lustre/mdt/mdt_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2010, 2016, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/mdt/mdt_handler.c
  33  *
  34  * Lustre Metadata Target (mdt) request handler
  35  *
  36  * Author: Peter Braam <braam@clusterfs.com>
  37  * Author: Andreas Dilger <adilger@clusterfs.com>
  38  * Author: Phil Schwan <phil@clusterfs.com>
  39  * Author: Mike Shaver <shaver@clusterfs.com>
  40  * Author: Nikita Danilov <nikita@clusterfs.com>
  41  * Author: Huang Hua <huanghua@clusterfs.com>
  42  * Author: Yury Umanets <umka@clusterfs.com>
  43  */
  44
  45 #define DEBUG_SUBSYSTEM S_MDS
  46
  47 #include <linux/module.h>
  48 #include <linux/pagemap.h>
  49
  50 #include <dt_object.h>
  51 #include <lustre_acl.h>
  52 #include <lustre_export.h>
  53 #include <lustre_ioctl.h>
  54 #include <lustre_lfsck.h>
  55 #include <lustre_log.h>
  56 #include <lustre_net.h>
  57 #include <lustre_nodemap.h>
  58 #include <lustre_mds.h>
  59 #include <lustre_param.h>
  60 #include <lustre_quota.h>
  61 #include <lustre_swab.h>
  62 #include <obd.h>
  63 #include <obd_support.h>
  64 #include <lustre_barrier.h>
  65
  66 #include <llog_swab.h>
  67
  68 #include "mdt_internal.h"
  69
  70
  71 static unsigned int max_mod_rpcs_per_client = 8;
  72 module_param(max_mod_rpcs_per_client, uint, 0644);
  73 MODULE_PARM_DESC(max_mod_rpcs_per_client, "maximum number of modify RPCs in flight allowed per client");
  74
  75 mdl_mode_t mdt_mdl_lock_modes[] = {
  76         [LCK_MINMODE] = MDL_MINMODE,
  77         [LCK_EX]      = MDL_EX,
  78         [LCK_PW]      = MDL_PW,
  79         [LCK_PR]      = MDL_PR,
  80         [LCK_CW]      = MDL_CW,
  81         [LCK_CR]      = MDL_CR,
  82         [LCK_NL]      = MDL_NL,
  83         [LCK_GROUP]   = MDL_GROUP
  84 };
  85
  86 enum ldlm_mode mdt_dlm_lock_modes[] = {
  87         [MDL_MINMODE]   = LCK_MINMODE,
  88         [MDL_EX]        = LCK_EX,
  89         [MDL_PW]        = LCK_PW,
  90         [MDL_PR]        = LCK_PR,
  91         [MDL_CW]        = LCK_CW,
  92         [MDL_CR]        = LCK_CR,
  93         [MDL_NL]        = LCK_NL,
  94         [MDL_GROUP]     = LCK_GROUP
  95 };
  96
  97 static struct mdt_device *mdt_dev(struct lu_device *d);
  98 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
  99
 100 static const struct lu_object_operations mdt_obj_ops;
 101
 102 /* Slab for MDT object allocation */
 103 static struct kmem_cache *mdt_object_kmem;
 104
 105 /* For HSM restore handles */
 106 struct kmem_cache *mdt_hsm_cdt_kmem;
 107
 108 /* For HSM request handles */
 109 struct kmem_cache *mdt_hsm_car_kmem;
 110
 111 static struct lu_kmem_descr mdt_caches[] = {
 112         {
 113                 .ckd_cache = &mdt_object_kmem,
 114                 .ckd_name  = "mdt_obj",
 115                 .ckd_size  = sizeof(struct mdt_object)
 116         },
 117         {
 118                 .ckd_cache      = &mdt_hsm_cdt_kmem,
 119                 .ckd_name       = "mdt_cdt_restore_handle",
 120                 .ckd_size       = sizeof(struct cdt_restore_handle)
 121         },
 122         {
 123                 .ckd_cache      = &mdt_hsm_car_kmem,
 124                 .ckd_name       = "mdt_cdt_agent_req",
 125                 .ckd_size       = sizeof(struct cdt_agent_req)
 126         },
 127         {
 128                 .ckd_cache = NULL
 129         }
 130 };
 131
 132 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag)
 133 {
 134         if (!rep)
 135                 return 0;
 136         return rep->lock_policy_res1 & op_flag;
 137 }
 138
 139 void mdt_clear_disposition(struct mdt_thread_info *info,
 140                            struct ldlm_reply *rep, __u64 op_flag)
 141 {
 142         if (info) {
 143                 info->mti_opdata &= ~op_flag;
 144                 tgt_opdata_clear(info->mti_env, op_flag);
 145         }
 146         if (rep)
 147                 rep->lock_policy_res1 &= ~op_flag;
 148 }
 149
 150 void mdt_set_disposition(struct mdt_thread_info *info,
 151                          struct ldlm_reply *rep, __u64 op_flag)
 152 {
 153         if (info) {
 154                 info->mti_opdata |= op_flag;
 155                 tgt_opdata_set(info->mti_env, op_flag);
 156         }
 157         if (rep)
 158                 rep->lock_policy_res1 |= op_flag;
 159 }
 160
 161 void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm)
 162 {
 163         lh->mlh_pdo_hash = 0;
 164         lh->mlh_reg_mode = lm;
 165         lh->mlh_rreg_mode = lm;
 166         lh->mlh_type = MDT_REG_LOCK;
 167 }
 168
 169 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode,
 170                        const struct lu_name *lname)
 171 {
 172         lh->mlh_reg_mode = lock_mode;
 173         lh->mlh_pdo_mode = LCK_MINMODE;
 174         lh->mlh_rreg_mode = lock_mode;
 175         lh->mlh_type = MDT_PDO_LOCK;
 176
 177         if (lu_name_is_valid(lname)) {
 178                 lh->mlh_pdo_hash = full_name_hash(lname->ln_name,
 179                                                   lname->ln_namelen);
 180                 /* XXX Workaround for LU-2856
 181                  *
 182                  * Zero is a valid return value of full_name_hash, but
 183                  * several users of mlh_pdo_hash assume a non-zero
 184                  * hash value. We therefore map zero onto an
 185                  * arbitrary, but consistent value (1) to avoid
 186                  * problems further down the road. */
 187                 if (unlikely(lh->mlh_pdo_hash == 0))
 188                         lh->mlh_pdo_hash = 1;
 189         } else {
 190                 lh->mlh_pdo_hash = 0;
 191         }
 192 }
 193
 194 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 195                               struct mdt_lock_handle *lh)
 196 {
 197         mdl_mode_t mode;
 198         ENTRY;
 199
 200         /*
 201          * Any dir access needs couple of locks:
 202          *
 203          * 1) on part of dir we gonna take lookup/modify;
 204          *
 205          * 2) on whole dir to protect it from concurrent splitting and/or to
 206          * flush client's cache for readdir().
 207          *
 208          * so, for a given mode and object this routine decides what lock mode
 209          * to use for lock #2:
 210          *
 211          * 1) if caller's gonna lookup in dir then we need to protect dir from
 212          * being splitted only - LCK_CR
 213          *
 214          * 2) if caller's gonna modify dir then we need to protect dir from
 215          * being splitted and to flush cache - LCK_CW
 216          *
 217          * 3) if caller's gonna modify dir and that dir seems ready for
 218          * splitting then we need to protect it from any type of access
 219          * (lookup/modify/split) - LCK_EX --bzzz
 220          */
 221
 222         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 223         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 224
 225         /*
 226          * Ask underlaying level its opinion about preferable PDO lock mode
 227          * having access type passed as regular lock mode:
 228          *
 229          * - MDL_MINMODE means that lower layer does not want to specify lock
 230          * mode;
 231          *
 232          * - MDL_NL means that no PDO lock should be taken. This is used in some
 233          * cases. Say, for non-splittable directories no need to use PDO locks
 234          * at all.
 235          */
 236         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 237                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 238
 239         if (mode != MDL_MINMODE) {
 240                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 241         } else {
 242                 /*
 243                  * Lower layer does not want to specify locking mode. We do it
 244                  * our selves. No special protection is needed, just flush
 245                  * client's cache on modification and allow concurrent
 246                  * mondification.
 247                  */
 248                 switch (lh->mlh_reg_mode) {
 249                 case LCK_EX:
 250                         lh->mlh_pdo_mode = LCK_EX;
 251                         break;
 252                 case LCK_PR:
 253                         lh->mlh_pdo_mode = LCK_CR;
 254                         break;
 255                 case LCK_PW:
 256                         lh->mlh_pdo_mode = LCK_CW;
 257                         break;
 258                 default:
 259                         CERROR("Not expected lock type (0x%x)\n",
 260                                (int)lh->mlh_reg_mode);
 261                         LBUG();
 262                 }
 263         }
 264
 265         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 266         EXIT;
 267 }
 268
 269 static int mdt_lookup_fileset(struct mdt_thread_info *info, const char *fileset,
 270                               struct lu_fid *fid)
 271 {
 272         struct mdt_device *mdt = info->mti_mdt;
 273         struct lu_name *lname = &info->mti_name;
 274         char *name = NULL;
 275         struct mdt_object *parent;
 276         u32 mode;
 277         int rc = 0;
 278
 279         LASSERT(!info->mti_cross_ref);
 280
 281         OBD_ALLOC(name, NAME_MAX + 1);
 282         if (name == NULL)
 283                 return -ENOMEM;
 284         lname->ln_name = name;
 285
 286         /*
 287          * We may want to allow this to mount a completely separate
 288          * fileset from the MDT in the future, but keeping it to
 289          * ROOT/ only for now avoid potential security issues.
