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