if (nonblock)
2630                 dlmflags |= LDLM_FL_BLOCK_NOWAIT;
2631
2632         /*
2633          * Take PDO lock on whole directory and build correct @res_id for lock
2634          * on part of directory.
2635          */
2636         if (lh->mlh_pdo_hash != 0) {
2637                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2638                 mdt_lock_pdo_mode(info, o, lh);
2639                 if (lh->mlh_pdo_mode != LCK_NL) {
2640                         /*
2641                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2642                          * is never going to be sent to client and we do not
2643                          * want it slowed down due to possible cancels.
2644                          */
2645                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2646                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2647                                           policy, res_id, dlmflags,
2648                                           info->mti_exp == NULL ? NULL :
2649                                           &info->mti_exp->exp_handle.h_cookie);
2650                         if (unlikely(rc != 0))
2651                                 GOTO(out_unlock, rc);
2652                 }
2653
2654                 /*
2655                  * Finish res_id initializing by name hash marking part of
2656                  * directory which is taking modification.
2657                  */
2658                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2659         }
2660
2661         policy->l_inodebits.bits = ibits;
2662
2663         /*
2664          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2665          * going to be sent to client. If it is - mdt_intent_policy() path will
2666          * fix it up and turn FL_LOCAL flag off.
2667          */
2668         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2669                           res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2670                           info->mti_exp == NULL ? NULL :
2671                           &info->mti_exp->exp_handle.h_cookie);
2672 out_unlock:
2673         if (rc != 0)
2674                 mdt_object_unlock(info, o, lh, 1);
2675         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2676                    lh->mlh_pdo_hash != 0 &&
2677                    (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
2678                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2679
2680         RETURN(rc);
2681 }
2682
2683 static int
2684 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
2685                          struct mdt_lock_handle *lh, __u64 ibits, bool nonblock,
2686                          bool cos_incompat)
2687 {
2688         struct mdt_lock_handle *local_lh = NULL;
2689         int rc;
2690         ENTRY;
2691
2692         if (!mdt_object_remote(o)) {
2693                 rc = mdt_object_local_lock(info, o, lh, ibits, nonblock,
2694                                            cos_incompat);
2695                 RETURN(rc);
2696         }
2697
2698         /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
2699         ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
2700                    MDS_INODELOCK_XATTR);
2701
2702         /* Only enqueue LOOKUP lock for remote object */
2703         if (ibits & MDS_INODELOCK_LOOKUP) {
2704                 rc = mdt_object_local_lock(info, o, lh, MDS_INODELOCK_LOOKUP,
2705                                            nonblock, cos_incompat);
2706                 if (rc != ELDLM_OK)
2707                         RETURN(rc);
2708
2709                 local_lh = lh;
2710         }
2711
2712         if (ibits & MDS_INODELOCK_UPDATE) {
2713                 /* Sigh, PDO needs to enqueue 2 locks right now, but
2714                  * enqueue RPC can only request 1 lock, to avoid extra
2715                  * RPC, so it will instead enqueue EX lock for remote
2716                  * object anyway XXX*/
2717                 if (lh->mlh_type == MDT_PDO_LOCK &&
2718                     lh->mlh_pdo_hash != 0) {
2719                         CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
2720                                "EX lock.\n", mdt_obd_name(info->mti_mdt),
2721                                PFID(mdt_object_fid(o)));
2722                         lh->mlh_pdo_hash = 0;
2723                         lh->mlh_rreg_mode = LCK_EX;
2724                         lh->mlh_type = MDT_REG_LOCK;
2725                 }
2726                 rc = mdt_remote_object_lock(info, o, mdt_object_fid(o),
2727                                             &lh->mlh_rreg_lh,
2728                                             lh->mlh_rreg_mode,
2729                                             MDS_INODELOCK_UPDATE, nonblock,
2730                                             false);
2731                 if (rc != ELDLM_OK) {
2732                         if (local_lh != NULL)
2733                                 mdt_object_unlock(info, o, local_lh, rc);
2734                         RETURN(rc);
2735                 }
2736         }
2737
2738         RETURN(0);
2739 }
2740
2741 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2742                     struct mdt_lock_handle *lh, __u64 ibits)
2743 {
2744         return mdt_object_lock_internal(info, o, lh, ibits, false, false);
2745 }
2746
2747 int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2748                           struct mdt_lock_handle *lh, __u64 ibits,
2749                           bool cos_incompat)
2750 {
2751         LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
2752         return mdt_object_lock_internal(info, o, lh, ibits, false,
2753                                         cos_incompat);
2754 }
2755
2756 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
2757                         struct mdt_lock_handle *lh, __u64 ibits)
2758 {
2759         struct mdt_lock_handle tmp = *lh;
2760         int rc;
2761
2762         rc = mdt_object_lock_internal(info, o, &tmp, ibits, true, false);
2763         if (rc == 0)
2764                 *lh = tmp;
2765
2766         return rc == 0;
2767 }
2768
2769 int mdt_reint_object_lock_try(struct mdt_thread_info *info,
2770                               struct mdt_object *o, struct mdt_lock_handle *lh,
2771                               __u64 ibits, bool cos_incompat)
2772 {
2773         struct mdt_lock_handle tmp = *lh;
2774         int rc;
2775
2776         LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
2777         rc = mdt_object_lock_internal(info, o, &tmp, ibits, true, cos_incompat);
2778         if (rc == 0)
2779                 *lh = tmp;
2780
2781         return rc == 0;
2782 }
2783
2784 /**
2785  * Save a lock within request object.
2786  *
2787  * Keep the lock referenced until whether client ACK or transaction
2788  * commit happens or release the lock immediately depending on input
2789  * parameters. If COS is ON, a write lock is converted to COS lock
2790  * before saving.
2791  *
2792  * \param info thead info object
2793  * \param h lock handle
2794  * \param mode lock mode
2795  * \param decref force immediate lock releasing
2796  */
2797 static void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2798                           enum ldlm_mode mode, int decref)
2799 {
2800         ENTRY;
2801
2802         if (lustre_handle_is_used(h)) {
2803                 if (decref || !info->mti_has_trans ||
2804                     !(mode & (LCK_PW | LCK_EX))) {
2805                         mdt_fid_unlock(h, mode);
2806                 } else {
2807                         struct mdt_device *mdt = info->mti_mdt;
2808                         struct ldlm_lock *lock = ldlm_handle2lock(h);
2809                         struct ptlrpc_request *req = mdt_info_req(info);
2810                         int cos;
2811
2812                         cos = (mdt_cos_is_enabled(mdt) ||
2813                                mdt_slc_is_enabled(mdt));
2814
2815                         LASSERTF(lock != NULL, "no lock for cookie %#llx\n",
2816                                  h->cookie);
2817
2818                         /* there is no request if mdt_object_unlock() is called
2819                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
2820                         if (likely(req != NULL)) {
2821                                 CDEBUG(D_HA, "request = %p reply state = %p"
2822                                        " transno = %lld\n", req,
2823                                        req->rq_reply_state, req->rq_transno);
2824                                 if (cos) {
2825                                         ldlm_lock_downgrade(lock, LCK_COS);
2826                                         mode = LCK_COS;
2827                                 }
2828                                 ptlrpc_save_lock(req, h, mode, cos);
2829                         } else {
2830                                 mdt_fid_unlock(h, mode);
2831                         }
2832                         if (mdt_is_lock_sync(lock)) {
2833                                 CDEBUG(D_HA, "found sync-lock,"
2834                                        " async commit started\n");
2835                                 mdt_device_commit_async(info->mti_env,
2836                                                         mdt);
2837                         }
2838                         LDLM_LOCK_PUT(lock);
2839                 }
2840                 h->cookie = 0ull;
2841         }
2842
2843         EXIT;
2844 }
2845
2846 /**
2847  * Save cross-MDT lock in uncommitted_slc_locks
2848  *
2849  * Keep the lock referenced until transaction commit happens or release the lock
2850  * immediately depending on input parameters.
2851  *
2852  * \param info thead info object
2853  * \param h lock handle
2854  * \param mode lock mode
2855  * \param decref force immediate lock releasing
2856  */
2857 static void mdt_save_remote_lock(struct mdt_thread_info *info,
2858                                  struct mdt_object *o, struct lustre_handle *h,
2859                                  enum ldlm_mode mode, int decref)
2860 {
2861         ENTRY;
2862
2863         if (lustre_handle_is_used(h)) {
2864                 struct ldlm_lock *lock = ldlm_handle2lock(h);
2865
2866                 if (o != NULL &&
2867                     (lock->l_policy_data.l_inodebits.bits &
2868                      (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)))
2869                         mo_invalidate(info->mti_env, mdt_object_child(o));
2870
2871                 if (decref || !info->mti_has_trans ||
2872                     !(mode & (LCK_PW | LCK_EX))) {
2873                         ldlm_lock_decref_and_cancel(h, mode);
2874                         LDLM_LOCK_PUT(lock);
2875                 } else {
2876                         struct ptlrpc_request *req = mdt_info_req(info);
2877
2878                         LASSERT(req != NULL);
2879                         tgt_save_slc_lock(&info->mti_mdt->mdt_lut, lock,
2880                                           req->rq_transno);
2881                         ldlm_lock_decref(h, mode);
2882                 }
2883                 h->cookie = 0ull;
2884         }
2885
2886         EXIT;
2887 }
2888
2889 /**
2890  * Unlock mdt object.
2891  *
2892  * Immeditely release the regular lock and the PDO lock or save the
2893  * lock in request and keep them referenced until client ACK or
2894  * transaction commit.
2895  *
2896  * \param info thread info object
2897  * \param o mdt object
2898  * \param lh mdt lock handle referencing regular and PDO locks
2899  * \param decref force immediate lock releasing
2900  *
2901  * XXX o is not used and may be NULL, see hsm_cdt_request_completed().
2902  */
2903 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2904                        struct mdt_lock_handle *lh, int decref)
2905 {
2906         ENTRY;
2907
2908         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2909         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2910         mdt_save_remote_lock(info, o, &lh->mlh_rreg_lh, lh->mlh_rreg_mode,
2911                              decref);
2912
2913         EXIT;
2914 }
2915
2916 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2917                                         const struct lu_fid *f,
2918                                         struct mdt_lock_handle *lh,
2919                                         __u64 ibits)
2920 {
2921         struct mdt_object *o;
2922
2923         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2924         if (!IS_ERR(o)) {
2925                 int rc;
2926
2927                 rc = mdt_object_lock(info, o, lh, ibits);
2928                 if (rc != 0) {
2929                         mdt_object_put(info->mti_env, o);
2930                         o = ERR_PTR(rc);
2931                 }
2932         }
2933         return o;
2934 }
2935
2936 void mdt_object_unlock_put(struct mdt_thread_info * info,
2937                            struct mdt_object * o,
2938                            struct mdt_lock_handle *lh,
2939                            int decref)
2940 {
2941         mdt_object_unlock(info, o, lh, decref);
2942         mdt_object_put(info->mti_env, o);
2943 }
2944
2945 /*
2946  * Generic code handling requests that have struct mdt_body passed in:
2947  *
2948  *  - extract mdt_body from request and save it in @info, if present;
2949  *
2950  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
2951  *  @info;
2952  *
2953  *  - if HABEO_CORPUS flag is set for this request type check whether object
2954  *  actually exists on storage (lu_object_exists()).
2955  *
2956  */
2957 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2958 {
2959         const struct mdt_body    *body;
2960         struct mdt_object        *obj;
2961         const struct lu_env      *env;
2962         struct req_capsule       *pill;
2963         int                       rc;
2964         ENTRY;
2965
2966         env = info->mti_env;
2967         pill = info->mti_pill;
2968
2969         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2970         if (body == NULL)
2971                 RETURN(-EFAULT);
2972
2973         if (!(body->mbo_valid & OBD_MD_FLID))
2974                 RETURN(0);
2975
2976         if (!fid_is_sane(&body->mbo_fid1)) {
2977                 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
2978                 RETURN(-EINVAL);
2979         }
2980
2981         obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
2982         if (!IS_ERR(obj)) {
2983                 if ((flags & HABEO_CORPUS) && !mdt_object_exists(obj)) {
2984                         mdt_object_put(env, obj);
2985                         rc = -ENOENT;
2986                 } else {
2987                         info->mti_object = obj;
2988                         rc = 0;
2989                 }
2990         } else
2991                 rc = PTR_ERR(obj);
2992
2993         RETURN(rc);
2994 }
2995
2996 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2997 {
2998         struct req_capsule *pill = info->mti_pill;
2999         int rc;
3000         ENTRY;
3001
3002         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
3003                 rc = mdt_body_unpack(info, flags);
3004         else
3005                 rc = 0;
3006
3007         if (rc == 0 && (flags & HABEO_REFERO)) {
3008                 /* Pack reply. */
3009                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
3010                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
3011                                              DEF_REP_MD_SIZE);
3012                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
3013                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
3014                                              RCL_SERVER, 0);
3015
3016                 rc = req_capsule_server_pack(pill);
3017         }
3018         RETURN(rc);
3019 }
3020
3021 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
3022 {
3023         lh->mlh_type = MDT_NUL_LOCK;
3024         lh->mlh_reg_lh.cookie = 0ull;
3025         lh->mlh_reg_mode = LCK_MINMODE;
3026         lh->mlh_pdo_lh.cookie = 0ull;
3027         lh->mlh_pdo_mode = LCK_MINMODE;
3028         lh->mlh_rreg_lh.cookie = 0ull;
3029         lh->mlh_rreg_mode = LCK_MINMODE;
3030 }
3031
3032 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
3033 {
3034         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
3035         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
3036 }
3037
3038 /*
3039  * Initialize fields of struct mdt_thread_info. Other fields are left in
3040  * uninitialized state, because it's too expensive to zero out whole
3041  * mdt_thread_info (> 1K) on each request arrival.