 290          */
 291         *fid = mdt->mdt_md_root_fid;
 292
 293         while (rc == 0 && fileset != NULL && *fileset != '\0') {
 294                 const char *s1 = fileset;
 295                 const char *s2;
 296
 297                 while (*++s1 == '/')
 298                         ;
 299                 s2 = s1;
 300                 while (*s2 != '/' && *s2 != '\0')
 301                         s2++;
 302
 303                 if (s2 == s1)
 304                         break;
 305
 306                 fileset = s2;
 307
 308                 lname->ln_namelen = s2 - s1;
 309                 if (lname->ln_namelen > NAME_MAX) {
 310                         rc = -EINVAL;
 311                         break;
 312                 }
 313
 314                 /* reject .. as a path component */
 315                 if (lname->ln_namelen == 2 &&
 316                     strncmp(s1, "..", 2) == 0) {
 317                         rc = -EINVAL;
 318                         break;
 319                 }
 320
 321                 strncpy(name, s1, lname->ln_namelen);
 322                 name[lname->ln_namelen] = '\0';
 323
 324                 parent = mdt_object_find(info->mti_env, mdt, fid);
 325                 if (IS_ERR(parent)) {
 326                         rc = PTR_ERR(parent);
 327                         break;
 328                 }
 329                 /* Only got the fid of this obj by name */
 330                 fid_zero(fid);
 331                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
 332                                 fid, &info->mti_spec);
 333                 mdt_object_put(info->mti_env, parent);
 334         }
 335         if (!rc) {
 336                 parent = mdt_object_find(info->mti_env, mdt, fid);
 337                 if (IS_ERR(parent))
 338                         rc = PTR_ERR(parent);
 339                 else {
 340                         mode = lu_object_attr(&parent->mot_obj);
 341                         mdt_object_put(info->mti_env, parent);
 342                         if (!S_ISDIR(mode))
 343                                 rc = -ENOTDIR;
 344                 }
 345         }
 346
 347         OBD_FREE(name, NAME_MAX + 1);
 348
 349         return rc;
 350 }
 351
 352 static int mdt_get_root(struct tgt_session_info *tsi)
 353 {
 354         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
 355         struct mdt_device       *mdt = info->mti_mdt;
 356         struct mdt_body         *repbody;
 357         char                    *fileset = NULL, *buffer = NULL;
 358         int                      rc;
 359         struct obd_export       *exp = info->mti_exp;
 360         char                    *nodemap_fileset;
 361
 362         ENTRY;
 363
 364         rc = mdt_check_ucred(info);
 365         if (rc)
 366                 GOTO(out, rc = err_serious(rc));
 367
 368         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GET_ROOT_PACK))
 369                 GOTO(out, rc = err_serious(-ENOMEM));
 370
 371         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 372         if (req_capsule_get_size(info->mti_pill, &RMF_NAME, RCL_CLIENT) > 0) {
 373                 fileset = req_capsule_client_get(info->mti_pill, &RMF_NAME);
 374                 if (fileset == NULL)
 375                         GOTO(out, rc = err_serious(-EFAULT));
 376         }
 377
 378         nodemap_fileset = nodemap_get_fileset(exp->exp_target_data.ted_nodemap);
 379         if (nodemap_fileset && nodemap_fileset[0]) {
 380                 CDEBUG(D_INFO, "nodemap fileset is %s\n", nodemap_fileset);
 381                 if (fileset) {
 382                         /* consider fileset from client as a sub-fileset
 383                          * of the nodemap one */
 384                         OBD_ALLOC(buffer, PATH_MAX + 1);
 385                         if (buffer == NULL)
 386                                 GOTO(out, rc = err_serious(-ENOMEM));
 387                         if (snprintf(buffer, PATH_MAX + 1, "%s/%s",
 388                                      nodemap_fileset, fileset) >= PATH_MAX + 1)
 389                                 GOTO(out, rc = err_serious(-EINVAL));
 390                         fileset = buffer;
 391                 } else {
 392                         /* enforce fileset as specified in the nodemap */
 393                         fileset = nodemap_fileset;
 394                 }
 395         }
 396
 397         if (fileset) {
 398                 CDEBUG(D_INFO, "Getting fileset %s\n", fileset);
 399                 rc = mdt_lookup_fileset(info, fileset, &repbody->mbo_fid1);
 400                 if (rc < 0)
 401                         GOTO(out, rc = err_serious(rc));
 402         } else {
 403                 repbody->mbo_fid1 = mdt->mdt_md_root_fid;
 404         }
 405         repbody->mbo_valid |= OBD_MD_FLID;
 406
 407         EXIT;
 408 out:
 409         mdt_thread_info_fini(info);
 410         if (buffer)
 411                 OBD_FREE(buffer, PATH_MAX+1);
 412         return rc;
 413 }
 414
 415 static int mdt_statfs(struct tgt_session_info *tsi)
 416 {
 417         struct ptlrpc_request           *req = tgt_ses_req(tsi);
 418         struct mdt_thread_info          *info = tsi2mdt_info(tsi);
 419         struct md_device                *next = info->mti_mdt->mdt_child;
 420         struct ptlrpc_service_part      *svcpt;
 421         struct obd_statfs               *osfs;
 422         int                             rc;
 423
 424         ENTRY;
 425
 426         svcpt = req->rq_rqbd->rqbd_svcpt;
 427
 428         /* This will trigger a watchdog timeout */
 429         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 430                          (MDT_SERVICE_WATCHDOG_FACTOR *
 431                           at_get(&svcpt->scp_at_estimate)) + 1);
 432
 433         rc = mdt_check_ucred(info);
 434         if (rc)
 435                 GOTO(out, rc = err_serious(rc));
 436
 437         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
 438                 GOTO(out, rc = err_serious(-ENOMEM));
 439
 440         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 441         if (!osfs)
 442                 GOTO(out, rc = -EPROTO);
 443
 444         /** statfs information are cached in the mdt_device */
 445         if (cfs_time_before_64(info->mti_mdt->mdt_osfs_age,
 446                                cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS))) {
 447                 /** statfs data is too old, get up-to-date one */
 448                 rc = next->md_ops->mdo_statfs(info->mti_env, next, osfs);
 449                 if (rc)
 450                         GOTO(out, rc);
 451                 spin_lock(&info->mti_mdt->mdt_lock);
 452                 info->mti_mdt->mdt_osfs = *osfs;
 453                 info->mti_mdt->mdt_osfs_age = cfs_time_current_64();
 454                 spin_unlock(&info->mti_mdt->mdt_lock);
 455         } else {
 456                 /** use cached statfs data */
 457                 spin_lock(&info->mti_mdt->mdt_lock);
 458                 *osfs = info->mti_mdt->mdt_osfs;
 459                 spin_unlock(&info->mti_mdt->mdt_lock);
 460         }
 461
 462         if (rc == 0)
 463                 mdt_counter_incr(req, LPROC_MDT_STATFS);
 464 out:
 465         mdt_thread_info_fini(info);
 466         RETURN(rc);
 467 }
 468
 469 #ifdef CONFIG_FS_POSIX_ACL
 470 /*
 471  * Pack ACL data into the reply. UIDs/GIDs are mapped and filtered by nodemap.
 472  *
 473  * \param       info    thread info object
 474  * \param       repbody reply to pack ACLs into
 475  * \param       o       mdt object of file to examine
 476  * \param       nodemap nodemap of client to reply to
 477  * \retval      0       success
 478  * \retval      -errno  error getting or parsing ACL from disk
 479  */
 480 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
 481                       struct mdt_object *o, struct lu_nodemap *nodemap)
 482 {
 483         const struct lu_env     *env = info->mti_env;
 484         struct md_object        *next = mdt_object_child(o);
 485         struct lu_buf           *buf = &info->mti_buf;
 486         int rc;
 487
 488         buf->lb_buf = req_capsule_server_get(info->mti_pill, &RMF_ACL);
 489         buf->lb_len = req_capsule_get_size(info->mti_pill, &RMF_ACL,
 490                                            RCL_SERVER);
 491         if (buf->lb_len == 0)
 492                 return 0;
 493
 494         rc = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_ACCESS);
 495         if (rc < 0) {
 496                 if (rc == -ENODATA) {
 497                         repbody->mbo_aclsize = 0;
 498                         repbody->mbo_valid |= OBD_MD_FLACL;
 499                         rc = 0;
 500                 } else if (rc == -EOPNOTSUPP) {
 501                         rc = 0;
 502                 } else {
 503                         CERROR("%s: unable to read "DFID" ACL: rc = %d\n",
 504                                mdt_obd_name(info->mti_mdt),
 505                                PFID(mdt_object_fid(o)), rc);
 506                 }
 507         } else {
 508                 rc = nodemap_map_acl(nodemap, buf->lb_buf,
 509                                      rc, NODEMAP_FS_TO_CLIENT);
 510                 /* if all ACLs mapped out, rc is still >= 0 */
 511                 if (rc < 0) {
 512                         CERROR("%s: nodemap_map_acl unable to parse "DFID
 513                                " ACL: rc = %d\n", mdt_obd_name(info->mti_mdt),
 514                                PFID(mdt_object_fid(o)), rc);
 515                 } else {
 516                         repbody->mbo_aclsize = rc;
 517                         repbody->mbo_valid |= OBD_MD_FLACL;
 518                         rc = 0;
 519                 }
 520         }
 521         return rc;
 522 }
 523 #endif
 524
 525 /* XXX Look into layout in MDT layer. */
 526 static inline bool mdt_hsm_is_released(struct lov_mds_md *lmm)
 527 {
 528         struct lov_comp_md_v1   *comp_v1;
 529         struct lov_mds_md       *v1;
 530         int                      i;
 531
 532         if (lmm->lmm_magic == LOV_MAGIC_COMP_V1) {
 533                 comp_v1 = (struct lov_comp_md_v1 *)lmm;
 534
 535                 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
 536                         v1 = (struct lov_mds_md *)((char *)comp_v1 +
 537                                 comp_v1->lcm_entries[i].lcme_offset);
 538                         /* We don't support partial release for now */
 539                         if (!(v1->lmm_pattern & LOV_PATTERN_F_RELEASED))
 540                                 return false;
 541                 }
 542                 return true;
 543         } else {
 544                 return (lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) ?
