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