3042  */
3043 void mdt_thread_info_init(struct ptlrpc_request *req,
3044                           struct mdt_thread_info *info)
3045 {
3046         int i;
3047
3048         info->mti_pill = &req->rq_pill;
3049
3050         /* lock handle */
3051         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3052                 mdt_lock_handle_init(&info->mti_lh[i]);
3053
3054         /* mdt device: it can be NULL while CONNECT */
3055         if (req->rq_export) {
3056                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
3057                 info->mti_exp = req->rq_export;
3058         } else
3059                 info->mti_mdt = NULL;
3060         info->mti_env = req->rq_svc_thread->t_env;
3061         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
3062
3063         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
3064         info->mti_big_buf = LU_BUF_NULL;
3065         info->mti_body = NULL;
3066         info->mti_object = NULL;
3067         info->mti_dlm_req = NULL;
3068         info->mti_has_trans = 0;
3069         info->mti_cross_ref = 0;
3070         info->mti_opdata = 0;
3071         info->mti_big_lmm_used = 0;
3072
3073         info->mti_spec.no_create = 0;
3074         info->mti_spec.sp_rm_entry = 0;
3075         info->mti_spec.sp_permitted = 0;
3076         info->mti_spec.sp_migrate_close = 0;
3077
3078         info->mti_spec.u.sp_ea.eadata = NULL;
3079         info->mti_spec.u.sp_ea.eadatalen = 0;
3080 }
3081
3082 void mdt_thread_info_fini(struct mdt_thread_info *info)
3083 {
3084         int i;
3085
3086         if (info->mti_object != NULL) {
3087                 mdt_object_put(info->mti_env, info->mti_object);
3088                 info->mti_object = NULL;
3089         }
3090
3091         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3092                 mdt_lock_handle_fini(&info->mti_lh[i]);
3093         info->mti_env = NULL;
3094         info->mti_pill = NULL;
3095         info->mti_exp = NULL;
3096
3097         if (unlikely(info->mti_big_buf.lb_buf != NULL))
3098                 lu_buf_free(&info->mti_big_buf);
3099 }
3100
3101 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
3102 {
3103         struct mdt_thread_info  *mti;
3104
3105         mti = mdt_th_info(tsi->tsi_env);
3106         LASSERT(mti != NULL);
3107
3108         mdt_thread_info_init(tgt_ses_req(tsi), mti);
3109         if (tsi->tsi_corpus != NULL) {
3110                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
3111                 lu_object_get(tsi->tsi_corpus);
3112         }
3113         mti->mti_body = tsi->tsi_mdt_body;
3114         mti->mti_dlm_req = tsi->tsi_dlm_req;
3115
3116         return mti;
3117 }
3118
3119 static int mdt_tgt_connect(struct tgt_session_info *tsi)
3120 {
3121         if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
3122             cfs_fail_val ==
3123             tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id) {
3124                 set_current_state(TASK_UNINTERRUPTIBLE);
3125                 schedule_timeout(msecs_to_jiffies(3 * MSEC_PER_SEC));
3126         }
3127
3128         return tgt_connect(tsi);
3129 }
3130
3131 enum mdt_it_code {
3132         MDT_IT_OPEN,
3133         MDT_IT_OCREAT,
3134         MDT_IT_CREATE,
3135         MDT_IT_GETATTR,
3136         MDT_IT_READDIR,
3137         MDT_IT_LOOKUP,
3138         MDT_IT_UNLINK,
3139         MDT_IT_TRUNC,
3140         MDT_IT_GETXATTR,
3141         MDT_IT_LAYOUT,
3142         MDT_IT_QUOTA,
3143         MDT_IT_NR
3144 };
3145
3146 static int mdt_intent_getattr(enum mdt_it_code opcode,
3147                               struct mdt_thread_info *info,
3148                               struct ldlm_lock **,
3149                               __u64);
3150
3151 static int mdt_intent_getxattr(enum mdt_it_code opcode,
3152                                 struct mdt_thread_info *info,
3153                                 struct ldlm_lock **lockp,
3154                                 __u64 flags);
3155
3156 static int mdt_intent_layout(enum mdt_it_code opcode,
3157                              struct mdt_thread_info *info,
3158                              struct ldlm_lock **,
3159                              __u64);
3160 static int mdt_intent_reint(enum mdt_it_code opcode,
3161                             struct mdt_thread_info *info,
3162                             struct ldlm_lock **,
3163                             __u64);
3164
3165 static struct mdt_it_flavor {
3166         const struct req_format *it_fmt;
3167         __u32                    it_flags;
3168         int                    (*it_act)(enum mdt_it_code ,
3169                                          struct mdt_thread_info *,
3170                                          struct ldlm_lock **,
3171                                          __u64);
3172         long                     it_reint;
3173 } mdt_it_flavor[] = {
3174         [MDT_IT_OPEN]     = {
3175                 .it_fmt   = &RQF_LDLM_INTENT,
3176                 /*.it_flags = HABEO_REFERO,*/
3177                 .it_flags = 0,
3178                 .it_act   = mdt_intent_reint,
3179                 .it_reint = REINT_OPEN
3180         },
3181         [MDT_IT_OCREAT]   = {
3182                 .it_fmt   = &RQF_LDLM_INTENT,
3183                 /*
3184                  * OCREAT is not a MUTABOR request as if the file
3185                  * already exists.
3186                  * We do the extra check of OBD_CONNECT_RDONLY in
3187                  * mdt_reint_open() when we really need to create
3188                  * the object.
3189                  */
3190                 .it_flags = 0,
3191                 .it_act   = mdt_intent_reint,
3192                 .it_reint = REINT_OPEN
3193         },
3194         [MDT_IT_CREATE]   = {
3195                 .it_fmt   = &RQF_LDLM_INTENT,
3196                 .it_flags = MUTABOR,
3197                 .it_act   = mdt_intent_reint,
3198                 .it_reint = REINT_CREATE
3199         },
3200         [MDT_IT_GETATTR]  = {
3201                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3202                 .it_flags = HABEO_REFERO,
3203                 .it_act   = mdt_intent_getattr
3204         },
3205         [MDT_IT_READDIR]  = {
3206                 .it_fmt   = NULL,
3207                 .it_flags = 0,
3208                 .it_act   = NULL
3209         },
3210         [MDT_IT_LOOKUP]   = {
3211                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3212                 .it_flags = HABEO_REFERO,
3213                 .it_act   = mdt_intent_getattr
3214         },
3215         [MDT_IT_UNLINK]   = {
3216                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
3217                 .it_flags = MUTABOR,
3218                 .it_act   = NULL,
3219                 .it_reint = REINT_UNLINK
3220         },
3221         [MDT_IT_TRUNC]    = {
3222                 .it_fmt   = NULL,
3223                 .it_flags = MUTABOR,
3224                 .it_act   = NULL
3225         },
3226         [MDT_IT_GETXATTR] = {
3227                 .it_fmt   = &RQF_LDLM_INTENT_GETXATTR,
3228                 .it_flags = HABEO_CORPUS,
3229                 .it_act   = mdt_intent_getxattr
3230         },
3231         [MDT_IT_LAYOUT] = {
3232                 .it_fmt   = &RQF_LDLM_INTENT_LAYOUT,
3233                 .it_flags = 0,
3234                 .it_act   = mdt_intent_layout
3235         }
3236 };
3237
3238 static int
3239 mdt_intent_lock_replace(struct mdt_thread_info *info,
3240                         struct ldlm_lock **lockp,
3241                         struct mdt_lock_handle *lh,
3242                         __u64 flags, int result)
3243 {
3244         struct ptlrpc_request  *req = mdt_info_req(info);
3245         struct ldlm_lock       *lock = *lockp;
3246         struct ldlm_lock       *new_lock;
3247
3248         /* If possible resent found a lock, @lh is set to its handle */
3249         new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3250
3251         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3252                 lh->mlh_reg_lh.cookie = 0;
3253                 RETURN(0);
3254         }
3255
3256         if (new_lock == NULL && (flags & LDLM_FL_RESENT)) {
3257                 /* Lock is pinned by ldlm_handle_enqueue0() as it is
3258                  * a resend case, however, it could be already destroyed
3259                  * due to client eviction or a raced cancel RPC. */
3260                 LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx\n",
3261                                   lh->mlh_reg_lh.cookie);
3262                 lh->mlh_reg_lh.cookie = 0;
3263                 RETURN(-ESTALE);
3264         }
3265
3266         LASSERTF(new_lock != NULL,
3267                  "lockh %#llx flags %#llx : rc = %d\n",
3268                  lh->mlh_reg_lh.cookie, flags, result);
3269
3270         /*
3271          * If we've already given this lock to a client once, then we should
3272          * have no readers or writers.  Otherwise, we should have one reader
3273          * _or_ writer ref (which will be zeroed below) before returning the
3274          * lock to a client.
3275          */
3276         if (new_lock->l_export == req->rq_export) {
3277                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3278         } else {
3279                 LASSERT(new_lock->l_export == NULL);
3280                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3281         }
3282
3283         *lockp = new_lock;
3284
3285         if (new_lock->l_export == req->rq_export) {
3286                 /*
3287                  * Already gave this to the client, which means that we
3288                  * reconstructed a reply.
3289                  */
3290                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3291                         MSG_RESENT);
3292
3293                 LDLM_LOCK_RELEASE(new_lock);
3294                 lh->mlh_reg_lh.cookie = 0;
3295                 RETURN(ELDLM_LOCK_REPLACED);
3296         }
3297
3298         /*
3299          * Fixup the lock to be given to the client.
3300          */
3301         lock_res_and_lock(new_lock);
3302         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3303          * possible blocking AST. */
3304         while (new_lock->l_readers > 0) {
3305                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3306                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3307                 new_lock->l_readers--;
3308         }
3309         while (new_lock->l_writers > 0) {
3310                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3311                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3312                 new_lock->l_writers--;
3313         }
3314
3315         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3316         new_lock->l_blocking_ast = lock->l_blocking_ast;
3317         new_lock->l_completion_ast = lock->l_completion_ast;
3318         new_lock->l_remote_handle = lock->l_remote_handle;
3319         new_lock->l_flags &= ~LDLM_FL_LOCAL;
3320
3321         unlock_res_and_lock(new_lock);
3322
3323         cfs_hash_add(new_lock->l_export->exp_lock_hash,
3324                      &new_lock->l_remote_handle,
3325                      &new_lock->l_exp_hash);
3326
3327         LDLM_LOCK_RELEASE(new_lock);
3328         lh->mlh_reg_lh.cookie = 0;
3329
3330         RETURN(ELDLM_LOCK_REPLACED);
3331 }
3332
3333 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3334                                     struct ldlm_lock *new_lock,
3335                                     struct mdt_lock_handle *lh,
3336                                     __u64 flags)
3337 {
3338         struct ptlrpc_request  *req = mdt_info_req(info);
3339         struct ldlm_request    *dlmreq;
3340
3341         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3342                 return;
3343
3344         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3345
3346         /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
3347          * lock was found by ldlm_handle_enqueue(); if so @lh must be
3348          * initialized. */
3349         if (flags & LDLM_FL_RESENT) {
3350                 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
3351                 lh->mlh_reg_mode = new_lock->l_granted_mode;
3352
3353                 LDLM_DEBUG(new_lock, "Restoring lock cookie");
3354                 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie %#llx",
3355                           lh->mlh_reg_lh.cookie);
3356                 return;
3357         }
3358
3359         /*
3360          * If the xid matches, then we know this is a resent request, and allow
3361          * it. (It's probably an OPEN, for which we don't send a lock.
3362          */
3363         if (req_can_reconstruct(req, NULL))
3364                 return;
3365
3366         /*
3367          * This remote handle isn't enqueued, so we never received or processed
3368          * this request.  Clear MSG_RESENT, because it can be handled like any
3369          * normal request now.
3370          */
3371         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3372
3373         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle %#llx",
3374                   dlmreq->lock_handle[0].cookie);
3375 }
3376
3377 static int mdt_intent_getxattr(enum mdt_it_code opcode,
3378                                 struct mdt_thread_info *info,
3379                                 struct ldlm_lock **lockp,
3380                                 __u64 flags)
3381 {
3382         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3383         struct ldlm_reply      *ldlm_rep = NULL;
3384         int rc;
3385         ENTRY;
3386
3387         /*
3388          * Initialize lhc->mlh_reg_lh either from a previously granted lock
3389          * (for the resend case) or a new lock. Below we will use it to
3390          * replace the original lock.
3391          */
3392         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3393         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3394                 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3395                 rc = mdt_object_lock(info, info->mti_object, lhc,
3396                                      MDS_INODELOCK_XATTR);
3397                 if (rc)
3398                         return rc;
3399         }
3400
3401         rc = mdt_getxattr(info);
3402
3403         if (mdt_info_req(info)->rq_repmsg != NULL)
3404                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3405
3406         if (ldlm_rep == NULL ||
3407             OBD_FAIL_CHECK(OBD_FAIL_MDS_XATTR_REP)) {
3408                 mdt_object_unlock(info,  info->mti_object, lhc, 1);
3409                 RETURN(err_serious(-EFAULT));
3410         }
3411
3412         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3413
3414         /* This is left for interop instead of adding a new interop flag.
3415          * LU-7433 */
3416 #if LUSTRE_VERSION_CODE > OBD_OCD_VERSION(3, 0, 0, 0)
3417         if (ldlm_rep->lock_policy_res2) {
3418                 mdt_object_unlock(info, info->mti_object, lhc, 1);
3419                 RETURN(ELDLM_LOCK_ABORTED);
3420         }
3421 #endif
3422
3423         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3424         RETURN(rc);
3425 }
3426
3427 static int mdt_intent_getattr(enum mdt_it_code opcode,
3428                               struct mdt_thread_info *info,
3429                               struct ldlm_lock **lockp,
3430                               __u64 flags)
3431 {
3432         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3433         __u64                   child_bits;
3434         struct ldlm_reply      *ldlm_rep;
3435         struct mdt_body        *reqbody;
3436         struct mdt_body        *repbody;
3437         int                     rc, rc2;
3438         ENTRY;
3439
3440         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3441         LASSERT(reqbody);
3442
3443         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3444         LASSERT(repbody);
3445
3446         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
3447         repbody->mbo_eadatasize = 0;
3448         repbody->mbo_aclsize = 0;
3449
3450         switch (opcode) {
3451         case MDT_IT_LOOKUP:
3452                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3453                 break;
3454         case MDT_IT_GETATTR:
3455                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3456                              MDS_INODELOCK_PERM;
3457                 break;
3458         default:
3459                 CERROR("Unsupported intent (%d)\n", opcode);
3460                 GOTO(out_shrink, rc = -EINVAL);
3461         }
3462
3463         rc = mdt_init_ucred_intent_getattr(info, reqbody);
3464         if (rc)
3465                 GOTO(out_shrink, rc);
3466
3467         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3468         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3469
3470         /* Get lock from request for possible resent case. */
3471         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3472
3473         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3474         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3475
3476         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3477                 ldlm_rep->lock_policy_res2 = 0;
3478         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3479             ldlm_rep->lock_policy_res2) {
3480                 lhc->mlh_reg_lh.cookie = 0ull;
3481                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3482         }
3483
3484         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3485         EXIT;
3486 out_ucred:
3487         mdt_exit_ucred(info);
3488 out_shrink:
3489         mdt_client_compatibility(info);
3490         rc2 = mdt_fix_reply(info);
3491         if (rc == 0)
3492                 rc = rc2;
3493         return rc;
3494 }
3495
3496 static int mdt_intent_layout(enum mdt_it_code opcode,
3497                              struct mdt_thread_info *info,
3498                              struct ldlm_lock **lockp,
3499                              __u64 flags)
3500 {
3501         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_LAYOUT];
3502         struct layout_intent *layout;
3503         struct lu_fid *fid;
3504         struct mdt_object *obj = NULL;
3505         bool layout_change = false;
3506         int layout_size = 0;
3507         int rc = 0;
3508         ENTRY;
3509
3510         if (opcode != MDT_IT_LAYOUT) {
3511                 CERROR("%s: Unknown intent (%d)\n", mdt_obd_name(info->mti_mdt),
3512                         opcode);
3513                 RETURN(-EINVAL);
3514         }
3515
3516         layout = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3517         if (layout == NULL)
3518                 RETURN(-EPROTO);
3519
3520         switch (layout->li_opc) {
3521         case LAYOUT_INTENT_TRUNC:
3522         case LAYOUT_INTENT_WRITE:
3523                 layout_change = true;
3524                 break;
3525         case LAYOUT_INTENT_ACCESS:
3526                 break;
3527         case LAYOUT_INTENT_READ:
3528         case LAYOUT_INTENT_GLIMPSE:
3529         case LAYOUT_INTENT_RELEASE:
3530         case LAYOUT_INTENT_RESTORE:
3531                 CERROR("%s: Unsupported layout intent opc %d\n",
3532                        mdt_obd_name(info->mti_mdt), layout->li_opc);
3533                 rc = -ENOTSUPP;
3534                 break;
3535         default:
3536                 CERROR("%s: Unknown layout intent opc %d\n",
3537                        mdt_obd_name(info->mti_mdt), layout->li_opc);
3538                 rc = -EINVAL;
3539                 break;
3540         }
3541         if (rc < 0)
3542                 RETURN(rc);
3543
3544         fid = &info->mti_tmp_fid2;
3545         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3546
3547         /* Get lock from request for possible resent case. */
3548         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3549
3550         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3551         if (IS_ERR(obj))
3552                 GOTO(out, rc = PTR_ERR(obj));
3553
3554         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3555                 layout_size = mdt_attr_get_eabuf_size(info, obj);
3556                 if (layout_size < 0)
3557                         GOTO(out_obj, rc = layout_size);
3558
3559                 if (layout_size > info->mti_mdt->mdt_max_mdsize)
3560                         info->mti_mdt->mdt_max_mdsize = layout_size;
3561         }
3562
3563         /*
3564          * set reply buffer size, so that ldlm_handle_enqueue0()->
3565          * ldlm_lvbo_fill() will fill the reply buffer with lovea.