 545                         true : false;
 546         }
 547 }
 548
 549 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 550                         const struct lu_attr *attr, const struct lu_fid *fid)
 551 {
 552         struct md_attr *ma = &info->mti_attr;
 553         struct obd_export *exp = info->mti_exp;
 554         struct lu_nodemap *nodemap = NULL;
 555
 556         LASSERT(ma->ma_valid & MA_INODE);
 557
 558         if (attr->la_valid & LA_ATIME) {
 559                 b->mbo_atime = attr->la_atime;
 560                 b->mbo_valid |= OBD_MD_FLATIME;
 561         }
 562         if (attr->la_valid & LA_MTIME) {
 563                 b->mbo_mtime = attr->la_mtime;
 564                 b->mbo_valid |= OBD_MD_FLMTIME;
 565         }
 566         if (attr->la_valid & LA_CTIME) {
 567                 b->mbo_ctime = attr->la_ctime;
 568                 b->mbo_valid |= OBD_MD_FLCTIME;
 569         }
 570         if (attr->la_valid & LA_FLAGS) {
 571                 b->mbo_flags = attr->la_flags;
 572                 b->mbo_valid |= OBD_MD_FLFLAGS;
 573         }
 574         if (attr->la_valid & LA_NLINK) {
 575                 b->mbo_nlink = attr->la_nlink;
 576                 b->mbo_valid |= OBD_MD_FLNLINK;
 577         }
 578         if (attr->la_valid & (LA_UID|LA_GID)) {
 579                 nodemap = nodemap_get_from_exp(exp);
 580                 if (IS_ERR(nodemap))
 581                         goto out;
 582         }
 583         if (attr->la_valid & LA_UID) {
 584                 b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 585                                             NODEMAP_FS_TO_CLIENT,
 586                                             attr->la_uid);
 587                 b->mbo_valid |= OBD_MD_FLUID;
 588         }
 589         if (attr->la_valid & LA_GID) {
 590                 b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 591                                             NODEMAP_FS_TO_CLIENT,
 592                                             attr->la_gid);
 593                 b->mbo_valid |= OBD_MD_FLGID;
 594         }
 595
 596         b->mbo_mode = attr->la_mode;
 597         if (attr->la_valid & LA_MODE)
 598                 b->mbo_valid |= OBD_MD_FLMODE;
 599         if (attr->la_valid & LA_TYPE)
 600                 b->mbo_valid |= OBD_MD_FLTYPE;
 601
 602         if (fid != NULL) {
 603                 b->mbo_fid1 = *fid;
 604                 b->mbo_valid |= OBD_MD_FLID;
 605                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid=%#llx\n",
 606                        PFID(fid), b->mbo_nlink, b->mbo_mode, b->mbo_valid);
 607         }
 608
 609         if (!(attr->la_valid & LA_TYPE))
 610                 return;
 611
 612         b->mbo_rdev   = attr->la_rdev;
 613         b->mbo_size   = attr->la_size;
 614         b->mbo_blocks = attr->la_blocks;
 615
 616         if (!S_ISREG(attr->la_mode)) {
 617                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 618         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
 619                 /* means no objects are allocated on osts. */
 620                 LASSERT(!(ma->ma_valid & MA_LOV));
 621                 /* just ignore blocks occupied by extend attributes on MDS */
 622                 b->mbo_blocks = 0;
 623                 /* if no object is allocated on osts, the size on mds is valid.
 624                  * b=22272 */
 625                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 626         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL &&
 627                    mdt_hsm_is_released(ma->ma_lmm)) {
 628                 /* A released file stores its size on MDS. */
 629                 /* But return 1 block for released file, unless tools like tar
 630                  * will consider it fully sparse. (LU-3864)
 631                  */
 632                 if (unlikely(b->mbo_size == 0))
 633                         b->mbo_blocks = 0;
 634                 else
 635                         b->mbo_blocks = 1;
 636                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 637         }
 638
 639         if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE))
 640                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 641                        PFID(fid), (unsigned long long)b->mbo_size);
 642
 643 out:
 644         if (!IS_ERR_OR_NULL(nodemap))
 645                 nodemap_putref(nodemap);
 646 }
 647
 648 static inline int mdt_body_has_lov(const struct lu_attr *la,
 649                                    const struct mdt_body *body)
 650 {
 651         return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
 652                (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
 653 }
 654
 655 void mdt_client_compatibility(struct mdt_thread_info *info)
 656 {
 657         struct mdt_body       *body;
 658         struct ptlrpc_request *req = mdt_info_req(info);
 659         struct obd_export     *exp = req->rq_export;
 660         struct md_attr        *ma = &info->mti_attr;
 661         struct lu_attr        *la = &ma->ma_attr;
 662         ENTRY;
 663
 664         if (exp_connect_layout(exp))
 665                 /* the client can deal with 16-bit lmm_stripe_count */
 666                 RETURN_EXIT;
 667
 668         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 669
 670         if (!mdt_body_has_lov(la, body))
 671                 RETURN_EXIT;
 672
 673         /* now we have a reply with a lov for a client not compatible with the
 674          * layout lock so we have to clean the layout generation number */
 675         if (S_ISREG(la->la_mode))
 676                 ma->ma_lmm->lmm_layout_gen = 0;
 677         EXIT;
 678 }
 679
 680 static int mdt_attr_get_eabuf_size(struct mdt_thread_info *info,
 681                                    struct mdt_object *o)
 682 {
 683         const struct lu_env *env = info->mti_env;
 684         int rc, rc2;
 685
 686         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 687                           XATTR_NAME_LOV);
 688
 689         if (rc == -ENODATA)
 690                 rc = 0;
 691
 692         if (rc < 0)
 693                 goto out;
 694
 695         /* Is it a directory? Let's check for the LMV as well */
 696         if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
 697                 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 698                                    XATTR_NAME_LMV);
 699
 700                 if (rc2 == -ENODATA)
 701                         rc2 = mo_xattr_get(env, mdt_object_child(o),
 702                                            &LU_BUF_NULL,
 703                                            XATTR_NAME_DEFAULT_LMV);
 704
 705                 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
 706                         rc = rc2;
 707         }
 708
 709 out:
 710         return rc;
 711 }
 712
 713 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
 714                       const char *name)
 715 {
 716         const struct lu_env *env = info->mti_env;
 717         int rc;
 718         ENTRY;
 719
 720         LASSERT(info->mti_big_lmm_used == 0);
 721         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
 722         if (rc < 0)
 723                 RETURN(rc);
 724
 725         /* big_lmm may need to be grown */
 726         if (info->mti_big_lmmsize < rc) {
 727                 int size = size_roundup_power2(rc);
 728
 729                 if (info->mti_big_lmmsize > 0) {
 730                         /* free old buffer */
 731                         LASSERT(info->mti_big_lmm);
 732                         OBD_FREE_LARGE(info->mti_big_lmm,
 733                                        info->mti_big_lmmsize);
 734                         info->mti_big_lmm = NULL;
 735                         info->mti_big_lmmsize = 0;
 736                 }
 737
 738                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
 739                 if (info->mti_big_lmm == NULL)
 740                         RETURN(-ENOMEM);
 741                 info->mti_big_lmmsize = size;
 742         }
 743         LASSERT(info->mti_big_lmmsize >= rc);
 744
 745         info->mti_buf.lb_buf = info->mti_big_lmm;
 746         info->mti_buf.lb_len = info->mti_big_lmmsize;
 747         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
 748
 749         RETURN(rc);
 750 }
 751
 752 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
 753                    struct md_attr *ma, const char *name)
 754 {
 755         struct md_object *next = mdt_object_child(o);
 756         struct lu_buf    *buf = &info->mti_buf;
 757         int rc;
 758
 759         if (strcmp(name, XATTR_NAME_LOV) == 0) {
 760                 buf->lb_buf = ma->ma_lmm;
 761                 buf->lb_len = ma->ma_lmm_size;
 762                 LASSERT(!(ma->ma_valid & MA_LOV));
 763         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 764                 buf->lb_buf = ma->ma_lmv;
 765                 buf->lb_len = ma->ma_lmv_size;
 766                 LASSERT(!(ma->ma_valid & MA_LMV));
 767         } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 768                 buf->lb_buf = ma->ma_lmv;
 769                 buf->lb_len = ma->ma_lmv_size;
 770                 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
 771         } else {
 772                 return -EINVAL;
 773         }
 774
 775         rc = mo_xattr_get(info->mti_env, next, buf, name);
 776         if (rc > 0) {
 777
 778 got:
 779                 if (strcmp(name, XATTR_NAME_LOV) == 0) {
 780                         if (info->mti_big_lmm_used)
 781                                 ma->ma_lmm = info->mti_big_lmm;
 782
 783                         /* NOT return LOV EA with hole to old client. */
 784                         if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
 785                                      LOV_PATTERN_F_HOLE) &&
 786                             !(exp_connect_flags(info->mti_exp) &
 787                               OBD_CONNECT_LFSCK)) {
 788                                 return -EIO;
 789                         } else {
 790                                 ma->ma_lmm_size = rc;
 791                                 ma->ma_valid |= MA_LOV;
 792                         }
 793                 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
 794                         if (info->mti_big_lmm_used)
 795                                 ma->ma_lmv = info->mti_big_lmm;
 796
 797                         ma->ma_lmv_size = rc;
 798                         ma->ma_valid |= MA_LMV;
 799                 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
 800                         ma->ma_lmv_size = rc;
 801                         ma->ma_valid |= MA_LMV_DEF;
 802                 }
 803
 804                 /* Update mdt_max_mdsize so all clients will be aware that */
 805                 if (info->mti_mdt->mdt_max_mdsize < rc)
 806                         info->mti_mdt->mdt_max_mdsize = rc;
 807
 808                 rc = 0;
 809         } else if (rc == -ENODATA) {
 810                 /* no LOV EA */
 811                 rc = 0;
 812         } else if (rc == -ERANGE) {
 813                 /* Default LMV has fixed size, so it must be able to fit
 814                  * in the original buffer */
 815                 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
 816                         return rc;
 817                 rc = mdt_big_xattr_get(info, o, name);
 818                 if (rc > 0) {
 819                         info->mti_big_lmm_used = 1;
 820                         goto got;
 821                 }
 822         }
 823
 824         return rc;
 825 }
 826
 827 static int mdt_attr_get_pfid(struct mdt_thread_info *info,
 828                              struct mdt_object *o, struct lu_fid *pfid)
 829 {
 830         struct lu_buf           *buf = &info->mti_buf;
 831         struct link_ea_header   *leh;
 832         struct link_ea_entry    *lee;
 833         int                      rc;
 834         ENTRY;
 835
 836         buf->lb_buf = info->mti_big_lmm;
 837         buf->lb_len = info->mti_big_lmmsize;
 838         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
 839                           buf, XATTR_NAME_LINK);
 840         /* ignore errors, MA_PFID won't be set and it is
 841          * up to the caller to treat this as an error */
 842         if (rc == -ERANGE || buf->lb_len == 0) {
 843                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
 844                 buf->lb_buf = info->mti_big_lmm;
 845                 buf->lb_len = info->mti_big_lmmsize;
 846         }
 847
 848         if (rc < 0)
 849                 RETURN(rc);
 850         if (rc < sizeof(*leh)) {
 851                 CERROR("short LinkEA on "DFID": rc = %d\n",
 852                        PFID(mdt_object_fid(o)), rc);
 853                 RETURN(-ENODATA);
 854         }
 855
 856         leh = (struct link_ea_header *) buf->lb_buf;
 857         lee = (struct link_ea_entry *)(leh + 1);
 858         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
 859                 leh->leh_magic = LINK_EA_MAGIC;
 860                 leh->leh_reccount = __swab32(leh->leh_reccount);
 861                 leh->leh_len = __swab64(leh->leh_len);
 862         }
 863         if (leh->leh_magic != LINK_EA_MAGIC)
 864                 RETURN(-EINVAL);
 865         if (leh->leh_reccount == 0)
 866                 RETURN(-ENODATA);
 867
 868         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
 869         fid_be_to_cpu(pfid, pfid);
 870
 871         RETURN(0);
 872 }
 873
 874 int mdt_attr_get_complex(struct mdt_thread_info *info,
 875                          struct mdt_object *o, struct md_attr *ma)
 876 {
 877         const struct lu_env *env = info->mti_env;
