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