3566          */
3567         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3568         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
3569                              layout_size);
3570         rc = req_capsule_server_pack(info->mti_pill);
3571         if (rc)
3572                 GOTO(out_obj, rc);
3573
3574
3575         if (layout_change) {
3576                 struct lu_buf *buf = &info->mti_buf;
3577
3578                 buf->lb_buf = NULL;
3579                 buf->lb_len = 0;
3580                 if (unlikely(req_is_replay(mdt_info_req(info)))) {
3581                         buf->lb_buf = req_capsule_client_get(info->mti_pill,
3582                                         &RMF_EADATA);
3583                         buf->lb_len = req_capsule_get_size(info->mti_pill,
3584                                         &RMF_EADATA, RCL_CLIENT);
3585                         /*
3586                          * If it's a replay of layout write intent RPC, the
3587                          * client has saved the extended lovea when
3588                          * it get reply then.
3589                          */
3590                         if (buf->lb_len > 0)
3591                                 mdt_fix_lov_magic(info, buf->lb_buf);
3592                 }
3593
3594                 /*
3595                  * Instantiate some layout components, if @buf contains
3596                  * lovea, then it's a replay of the layout intent write
3597                  * RPC.
3598                  */
3599                 rc = mdt_layout_change(info, obj, layout, buf);
3600                 if (rc)
3601                         GOTO(out_obj, rc);
3602         }
3603 out_obj:
3604         mdt_object_put(info->mti_env, obj);
3605
3606         if (rc == 0 && lustre_handle_is_used(&lhc->mlh_reg_lh))
3607                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3608
3609 out:
3610         lhc->mlh_reg_lh.cookie = 0;
3611
3612         return rc;
3613 }
3614
3615 static int mdt_intent_reint(enum mdt_it_code opcode,
3616                             struct mdt_thread_info *info,
3617                             struct ldlm_lock **lockp,
3618                             __u64 flags)
3619 {
3620         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3621         struct ldlm_reply      *rep = NULL;
3622         long                    opc;
3623         int                     rc;
3624
3625         static const struct req_format *intent_fmts[REINT_MAX] = {
3626                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3627                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3628         };
3629
3630         ENTRY;
3631
3632         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
3633         if (opc < 0)
3634                 RETURN(opc);
3635
3636         if (mdt_it_flavor[opcode].it_reint != opc) {
3637                 CERROR("Reint code %ld doesn't match intent: %d\n",
3638                        opc, opcode);
3639                 RETURN(err_serious(-EPROTO));
3640         }
3641
3642         /* Get lock from request for possible resent case. */
3643         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3644
3645         rc = mdt_reint_internal(info, lhc, opc);
3646
3647         /* Check whether the reply has been packed successfully. */
3648         if (mdt_info_req(info)->rq_repmsg != NULL)
3649                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3650         if (rep == NULL)
3651                 RETURN(err_serious(-EFAULT));
3652
3653         /* MDC expects this in any case */
3654         if (rc != 0)
3655                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3656
3657         /* the open lock or the lock for cross-ref object should be
3658          * returned to the client */
3659         if (lustre_handle_is_used(&lhc->mlh_reg_lh) &&
3660             (rc == 0 || rc == -MDT_EREMOTE_OPEN)) {
3661                 rep->lock_policy_res2 = 0;
3662                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3663                 RETURN(rc);
3664         }
3665
3666         rep->lock_policy_res2 = clear_serious(rc);
3667
3668         if (rep->lock_policy_res2 == -ENOENT &&
3669             mdt_get_disposition(rep, DISP_LOOKUP_NEG) &&
3670             !mdt_get_disposition(rep, DISP_OPEN_CREATE))
3671                 rep->lock_policy_res2 = 0;
3672
3673         lhc->mlh_reg_lh.cookie = 0ull;
3674         if (rc == -ENOTCONN || rc == -ENODEV ||
3675             rc == -EOVERFLOW) { /**< if VBR failure then return error */
3676                 /*
3677                  * If it is the disconnect error (ENODEV & ENOCONN), the error
3678                  * will be returned by rq_status, and client at ptlrpc layer
3679                  * will detect this, then disconnect, reconnect the import
3680                  * immediately, instead of impacting the following the rpc.
3681                  */
3682                 RETURN(rc);
3683         }
3684         /*
3685          * For other cases, the error will be returned by intent, and client
3686          * will retrieve the result from intent.
3687          */
3688         RETURN(ELDLM_LOCK_ABORTED);
3689 }
3690
3691 static int mdt_intent_code(enum ldlm_intent_flags itcode)
3692 {
3693         int rc;
3694
3695         switch (itcode) {
3696         case IT_OPEN:
3697                 rc = MDT_IT_OPEN;
3698                 break;
3699         case IT_OPEN|IT_CREAT:
3700                 rc = MDT_IT_OCREAT;
3701                 break;
3702         case IT_CREAT:
3703                 rc = MDT_IT_CREATE;
3704                 break;
3705         case IT_READDIR:
3706                 rc = MDT_IT_READDIR;
3707                 break;
3708         case IT_GETATTR:
3709                 rc = MDT_IT_GETATTR;
3710                 break;
3711         case IT_LOOKUP:
3712                 rc = MDT_IT_LOOKUP;
3713                 break;
3714         case IT_UNLINK:
3715                 rc = MDT_IT_UNLINK;
3716                 break;
3717         case IT_TRUNC:
3718                 rc = MDT_IT_TRUNC;
3719                 break;
3720         case IT_GETXATTR:
3721                 rc = MDT_IT_GETXATTR;
3722                 break;
3723         case IT_LAYOUT:
3724                 rc = MDT_IT_LAYOUT;
3725                 break;
3726         case IT_QUOTA_DQACQ:
3727         case IT_QUOTA_CONN:
3728                 rc = MDT_IT_QUOTA;
3729                 break;
3730         default:
3731                 CERROR("Unknown intent opcode: 0x%08x\n", itcode);
3732                 rc = -EINVAL;
3733                 break;
3734         }
3735         return rc;
3736 }
3737
3738 static int mdt_intent_opc(enum ldlm_intent_flags itopc,
3739                           struct mdt_thread_info *info,
3740                           struct ldlm_lock **lockp, __u64 flags)
3741 {
3742         struct req_capsule      *pill = info->mti_pill;
3743         struct ptlrpc_request   *req = mdt_info_req(info);
3744         struct mdt_it_flavor    *flv;
3745         int opc;
3746         int rc;
3747         ENTRY;
3748
3749         opc = mdt_intent_code(itopc);
3750         if (opc < 0)
3751                 RETURN(-EINVAL);
3752
3753         if (opc == MDT_IT_QUOTA) {
3754                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
3755
3756                 if (qmt == NULL)
3757                         RETURN(-EOPNOTSUPP);
3758
3759                 if (mdt_rdonly(req->rq_export))
3760                         RETURN(-EROFS);
3761
3762                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
3763                 /* pass the request to quota master */
3764                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
3765                                                  mdt_info_req(info), lockp,
3766                                                  flags);
3767                 RETURN(rc);
3768         }
3769
3770         flv = &mdt_it_flavor[opc];
3771         if (flv->it_fmt != NULL)
3772                 req_capsule_extend(pill, flv->it_fmt);
3773
3774         rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3775         if (rc < 0)
3776                 RETURN(rc);
3777
3778         if (flv->it_flags & MUTABOR && mdt_rdonly(req->rq_export))
3779                 RETURN(-EROFS);
3780
3781         if (flv->it_act != NULL) {
3782                 struct ldlm_reply *rep;
3783
3784                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_INTENT_DELAY, 10);
3785
3786                 /* execute policy */
3787                 rc = flv->it_act(opc, info, lockp, flags);
3788
3789                 /* Check whether the reply has been packed successfully. */
3790                 if (req->rq_repmsg != NULL) {
3791                         rep = req_capsule_server_get(info->mti_pill,
3792                                                      &RMF_DLM_REP);
3793                         rep->lock_policy_res2 =
3794                                 ptlrpc_status_hton(rep->lock_policy_res2);
3795                 }
3796         }
3797
3798         RETURN(rc);
3799 }
3800
3801 static int mdt_intent_policy(struct ldlm_namespace *ns,
3802                              struct ldlm_lock **lockp, void *req_cookie,
3803                              enum ldlm_mode mode, __u64 flags, void *data)
3804 {
3805         struct tgt_session_info *tsi;
3806         struct mdt_thread_info  *info;
3807         struct ptlrpc_request   *req  =  req_cookie;
3808         struct ldlm_intent      *it;
3809         struct req_capsule      *pill;
3810         int rc;
3811
3812         ENTRY;
3813
3814         LASSERT(req != NULL);
3815
3816         tsi = tgt_ses_info(req->rq_svc_thread->t_env);
3817
3818         info = tsi2mdt_info(tsi);
3819         LASSERT(info != NULL);
3820         pill = info->mti_pill;
3821         LASSERT(pill->rc_req == req);
3822
3823         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3824                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
3825                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3826                 if (it != NULL) {
3827                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
3828                         if (rc == 0)
3829                                 rc = ELDLM_OK;
3830
3831                         /* Lock without inodebits makes no sense and will oops
3832                          * later in ldlm. Let's check it now to see if we have
3833                          * ibits corrupted somewhere in mdt_intent_opc().
3834                          * The case for client miss to set ibits has been
3835                          * processed by others. */
3836                         LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
3837                                         lr_type == LDLM_IBITS,
3838                                      info->mti_dlm_req->lock_desc.\
3839                                         l_policy_data.l_inodebits.bits != 0));
3840                 } else
3841                         rc = err_serious(-EFAULT);
3842         } else {
3843                 /* No intent was provided */
3844                 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3845                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
3846                 rc = req_capsule_server_pack(pill);
3847                 if (rc)
3848                         rc = err_serious(rc);
3849         }
3850         mdt_thread_info_fini(info);
3851         RETURN(rc);
3852 }
3853
3854 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
3855 {
3856         struct seq_server_site  *ss = mdt_seq_site(mdt);
3857
3858         if (ss->ss_node_id == 0)
3859                 return;
3860
3861         if (ss->ss_client_seq != NULL) {
3862                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3863                 ss->ss_client_seq->lcs_exp = NULL;
3864         }
3865
3866         if (ss->ss_server_fld != NULL) {
3867                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
3868                 ss->ss_server_fld->lsf_control_exp = NULL;
3869         }
3870 }
3871
3872 static void mdt_seq_fini_cli(struct mdt_device *mdt)
3873 {
3874         struct seq_server_site *ss = mdt_seq_site(mdt);
3875
3876         if (ss == NULL)
3877                 return;
3878
3879         if (ss->ss_server_seq != NULL)
3880                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
3881 }
3882
3883 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
3884 {
3885         mdt_seq_fini_cli(mdt);
3886         mdt_deregister_seq_exp(mdt);
3887
3888         return seq_site_fini(env, mdt_seq_site(mdt));
3889 }
3890
3891 /**
3892  * It will retrieve its FLDB entries from MDT0, and it only happens
3893  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
3894  * and it needs to refresh FLDB from the MDT0.
3895  **/
3896 static int mdt_register_lwp_callback(void *data)
3897 {
3898         struct lu_env           env;
3899         struct mdt_device       *mdt = data;
3900         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
3901         int                     rc;
3902         ENTRY;
3903
3904         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
3905
3906         rc = lu_env_init(&env, LCT_MD_THREAD);
3907         if (rc < 0) {
3908                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
3909                 RETURN(rc);
3910         }
3911
3912         /* Allocate new sequence now to avoid creating local transaction
3913          * in the normal transaction process */
3914         rc = seq_server_check_and_alloc_super(&env,
3915                                               mdt_seq_site(mdt)->ss_server_seq);
3916         if (rc < 0)
3917                 GOTO(out, rc);
3918
3919         if (fld->lsf_new) {
3920                 rc = fld_update_from_controller(&env, fld);
3921                 if (rc != 0) {
3922                         CERROR("%s: cannot update controller: rc = %d\n",
3923                                mdt_obd_name(mdt), rc);
3924                         GOTO(out, rc);
3925                 }
3926         }
3927 out:
3928         lu_env_fini(&env);
3929         RETURN(rc);
3930 }
3931
3932 static int mdt_register_seq_exp(struct mdt_device *mdt)
3933 {
3934         struct seq_server_site  *ss = mdt_seq_site(mdt);
3935         char                    *lwp_name = NULL;
3936         int                     rc;
3937
3938         if (ss->ss_node_id == 0)
3939                 return 0;
3940
3941         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
3942         if (lwp_name == NULL)
3943                 GOTO(out_free, rc = -ENOMEM);
3944
3945         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
3946         if (rc != 0)
3947                 GOTO(out_free, rc);
3948
3949         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
3950                                       NULL, NULL);
3951         if (rc != 0)
3952                 GOTO(out_free, rc);
3953
3954         rc = lustre_register_lwp_item(lwp_name,
3955                                       &ss->ss_server_fld->lsf_control_exp,
3956                                       mdt_register_lwp_callback, mdt);
3957         if (rc != 0) {
3958                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3959                 ss->ss_client_seq->lcs_exp = NULL;
3960                 GOTO(out_free, rc);
3961         }
3962 out_free:
3963         if (lwp_name != NULL)
3964                 OBD_FREE(lwp_name, MAX_OBD_NAME);
3965
3966         return rc;
3967 }
3968
3969 /*
3970  * Init client sequence manager which is used by local MDS to talk to sequence
3971  * controller on remote node.