 878         struct md_object    *next = mdt_object_child(o);
 879         struct lu_buf       *buf = &info->mti_buf;
 880         int                  need = ma->ma_need;
 881         int                  rc = 0, rc2;
 882         u32                  mode;
 883         ENTRY;
 884
 885         ma->ma_valid = 0;
 886
 887         if (mdt_object_exists(o) == 0)
 888                 GOTO(out, rc = -ENOENT);
 889         mode = lu_object_attr(&next->mo_lu);
 890
 891         if (need & MA_INODE) {
 892                 ma->ma_need = MA_INODE;
 893                 rc = mo_attr_get(env, next, ma);
 894                 if (rc)
 895                         GOTO(out, rc);
 896                 ma->ma_valid |= MA_INODE;
 897         }
 898
 899         if (need & MA_PFID) {
 900                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
 901                 if (rc == 0)
 902                         ma->ma_valid |= MA_PFID;
 903                 /* ignore this error, parent fid is not mandatory */
 904                 rc = 0;
 905         }
 906
 907         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
 908                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
 909                 if (rc)
 910                         GOTO(out, rc);
 911         }
 912
 913         if (need & MA_LMV && S_ISDIR(mode)) {
 914                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
 915                 if (rc != 0)
 916                         GOTO(out, rc);
 917         }
 918
 919         if (need & MA_LMV_DEF && S_ISDIR(mode)) {
 920                 rc = mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
 921                 if (rc != 0)
 922                         GOTO(out, rc);
 923         }
 924
 925         if (need & MA_HSM && S_ISREG(mode)) {
 926                 buf->lb_buf = info->mti_xattr_buf;
 927                 buf->lb_len = sizeof(info->mti_xattr_buf);
 928                 CLASSERT(sizeof(struct hsm_attrs) <=
 929                          sizeof(info->mti_xattr_buf));
 930                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
 931                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
 932                 if (rc2 == 0)
 933                         ma->ma_valid |= MA_HSM;
 934                 else if (rc2 < 0 && rc2 != -ENODATA)
 935                         GOTO(out, rc = rc2);
 936         }
 937
 938 #ifdef CONFIG_FS_POSIX_ACL
 939         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
 940                 buf->lb_buf = ma->ma_acl;
 941                 buf->lb_len = ma->ma_acl_size;
 942                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
 943                 if (rc2 > 0) {
 944                         ma->ma_acl_size = rc2;
 945                         ma->ma_valid |= MA_ACL_DEF;
 946                 } else if (rc2 == -ENODATA) {
 947                         /* no ACLs */
 948                         ma->ma_acl_size = 0;
 949                 } else
 950                         GOTO(out, rc = rc2);
 951         }
 952 #endif
 953 out:
 954         ma->ma_need = need;
 955         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = %#llx ma_lmm=%p\n",
 956                rc, ma->ma_valid, ma->ma_lmm);
 957         RETURN(rc);
 958 }
 959
 960 static int mdt_getattr_internal(struct mdt_thread_info *info,
 961                                 struct mdt_object *o, int ma_need)
 962 {
 963         struct md_object        *next = mdt_object_child(o);
 964         const struct mdt_body   *reqbody = info->mti_body;
 965         struct ptlrpc_request   *req = mdt_info_req(info);
 966         struct md_attr          *ma = &info->mti_attr;
 967         struct lu_attr          *la = &ma->ma_attr;
 968         struct req_capsule      *pill = info->mti_pill;
 969         const struct lu_env     *env = info->mti_env;
 970         struct mdt_body         *repbody;
 971         struct lu_buf           *buffer = &info->mti_buf;
 972         struct obd_export       *exp = info->mti_exp;
 973         int                      rc;
 974         ENTRY;
 975
 976         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
 977                 RETURN(err_serious(-ENOMEM));
 978
 979         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 980
 981         ma->ma_valid = 0;
 982
 983         if (mdt_object_remote(o)) {
 984                 /* This object is located on remote node.*/
 985                 /* Return -ENOTSUPP for old client */
 986                 if (!mdt_is_dne_client(req->rq_export))
 987                         GOTO(out, rc = -ENOTSUPP);
 988
 989                 repbody->mbo_fid1 = *mdt_object_fid(o);
 990                 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
 991                 GOTO(out, rc = 0);
 992         }
 993
 994         if (reqbody->mbo_eadatasize > 0) {
 995                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
 996                 if (buffer->lb_buf == NULL)
 997                         GOTO(out, rc = -EPROTO);
 998                 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
 999                                                       RCL_SERVER);
1000         } else {
1001                 buffer->lb_buf = NULL;
1002                 buffer->lb_len = 0;
1003                 ma_need &= ~(MA_LOV | MA_LMV);
1004                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
1005                        mdt_obd_name(info->mti_mdt),
1006                        req->rq_export->exp_client_uuid.uuid);
1007         }
1008
1009         /* If it is dir object and client require MEA, then we got MEA */
1010         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1011             (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
1012                 /* Assumption: MDT_MD size is enough for lmv size. */
1013                 ma->ma_lmv = buffer->lb_buf;
1014                 ma->ma_lmv_size = buffer->lb_len;
1015                 ma->ma_need = MA_INODE;
1016                 if (ma->ma_lmv_size > 0) {
1017                         if (reqbody->mbo_valid & OBD_MD_MEA)
1018                                 ma->ma_need |= MA_LMV;
1019                         else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA)
1020                                 ma->ma_need |= MA_LMV_DEF;
1021                 }
1022         } else {
1023                 ma->ma_lmm = buffer->lb_buf;
1024                 ma->ma_lmm_size = buffer->lb_len;
1025                 ma->ma_need = MA_INODE | MA_HSM;
1026                 if (ma->ma_lmm_size > 0)
1027                         ma->ma_need |= MA_LOV;
1028         }
1029
1030         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1031             reqbody->mbo_valid & OBD_MD_FLDIREA  &&
1032             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
1033                 /* get default stripe info for this dir. */
1034                 ma->ma_need |= MA_LOV_DEF;
1035         }
1036         ma->ma_need |= ma_need;
1037
1038         rc = mdt_attr_get_complex(info, o, ma);
1039         if (unlikely(rc)) {
1040                 CDEBUG(rc == -ENOENT ? D_OTHER : D_ERROR,
1041                        "%s: getattr error for "DFID": rc = %d\n",
1042                        mdt_obd_name(info->mti_mdt),
1043                        PFID(mdt_object_fid(o)), rc);
1044                 RETURN(rc);
1045         }
1046
1047         /* if file is released, check if a restore is running */
1048         if (ma->ma_valid & MA_HSM) {
1049                 repbody->mbo_valid |= OBD_MD_TSTATE;
1050                 if ((ma->ma_hsm.mh_flags & HS_RELEASED) &&
1051                     mdt_hsm_restore_is_running(info, mdt_object_fid(o)))
1052                         repbody->mbo_t_state = MS_RESTORE;
1053         }
1054
1055         if (likely(ma->ma_valid & MA_INODE))
1056                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
1057         else
1058                 RETURN(-EFAULT);
1059
1060         if (mdt_body_has_lov(la, reqbody)) {
1061                 if (ma->ma_valid & MA_LOV) {
1062                         LASSERT(ma->ma_lmm_size);
1063                         repbody->mbo_eadatasize = ma->ma_lmm_size;
1064                         if (S_ISDIR(la->la_mode))
1065                                 repbody->mbo_valid |= OBD_MD_FLDIREA;
1066                         else
1067                                 repbody->mbo_valid |= OBD_MD_FLEASIZE;
1068                         mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
1069                 }
1070                 if (ma->ma_valid & MA_LMV) {
1071                         /* Return -ENOTSUPP for old client */
1072                         if (!mdt_is_striped_client(req->rq_export))
1073                                 RETURN(-ENOTSUPP);
1074
1075                         LASSERT(S_ISDIR(la->la_mode));
1076                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1077                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1078                         repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
1079                 }
1080                 if (ma->ma_valid & MA_LMV_DEF) {
1081                         /* Return -ENOTSUPP for old client */
1082                         if (!mdt_is_striped_client(req->rq_export))
1083                                 RETURN(-ENOTSUPP);
1084                         LASSERT(S_ISDIR(la->la_mode));
1085                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1086                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1087                         repbody->mbo_valid |= (OBD_MD_FLDIREA |
1088                                                OBD_MD_DEFAULT_MEA);
1089                 }
1090         } else if (S_ISLNK(la->la_mode) &&
1091                    reqbody->mbo_valid & OBD_MD_LINKNAME) {
1092                 buffer->lb_buf = ma->ma_lmm;
1093                 /* eadatasize from client includes NULL-terminator, so
1094                  * there is no need to read it */
1095                 buffer->lb_len = reqbody->mbo_eadatasize - 1;
1096                 rc = mo_readlink(env, next, buffer);
1097                 if (unlikely(rc <= 0)) {
1098                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
1099                                mdt_obd_name(info->mti_mdt),
1100                                PFID(mdt_object_fid(o)), rc);
1101                         rc = -EFAULT;
1102                 } else {
1103                         int print_limit = min_t(int, PAGE_SIZE - 128, rc);
1104
1105                         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
1106                                 rc -= 2;
1107                         repbody->mbo_valid |= OBD_MD_LINKNAME;
1108                         /* we need to report back size with NULL-terminator
1109                          * because client expects that */
1110                         repbody->mbo_eadatasize = rc + 1;
1111                         if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
1112                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d "
1113                                        "on "DFID ", expected %d\n",
1114                                        mdt_obd_name(info->mti_mdt),
1115                                        rc, PFID(mdt_object_fid(o)),
1116                                        reqbody->mbo_eadatasize - 1);
1117                         /* NULL terminate */
1118                         ((char *)ma->ma_lmm)[rc] = 0;
1119
1120                         /* If the total CDEBUG() size is larger than a page, it
1121                          * will print a warning to the console, avoid this by
1122                          * printing just the last part of the symlink. */
1123                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
1124                                print_limit < rc ? "..." : "", print_limit,
1125                                (char *)ma->ma_lmm + rc - print_limit, rc);
1126                         rc = 0;
1127                 }
1128         }
1129
1130         if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
1131                 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1132                 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1133                 CDEBUG(D_INODE, "changing the max MD size to %u\n",
1134                        repbody->mbo_max_mdsize);
1135         }
1136
1137 #ifdef CONFIG_FS_POSIX_ACL
1138         if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1139                  (reqbody->mbo_valid & OBD_MD_FLACL)) {
1140                 struct lu_nodemap *nodemap = nodemap_get_from_exp(exp);
1141                 if (IS_ERR(nodemap))
1142                         RETURN(PTR_ERR(nodemap));
1143
1144                 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
1145                 nodemap_putref(nodemap);
1146         }
1147 #endif
1148
1149 out:
1150         if (rc == 0)
1151                 mdt_counter_incr(req, LPROC_MDT_GETATTR);
1152
1153         RETURN(rc);
1154 }
1155
1156 static int mdt_getattr(struct tgt_session_info *tsi)
1157 {
1158         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1159         struct mdt_object       *obj = info->mti_object;
1160         struct req_capsule      *pill = info->mti_pill;
1161         struct mdt_body         *reqbody;
1162         struct mdt_body         *repbody;
1163         int rc, rc2;
1164         ENTRY;
1165
1166         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1167         LASSERT(reqbody);
1168         LASSERT(obj != NULL);
1169         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1170
1171         /* Unlike intent case where we need to pre-fill out buffers early on
1172          * in intent policy for ldlm reasons, here we can have a much better
1173          * guess at EA size by just reading it from disk.