3972  */
3973 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
3974 {
3975         struct seq_server_site  *ss = mdt_seq_site(mdt);
3976         int                     rc;
3977         char                    *prefix;
3978         ENTRY;
3979
3980         /* check if this is adding the first MDC and controller is not yet
3981          * initialized. */
3982         OBD_ALLOC_PTR(ss->ss_client_seq);
3983         if (ss->ss_client_seq == NULL)
3984                 RETURN(-ENOMEM);
3985
3986         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3987         if (prefix == NULL) {
3988                 OBD_FREE_PTR(ss->ss_client_seq);
3989                 ss->ss_client_seq = NULL;
3990                 RETURN(-ENOMEM);
3991         }
3992
3993         /* Note: seq_client_fini will be called in seq_site_fini */
3994         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
3995         rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
3996                              prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
3997                                                             NULL);
3998         OBD_FREE(prefix, MAX_OBD_NAME + 5);
3999         if (rc != 0) {
4000                 OBD_FREE_PTR(ss->ss_client_seq);
4001                 ss->ss_client_seq = NULL;
4002                 RETURN(rc);
4003         }
4004
4005         rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
4006
4007         RETURN(rc);
4008 }
4009
4010 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
4011 {
4012         struct seq_server_site  *ss;
4013         int                     rc;
4014         ENTRY;
4015
4016         ss = mdt_seq_site(mdt);
4017         /* init sequence controller server(MDT0) */
4018         if (ss->ss_node_id == 0) {
4019                 OBD_ALLOC_PTR(ss->ss_control_seq);
4020                 if (ss->ss_control_seq == NULL)
4021                         RETURN(-ENOMEM);
4022
4023                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
4024                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
4025                                      ss);
4026                 if (rc)
4027                         GOTO(out_seq_fini, rc);
4028         }
4029
4030         /* Init normal sequence server */
4031         OBD_ALLOC_PTR(ss->ss_server_seq);
4032         if (ss->ss_server_seq == NULL)
4033                 GOTO(out_seq_fini, rc = -ENOMEM);
4034
4035         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
4036                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
4037         if (rc)
4038                 GOTO(out_seq_fini, rc);
4039
4040         /* init seq client for seq server to talk to seq controller(MDT0) */
4041         rc = mdt_seq_init_cli(env, mdt);
4042         if (rc != 0)
4043                 GOTO(out_seq_fini, rc);
4044
4045         if (ss->ss_node_id != 0)
4046                 /* register controller export through lwp */
4047                 rc = mdt_register_seq_exp(mdt);
4048
4049         EXIT;
4050 out_seq_fini:
4051         if (rc)
4052                 mdt_seq_fini(env, mdt);
4053
4054         return rc;
4055 }
4056
4057 /*
4058  * FLD wrappers
4059  */
4060 static int mdt_fld_fini(const struct lu_env *env,
4061                         struct mdt_device *m)
4062 {
4063         struct seq_server_site *ss = mdt_seq_site(m);
4064         ENTRY;
4065
4066         if (ss && ss->ss_server_fld) {
4067                 fld_server_fini(env, ss->ss_server_fld);
4068                 OBD_FREE_PTR(ss->ss_server_fld);
4069                 ss->ss_server_fld = NULL;
4070         }
4071
4072         RETURN(0);
4073 }
4074
4075 static int mdt_fld_init(const struct lu_env *env,
4076                         const char *uuid,
4077                         struct mdt_device *m)
4078 {
4079         struct seq_server_site *ss;
4080         int rc;
4081         ENTRY;
4082
4083         ss = mdt_seq_site(m);
4084
4085         OBD_ALLOC_PTR(ss->ss_server_fld);
4086         if (ss->ss_server_fld == NULL)
4087                 RETURN(rc = -ENOMEM);
4088
4089         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
4090                              LU_SEQ_RANGE_MDT);
4091         if (rc) {
4092                 OBD_FREE_PTR(ss->ss_server_fld);
4093                 ss->ss_server_fld = NULL;
4094                 RETURN(rc);
4095         }
4096
4097         RETURN(0);
4098 }
4099
4100 static void mdt_stack_pre_fini(const struct lu_env *env,
4101                            struct mdt_device *m, struct lu_device *top)
4102 {
4103         struct lustre_cfg_bufs  *bufs;
4104         struct lustre_cfg       *lcfg;
4105         struct mdt_thread_info  *info;
4106         ENTRY;
4107
4108         LASSERT(top);
4109
4110         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4111         LASSERT(info != NULL);
4112
4113         bufs = &info->mti_u.bufs;
4114
4115         LASSERT(m->mdt_child_exp);
4116         LASSERT(m->mdt_child_exp->exp_obd);
4117
4118         /* process cleanup, pass mdt obd name to get obd umount flags */
4119         /* XXX: this is needed because all layers are referenced by
4120          * objects (some of them are pinned by osd, for example *
4121          * the proper solution should be a model where object used
4122          * by osd only doesn't have mdt/mdd slices -bzzz */
4123         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4124         lustre_cfg_bufs_set_string(bufs, 1, NULL);
4125         lcfg = lustre_cfg_new(LCFG_PRE_CLEANUP, bufs);
4126         if (lcfg == NULL)
4127                 RETURN_EXIT;
4128
4129         top->ld_ops->ldo_process_config(env, top, lcfg);
4130         lustre_cfg_free(lcfg);
4131         EXIT;
4132 }
4133
4134 static void mdt_stack_fini(const struct lu_env *env,
4135                            struct mdt_device *m, struct lu_device *top)
4136 {
4137         struct obd_device       *obd = mdt2obd_dev(m);
4138         struct lustre_cfg_bufs  *bufs;
4139         struct lustre_cfg       *lcfg;
4140         struct mdt_thread_info  *info;
4141         char                     flags[3] = "";
4142         ENTRY;
4143
4144         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4145         LASSERT(info != NULL);
4146
4147         lu_dev_del_linkage(top->ld_site, top);
4148
4149         lu_site_purge(env, top->ld_site, -1);
4150
4151         bufs = &info->mti_u.bufs;
4152         /* process cleanup, pass mdt obd name to get obd umount flags */
4153         /* another purpose is to let all layers to release their objects */
4154         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4155         if (obd->obd_force)
4156                 strcat(flags, "F");
4157         if (obd->obd_fail)
4158                 strcat(flags, "A");
4159         lustre_cfg_bufs_set_string(bufs, 1, flags);
4160         lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
4161         if (lcfg == NULL)
4162                 RETURN_EXIT;
4163
4164         LASSERT(top);
4165         top->ld_ops->ldo_process_config(env, top, lcfg);
4166         lustre_cfg_free(lcfg);
4167
4168         lu_site_purge(env, top->ld_site, -1);
4169
4170         m->mdt_child = NULL;
4171         m->mdt_bottom = NULL;
4172
4173         obd_disconnect(m->mdt_child_exp);
4174         m->mdt_child_exp = NULL;
4175
4176         obd_disconnect(m->mdt_bottom_exp);
4177         m->mdt_child_exp = NULL;
4178 }
4179
4180 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
4181                                const char *next, struct obd_export **exp)
4182 {
4183         struct obd_connect_data *data = NULL;
4184         struct obd_device       *obd;
4185         int                      rc;
4186         ENTRY;
4187
4188         OBD_ALLOC_PTR(data);
4189         if (data == NULL)
4190                 GOTO(out, rc = -ENOMEM);
4191
4192         obd = class_name2obd(next);
4193         if (obd == NULL) {
4194                 CERROR("%s: can't locate next device: %s\n",
4195                        mdt_obd_name(m), next);
4196                 GOTO(out, rc = -ENOTCONN);
4197         }
4198
4199         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4200         data->ocd_version = LUSTRE_VERSION_CODE;
4201
4202         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
4203         if (rc) {
4204                 CERROR("%s: cannot connect to next dev %s (%d)\n",
4205                        mdt_obd_name(m), next, rc);
4206                 GOTO(out, rc);
4207         }
4208
4209 out:
4210         if (data)
4211                 OBD_FREE_PTR(data);
4212         RETURN(rc);
4213 }
4214
4215 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
4216                           struct lustre_cfg *cfg)
4217 {
4218         char                   *dev = lustre_cfg_string(cfg, 0);
4219         int                     rc, name_size, uuid_size;
4220         char                   *name, *uuid, *p;
4221         struct lustre_cfg_bufs *bufs;
4222         struct lustre_cfg      *lcfg;
4223         struct obd_device      *obd;
4224         struct lustre_profile  *lprof;
4225         struct lu_site         *site;
4226         ENTRY;
4227
4228         /* in 1.8 we had the only device in the stack - MDS.
4229          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
4230          * in 2.3 OSD is instantiated by obd_mount.c, so we need
4231          * to generate names and setup MDT, MDD. MDT will be using
4232          * generated name to connect to MDD. for MDD the next device
4233          * will be LOD with name taken from so called "profile" which
4234          * is generated by mount_option line
4235          *
4236          * 1.8 MGS generates config. commands like this:
4237          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
4238          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
4239          * 2.0 MGS generates config. commands like this:
4240          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
4241          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4242          *                    3:lustre-MDT0000-mdtlov  4:f
4243          *
4244          * we generate MDD name from MDT one, just replacing T with D
4245          *
4246          * after all the preparations, the logical equivalent will be
4247          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
4248          *                    3:lustre-MDT0000-mdtlov  4:f
4249          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4250          *                    3:lustre-MDD0000  4:f
4251          *
4252          *  notice we build the stack from down to top: MDD first, then MDT */
4253
4254         name_size = MAX_OBD_NAME;
4255         uuid_size = MAX_OBD_NAME;
4256
4257         OBD_ALLOC(name, name_size);
4258         OBD_ALLOC(uuid, uuid_size);
4259         if (name == NULL || uuid == NULL)
4260                 GOTO(cleanup_mem, rc = -ENOMEM);
4261
4262         OBD_ALLOC_PTR(bufs);
4263         if (!bufs)
4264                 GOTO(cleanup_mem, rc = -ENOMEM);
4265
4266         strcpy(name, dev);
4267         p = strstr(name, "-MDT");
4268         if (p == NULL)
4269                 GOTO(free_bufs, rc = -ENOMEM);
4270         p[3] = 'D';
4271
4272         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
4273
4274         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4275         if (lprof == NULL || lprof->lp_dt == NULL) {
4276                 CERROR("can't find the profile: %s\n",
4277                        lustre_cfg_string(cfg, 0));
4278                 GOTO(free_bufs, rc = -EINVAL);
4279         }
4280
4281         lustre_cfg_bufs_reset(bufs, name);
4282         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
4283         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4284         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4285
4286         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4287         if (lcfg == NULL)
4288                 GOTO(put_profile, rc = -ENOMEM);
4289
4290         rc = class_attach(lcfg);
4291         if (rc)
4292                 GOTO(lcfg_cleanup, rc);
4293
4294         obd = class_name2obd(name);
4295         if (!obd) {
4296                 CERROR("Can not find obd %s (%s in config)\n",
4297                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
4298                 GOTO(lcfg_cleanup, rc = -EINVAL);
4299         }
4300
4301         lustre_cfg_free(lcfg);
4302
4303         lustre_cfg_bufs_reset(bufs, name);
4304         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4305         lustre_cfg_bufs_set_string(bufs, 2, dev);
4306         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4307
4308         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4309         if (lcfg == NULL)
4310                 GOTO(class_detach, rc = -ENOMEM);
4311
4312         rc = class_setup(obd, lcfg);
4313         if (rc)
4314                 GOTO(class_detach, rc);
4315
4316         /* connect to MDD we just setup */
4317         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
4318         if (rc)
4319                 GOTO(class_detach, rc);
4320
4321         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
4322         LASSERT(site);
4323         LASSERT(mdt_lu_site(mdt) == NULL);
4324         mdt->mdt_lu_dev.ld_site = site;
4325         site->ls_top_dev = &mdt->mdt_lu_dev;
4326         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
4327
4328         /* now connect to bottom OSD */
4329         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
4330         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
4331         if (rc)
4332                 GOTO(class_detach, rc);
4333         mdt->mdt_bottom =
4334                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
4335
4336         rc = lu_env_refill((struct lu_env *)env);
4337         if (rc != 0)
4338                 CERROR("Failure to refill session: '%d'\n", rc);
4339
4340         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
4341
4342         EXIT;
4343 class_detach:
4344         if (rc)
4345                 class_detach(obd, lcfg);
4346 lcfg_cleanup:
4347         lustre_cfg_free(lcfg);
4348 put_profile:
4349         class_put_profile(lprof);
4350 free_bufs:
4351         OBD_FREE_PTR(bufs);
4352 cleanup_mem:
4353         if (name)
4354                 OBD_FREE(name, name_size);
4355         if (uuid)
4356                 OBD_FREE(uuid, uuid_size);
4357         RETURN(rc);
4358 }
4359
4360 /* setup quota master target on MDT0 */
4361 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
4362                           struct lustre_cfg *cfg)
4363 {
4364         struct obd_device       *obd;
4365         char                    *dev = lustre_cfg_string(cfg, 0);
4366         char                    *qmtname, *uuid, *p;
4367         struct lustre_cfg_bufs  *bufs;
4368         struct lustre_cfg       *lcfg;
4369         struct lustre_profile   *lprof;
4370         struct obd_connect_data *data;
4371         int                      rc;
4372         ENTRY;
4373
4374         LASSERT(mdt->mdt_qmt_exp == NULL);
4375         LASSERT(mdt->mdt_qmt_dev == NULL);
4376
4377         /* quota master is on MDT0 only for now */
4378         if (mdt->mdt_seq_site.ss_node_id != 0)
4379                 RETURN(0);
4380
4381         /* MGS generates config commands which look as follows:
4382          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4383          *                    3:lustre-MDT0000-mdtlov  4:f
4384          *
4385          * We generate the QMT name from the MDT one, just replacing MD with QM
4386          * after all the preparations, the logical equivalent will be:
4387          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
4388          *                    3:lustre-MDT0000-osd  4:f */
4389         OBD_ALLOC(qmtname, MAX_OBD_NAME);
4390         OBD_ALLOC(uuid, UUID_MAX);
4391         OBD_ALLOC_PTR(bufs);
4392         OBD_ALLOC_PTR(data);
4393         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
4394                 GOTO(cleanup_mem, rc = -ENOMEM);
4395
4396         strcpy(qmtname, dev);
4397         p = strstr(qmtname, "-MDT");
4398         if (p == NULL)
4399                 GOTO(cleanup_mem, rc = -ENOMEM);
4400         /* replace MD with QM */
4401         p[1] = 'Q';
4402         p[2] = 'M';
4403
4404         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
4405
4406         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4407         if (lprof == NULL || lprof->lp_dt == NULL) {
4408                 CERROR("can't find profile for %s\n",
4409                        lustre_cfg_string(cfg, 0));
4410                 GOTO(cleanup_mem, rc = -EINVAL);
4411         }
4412
4413         lustre_cfg_bufs_reset(bufs, qmtname);
4414         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
4415         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4416         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4417
4418         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4419         if (lcfg == NULL)
4420                 GOTO(put_profile, rc = -ENOMEM);
4421
4422         rc = class_attach(lcfg);
4423         if (rc)
4424                 GOTO(lcfg_cleanup, rc);
4425
4426         obd = class_name2obd(qmtname);
4427         if (!obd) {
4428                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
4429                        lustre_cfg_string(cfg, 0));
4430                 GOTO(lcfg_cleanup, rc = -EINVAL);
4431         }
4432
4433         lustre_cfg_free(lcfg);
4434
4435         lustre_cfg_bufs_reset(bufs, qmtname);
4436         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4437         lustre_cfg_bufs_set_string(bufs, 2, dev);
4438
4439         /* for quota, the next device should be the OSD device */
4440         lustre_cfg_bufs_set_string(bufs, 3,
4441                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
4442
4443         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4444         if (lcfg == NULL)
4445                 GOTO(class_detach, rc = -ENOMEM);
4446
4447         rc = class_setup(obd, lcfg);
4448         if (rc)
4449                 GOTO(class_detach, rc);
4450
4451         mdt->mdt_qmt_dev = obd->obd_lu_dev;
4452
4453         /* configure local quota objects */
4454         rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
4455                                                    &mdt->mdt_lu_dev,
4456                                                    mdt->mdt_qmt_dev);
4457         if (rc)
4458                 GOTO(class_cleanup, rc);
4459
4460         /* connect to quota master target */
4461         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4462         data->ocd_version = LUSTRE_VERSION_CODE;
4463         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
4464                          data, NULL);
4465         if (rc) {
4466                 CERROR("cannot connect to quota master device %s (%d)\n",
4467                        qmtname, rc);
4468                 GOTO(class_cleanup, rc);
4469         }
4470
4471         EXIT;
4472 class_cleanup:
4473         if (rc) {
4474                 class_manual_cleanup(obd);
4475                 mdt->mdt_qmt_dev = NULL;
4476         }
4477 class_detach:
4478         if (rc)
4479                 class_detach(obd, lcfg);
4480 lcfg_cleanup:
4481         lustre_cfg_free(lcfg);
4482 put_profile:
4483         class_put_profile(lprof);
4484 cleanup_mem:
4485         if (bufs)
4486                 OBD_FREE_PTR(bufs);
4487         if (qmtname)
4488                 OBD_FREE(qmtname, MAX_OBD_NAME);
4489         if (uuid)
4490                 OBD_FREE(uuid, UUID_MAX);
4491         if (data)
4492                 OBD_FREE_PTR(data);
4493         return rc;
4494 }
4495
4496 /* Shutdown quota master target associated with mdt */
4497 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
4498 {
4499         ENTRY;
4500
4501         if (mdt->mdt_qmt_exp == NULL)