1174          * Exceptions are readdir and (missing) directory striping */
1175         /* Readlink */
1176         if (reqbody->mbo_valid & OBD_MD_LINKNAME) {
1177                 /* No easy way to know how long is the symlink, but it cannot
1178                  * be more than PATH_MAX, so we allocate +1 */
1179                 rc = PATH_MAX + 1;
1180
1181         /* A special case for fs ROOT: getattr there might fetch
1182          * default EA for entire fs, not just for this dir!
1183          */
1184         } else if (lu_fid_eq(mdt_object_fid(obj),
1185                              &info->mti_mdt->mdt_md_root_fid) &&
1186                    (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1187                    (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1188                                                                  MDS_GETATTR)) {
1189                 /* Should the default strping be bigger, mdt_fix_reply
1190                  * will reallocate */
1191                 rc = DEF_REP_MD_SIZE;
1192         } else {
1193                 /* Read the actual EA size from disk */
1194                 rc = mdt_attr_get_eabuf_size(info, obj);
1195         }
1196
1197         if (rc < 0)
1198                 GOTO(out, rc = err_serious(rc));
1199
1200         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1201
1202         rc = req_capsule_server_pack(pill);
1203         if (unlikely(rc != 0))
1204                 GOTO(out, rc = err_serious(rc));
1205
1206         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1207         LASSERT(repbody != NULL);
1208         repbody->mbo_eadatasize = 0;
1209         repbody->mbo_aclsize = 0;
1210
1211         rc = mdt_check_ucred(info);
1212         if (unlikely(rc))
1213                 GOTO(out_shrink, rc);
1214
1215         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1216
1217         rc = mdt_getattr_internal(info, obj, 0);
1218         EXIT;
1219 out_shrink:
1220         mdt_client_compatibility(info);
1221         rc2 = mdt_fix_reply(info);
1222         if (rc == 0)
1223                 rc = rc2;
1224 out:
1225         mdt_thread_info_fini(info);
1226         return rc;
1227 }
1228
1229 /**
1230  * Handler of layout intent RPC requiring the layout modification
1231  *
1232  * \param[in] info      thread environment
1233  * \param[in] obj       object
1234  * \param[in] layout    layout intent
1235  * \param[in] buf       buffer containing client's lovea, could be empty
1236  *
1237  * \retval 0    on success
1238  * \retval < 0  error code
1239  */
1240 static int mdt_layout_change(struct mdt_thread_info *info,
1241                              struct mdt_object *obj,
1242                              struct layout_intent *layout,
1243                              const struct lu_buf *buf)
1244 {
1245         struct mdt_lock_handle *lh = &info->mti_lh[MDT_LH_LOCAL];
1246         int rc;
1247         ENTRY;
1248
1249         if (layout->li_start >= layout->li_end) {
1250                 CERROR("Recieved an invalid layout change range [%llu, %llu) "
1251                        "for "DFID"\n", layout->li_start, layout->li_end,
1252                        PFID(mdt_object_fid(obj)));
1253                 RETURN(-EINVAL);
1254         }
1255
1256         if (!S_ISREG(lu_object_attr(&obj->mot_obj)))
1257                 GOTO(out, rc = -EINVAL);
1258
1259         rc = mo_permission(info->mti_env, NULL, mdt_object_child(obj), NULL,
1260                            MAY_WRITE);
1261         if (rc)
1262                 GOTO(out, rc);
1263
1264         /* take layout lock to prepare layout change */
1265         mdt_lock_reg_init(lh, LCK_EX);
1266         rc = mdt_object_lock(info, obj, lh,
1267                              MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR);
1268         if (rc)
1269                 GOTO(out, rc);
1270
1271         rc = mo_layout_change(info->mti_env, mdt_object_child(obj), layout,
1272                               buf);
1273
1274         mdt_object_unlock(info, obj, lh, 1);
1275 out:
1276         RETURN(rc);
1277 }
1278
1279 /**
1280  * Exchange MOF_LOV_CREATED flags between two objects after a
1281  * layout swap. No assumption is made on whether o1 or o2 have
1282  * created objects or not.
1283  *
1284  * \param[in,out] o1    First swap layout object
1285  * \param[in,out] o2    Second swap layout object
1286  */
1287 static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2)
1288 {
1289         unsigned int o1_lov_created = o1->mot_lov_created;
1290
1291         mutex_lock(&o1->mot_lov_mutex);
1292         mutex_lock(&o2->mot_lov_mutex);
1293
1294         o1->mot_lov_created = o2->mot_lov_created;
1295         o2->mot_lov_created = o1_lov_created;
1296
1297         mutex_unlock(&o2->mot_lov_mutex);
1298         mutex_unlock(&o1->mot_lov_mutex);
1299 }
1300
1301 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1302 {
1303         struct mdt_thread_info  *info;
1304         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1305         struct obd_export       *exp = req->rq_export;
1306         struct mdt_object       *o1, *o2, *o;
1307         struct mdt_lock_handle  *lh1, *lh2;
1308         struct mdc_swap_layouts *msl;
1309         int                      rc;
1310         ENTRY;
1311
1312         /* client does not support layout lock, so layout swaping
1313          * is disabled.
1314          * FIXME: there is a problem for old clients which don't support
1315          * layout lock yet. If those clients have already opened the file
1316          * they won't be notified at all so that old layout may still be
1317          * used to do IO. This can be fixed after file release is landed by
1318          * doing exclusive open and taking full EX ibits lock. - Jinshan */
1319         if (!exp_connect_layout(exp))
1320                 RETURN(-EOPNOTSUPP);
1321
1322         info = tsi2mdt_info(tsi);
1323
1324         if (info->mti_dlm_req != NULL)
1325                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1326
1327         o1 = info->mti_object;
1328         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1329                                 &info->mti_body->mbo_fid2);
1330         if (IS_ERR(o))
1331                 GOTO(out, rc = PTR_ERR(o));
1332
1333         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1334                 GOTO(put, rc = -ENOENT);
1335
1336         rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1337         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1338                 GOTO(put, rc);
1339
1340         if (rc < 0)
1341                 swap(o1, o2);
1342
1343         /* permission check. Make sure the calling process having permission
1344          * to write both files. */
1345         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1346                                 MAY_WRITE);
1347         if (rc < 0)
1348                 GOTO(put, rc);
1349
1350         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1351                                 MAY_WRITE);
1352         if (rc < 0)
1353                 GOTO(put, rc);
1354
1355         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1356         if (msl == NULL)
1357                 GOTO(put, rc = -EPROTO);
1358
1359         lh1 = &info->mti_lh[MDT_LH_NEW];
1360         mdt_lock_reg_init(lh1, LCK_EX);
1361         lh2 = &info->mti_lh[MDT_LH_OLD];
1362         mdt_lock_reg_init(lh2, LCK_EX);
1363
1364         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1365                              MDS_INODELOCK_XATTR);
1366         if (rc < 0)
1367                 GOTO(put, rc);
1368
1369         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1370                              MDS_INODELOCK_XATTR);
1371         if (rc < 0)
1372                 GOTO(unlock1, rc);
1373
1374         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1375                              mdt_object_child(o2), msl->msl_flags);
1376         if (rc < 0)
1377                 GOTO(unlock2, rc);
1378
1379         mdt_swap_lov_flag(o1, o2);
1380
1381 unlock2:
1382         mdt_object_unlock(info, o2, lh2, rc);
1383 unlock1:
1384         mdt_object_unlock(info, o1, lh1, rc);
1385 put:
1386         mdt_object_put(info->mti_env, o);
1387 out:
1388         mdt_thread_info_fini(info);
1389         RETURN(rc);
1390 }
1391
1392 static int mdt_raw_lookup(struct mdt_thread_info *info,
1393                           struct mdt_object *parent,
1394                           const struct lu_name *lname,
1395                           struct ldlm_reply *ldlm_rep)
1396 {
1397         struct lu_fid   *child_fid = &info->mti_tmp_fid1;
1398         int              rc;
1399         ENTRY;
1400
1401         LASSERT(!info->mti_cross_ref);
1402
1403         /* Only got the fid of this obj by name */
1404         fid_zero(child_fid);
1405         rc = mdo_lookup(info->mti_env, mdt_object_child(info->mti_object),
1406                         lname, child_fid, &info->mti_spec);
1407         if (rc == 0) {
1408                 struct mdt_body *repbody;
1409
1410                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1411                 repbody->mbo_fid1 = *child_fid;
1412                 repbody->mbo_valid = OBD_MD_FLID;
1413                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1414         } else if (rc == -ENOENT) {
1415                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1416         }
1417
1418         RETURN(rc);
1419 }
1420
1421 /*
1422  * UPDATE lock should be taken against parent, and be released before exit;
1423  * child_bits lock should be taken against child, and be returned back:
1424  *            (1)normal request should release the child lock;
1425  *            (2)intent request will grant the lock to client.