4502                 RETURN_EXIT;
4503         LASSERT(mdt->mdt_qmt_dev != NULL);
4504
4505         /* the qmt automatically shuts down when the mdt disconnects */
4506         obd_disconnect(mdt->mdt_qmt_exp);
4507         mdt->mdt_qmt_exp = NULL;
4508         mdt->mdt_qmt_dev = NULL;
4509         EXIT;
4510 }
4511
4512 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
4513  * for now. This will be removed along with converting rest of MDT code
4514  * to use tgt_session_info */
4515 static int mdt_tgt_getxattr(struct tgt_session_info *tsi)
4516 {
4517         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
4518         int                      rc;
4519
4520         rc = mdt_getxattr(info);
4521
4522         mdt_thread_info_fini(info);
4523         return rc;
4524 }
4525
4526 static struct tgt_handler mdt_tgt_handlers[] = {
4527 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4528                 0,                      MDS_CONNECT,    mdt_tgt_connect,
4529                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
4530 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4531                 0,                      MDS_DISCONNECT, tgt_disconnect,
4532                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
4533 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4534                 HABEO_REFERO,           MDS_SET_INFO,   mdt_set_info,
4535                 &RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION),
4536 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
4537 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_GET_ROOT,   mdt_get_root),
4538 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETATTR,    mdt_getattr),
4539 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_GETATTR_NAME,
4540                                                         mdt_getattr_name),
4541 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETXATTR,   mdt_tgt_getxattr),
4542 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_STATFS,     mdt_statfs),
4543 TGT_MDT_HDL(0           | MUTABOR,      MDS_REINT,      mdt_reint),
4544 TGT_MDT_HDL(HABEO_CORPUS,               MDS_CLOSE,      mdt_close),
4545 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_READPAGE,   mdt_readpage),
4546 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_SYNC,       mdt_sync),
4547 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
4548 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_PROGRESS,
4549                                                         mdt_hsm_progress),
4550 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_REGISTER,
4551                                                         mdt_hsm_ct_register),
4552 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_UNREGISTER,
4553                                                         mdt_hsm_ct_unregister),
4554 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_STATE_GET,
4555                                                         mdt_hsm_state_get),
4556 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_STATE_SET,
4557                                                         mdt_hsm_state_set),
4558 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_ACTION, mdt_hsm_action),
4559 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_REQUEST,
4560                                                         mdt_hsm_request),
4561 TGT_MDT_HDL(HABEO_CLAVIS | HABEO_CORPUS | HABEO_REFERO | MUTABOR,
4562             MDS_SWAP_LAYOUTS,
4563             mdt_swap_layouts),
4564 };
4565
4566 static struct tgt_handler mdt_sec_ctx_ops[] = {
4567 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
4568 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
4569 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
4570 };
4571
4572 static struct tgt_handler mdt_quota_ops[] = {
4573 TGT_QUOTA_HDL(HABEO_REFERO,             QUOTA_DQACQ,      mdt_quota_dqacq),
4574 };
4575
4576 static struct tgt_opc_slice mdt_common_slice[] = {
4577         {
4578                 .tos_opc_start  = MDS_FIRST_OPC,
4579                 .tos_opc_end    = MDS_LAST_OPC,
4580                 .tos_hs         = mdt_tgt_handlers
4581         },
4582         {
4583                 .tos_opc_start  = OBD_FIRST_OPC,
4584                 .tos_opc_end    = OBD_LAST_OPC,
4585                 .tos_hs         = tgt_obd_handlers
4586         },
4587         {
4588                 .tos_opc_start  = LDLM_FIRST_OPC,
4589                 .tos_opc_end    = LDLM_LAST_OPC,
4590                 .tos_hs         = tgt_dlm_handlers
4591         },
4592         {
4593                 .tos_opc_start  = SEC_FIRST_OPC,
4594                 .tos_opc_end    = SEC_LAST_OPC,
4595                 .tos_hs         = mdt_sec_ctx_ops
4596         },
4597         {
4598                 .tos_opc_start  = OUT_UPDATE_FIRST_OPC,
4599                 .tos_opc_end    = OUT_UPDATE_LAST_OPC,
4600                 .tos_hs         = tgt_out_handlers
4601         },
4602         {
4603                 .tos_opc_start  = FLD_FIRST_OPC,
4604                 .tos_opc_end    = FLD_LAST_OPC,
4605                 .tos_hs         = fld_handlers
4606         },
4607         {
4608                 .tos_opc_start  = SEQ_FIRST_OPC,
4609                 .tos_opc_end    = SEQ_LAST_OPC,
4610                 .tos_hs         = seq_handlers
4611         },
4612         {
4613                 .tos_opc_start  = QUOTA_DQACQ,
4614                 .tos_opc_end    = QUOTA_LAST_OPC,
4615                 .tos_hs         = mdt_quota_ops
4616         },
4617         {
4618                 .tos_opc_start  = LLOG_FIRST_OPC,
4619                 .tos_opc_end    = LLOG_LAST_OPC,
4620                 .tos_hs         = tgt_llog_handlers
4621         },
4622         {
4623                 .tos_opc_start  = LFSCK_FIRST_OPC,
4624                 .tos_opc_end    = LFSCK_LAST_OPC,
4625                 .tos_hs         = tgt_lfsck_handlers
4626         },
4627
4628         {
4629                 .tos_hs         = NULL
4630         }
4631 };
4632
4633 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4634 {
4635         struct md_device *next = m->mdt_child;
4636         struct lu_device *d = &m->mdt_lu_dev;
4637         struct obd_device *obd = mdt2obd_dev(m);
4638         struct lfsck_stop stop;
4639
4640         ENTRY;
4641         stop.ls_status = LS_PAUSED;
4642         stop.ls_flags = 0;
4643         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
4644
4645         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
4646         ping_evictor_stop();
4647
4648         if (m->mdt_opts.mo_coordinator)
4649                 mdt_hsm_cdt_stop(m);
4650
4651         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
4652         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4653
4654         if (m->mdt_namespace != NULL)
4655                 ldlm_namespace_free_prior(m->mdt_namespace, NULL,
4656                                           d->ld_obd->obd_force);
4657
4658         obd_exports_barrier(obd);
4659         obd_zombie_barrier();
4660
4661         mdt_quota_fini(env, m);
4662
4663         cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
4664
4665         /* Calling the cleanup functions in the same order as in the mdt_init0
4666          * error path
4667          */
4668         mdt_procfs_fini(m);
4669
4670         target_recovery_fini(obd);
4671         upcall_cache_cleanup(m->mdt_identity_cache);
4672         m->mdt_identity_cache = NULL;
4673
4674         mdt_fs_cleanup(env, m);
4675
4676         tgt_fini(env, &m->mdt_lut);
4677
4678         mdt_hsm_cdt_fini(m);
4679
4680         if (m->mdt_namespace != NULL) {
4681                 ldlm_namespace_free_post(m->mdt_namespace);
4682                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4683         }
4684
4685         if (m->mdt_md_root != NULL) {
4686                 mdt_object_put(env, m->mdt_md_root);
4687                 m->mdt_md_root = NULL;
4688         }
4689
4690         mdt_seq_fini(env, m);
4691
4692         mdt_fld_fini(env, m);
4693
4694         /*
4695          * Finish the stack
4696          */
4697         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4698
4699         LASSERT(atomic_read(&d->ld_ref) == 0);
4700
4701         server_put_mount(mdt_obd_name(m), true);
4702
4703         EXIT;
4704 }
4705
4706 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4707
4708 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4709                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
4710 {
4711         struct mdt_thread_info    *info;
4712         struct obd_device         *obd;
4713         const char                *dev = lustre_cfg_string(cfg, 0);
4714         const char                *num = lustre_cfg_string(cfg, 2);
4715         struct lustre_mount_info  *lmi = NULL;
4716         struct lustre_sb_info     *lsi;
4717         struct lu_site            *s;
4718         struct seq_server_site    *ss_site;
4719         const char                *identity_upcall = "NONE";
4720         struct md_device          *next;
4721         int                        rc;
4722         long                       node_id;
4723         mntopt_t                   mntopts;
4724         ENTRY;
4725
4726         lu_device_init(&m->mdt_lu_dev, ldt);
4727         /*
4728          * Environment (env) might be missing mdt_thread_key values at that
4729          * point, if device is allocated when mdt_thread_key is in QUIESCENT
4730          * mode.
4731          *
4732          * Usually device allocation path doesn't use module key values, but
4733          * mdt has to do a lot of work here, so allocate key value.
4734          */
4735         rc = lu_env_refill((struct lu_env *)env);
4736         if (rc != 0)
4737                 RETURN(rc);
4738
4739         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4740         LASSERT(info != NULL);
4741
4742         obd = class_name2obd(dev);
4743         LASSERT(obd != NULL);
4744
4745         m->mdt_max_mdsize = MAX_MD_SIZE; /* 4 stripes */
4746         m->mdt_opts.mo_evict_tgt_nids = 1;
4747         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4748
4749         /* default is coordinator off, it is started through conf_param
4750          * or /proc */
4751         m->mdt_opts.mo_coordinator = 0;
4752
4753         lmi = server_get_mount(dev);
4754         if (lmi == NULL) {
4755                 CERROR("Cannot get mount info for %s!\n", dev);
4756                 RETURN(-EFAULT);
4757         } else {
4758                 lsi = s2lsi(lmi->lmi_sb);
4759                 /* CMD is supported only in IAM mode */
4760                 LASSERT(num);
4761                 node_id = simple_strtol(num, NULL, 10);
4762                 obd->u.obt.obt_magic = OBT_MAGIC;
4763                 if (lsi->lsi_lmd != NULL &&
4764                     lsi->lsi_lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
4765                         m->mdt_skip_lfsck = 1;
4766         }
4767
4768         m->mdt_squash.rsi_uid = 0;
4769         m->mdt_squash.rsi_gid = 0;
4770         INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
4771         init_rwsem(&m->mdt_squash.rsi_sem);
4772         spin_lock_init(&m->mdt_lock);
4773         m->mdt_osfs_age = cfs_time_shift_64(-1000);
4774         m->mdt_enable_remote_dir = 0;
4775         m->mdt_enable_remote_dir_gid = 0;
4776
4777         atomic_set(&m->mdt_mds_mds_conns, 0);
4778         atomic_set(&m->mdt_async_commit_count, 0);
4779
4780         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
4781         m->mdt_lu_dev.ld_obd = obd;
4782         /* Set this lu_device to obd for error handling purposes. */
4783         obd->obd_lu_dev = &m->mdt_lu_dev;
4784
4785         /* init the stack */
4786         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
4787         if (rc) {
4788                 CERROR("%s: Can't init device stack, rc %d\n",
4789                        mdt_obd_name(m), rc);
4790                 GOTO(err_lmi, rc);
4791         }
4792
4793         s = mdt_lu_site(m);
4794         ss_site = mdt_seq_site(m);
4795         s->ld_seq_site = ss_site;
4796         ss_site->ss_lu = s;
4797
4798         /* set server index */
4799         ss_site->ss_node_id = node_id;
4800
4801         /* failover is the default
4802          * FIXME: we do not failout mds0/mgs, which may cause some problems.
4803          * assumed whose ss_node_id == 0 XXX
4804          * */
4805         obd->obd_replayable = 1;
4806         /* No connection accepted until configurations will finish */
4807         obd->obd_no_conn = 1;
4808
4809         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4810                 char *str = lustre_cfg_string(cfg, 4);
4811                 if (strchr(str, 'n')) {
4812                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
4813                         obd->obd_replayable = 0;
4814                 }
4815         }
4816
4817         rc = mdt_fld_init(env, mdt_obd_name(m), m);
4818         if (rc)
4819                 GOTO(err_fini_stack, rc);
4820
4821         rc = mdt_seq_init(env, m);
4822         if (rc)
4823                 GOTO(err_fini_fld, rc);
4824
4825         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
4826                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
4827         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4828                                               LDLM_NAMESPACE_SERVER,
4829                                               LDLM_NAMESPACE_GREEDY,
4830                                               LDLM_NS_TYPE_MDT);
4831         if (m->mdt_namespace == NULL)
4832                 GOTO(err_fini_seq, rc = -ENOMEM);
4833
4834         m->mdt_namespace->ns_lvbp = m;
4835         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
4836
4837         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4838         /* set obd_namespace for compatibility with old code */
4839         obd->obd_namespace = m->mdt_namespace;
4840
4841         rc = mdt_hsm_cdt_init(m);
4842         if (rc != 0) {
4843                 CERROR("%s: error initializing coordinator, rc %d\n",
4844                        mdt_obd_name(m), rc);
4845                 GOTO(err_free_ns, rc);
4846         }
4847
4848         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
4849                       OBD_FAIL_MDS_ALL_REQUEST_NET,
4850                       OBD_FAIL_MDS_ALL_REPLY_NET);
4851         if (rc)
4852                 GOTO(err_free_hsm, rc);
4853
4854         rc = mdt_fs_setup(env, m, obd, lsi);
4855         if (rc)
4856                 GOTO(err_tgt, rc);
4857
4858         tgt_adapt_sptlrpc_conf(&m->mdt_lut);
4859
4860         next = m->mdt_child;
4861         rc = next->md_ops->mdo_iocontrol(env, next, OBD_IOC_GET_MNTOPT, 0,
4862                                          &mntopts);
4863         if (rc)
4864                 GOTO(err_fs_cleanup, rc);
4865
4866         if (mntopts & MNTOPT_USERXATTR)
4867                 m->mdt_opts.mo_user_xattr = 1;
4868         else
4869                 m->mdt_opts.mo_user_xattr = 0;
4870
4871         rc = next->md_ops->mdo_maxeasize_get(env, next, &m->mdt_max_ea_size);
4872         if (rc)
4873                 GOTO(err_fs_cleanup, rc);
4874
4875         if (mntopts & MNTOPT_ACL)
4876                 m->mdt_opts.mo_acl = 1;
4877         else
4878                 m->mdt_opts.mo_acl = 0;
4879
4880         /* XXX: to support suppgid for ACL, we enable identity_upcall
4881          * by default, otherwise, maybe got unexpected -EACCESS. */
4882         if (m->mdt_opts.mo_acl)
4883                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4884
4885         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
4886                                                 identity_upcall,
4887                                                 &mdt_identity_upcall_cache_ops);
4888         if (IS_ERR(m->mdt_identity_cache)) {
4889                 rc = PTR_ERR(m->mdt_identity_cache);
4890                 m->mdt_identity_cache = NULL;
4891                 GOTO(err_fs_cleanup, rc);
4892         }
4893
4894         rc = mdt_procfs_init(m, dev);
4895         if (rc) {
4896                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4897                 GOTO(err_recovery, rc);
4898         }
4899
4900         rc = mdt_quota_init(env, m, cfg);
4901         if (rc)
4902                 GOTO(err_procfs, rc);
4903
4904         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
4905         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
4906                            "mdt_ldlm_client", m->mdt_ldlm_client);
4907
4908         ping_evictor_start();
4909
4910         /* recovery will be started upon mdt_prepare()
4911          * when the whole stack is complete and ready
4912          * to serve the requests */
4913
4914         /* Reduce the initial timeout on an MDS because it doesn't need such
4915          * a long timeout as an OST does. Adaptive timeouts will adjust this
4916          * value appropriately. */
4917         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4918                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4919
4920         RETURN(0);
4921 err_procfs:
4922         mdt_procfs_fini(m);
4923 err_recovery:
4924         target_recovery_fini(obd);
4925         upcall_cache_cleanup(m->mdt_identity_cache);
4926         m->mdt_identity_cache = NULL;
4927 err_fs_cleanup:
4928         mdt_fs_cleanup(env, m);
4929 err_tgt:
4930         tgt_fini(env, &m->mdt_lut);
4931 err_free_hsm:
4932         mdt_hsm_cdt_fini(m);
4933 err_free_ns:
4934         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4935         obd->obd_namespace = m->mdt_namespace = NULL;
4936 err_fini_seq:
4937         mdt_seq_fini(env, m);
4938 err_fini_fld:
4939         mdt_fld_fini(env, m);
4940 err_fini_stack:
4941         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4942 err_lmi:
4943         if (lmi)
4944                 server_put_mount(dev, true);
4945         return(rc);
4946 }
4947
4948 /* For interoperability, the left element is old parameter, the right one
4949  * is the new version of the parameter, if some parameter is deprecated,
4950  * the new version should be set as NULL. */
4951 static struct cfg_interop_param mdt_interop_param[] = {
4952         { "mdt.group_upcall",   NULL },
4953         { "mdt.quota_type",     NULL },
4954         { "mdd.quota_type",     NULL },
4955         { "mdt.som",            NULL },
4956         { "mdt.rootsquash",     "mdt.root_squash" },
4957         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
4958         { NULL }
4959 };
4960
4961 /* used by MGS to process specific configurations */
4962 static int mdt_process_config(const struct lu_env *env,
4963                               struct lu_device *d, struct lustre_cfg *cfg)
4964 {
4965         struct mdt_device *m = mdt_dev(d);
4966         struct md_device *md_next = m->mdt_child;
4967         struct lu_device *next = md2lu_dev(md_next);
4968         int rc;
4969         ENTRY;
4970
4971         switch (cfg->lcfg_command) {
4972         case LCFG_PARAM: {
4973                 struct obd_device          *obd = d->ld_obd;
4974
4975                 /* For interoperability */
4976                 struct cfg_interop_param   *ptr = NULL;
4977                 struct lustre_cfg          *old_cfg = NULL;
4978                 char                       *param = NULL;
4979
4980                 param = lustre_cfg_string(cfg, 1);
4981                 if (param == NULL) {
4982                         CERROR("param is empty\n");
4983                         rc = -EINVAL;
4984                         break;
4985                 }
4986
4987                 ptr = class_find_old_param(param, mdt_interop_param);
4988                 if (ptr != NULL) {
4989                         if (ptr->new_param == NULL) {
4990                                 rc = 0;