1426  */
1427 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
1428                                  struct mdt_lock_handle *lhc,
1429                                  __u64 child_bits,
1430                                  struct ldlm_reply *ldlm_rep)
1431 {
1432         struct ptlrpc_request  *req       = mdt_info_req(info);
1433         struct mdt_body        *reqbody   = NULL;
1434         struct mdt_object      *parent    = info->mti_object;
1435         struct mdt_object      *child;
1436         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
1437         struct lu_name         *lname     = NULL;
1438         struct mdt_lock_handle *lhp       = NULL;
1439         struct ldlm_lock       *lock;
1440         bool                    is_resent;
1441         bool                    try_layout;
1442         int                     ma_need = 0;
1443         int                     rc;
1444         ENTRY;
1445
1446         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
1447         LASSERT(ergo(is_resent,
1448                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
1449
1450         if (parent == NULL)
1451                 RETURN(-ENOENT);
1452
1453         if (info->mti_cross_ref) {
1454                 /* Only getattr on the child. Parent is on another node. */
1455                 mdt_set_disposition(info, ldlm_rep,
1456                                     DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
1457                 child = parent;
1458                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", "
1459                        "ldlm_rep = %p\n",
1460                        PFID(mdt_object_fid(child)), ldlm_rep);
1461
1462                 rc = mdt_check_resent_lock(info, child, lhc);
1463                 if (rc < 0) {
1464                         RETURN(rc);
1465                 } else if (rc > 0) {
1466                         mdt_lock_handle_init(lhc);
1467                         mdt_lock_reg_init(lhc, LCK_PR);
1468
1469                         /*
1470                          * Object's name is on another MDS, no lookup or layout
1471                          * lock is needed here but update lock is.
1472                          */
1473                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
1474                                         MDS_INODELOCK_LAYOUT);
1475                         child_bits |= MDS_INODELOCK_PERM | MDS_INODELOCK_UPDATE;
1476
1477                         rc = mdt_object_lock(info, child, lhc, child_bits);
1478                         if (rc < 0)
1479                                 RETURN(rc);
1480                 }
1481
1482                 /* Finally, we can get attr for child. */
1483                 if (!mdt_object_exists(child)) {
1484                         LU_OBJECT_DEBUG(D_INFO, info->mti_env,
1485                                         &child->mot_obj,
1486                                         "remote object doesn't exist.");
1487                         mdt_object_unlock(info, child, lhc, 1);
1488                         RETURN(-ENOENT);
1489                 }
1490
1491                 rc = mdt_getattr_internal(info, child, 0);
1492                 if (unlikely(rc != 0))
1493                         mdt_object_unlock(info, child, lhc, 1);
1494
1495                 RETURN(rc);
1496         }
1497
1498         lname = &info->mti_name;
1499         mdt_name_unpack(info->mti_pill, &RMF_NAME, lname, MNF_FIX_ANON);
1500
1501         if (lu_name_is_valid(lname)) {
1502                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", "
1503                        "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)),
1504                        PNAME(lname), ldlm_rep);
1505         } else {
1506                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1507                 if (unlikely(reqbody == NULL))
1508                         RETURN(err_serious(-EPROTO));
1509
1510                 *child_fid = reqbody->mbo_fid2;
1511
1512                 if (unlikely(!fid_is_sane(child_fid)))
1513                         RETURN(err_serious(-EINVAL));
1514
1515                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", "
1516                        "ldlm_rep = %p\n",
1517                        PFID(mdt_object_fid(parent)),
1518                        PFID(&reqbody->mbo_fid2), ldlm_rep);
1519         }
1520
1521         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
1522
1523         if (unlikely(!mdt_object_exists(parent)) && lu_name_is_valid(lname)) {
1524                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1525                                 &parent->mot_obj,
1526                                 "Parent doesn't exist!");
1527                 RETURN(-ESTALE);
1528         }
1529
1530         if (mdt_object_remote(parent)) {
1531                 CERROR("%s: parent "DFID" is on remote target\n",
1532                        mdt_obd_name(info->mti_mdt),
1533                        PFID(mdt_object_fid(parent)));
1534                 RETURN(-EIO);
1535         }
1536
1537         if (lu_name_is_valid(lname)) {
1538                 /* Always allow to lookup ".." */
1539                 if (unlikely(lname->ln_namelen == 2 &&
1540                              lname->ln_name[0] == '.' &&
1541                              lname->ln_name[1] == '.'))
1542                         info->mti_spec.sp_permitted = 1;
1543
1544                 if (info->mti_body->mbo_valid == OBD_MD_FLID) {
1545                         rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
1546
1547                         RETURN(rc);
1548                 }
1549
1550                 /* step 1: lock parent only if parent is a directory */
1551                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
1552                         lhp = &info->mti_lh[MDT_LH_PARENT];
1553                         mdt_lock_pdo_init(lhp, LCK_PR, lname);
1554                         rc = mdt_object_lock(info, parent, lhp,
1555                                              MDS_INODELOCK_UPDATE);
1556                         if (unlikely(rc != 0))
1557                                 RETURN(rc);
1558                 }
1559
1560                 /* step 2: lookup child's fid by name */
1561                 fid_zero(child_fid);
1562                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
1563                                 child_fid, &info->mti_spec);
1564                 if (rc == -ENOENT)
1565                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1566
1567                 if (rc != 0)
1568                         GOTO(out_parent, rc);
1569         }
1570
1571         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1572
1573         /*
1574          *step 3: find the child object by fid & lock it.
1575          *        regardless if it is local or remote.
1576          *
1577          *Note: LU-3240 (commit 762f2114d282a98ebfa4dbbeea9298a8088ad24e)
1578          *      set parent dir fid the same as child fid in getattr by fid case
1579          *      we should not lu_object_find() the object again, could lead
1580          *      to hung if there is a concurrent unlink destroyed the object.
1581          */
1582         if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
1583                 mdt_object_get(info->mti_env, parent);
1584                 child = parent;
1585         } else {
1586                 child = mdt_object_find(info->mti_env, info->mti_mdt,
1587                                         child_fid);
1588         }
1589
1590         if (unlikely(IS_ERR(child)))
1591                 GOTO(out_parent, rc = PTR_ERR(child));
1592
1593         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
1594         if (!mdt_object_exists(child)) {
1595                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
1596                                 &child->mot_obj,
1597                                 "Object doesn't exist!");
1598                 GOTO(out_child, rc = -ENOENT);
1599         }
1600
1601         rc = mdt_check_resent_lock(info, child, lhc);
1602         if (rc < 0) {
1603                 GOTO(out_child, rc);
1604         } else if (rc > 0) {
1605                 mdt_lock_handle_init(lhc);
1606                 mdt_lock_reg_init(lhc, LCK_PR);
1607                 try_layout = false;
1608
1609                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
1610                       mdt_object_exists(child) && !mdt_object_remote(child)) {
1611                         struct md_attr *ma = &info->mti_attr;
1612
1613                         ma->ma_valid = 0;
1614                         ma->ma_need = MA_INODE;
1615                         rc = mdt_attr_get_complex(info, child, ma);
1616                         if (unlikely(rc != 0))
1617                                 GOTO(out_child, rc);
1618
1619                         /* If the file has not been changed for some time, we
1620                          * return not only a LOOKUP lock, but also an UPDATE
1621                          * lock and this might save us RPC on later STAT. For
1622                          * directories, it also let negative dentry cache start
1623                          * working for this dir. */
1624                         if (ma->ma_valid & MA_INODE &&
1625                             ma->ma_attr.la_valid & LA_CTIME &&
1626                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
1627                                 ma->ma_attr.la_ctime < cfs_time_current_sec())
1628                                 child_bits |= MDS_INODELOCK_UPDATE;
1629                 }
1630
1631                 /* layout lock must be granted in a best-effort way
1632                  * for IT operations */
1633                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
1634                 if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
1635                     exp_connect_layout(info->mti_exp) &&
1636                     S_ISREG(lu_object_attr(&child->mot_obj)) &&
1637                     !mdt_object_remote(child) && ldlm_rep != NULL) {
1638                         /* try to grant layout lock for regular file. */
1639                         try_layout = true;
1640                 }
1641
1642                 rc = 0;
1643                 if (try_layout) {
1644                         child_bits |= MDS_INODELOCK_LAYOUT;
1645                         /* try layout lock, it may fail to be granted due to
1646                          * contention at LOOKUP or UPDATE */
1647                         if (!mdt_object_lock_try(info, child, lhc,
1648                                                  child_bits)) {
1649                                 child_bits &= ~MDS_INODELOCK_LAYOUT;
1650                                 LASSERT(child_bits != 0);
1651                                 rc = mdt_object_lock(info, child, lhc,
1652                                                      child_bits);
1653                         } else {
1654                                 ma_need |= MA_LOV;
1655                         }
1656                 } else {
1657                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
1658                          * client will enqueue the lock to the remote MDT */
1659                         if (mdt_object_remote(child))
1660                                 child_bits &= ~MDS_INODELOCK_UPDATE;
1661                         rc = mdt_object_lock(info, child, lhc, child_bits);
1662                 }
1663                 if (unlikely(rc != 0))
1664                         GOTO(out_child, rc);
1665         }
1666
1667         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1668
1669         /* finally, we can get attr for child. */
1670         rc = mdt_getattr_internal(info, child, ma_need);
1671         if (unlikely(rc != 0)) {
1672                 mdt_object_unlock(info, child, lhc, 1);
1673         } else if (lock) {
1674                 /* Debugging code. */
1675                 LDLM_DEBUG(lock, "Returning lock to client");
1676                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
1677                                          &lock->l_resource->lr_name),
1678                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1679                          PLDLMRES(lock->l_resource),
1680                          PFID(mdt_object_fid(child)));
1681         }
1682         if (lock)
1683                 LDLM_LOCK_PUT(lock);
1684
1685         EXIT;
1686 out_child:
1687         mdt_object_put(info->mti_env, child);
1688 out_parent:
1689         if (lhp)
1690                 mdt_object_unlock(info, parent, lhp, 1);
1691         return rc;
1692 }
1693
1694 /* normal handler: should release the child lock */
1695 static int mdt_getattr_name(struct tgt_session_info *tsi)
1696 {