4991                                 CWARN("For interoperability, skip this %s."
4992                                       " It is obsolete.\n", ptr->old_param);
4993                                 break;
4994                         }
4995
4996                         CWARN("Found old param %s, changed it to %s.\n",
4997                               ptr->old_param, ptr->new_param);
4998
4999                         old_cfg = cfg;
5000                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
5001                         if (IS_ERR(cfg)) {
5002                                 rc = PTR_ERR(cfg);
5003                                 break;
5004                         }
5005                 }
5006
5007                 rc = class_process_proc_param(PARAM_MDT, obd->obd_vars,
5008                                               cfg, obd);
5009                 if (rc > 0 || rc == -ENOSYS) {
5010                         /* is it an HSM var ? */
5011                         rc = class_process_proc_param(PARAM_HSM,
5012                                                       hsm_cdt_get_proc_vars(),
5013                                                       cfg, obd);
5014                         if (rc > 0 || rc == -ENOSYS)
5015                                 /* we don't understand; pass it on */
5016                                 rc = next->ld_ops->ldo_process_config(env, next,
5017                                                                       cfg);
5018                 }
5019
5020                 if (old_cfg != NULL)
5021                         lustre_cfg_free(cfg);
5022
5023                 break;
5024         }
5025         default:
5026                 /* others are passed further */
5027                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
5028                 break;
5029         }
5030         RETURN(rc);
5031 }
5032
5033 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
5034                                           const struct lu_object_header *hdr,
5035                                           struct lu_device *d)
5036 {
5037         struct mdt_object *mo;
5038
5039         ENTRY;
5040
5041         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
5042         if (mo != NULL) {
5043                 struct lu_object *o;
5044                 struct lu_object_header *h;
5045
5046                 o = &mo->mot_obj;
5047                 h = &mo->mot_header;
5048                 lu_object_header_init(h);
5049                 lu_object_init(o, h, d);
5050                 lu_object_add_top(h, o);
5051                 o->lo_ops = &mdt_obj_ops;
5052                 spin_lock_init(&mo->mot_write_lock);
5053                 mutex_init(&mo->mot_lov_mutex);
5054                 init_rwsem(&mo->mot_open_sem);
5055                 atomic_set(&mo->mot_open_count, 0);
5056                 RETURN(o);
5057         }
5058         RETURN(NULL);
5059 }
5060
5061 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
5062                            const struct lu_object_conf *unused)
5063 {
5064         struct mdt_device *d = mdt_dev(o->lo_dev);
5065         struct lu_device  *under;
5066         struct lu_object  *below;
5067         int                rc = 0;
5068         ENTRY;
5069
5070         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
5071                PFID(lu_object_fid(o)));
5072
5073         under = &d->mdt_child->md_lu_dev;
5074         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
5075         if (below != NULL) {
5076                 lu_object_add(o, below);
5077         } else
5078                 rc = -ENOMEM;
5079
5080         RETURN(rc);
5081 }
5082
5083 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
5084 {
5085         struct mdt_object *mo = mdt_obj(o);
5086         struct lu_object_header *h;
5087         ENTRY;
5088
5089         h = o->lo_header;
5090         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
5091                PFID(lu_object_fid(o)));
5092
5093         LASSERT(atomic_read(&mo->mot_open_count) == 0);
5094         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
5095
5096         lu_object_fini(o);
5097         lu_object_header_fini(h);
5098         OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
5099
5100         EXIT;
5101 }
5102
5103 static int mdt_object_print(const struct lu_env *env, void *cookie,
5104                             lu_printer_t p, const struct lu_object *o)
5105 {
5106         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
5107
5108         return (*p)(env, cookie,
5109                     LUSTRE_MDT_NAME"-object@%p(%s %s, writecount=%d)",
5110                     mdto, mdto->mot_lov_created ? "lov_created" : "",
5111                     mdto->mot_cache_attr ? "cache_attr" : "",
5112                     mdto->mot_write_count);
5113 }
5114
5115 static int mdt_prepare(const struct lu_env *env,
5116                 struct lu_device *pdev,
5117                 struct lu_device *cdev)
5118 {
5119         struct mdt_device *mdt = mdt_dev(cdev);
5120         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
5121         struct obd_device *obd = cdev->ld_obd;
5122         int rc;
5123
5124         ENTRY;
5125
5126         LASSERT(obd);
5127
5128         rc = next->ld_ops->ldo_prepare(env, cdev, next);
5129         if (rc)
5130                 RETURN(rc);
5131
5132         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
5133         if (rc)
5134                 RETURN(rc);
5135
5136         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
5137         if (rc)
5138                 RETURN(rc);
5139
5140         rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
5141         /* The LFSCK instance is registered just now, so it must be there when
5142          * register the namespace to such instance. */
5143         LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
5144
5145         if (mdt->mdt_seq_site.ss_node_id == 0) {
5146                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
5147                                                          &mdt->mdt_md_root_fid);
5148                 if (rc)
5149                         RETURN(rc);
5150         }
5151
5152         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
5153
5154         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
5155         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
5156         LASSERT(obd->obd_no_conn);
5157         spin_lock(&obd->obd_dev_lock);
5158         obd->obd_no_conn = 0;
5159         spin_unlock(&obd->obd_dev_lock);
5160
5161         if (obd->obd_recovering == 0)
5162                 mdt_postrecov(env, mdt);
5163
5164         RETURN(rc);
5165 }
5166
5167 const struct lu_device_operations mdt_lu_ops = {
5168         .ldo_object_alloc   = mdt_object_alloc,
5169         .ldo_process_config = mdt_process_config,
5170         .ldo_prepare        = mdt_prepare,
5171 };
5172
5173 static const struct lu_object_operations mdt_obj_ops = {
5174         .loo_object_init    = mdt_object_init,
5175         .loo_object_free    = mdt_object_free,
5176         .loo_object_print   = mdt_object_print
5177 };
5178
5179 static int mdt_obd_set_info_async(const struct lu_env *env,
5180                                   struct obd_export *exp,
5181                                   __u32 keylen, void *key,
5182                                   __u32 vallen, void *val,
5183                                   struct ptlrpc_request_set *set)
5184 {
5185         int rc;
5186
5187         ENTRY;
5188
5189         if (KEY_IS(KEY_SPTLRPC_CONF)) {
5190                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp));
5191                 RETURN(rc);
5192         }
5193
5194         RETURN(0);
5195 }
5196
5197 /**
5198  * Match client and server connection feature flags.
5199  *
5200  * Compute the compatibility flags for a connection request based on
5201  * features mutually supported by client and server.
5202  *
5203  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
5204  * must not be updated here, otherwise a partially initialized value may
5205  * be exposed. After the connection request is successfully processed,
5206  * the top-level MDT connect request handler atomically updates the export
5207  * connect flags from the obd_connect_data::ocd_connect_flags field of the
5208  * reply. \see mdt_connect().
5209  *
5210  * Before 2.7.50 clients will send a struct obd_connect_data_v1 rather than a
5211  * full struct obd_connect_data. So care must be taken when accessing fields
5212  * that are not present in struct obd_connect_data_v1. See LU-16.
5213  *
5214  * \param exp   the obd_export associated with this client/target pair
5215  * \param mdt   the target device for the connection
5216  * \param data  stores data for this connect request
5217  *
5218  * \retval 0       success
5219  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
5220  * \retval -EBADE  client and server feature requirements are incompatible
5221  */
5222 static int mdt_connect_internal(struct obd_export *exp,
5223                                 struct mdt_device *mdt,
5224                                 struct obd_connect_data *data)
5225 {
5226         LASSERT(data != NULL);
5227
5228         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
5229
5230         if (mdt->mdt_bottom->dd_rdonly &&
5231             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5232             !(data->ocd_connect_flags & OBD_CONNECT_RDONLY))
5233                 RETURN(-EACCES);
5234
5235         if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
5236                 data->ocd_connect_flags2 &= MDT_CONNECT_SUPPORTED2;
5237
5238         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
5239
5240         if (!(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5241             !(data->ocd_connect_flags & OBD_CONNECT_IBITS)) {
5242                 CWARN("%s: client %s does not support ibits lock, either "
5243                       "very old or an invalid client: flags %#llx\n",
5244                       mdt_obd_name(mdt), exp->exp_client_uuid.uuid,
5245                       data->ocd_connect_flags);
5246                 return -EBADE;
5247         }
5248
5249         if (!mdt->mdt_opts.mo_acl)
5250                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
5251
5252         if (!mdt->mdt_opts.mo_user_xattr)
5253                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
5254
5255         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
5256                 data->ocd_brw_size = min(data->ocd_brw_size,
5257                                          (__u32)MD_MAX_BRW_SIZE);
5258                 if (data->ocd_brw_size == 0) {
5259                         CERROR("%s: cli %s/%p ocd_connect_flags: %#llx"
5260                                " ocd_version: %x ocd_grant: %d "
5261                                "ocd_index: %u ocd_brw_size is "
5262                                "unexpectedly zero, network data "
5263                                "corruption? Refusing connection of this"
5264                                " client\n",
5265                                mdt_obd_name(mdt),
5266                                exp->exp_client_uuid.uuid,
5267                                exp, data->ocd_connect_flags, data->ocd_version,
5268                                data->ocd_grant, data->ocd_index);
5269                         return -EPROTO;
5270                 }
5271         }
5272
5273         /* NB: Disregard the rule against updating
5274          * exp_connect_data.ocd_connect_flags in this case, since
5275          * tgt_client_new() needs to know if this is a lightweight
5276          * connection, and it is safe to expose this flag before
5277          * connection processing completes. */
5278         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
5279                 spin_lock(&exp->exp_lock);
5280                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
5281                 spin_unlock(&exp->exp_lock);
5282         }
5283
5284         data->ocd_version = LUSTRE_VERSION_CODE;
5285
5286         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
5287                 CWARN("%s: MDS requires FID support, but client not\n",
5288                       mdt_obd_name(mdt));
5289                 return -EBADE;
5290         }
5291
5292         if (OCD_HAS_FLAG(data, PINGLESS)) {
5293                 if (ptlrpc_pinger_suppress_pings()) {
5294                         spin_lock(&exp->exp_obd->obd_dev_lock);
5295                         list_del_init(&exp->exp_obd_chain_timed);
5296                         spin_unlock(&exp->exp_obd->obd_dev_lock);
5297                 } else {
5298                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
5299                 }
5300         }
5301
5302         data->ocd_max_easize = mdt->mdt_max_ea_size;
5303
5304         /* NB: Disregard the rule against updating
5305          * exp_connect_data.ocd_connect_flags in this case, since
5306          * tgt_client_new() needs to know if this is client supports
5307          * multiple modify RPCs, and it is safe to expose this flag before
5308          * connection processing completes. */
5309         if (data->ocd_connect_flags & OBD_CONNECT_MULTIMODRPCS) {
5310                 data->ocd_maxmodrpcs = max_mod_rpcs_per_client;
5311                 spin_lock(&exp->exp_lock);
5312                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_MULTIMODRPCS;
5313                 spin_unlock(&exp->exp_lock);
5314         }
5315
5316         return 0;
5317 }
5318
5319 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
5320                                    struct mdt_thread_info *info,
5321                                    struct mdt_file_data *mfd)
5322 {
5323         struct lu_context ses;
5324         int rc;
5325         ENTRY;
5326
5327         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
5328         if (rc)
5329                 RETURN(rc);
5330
5331         env->le_ses = &ses;
5332         lu_context_enter(&ses);
5333
5334         mdt_ucred(info)->uc_valid = UCRED_OLD;
5335         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
5336
5337         lu_context_exit(&ses);
5338         lu_context_fini(&ses);
5339         env->le_ses = NULL;
5340
5341         RETURN(rc);
5342 }
5343
5344 static int mdt_export_cleanup(struct obd_export *exp)
5345 {
5346         struct list_head         closing_list;
5347         struct mdt_export_data  *med = &exp->exp_mdt_data;
5348         struct obd_device       *obd = exp->exp_obd;
5349         struct mdt_device       *mdt;
5350         struct mdt_thread_info  *info;
5351         struct lu_env            env;
5352         struct mdt_file_data    *mfd, *n;
5353         int rc = 0;
5354         ENTRY;
5355
5356         INIT_LIST_HEAD(&closing_list);
5357         spin_lock(&med->med_open_lock);
5358         while (!list_empty(&med->med_open_head)) {
5359                 struct list_head *tmp = med->med_open_head.next;
5360                 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
5361
5362                 /* Remove mfd handle so it can't be found again.
5363                  * We are consuming the mfd_list reference here. */
5364                 class_handle_unhash(&mfd->mfd_handle);
5365                 list_move_tail(&mfd->mfd_list, &closing_list);
5366         }
5367         spin_unlock(&med->med_open_lock);
5368         mdt = mdt_dev(obd->obd_lu_dev);
5369         LASSERT(mdt != NULL);
5370
5371         rc = lu_env_init(&env, LCT_MD_THREAD);
5372         if (rc)
5373                 RETURN(rc);
5374
5375         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5376         LASSERT(info != NULL);
5377         memset(info, 0, sizeof *info);
5378         info->mti_env = &env;
5379         info->mti_mdt = mdt;
5380         info->mti_exp = exp;
5381
5382         if (!list_empty(&closing_list)) {
5383                 struct md_attr *ma = &info->mti_attr;
5384
5385                 /* Close any open files (which may also cause orphan
5386                  * unlinking). */
5387                 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5388                         list_del_init(&mfd->mfd_list);
5389                         ma->ma_need = ma->ma_valid = 0;
5390
5391                         /* This file is being closed due to an eviction, it
5392                          * could have been modified and now dirty regarding to
5393                          * HSM archive, check this!
5394                          * The logic here is to mark a file dirty if there's a
5395                          * chance it was dirtied before the client was evicted,
5396                          * so that we don't have to wait for a release attempt
5397                          * before finding out the file was actually dirty and
5398                          * fail the release. Aggressively marking it dirty here
5399                          * will cause the policy engine to attempt to
5400                          * re-archive it; when rearchiving, we can compare the
5401                          * current version to the HSM data_version and make the
5402                          * archive request into a noop if it's not actually
5403                          * dirty.