1697         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1698         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
1699         struct mdt_body        *reqbody;
1700         struct mdt_body        *repbody;
1701         int rc, rc2;
1702         ENTRY;
1703
1704         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
1705         LASSERT(reqbody != NULL);
1706         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1707         LASSERT(repbody != NULL);
1708
1709         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1710         repbody->mbo_eadatasize = 0;
1711         repbody->mbo_aclsize = 0;
1712
1713         rc = mdt_init_ucred_intent_getattr(info, reqbody);
1714         if (unlikely(rc))
1715                 GOTO(out_shrink, rc);
1716
1717         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
1718         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1719                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
1720                 lhc->mlh_reg_lh.cookie = 0;
1721         }
1722         mdt_exit_ucred(info);
1723         EXIT;
1724 out_shrink:
1725         mdt_client_compatibility(info);
1726         rc2 = mdt_fix_reply(info);
1727         if (rc == 0)
1728                 rc = rc2;
1729         mdt_thread_info_fini(info);
1730         return rc;
1731 }
1732
1733 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1734                          void *karg, void __user *uarg);
1735
1736 static int mdt_set_info(struct tgt_session_info *tsi)
1737 {
1738         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1739         char                    *key;
1740         void                    *val;
1741         int                      keylen, vallen, rc = 0;
1742
1743         ENTRY;
1744
1745         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
1746         if (key == NULL) {
1747                 DEBUG_REQ(D_HA, req, "no set_info key");
1748                 RETURN(err_serious(-EFAULT));
1749         }
1750
1751         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
1752                                       RCL_CLIENT);
1753
1754         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
1755         if (val == NULL) {
1756                 DEBUG_REQ(D_HA, req, "no set_info val");
1757                 RETURN(err_serious(-EFAULT));
1758         }
1759
1760         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
1761                                       RCL_CLIENT);
1762
1763         /* Swab any part of val you need to here */
1764         if (KEY_IS(KEY_READ_ONLY)) {
1765                 spin_lock(&req->rq_export->exp_lock);
1766                 if (*(__u32 *)val)
1767                         *exp_connect_flags_ptr(req->rq_export) |=
1768                                 OBD_CONNECT_RDONLY;
1769                 else
1770                         *exp_connect_flags_ptr(req->rq_export) &=
1771                                 ~OBD_CONNECT_RDONLY;
1772                 spin_unlock(&req->rq_export->exp_lock);
1773         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
1774                 struct changelog_setinfo *cs = val;
1775
1776                 if (vallen != sizeof(*cs)) {
1777                         CERROR("%s: bad changelog_clear setinfo size %d\n",
1778                                tgt_name(tsi->tsi_tgt), vallen);
1779                         RETURN(-EINVAL);
1780                 }
1781                 if (ptlrpc_req_need_swab(req)) {
1782                         __swab64s(&cs->cs_recno);
1783                         __swab32s(&cs->cs_id);
1784                 }
1785
1786                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
1787                                    vallen, val, NULL);
1788         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
1789                 if (vallen > 0)
1790                         obd_export_evict_by_nid(req->rq_export->exp_obd, val);
1791         } else {
1792                 RETURN(-EINVAL);
1793         }
1794         RETURN(rc);
1795 }
1796
1797 static int mdt_readpage(struct tgt_session_info *tsi)
1798 {
1799         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
1800         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
1801         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
1802         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
1803         struct mdt_body         *repbody;
1804         int                      rc;
1805         int                      i;
1806
1807         ENTRY;
1808
1809         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
1810                 RETURN(err_serious(-ENOMEM));
1811
1812         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
1813         if (repbody == NULL || reqbody == NULL)
1814                 RETURN(err_serious(-EFAULT));
1815
1816         /*
1817          * prepare @rdpg before calling lower layers and transfer itself. Here
1818          * reqbody->size contains offset of where to start to read and
1819          * reqbody->nlink contains number bytes to read.
1820          */
1821         rdpg->rp_hash = reqbody->mbo_size;
1822         if (rdpg->rp_hash != reqbody->mbo_size) {
1823                 CERROR("Invalid hash: %#llx != %#llx\n",
1824                        rdpg->rp_hash, reqbody->mbo_size);
1825                 RETURN(-EFAULT);
1826         }
1827
1828         rdpg->rp_attrs = reqbody->mbo_mode;
1829         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
1830                 rdpg->rp_attrs |= LUDA_64BITHASH;
1831         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
1832                                 exp_max_brw_size(tsi->tsi_exp));
1833         rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >>
1834                           PAGE_SHIFT;
1835         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1836         if (rdpg->rp_pages == NULL)
1837                 RETURN(-ENOMEM);
1838
1839         for (i = 0; i < rdpg->rp_npages; ++i) {
1840                 rdpg->rp_pages[i] = alloc_page(GFP_NOFS);
1841                 if (rdpg->rp_pages[i] == NULL)
1842                         GOTO(free_rdpg, rc = -ENOMEM);
1843         }
1844
1845         /* call lower layers to fill allocated pages with directory data */
1846         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
1847         if (rc < 0)
1848                 GOTO(free_rdpg, rc);
1849
1850         /* send pages to client */
1851         rc = tgt_sendpage(tsi, rdpg, rc);
1852
1853         EXIT;
1854 free_rdpg:
1855
1856         for (i = 0; i < rdpg->rp_npages; i++)
1857                 if (rdpg->rp_pages[i] != NULL)
1858                         __free_page(rdpg->rp_pages[i]);
1859         OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
1860
1861         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
1862                 RETURN(0);
1863
1864         return rc;
1865 }
1866
1867 static int mdt_fix_attr_ucred(struct mdt_thread_info *info, __u32 op)
1868 {
1869         struct lu_ucred *uc = mdt_ucred_check(info);
1870         struct lu_attr *attr = &info->mti_attr.ma_attr;
1871
1872         if (uc == NULL)
1873                 return -EINVAL;
1874
1875         if (op != REINT_SETATTR) {
1876                 if ((attr->la_valid & LA_UID) && (attr->la_uid != -1))
1877                         attr->la_uid = uc->uc_fsuid;
1878                 /* for S_ISGID, inherit gid from his parent, such work will be
1879                  * done in cmm/mdd layer, here set all cases as uc->uc_fsgid. */
1880                 if ((attr->la_valid & LA_GID) && (attr->la_gid != -1))
1881                         attr->la_gid = uc->uc_fsgid;
1882         }
1883
1884         return 0;
1885 }
1886
1887 static int mdt_reint_internal(struct mdt_thread_info *info,
1888                               struct mdt_lock_handle *lhc,
1889                               __u32 op)
1890 {
1891         struct req_capsule      *pill = info->mti_pill;
1892         struct mdt_body         *repbody;
1893         int                      rc = 0, rc2;
1894
1895         ENTRY;
1896
1897         rc = mdt_reint_unpack(info, op);
1898         if (rc != 0) {
1899                 CERROR("Can't unpack reint, rc %d\n", rc);
1900                 RETURN(err_serious(rc));
1901         }
1902
1903         /* for replay (no_create) lmm is not needed, client has it already */
1904         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1905                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
1906                                      DEF_REP_MD_SIZE);
1907
1908         /* llog cookies are always 0, the field is kept for compatibility */
1909         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
1910                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
1911
1912         rc = req_capsule_server_pack(pill);
1913         if (rc != 0) {
1914                 CERROR("Can't pack response, rc %d\n", rc);
1915                 RETURN(err_serious(rc));
1916         }
1917
1918         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
1919                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1920                 LASSERT(repbody);
1921                 repbody->mbo_eadatasize = 0;
1922                 repbody->mbo_aclsize = 0;
1923         }
1924
1925         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
1926
1927         /* for replay no cookkie / lmm need, because client have this already */
1928         if (info->mti_spec.no_create)
1929                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
1930                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
1931
1932         rc = mdt_init_ucred_reint(info);
1933         if (rc)
1934                 GOTO(out_shrink, rc);
1935
1936         rc = mdt_fix_attr_ucred(info, op);
1937         if (rc != 0)
1938                 GOTO(out_ucred, rc = err_serious(rc));
1939
1940         rc = mdt_check_resent(info, mdt_reconstruct, lhc);
1941         if (rc < 0) {
1942                 GOTO(out_ucred, rc);
1943         } else if (rc == 1) {
1944                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
1945                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
1946                 GOTO(out_ucred, rc);
1947         }
1948         rc = mdt_reint_rec(info, lhc);
1949         EXIT;
1950 out_ucred:
1951         mdt_exit_ucred(info);
1952 out_shrink:
1953         mdt_client_compatibility(info);
1954         rc2 = mdt_fix_reply(info);
1955         if (rc == 0)
1956                 rc = rc2;
1957         return rc;
1958 }
1959
1960 static long mdt_reint_opcode(struct ptlrpc_request *req,
1961                              const struct req_format **fmt)
1962 {
1963         struct mdt_device       *mdt;
1964         struct mdt_rec_reint    *rec;
1965         long                     opc;
1966
1967         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
1968         if (rec != NULL) {
1969                 opc = rec->rr_opcode;
1970                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
1971                 if (opc < REINT_MAX && fmt[opc] != NULL)
1972                         req_capsule_extend(&req->rq_pill, fmt[opc]);
1973                 else {
1974                         mdt = mdt_exp2dev(req->rq_export);
1975                         CERROR("%s: Unsupported opcode '%ld' from client '%s':"
1976                                " rc = %d\n", req->rq_export->exp_obd->obd_name,
1977                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
1978                         opc = err_serious(-EFAULT);
1979                 }
1980         } else {
1981                 opc = err_serious(-EFAULT);
1982         }
1983         return opc;
1984 }
1985
1986 static int mdt_reint(struct tgt_session_info *tsi)
1987 {
1988         long opc;
1989         int  rc;
1990         static const struct req_format *reint_fmts[REINT_MAX] = {
1991                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
1992                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
1993                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
1994                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
1995                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
1996                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
1997                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
1998                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
1999                 [REINT_MIGRATE]  = &RQF_MDS_REINT_RENAME
2000         };
2001
2002         ENTRY;
2003
2004         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
2005         if (opc >= 0) {
2006                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2007                 /*
2008                  * No lock possible here from client to pass it to reint code
2009                  * path.