5404                          */
5405                         if (mfd->mfd_mode & FMODE_WRITE)
5406                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
5407
5408                         /* Don't unlink orphan on failover umount, LU-184 */
5409                         if (exp->exp_flags & OBD_OPT_FAILOVER) {
5410                                 ma->ma_valid = MA_FLAGS;
5411                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5412                         }
5413                         mdt_mfd_close(info, mfd);
5414                 }
5415         }
5416         info->mti_mdt = NULL;
5417         /* cleanup client slot early */
5418         /* Do not erase record for recoverable client. */
5419         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5420                 tgt_client_del(&env, exp);
5421         lu_env_fini(&env);
5422
5423         RETURN(rc);
5424 }
5425
5426 static inline void mdt_enable_slc(struct mdt_device *mdt)
5427 {
5428         if (mdt->mdt_lut.lut_sync_lock_cancel == NEVER_SYNC_ON_CANCEL)
5429                 mdt->mdt_lut.lut_sync_lock_cancel = BLOCKING_SYNC_ON_CANCEL;
5430 }
5431
5432 static inline void mdt_disable_slc(struct mdt_device *mdt)
5433 {
5434         if (mdt->mdt_lut.lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL)
5435                 mdt->mdt_lut.lut_sync_lock_cancel = NEVER_SYNC_ON_CANCEL;
5436 }
5437
5438 static int mdt_obd_disconnect(struct obd_export *exp)
5439 {
5440         int rc;
5441         ENTRY;
5442
5443         LASSERT(exp);
5444         class_export_get(exp);
5445
5446         if ((exp_connect_flags(exp) & OBD_CONNECT_MDS_MDS) &&
5447             !(exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT)) {
5448                 struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
5449
5450                 if (atomic_dec_and_test(&mdt->mdt_mds_mds_conns))
5451                         mdt_disable_slc(mdt);
5452         }
5453
5454         rc = server_disconnect_export(exp);
5455         if (rc != 0)
5456                 CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
5457
5458         rc = mdt_export_cleanup(exp);
5459         nodemap_del_member(exp);
5460         class_export_put(exp);
5461         RETURN(rc);
5462 }
5463
5464 /* mds_connect copy */
5465 static int mdt_obd_connect(const struct lu_env *env,
5466                            struct obd_export **exp, struct obd_device *obd,
5467                            struct obd_uuid *cluuid,
5468                            struct obd_connect_data *data,
5469                            void *localdata)
5470 {
5471         struct obd_export       *lexp;
5472         struct lustre_handle    conn = { 0 };
5473         struct mdt_device       *mdt;
5474         int                      rc;
5475         lnet_nid_t              *client_nid = localdata;
5476         ENTRY;
5477
5478         LASSERT(env != NULL);
5479         LASSERT(data != NULL);
5480
5481         if (!exp || !obd || !cluuid)
5482                 RETURN(-EINVAL);
5483
5484         mdt = mdt_dev(obd->obd_lu_dev);
5485
5486         if ((data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5487             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
5488                 atomic_inc(&mdt->mdt_mds_mds_conns);
5489                 mdt_enable_slc(mdt);
5490         }
5491
5492         /*
5493          * first, check whether the stack is ready to handle requests
5494          * XXX: probably not very appropriate method is used now
5495          *      at some point we should find a better one
5496          */
5497         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) &&
5498             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) &&
5499             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
5500                 rc = obd_get_info(env, mdt->mdt_child_exp,
5501                                   sizeof(KEY_OSP_CONNECTED),
5502                                   KEY_OSP_CONNECTED, NULL, NULL);
5503                 if (rc)
5504                         RETURN(-EAGAIN);
5505                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
5506         }
5507
5508         rc = class_connect(&conn, obd, cluuid);
5509         if (rc)
5510                 RETURN(rc);
5511
5512         lexp = class_conn2export(&conn);
5513         LASSERT(lexp != NULL);
5514
5515         rc = nodemap_add_member(*client_nid, lexp);
5516         if (rc != 0 && rc != -EEXIST)
5517                 GOTO(out, rc);
5518
5519         rc = mdt_connect_internal(lexp, mdt, data);
5520         if (rc == 0) {
5521                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5522
5523                 LASSERT(lcd);
5524                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5525                 rc = tgt_client_new(env, lexp);
5526                 if (rc == 0)
5527                         mdt_export_stats_init(obd, lexp, localdata);
5528         }
5529 out:
5530         if (rc != 0) {
5531                 class_disconnect(lexp);
5532                 nodemap_del_member(lexp);
5533                 *exp = NULL;
5534         } else {
5535                 *exp = lexp;
5536                 /* Because we do not want this export to be evicted by pinger,
5537                  * let's not add this export to the timed chain list. */
5538                 if (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) {
5539                         spin_lock(&lexp->exp_obd->obd_dev_lock);
5540                         list_del_init(&lexp->exp_obd_chain_timed);
5541                         spin_unlock(&lexp->exp_obd->obd_dev_lock);
5542                 }
5543         }
5544
5545         RETURN(rc);
5546 }
5547
5548 static int mdt_obd_reconnect(const struct lu_env *env,
5549                              struct obd_export *exp, struct obd_device *obd,
5550                              struct obd_uuid *cluuid,
5551                              struct obd_connect_data *data,
5552                              void *localdata)
5553 {
5554         lnet_nid_t             *client_nid = localdata;
5555         int                     rc;
5556         ENTRY;
5557
5558         if (exp == NULL || obd == NULL || cluuid == NULL)
5559                 RETURN(-EINVAL);
5560
5561         rc = nodemap_add_member(*client_nid, exp);
5562         if (rc != 0 && rc != -EEXIST)
5563                 RETURN(rc);
5564
5565         rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
5566         if (rc == 0)
5567                 mdt_export_stats_init(obd, exp, localdata);
5568         else
5569                 nodemap_del_member(exp);
5570
5571         RETURN(rc);
5572 }
5573
5574 /* FIXME: Can we avoid using these two interfaces? */
5575 static int mdt_init_export(struct obd_export *exp)
5576 {
5577         struct mdt_export_data *med = &exp->exp_mdt_data;
5578         int                     rc;
5579         ENTRY;
5580
5581         INIT_LIST_HEAD(&med->med_open_head);
5582         spin_lock_init(&med->med_open_lock);
5583         spin_lock(&exp->exp_lock);
5584         exp->exp_connecting = 1;
5585         spin_unlock(&exp->exp_lock);
5586
5587         /* self-export doesn't need client data and ldlm initialization */
5588         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5589                                      &exp->exp_client_uuid)))
5590                 RETURN(0);
5591
5592         rc = tgt_client_alloc(exp);
5593         if (rc)
5594                 GOTO(err, rc);
5595
5596         rc = ldlm_init_export(exp);
5597         if (rc)
5598                 GOTO(err_free, rc);
5599
5600         RETURN(rc);
5601
5602 err_free:
5603         tgt_client_free(exp);
5604 err:
5605         CERROR("%s: Failed to initialize export: rc = %d\n",
5606                exp->exp_obd->obd_name, rc);
5607         return rc;
5608 }
5609
5610 static int mdt_destroy_export(struct obd_export *exp)
5611 {
5612         ENTRY;
5613
5614         target_destroy_export(exp);
5615         /* destroy can be called from failed obd_setup, so
5616          * checking uuid is safer than obd_self_export */
5617         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5618                                      &exp->exp_client_uuid)))
5619                 RETURN(0);
5620
5621         ldlm_destroy_export(exp);
5622         tgt_client_free(exp);
5623
5624         LASSERT(list_empty(&exp->exp_outstanding_replies));
5625         LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
5626
5627         RETURN(0);
5628 }
5629
5630 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
5631                    struct linkea_data *ldata)
5632 {
5633         int rc;
5634
5635         LASSERT(ldata->ld_buf->lb_buf != NULL);
5636
5637         if (!mdt_object_exists(mdt_obj))
5638                 return -ENODATA;
5639
5640         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5641                           ldata->ld_buf, XATTR_NAME_LINK);
5642         if (rc == -ERANGE) {
5643                 /* Buf was too small, figure out what we need. */
5644                 lu_buf_free(ldata->ld_buf);
5645                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5646                                   ldata->ld_buf, XATTR_NAME_LINK);
5647                 if (rc < 0)
5648                         return rc;
5649                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
5650                 if (ldata->ld_buf->lb_buf == NULL)
5651                         return -ENOMEM;
5652                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5653                                   ldata->ld_buf, XATTR_NAME_LINK);
5654         }
5655         if (rc < 0)
5656                 return rc;
5657
5658         return linkea_init_with_rec(ldata);
5659 }
5660
5661 /**
5662  * Given an MDT object, try to look up the full path to the object.
5663  * Part of the MDT layer implementation of lfs fid2path.
5664  *
5665  * \param[in]     info  Per-thread common data shared by MDT level handlers.
5666  * \param[in]     obj   Object to do path lookup of
5667  * \param[in,out] fp    User-provided struct to store path information
5668  * \param[in]     root_fid Root FID of current path should reach
5669  *
5670  * \retval 0 Lookup successful, path information stored in fp
5671  * \retval -EAGAIN Lookup failed, usually because object is being moved
5672  * \retval negative errno if there was a problem
5673  */
5674 static int mdt_path_current(struct mdt_thread_info *info,
5675                             struct mdt_object *obj,
5676                             struct getinfo_fid2path *fp,
5677                             struct lu_fid *root_fid)
5678 {
5679         struct mdt_device       *mdt = info->mti_mdt;
5680         struct mdt_object       *mdt_obj;
5681         struct link_ea_header   *leh;
5682         struct link_ea_entry    *lee;
5683         struct lu_name          *tmpname = &info->mti_name;
5684         struct lu_fid           *tmpfid = &info->mti_tmp_fid1;
5685         struct lu_buf           *buf = &info->mti_big_buf;
5686         char                    *ptr;
5687         int                     reclen;
5688         struct linkea_data      ldata = { NULL };
5689         int                     rc = 0;
5690         bool                    first = true;
5691         ENTRY;
5692
5693         /* temp buffer for path element, the buffer will be finally freed
5694          * in mdt_thread_info_fini */
5695         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
5696         if (buf->lb_buf == NULL)
5697                 RETURN(-ENOMEM);
5698
5699         ldata.ld_buf = buf;
5700         ptr = fp->gf_u.gf_path + fp->gf_pathlen - 1;
5701         *ptr = 0;
5702         --ptr;
5703         *tmpfid = fp->gf_fid = *mdt_object_fid(obj);
5704
5705         while (!lu_fid_eq(root_fid, &fp->gf_fid)) {
5706                 struct lu_buf           lmv_buf;
5707
5708                 if (!lu_fid_eq(root_fid, &mdt->mdt_md_root_fid) &&
5709                     lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid))
5710                         GOTO(out, rc = -ENOENT);
5711
5712                 if (lu_fid_eq(mdt_object_fid(obj), tmpfid)) {
5713                         mdt_obj = obj;
5714                         mdt_object_get(info->mti_env, mdt_obj);
5715                 } else {
5716                         mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
5717                         if (IS_ERR(mdt_obj))
5718                                 GOTO(out, rc = PTR_ERR(mdt_obj));
5719                 }
5720
5721                 if (!mdt_object_exists(mdt_obj)) {
5722                         mdt_object_put(info->mti_env, mdt_obj);
5723                         GOTO(out, rc = -ENOENT);
5724                 }
5725
5726                 if (mdt_object_remote(mdt_obj)) {
5727                         mdt_object_put(info->mti_env, mdt_obj);
5728                         GOTO(remote_out, rc = -EREMOTE);
5729                 }
5730
5731                 rc = mdt_links_read(info, mdt_obj, &ldata);
5732                 if (rc != 0) {
5733                         mdt_object_put(info->mti_env, mdt_obj);
5734                         GOTO(out, rc);
5735                 }
5736
5737                 leh = buf->lb_buf;
5738                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
5739                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
5740                 /* If set, use link #linkno for path lookup, otherwise use
5741                    link #0.  Only do this for the final path element. */
5742                 if (first && fp->gf_linkno < leh->leh_reccount) {
5743                         int count;
5744                         for (count = 0; count < fp->gf_linkno; count++) {
5745                                 lee = (struct link_ea_entry *)
5746                                      ((char *)lee + reclen);
5747                                 linkea_entry_unpack(lee, &reclen, tmpname,
5748                                                     tmpfid);
5749                         }
5750                         if (fp->gf_linkno < leh->leh_reccount - 1)
5751                                 /* indicate to user there are more links */
5752                                 fp->gf_linkno++;
5753                 }
5754
5755                 lmv_buf.lb_buf = info->mti_xattr_buf;
5756                 lmv_buf.lb_len = sizeof(info->mti_xattr_buf);
5757                 /* Check if it is slave stripes */
5758                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5759                                   &lmv_buf, XATTR_NAME_LMV);
5760                 mdt_object_put(info->mti_env, mdt_obj);
5761                 if (rc > 0) {
5762                         union lmv_mds_md *lmm = lmv_buf.lb_buf;
5763
5764                         /* For slave stripes, get its master */
5765                         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE) {
5766                                 fp->gf_fid = *tmpfid;
5767                                 continue;
5768                         }
5769                 } else if (rc < 0 && rc != -ENODATA) {
5770                         GOTO(out, rc);
5771                 }
5772
5773                 rc = 0;
5774
5775                 /* Pack the name in the end of the buffer */
5776                 ptr -= tmpname->ln_namelen;
5777                 if (ptr - 1 <= fp->gf_u.gf_path)
5778                         GOTO(out, rc = -EOVERFLOW);
5779                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
5780                 *(--ptr) = '/';
5781
5782                 /* keep the last resolved fid to the client, so the
5783                  * client will build the left path on another MDT for
5784                  * remote object */
5785                 fp->gf_fid = *tmpfid;
5786
5787                 first = false;
5788         }
5789
5790 remote_out:
5791         ptr++; /* skip leading / */
5792         memmove(fp->gf_u.gf_path, ptr,
5793                 fp->gf_u.gf_path + fp->gf_pathlen - ptr);
5794
5795 out:
5796         RETURN(rc);
5797 }
5798
5799 /**
5800  * Given an MDT object, use mdt_path_current to get the path.
5801  * Essentially a wrapper to retry mdt_path_current a set number of times
5802  * if -EAGAIN is returned (usually because an object is being moved).
5803  *
5804  * Part of the MDT layer implementation of lfs fid2path.
5805  *
5806  * \param[in]     info  Per-thread common data shared by mdt level handlers.
5807  * \param[in]     obj   Object to do path lookup of
5808  * \param[in,out] fp    User-provided struct for arguments and to store path
5809  *                      information
5810  *
5811  * \retval 0 Lookup successful, path information stored in fp
5812  * \retval negative errno if there was a problem
5813  */
5814 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
5815                     struct getinfo_fid2path *fp, struct lu_fid *root_fid)
5816 {
5817         struct mdt_device       *mdt = info->mti_mdt;
5818         int                     tries = 3;
5819         int                     rc = -EAGAIN;
5820         ENTRY;
5821
5822         if (fp->gf_pathlen < 3)
5823                 RETURN(-EOVERFLOW);
5824
5825         if (root_fid == NULL)
5826                 root_fid = &mdt->mdt_md_root_fid;
5827
5828         if (lu_fid_eq(root_fid, mdt_object_fid(obj))) {
5829                 fp->gf_u.gf_path[0] = '\0';
5830                 RETURN(0);
5831         }
5832
5833         /* Retry multiple times in case file is being moved */
5834         while (tries-- && rc == -EAGAIN)
5835                 rc = mdt_path_current(info, obj, fp, root_fid);
5836
5837         RETURN(rc);
5838 }
5839
5840 /**
5841  * Get the full path of the provided FID, as of changelog record recno.
5842  *
5843  * This checks sanity and looks up object for user provided FID
5844  * before calling the actual path lookup code.
5845  *
5846  * Part of the MDT layer implementation of lfs fid2path.
5847  *
5848  * \param[in]     info  Per-thread common data shared by mdt level handlers.
5849  * \param[in,out] fp    User-provided struct for arguments and to store path
5850  *                      information
5851  *
5852  * \retval 0 Lookup successful, path information and recno stored in fp
5853  * \retval -ENOENT, object does not exist
5854  * \retval negative errno if there was a problem
5855  */
5856 static int mdt_fid2path(struct mdt_thread_info *info,
5857                         struct lu_fid *root_fid,
5858                         struct getinfo_fid2path *fp)
5859 {
5860         struct mdt_device *mdt = info->mti_mdt;
5861         struct mdt_object *obj;
5862         int    rc;
5863         ENTRY;
5864
5865         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
5866                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
5867
5868         if (!fid_is_sane(&fp->gf_fid))
5869                 RETURN(-EINVAL);
5870
5871         if (!fid_is_namespace_visible(&fp->gf_fid)) {
5872                 CDEBUG(D_INFO, "%s: "DFID" is invalid, f_seq should be >= %#llx"
5873                        ", or f_oid != 0, or f_ver == 0\n", mdt_obd_name(mdt),
5874                        PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
5875                 RETURN(-EINVAL);
5876         }
5877
5878         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
5879         if (IS_ERR(obj)) {
5880                 rc = PTR_ERR(obj);
5881                 CDEBUG(D_IOCTL, "cannot find "DFID": rc = %d\n",
5882                        PFID(&fp->gf_fid), rc);
5883                 RETURN(rc);
5884         }
5885
5886         if (mdt_object_remote(obj))
5887                 rc = -EREMOTE;
5888         else if (!mdt_object_exists(obj))
5889                 rc = -ENOENT;
5890         else
5891                 rc = 0;
5892
5893         if (rc < 0) {
5894                 mdt_object_put(info->mti_env, obj);
5895                 CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n",
5896                        PFID(&fp->gf_fid), rc);
5897                 RETURN(rc);
5898         }
5899
5900         rc = mdt_path(info, obj, fp, root_fid);
5901
5902         CDEBUG(D_INFO, "fid "DFID", path %s recno %#llx linkno %u\n",
5903                PFID(&fp->gf_fid), fp->gf_u.gf_path,
5904                fp->gf_recno, fp->gf_linkno);
5905
5906         mdt_object_put(info->mti_env, obj);
5907
5908         RETURN(rc);
5909 }
5910
5911 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, int keylen,
5912                             void *val, int vallen)
5913 {
5914         struct getinfo_fid2path *fpout, *fpin;
5915         struct lu_fid *root_fid = NULL;
5916         int rc = 0;
5917
5918         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
5919         fpout = val;
5920
5921         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5922                 lustre_swab_fid2path(fpin);
5923
5924         memcpy(fpout, fpin, sizeof(*fpin));
5925         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
5926                 RETURN(-EINVAL);
5927
5928         if (keylen >= cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*fpin) +
5929                       sizeof(struct lu_fid)) {
5930                 /* client sent its root FID, which is normally fileset FID */
5931                 root_fid = fpin->gf_u.gf_root_fid;
5932                 if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5933                         lustre_swab_lu_fid(root_fid);
5934
5935                 if (root_fid != NULL && !fid_is_sane(root_fid))
5936                         RETURN(-EINVAL);
5937         }
5938
5939         rc = mdt_fid2path(info, root_fid, fpout);
5940         RETURN(rc);
5941 }
5942
5943 int mdt_get_info(struct tgt_session_info *tsi)
5944 {
5945         char    *key;
5946         int      keylen;
5947         __u32   *vallen;
5948         void    *valout;
5949         int      rc;
5950
5951         ENTRY;
5952
5953         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
5954         if (key == NULL) {
5955                 CDEBUG(D_IOCTL, "No GETINFO key\n");
5956                 RETURN(err_serious(-EFAULT));
5957         }
5958         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
5959                                       RCL_CLIENT);
5960
5961         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
5962         if (vallen == NULL) {
5963                 CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n",
5964                                 tgt_name(tsi->tsi_tgt));
5965                 RETURN(err_serious(-EFAULT));
5966         }
5967
5968         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
5969                              *vallen);
5970         rc = req_capsule_server_pack(tsi->tsi_pill);
5971         if (rc)
5972                 RETURN(err_serious(rc));
5973
5974         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
5975         if (valout == NULL) {
5976                 CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n",
5977                                 tgt_name(tsi->tsi_tgt));
5978                 RETURN(err_serious(-EFAULT));
5979         }
5980
5981         if (KEY_IS(KEY_FID2PATH)) {
5982                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
5983
5984                 rc = mdt_rpc_fid2path(info, key, keylen, valout, *vallen);
5985                 mdt_thread_info_fini(info);
5986         } else {
5987                 rc = -EINVAL;
5988         }
5989         RETURN(rc);
5990 }
5991
5992 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
5993 {
5994         struct obd_ioctl_data *data = karg;
5995         struct lu_fid *fid;
5996         __u64 version;
5997         struct mdt_object *obj;
5998         struct mdt_lock_handle  *lh;
5999         int rc;
6000         ENTRY;
6001
6002         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
6003             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
6004                 RETURN(-EINVAL);
6005
6006         fid = (struct lu_fid *)data->ioc_inlbuf1;
6007
6008         if (!fid_is_sane(fid))
6009                 RETURN(-EINVAL);
6010
6011         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
6012
6013         lh = &mti->mti_lh[MDT_LH_PARENT];
6014         mdt_lock_reg_init(lh, LCK_CR);
6015
6016         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
6017         if (IS_ERR(obj))
6018                 RETURN(PTR_ERR(obj));
6019
6020         if (mdt_object_remote(obj)) {
6021                 rc = -EREMOTE;
6022                 /**
6023                  * before calling version get the correct MDS should be
6024                  * fid, this is error to find remote object here
6025                  */
6026                 CERROR("nonlocal object "DFID"\n", PFID(fid));
6027         } else if (!mdt_object_exists(obj)) {
6028                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
6029                 rc = -ENOENT;
6030         } else {
6031                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
6032                *(__u64 *)data->ioc_inlbuf2 = version;
6033                 rc = 0;
6034         }
6035         mdt_object_unlock_put(mti, obj, lh, 1);
6036         RETURN(rc);
6037 }
6038
6039 /* ioctls on obd dev */
6040 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
6041                          void *karg, void __user *uarg)
6042 {
6043         struct lu_env      env;
6044         struct obd_device *obd = exp->exp_obd;
6045         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
6046         struct dt_device  *dt = mdt->mdt_bottom;
6047         int rc;
6048
6049         ENTRY;
6050         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
6051         rc = lu_env_init(&env, LCT_MD_THREAD);
6052         if (rc)
6053                 RETURN(rc);
6054
6055         switch (cmd) {
6056         case OBD_IOC_SYNC:
6057                 rc = mdt_device_sync(&env, mdt);
6058                 break;
6059         case OBD_IOC_SET_READONLY:
6060                 rc = dt_sync(&env, dt);
6061                 if (rc == 0)
6062                         rc = dt_ro(&env, dt);
6063                 break;
6064         case OBD_IOC_ABORT_RECOVERY:
6065                 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
6066                 obd->obd_abort_recovery = 1;
6067                 target_stop_recovery_thread(obd);
6068                 rc = 0;
6069                 break;
6070         case OBD_IOC_CHANGELOG_REG:
6071         case OBD_IOC_CHANGELOG_DEREG:
6072         case OBD_IOC_CHANGELOG_CLEAR:
6073                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(&env,
6074                                                            mdt->mdt_child,
6075                                                            cmd, len, karg);
6076                 break;
6077         case OBD_IOC_START_LFSCK: {
6078                 struct md_device *next = mdt->mdt_child;
6079                 struct obd_ioctl_data *data = karg;
6080                 struct lfsck_start_param lsp;
6081
6082                 if (unlikely(data == NULL)) {
6083                         rc = -EINVAL;
6084                         break;
6085                 }
6086
6087                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
6088                 lsp.lsp_index_valid = 0;
6089                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
6090                 break;
6091         }
6092         case OBD_IOC_STOP_LFSCK: {
6093                 struct md_device        *next = mdt->mdt_child;
6094                 struct obd_ioctl_data   *data = karg;
6095                 struct lfsck_stop        stop;
6096
6097                 stop.ls_status = LS_STOPPED;
6098                 /* Old lfsck utils may pass NULL @stop. */
6099                 if (data->ioc_inlbuf1 == NULL)
6100                         stop.ls_flags = 0;
6101                 else
6102                         stop.ls_flags =
6103                         ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
6104
6105                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
6106                 break;
6107         }
6108         case OBD_IOC_QUERY_LFSCK: {
6109                 struct md_device        *next = mdt->mdt_child;
6110                 struct obd_ioctl_data   *data = karg;
6111
6112                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0,
6113                                                  data->ioc_inlbuf1);
6114                 break;
6115         }
6116         case OBD_IOC_GET_OBJ_VERSION: {
6117                 struct mdt_thread_info *mti;
6118                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6119                 memset(mti, 0, sizeof *mti);
6120                 mti->mti_env = &env;
6121                 mti->mti_mdt = mdt;
6122                 mti->mti_exp = exp;
6123
6124                 rc = mdt_ioc_version_get(mti, karg);
6125                 break;
6126         }
6127         case OBD_IOC_CATLOGLIST: {
6128                 struct mdt_thread_info *mti;
6129
6130                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6131                 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
6132                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
6133                                        &mti->mti_tmp_fid1);
6134                 break;
6135          }
6136         default:
6137                 rc = -EOPNOTSUPP;
6138                 CERROR("%s: Not supported cmd = %d, rc = %d\n",
6139                         mdt_obd_name(mdt), cmd, rc);
6140         }
6141
6142         lu_env_fini(&env);
6143         RETURN(rc);
6144 }
6145
6146 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
6147 {
6148         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
6149         int rc;
6150         ENTRY;
6151
6152         if (!mdt->mdt_skip_lfsck && !mdt->mdt_bottom->dd_rdonly) {
6153                 struct lfsck_start_param lsp;
6154
6155                 lsp.lsp_start = NULL;
6156                 lsp.lsp_index_valid = 0;
6157                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
6158                                                            OBD_IOC_START_LFSCK,
6159                                                            0, &lsp);
6160                 if (rc != 0 && rc != -EALREADY)
6161                         CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
6162                               mdt_obd_name(mdt), rc);
6163         }
6164
6165         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
6166         RETURN(rc);
6167 }
6168
6169 static int mdt_obd_postrecov(struct obd_device *obd)
6170 {
6171         struct lu_env env;
6172         int rc;
6173
6174         rc = lu_env_init(&env, LCT_MD_THREAD);
6175         if (rc)
6176                 RETURN(rc);
6177         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
6178         lu_env_fini(&env);
6179         return rc;
6180 }
6181
6182 static struct obd_ops mdt_obd_device_ops = {
6183         .o_owner          = THIS_MODULE,
6184         .o_set_info_async = mdt_obd_set_info_async,
6185         .o_connect        = mdt_obd_connect,
6186         .o_reconnect      = mdt_obd_reconnect,
6187         .o_disconnect     = mdt_obd_disconnect,
6188         .o_init_export    = mdt_init_export,
6189         .o_destroy_export = mdt_destroy_export,
6190         .o_iocontrol      = mdt_iocontrol,
6191         .o_postrecov      = mdt_obd_postrecov,
6192 };
6193
6194 static struct lu_device* mdt_device_fini(const struct lu_env *env,
6195                                          struct lu_device *d)
6196 {
6197         struct mdt_device *m = mdt_dev(d);
6198         ENTRY;
6199
6200         mdt_fini(env, m);
6201         RETURN(NULL);
6202 }
6203
6204 static struct lu_device *mdt_device_free(const struct lu_env *env,
6205                                          struct lu_device *d)
6206 {
6207         struct mdt_device *m = mdt_dev(d);
6208         ENTRY;
6209
6210         lu_device_fini(&m->mdt_lu_dev);
6211         OBD_FREE_PTR(m);
6212
6213         RETURN(NULL);
6214 }
6215
6216 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
6217                                           struct lu_device_type *t,
6218                                           struct lustre_cfg *cfg)
6219 {
6220         struct lu_device  *l;
6221         struct mdt_device *m;
6222
6223         OBD_ALLOC_PTR(m);
6224         if (m != NULL) {
6225                 int rc;
6226
6227                 l = &m->mdt_lu_dev;
6228                 rc = mdt_init0(env, m, t, cfg);
6229                 if (rc != 0) {
6230                         mdt_device_free(env, l);
6231                         l = ERR_PTR(rc);
6232                         return l;
6233                 }
6234         } else
6235                 l = ERR_PTR(-ENOMEM);
6236         return l;
6237 }
6238
6239 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
6240 LU_KEY_INIT(mdt, struct mdt_thread_info);
6241
6242 static void mdt_key_fini(const struct lu_context *ctx,
6243                          struct lu_context_key *key, void* data)
6244 {
6245         struct mdt_thread_info *info = data;
6246
6247         if (info->mti_big_lmm) {
6248                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
6249                 info->mti_big_lmm = NULL;
6250                 info->mti_big_lmmsize = 0;
6251         }
6252         OBD_FREE_PTR(info);
6253 }
6254
6255 /* context key: mdt_thread_key */
6256 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
6257
6258 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
6259 {
6260         return lu_ucred(info->mti_env);
6261 }
6262
6263 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
6264 {
6265         return lu_ucred_check(info->mti_env);
6266 }
6267
6268 /**
6269  * Enable/disable COS (Commit On Sharing).
6270  *
6271  * Set/Clear the COS flag in mdt options.
6272  *
6273  * \param mdt mdt device
6274  * \param val 0 disables COS, other values enable COS
6275  */
6276 void mdt_enable_cos(struct mdt_device *mdt, int val)
6277 {
6278         struct lu_env env;
6279         int rc;
6280
6281         mdt->mdt_opts.mo_cos = !!val;
6282         rc = lu_env_init(&env, LCT_LOCAL);
6283         if (unlikely(rc != 0)) {
6284                 CWARN("%s: lu_env initialization failed, cannot "
6285                       "sync: rc = %d\n", mdt_obd_name(mdt), rc);
6286                 return;
6287         }
6288         mdt_device_sync(&env, mdt);
6289         lu_env_fini(&env);
6290 }
6291
6292 /**
6293  * Check COS (Commit On Sharing) status.
6294  *
6295  * Return COS flag status.
6296  *
6297  * \param mdt mdt device
6298  */
6299 int mdt_cos_is_enabled(struct mdt_device *mdt)
6300 {
6301         return mdt->mdt_opts.mo_cos != 0;
6302 }
6303
6304 static struct lu_device_type_operations mdt_device_type_ops = {
6305         .ldto_device_alloc = mdt_device_alloc,
6306         .ldto_device_free  = mdt_device_free,
6307         .ldto_device_fini  = mdt_device_fini
6308 };
6309
6310 static struct lu_device_type mdt_device_type = {
6311         .ldt_tags     = LU_DEVICE_MD,
6312         .ldt_name     = LUSTRE_MDT_NAME,
6313         .ldt_ops      = &mdt_device_type_ops,
6314         .ldt_ctx_tags = LCT_MD_THREAD
6315 };
6316
6317 static int __init mdt_init(void)
6318 {
6319         int rc;
6320
6321         CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") ==
6322                  FID_NOBRACE_LEN + 1);
6323         CLASSERT(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") ==
6324                  FID_LEN + 1);
6325         rc = lu_kmem_init(mdt_caches);
6326         if (rc)
6327                 return rc;
6328
6329         rc = mds_mod_init();
6330         if (rc)
6331                 GOTO(lu_fini, rc);
6332
6333         rc = class_register_type(&mdt_obd_device_ops, NULL, true, NULL,
6334                                  LUSTRE_MDT_NAME, &mdt_device_type);
6335         if (rc)
6336                 GOTO(mds_fini, rc);
6337 lu_fini:
6338         if (rc)
6339                 lu_kmem_fini(mdt_caches);
6340 mds_fini:
6341         if (rc)
6342                 mds_mod_exit();
6343         return rc;
6344 }
6345
6346 static void __exit mdt_exit(void)
6347 {
6348         class_unregister_type(LUSTRE_MDT_NAME);
6349         mds_mod_exit();
6350         lu_kmem_fini(mdt_caches);
6351 }
6352
6353 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
6354 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
6355 MODULE_VERSION(LUSTRE_VERSION_STRING);
6356 MODULE_LICENSE("GPL");
6357
6358 module_init(mdt_init);
6359 module_exit(mdt_exit);