2010                  */
2011                 rc = mdt_reint_internal(info, NULL, opc);
2012                 mdt_thread_info_fini(info);
2013         } else {
2014                 rc = opc;
2015         }
2016
2017         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
2018         RETURN(rc);
2019 }
2020
2021 /* this should sync the whole device */
2022 static int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
2023 {
2024         struct dt_device *dt = mdt->mdt_bottom;
2025         int rc;
2026         ENTRY;
2027
2028         rc = dt->dd_ops->dt_sync(env, dt);
2029         RETURN(rc);
2030 }
2031
2032 /* this should sync this object */
2033 static int mdt_object_sync(struct mdt_thread_info *info)
2034 {
2035         struct md_object *next;
2036         int rc;
2037         ENTRY;
2038
2039         if (!mdt_object_exists(info->mti_object)) {
2040                 CWARN("%s: non existing object "DFID": rc = %d\n",
2041                       mdt_obd_name(info->mti_mdt),
2042                       PFID(mdt_object_fid(info->mti_object)), -ESTALE);
2043                 RETURN(-ESTALE);
2044         }
2045         next = mdt_object_child(info->mti_object);
2046         rc = mo_object_sync(info->mti_env, next);
2047
2048         RETURN(rc);
2049 }
2050
2051 static int mdt_sync(struct tgt_session_info *tsi)
2052 {
2053         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2054         struct req_capsule      *pill = tsi->tsi_pill;
2055         struct mdt_body         *body;
2056         int                      rc;
2057
2058         ENTRY;
2059
2060         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
2061                 RETURN(err_serious(-ENOMEM));
2062
2063         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
2064                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
2065         } else {
2066                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2067
2068                 /* sync an object */
2069                 rc = mdt_object_sync(info);
2070                 if (rc == 0) {
2071                         const struct lu_fid *fid;
2072                         struct lu_attr *la = &info->mti_attr.ma_attr;
2073
2074                         info->mti_attr.ma_need = MA_INODE;
2075                         info->mti_attr.ma_valid = 0;
2076                         rc = mdt_attr_get_complex(info, info->mti_object,
2077                                                   &info->mti_attr);
2078                         if (rc == 0) {
2079                                 body = req_capsule_server_get(pill,
2080                                                               &RMF_MDT_BODY);
2081                                 fid = mdt_object_fid(info->mti_object);
2082                                 mdt_pack_attr2body(info, body, la, fid);
2083                         }
2084                 }
2085                 mdt_thread_info_fini(info);
2086         }
2087         if (rc == 0)
2088                 mdt_counter_incr(req, LPROC_MDT_SYNC);
2089
2090         RETURN(rc);
2091 }
2092
2093 /*
2094  * Handle quota control requests to consult current usage/limit, but also
2095  * to configure quota enforcement
2096  */
2097 static int mdt_quotactl(struct tgt_session_info *tsi)
2098 {
2099         struct obd_export       *exp  = tsi->tsi_exp;
2100         struct req_capsule      *pill = tsi->tsi_pill;
2101         struct obd_quotactl     *oqctl, *repoqc;
2102         int                      id, rc;
2103         struct mdt_device       *mdt = mdt_exp2dev(exp);
2104         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2105         struct lu_nodemap       *nodemap;
2106         ENTRY;
2107
2108         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
2109         if (oqctl == NULL)
2110                 RETURN(err_serious(-EPROTO));
2111
2112         rc = req_capsule_server_pack(pill);
2113         if (rc)
2114                 RETURN(err_serious(rc));
2115
2116         nodemap = nodemap_get_from_exp(exp);
2117         if (IS_ERR(nodemap))
2118                 RETURN(PTR_ERR(nodemap));
2119
2120         switch (oqctl->qc_cmd) {
2121                 /* master quotactl */
2122         case Q_SETINFO:
2123         case Q_SETQUOTA:
2124                 if (!nodemap_can_setquota(nodemap))
2125                         GOTO(out_nodemap, rc = -EPERM);
2126         case Q_GETINFO:
2127         case Q_GETQUOTA:
2128                 if (qmt == NULL)
2129                         GOTO(out_nodemap, rc = -EOPNOTSUPP);
2130                 /* slave quotactl */
2131         case Q_GETOINFO:
2132         case Q_GETOQUOTA:
2133                 break;
2134         default:
2135                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2136                 GOTO(out_nodemap, rc = -EFAULT);
2137         }
2138
2139         id = oqctl->qc_id;
2140         switch (oqctl->qc_type) {
2141         case USRQUOTA:
2142                 id = nodemap_map_id(nodemap, NODEMAP_UID,
2143                                     NODEMAP_CLIENT_TO_FS, id);
2144                 break;
2145         case GRPQUOTA:
2146                 id = nodemap_map_id(nodemap, NODEMAP_UID,
2147                                     NODEMAP_CLIENT_TO_FS, id);
2148                 break;
2149         default:
2150                 GOTO(out_nodemap, rc = -EOPNOTSUPP);
2151         }
2152         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
2153         if (repoqc == NULL)
2154                 GOTO(out_nodemap, rc = err_serious(-EFAULT));
2155
2156         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA)
2157                 barrier_exit(tsi->tsi_tgt->lut_bottom);
2158
2159         if (oqctl->qc_id != id)
2160                 swap(oqctl->qc_id, id);
2161
2162         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA) {
2163                 if (unlikely(!barrier_entry(tsi->tsi_tgt->lut_bottom)))
2164                         RETURN(-EINPROGRESS);
2165         }
2166
2167         switch (oqctl->qc_cmd) {
2168
2169         case Q_GETINFO:
2170         case Q_SETINFO:
2171         case Q_SETQUOTA:
2172         case Q_GETQUOTA:
2173                 /* forward quotactl request to QMT */
2174                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
2175                 break;
2176
2177         case Q_GETOINFO:
2178         case Q_GETOQUOTA:
2179                 /* slave quotactl */
2180                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
2181                                    oqctl);
2182                 break;
2183
2184         default:
2185                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
2186                 GOTO(out_nodemap, rc = -EFAULT);
2187         }
2188
2189         if (oqctl->qc_id != id)
2190                 swap(oqctl->qc_id, id);
2191
2192         *repoqc = *oqctl;
2193
2194         EXIT;
2195
2196 out_nodemap:
2197         nodemap_putref(nodemap);
2198
2199         return rc;
2200 }
2201
2202 /** clone llog ctxt from child (mdd)
2203  * This allows remote llog (replicator) access.
2204  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
2205  * context was originally set up, or we can handle them directly.
2206  * I choose the latter, but that means I need any llog
2207  * contexts set up by child to be accessable by the mdt.  So we clone the
2208  * context into our context list here.
2209  */
2210 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
2211                                int idx)
2212 {
2213         struct md_device  *next = mdt->mdt_child;
2214         struct llog_ctxt *ctxt;
2215         int rc;
2216
2217         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
2218                 return 0;
2219
2220         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
2221         if (rc || ctxt == NULL) {
2222                 return 0;
2223         }
2224
2225         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
2226         if (rc)
2227                 CERROR("Can't set mdt ctxt %d\n", rc);
2228
2229         return rc;
2230 }
2231
2232 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
2233                                  struct mdt_device *mdt, int idx)
2234 {
2235         struct llog_ctxt *ctxt;
2236
2237         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
2238         if (ctxt == NULL)
2239                 return 0;
2240         /* Put once for the get we just did, and once for the clone */
2241         llog_ctxt_put(ctxt);
2242         llog_ctxt_put(ctxt);
2243         return 0;
2244 }
2245
2246 /*
2247  * sec context handlers
2248  */
2249 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
2250 {
2251         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
2252
2253         return 0;
2254 }
2255
2256 /*
2257  * quota request handlers
2258  */
2259 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
2260 {
2261         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
2262         struct lu_device        *qmt = mdt->mdt_qmt_dev;
2263         int                      rc;
2264         ENTRY;
2265
2266         if (qmt == NULL)
2267                 RETURN(err_serious(-EOPNOTSUPP));
2268
2269         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
2270         RETURN(rc);
2271 }
2272
2273 struct mdt_object *mdt_object_new(const struct lu_env *env,
2274                                   struct mdt_device *d,
2275                                   const struct lu_fid *f)
2276 {
2277         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
2278         struct lu_object *o;
2279         struct mdt_object *m;
2280         ENTRY;
2281
2282         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
2283         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
2284         if (unlikely(IS_ERR(o)))
2285                 m = (struct mdt_object *)o;
2286         else
2287                 m = mdt_obj(o);
2288         RETURN(m);
2289 }
2290
2291 struct mdt_object *mdt_object_find(const struct lu_env *env,
2292                                    struct mdt_device *d,
2293                                    const struct lu_fid *f)
2294 {
2295         struct lu_object *o;
2296         struct mdt_object *m;
2297         ENTRY;
2298
2299         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
2300         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
2301         if (unlikely(IS_ERR(o)))
2302                 m = (struct mdt_object *)o;
2303         else
2304                 m = mdt_obj(o);
2305
2306         RETURN(m);
2307 }
2308
2309 /**
2310  * Asyncronous commit for mdt device.
2311  *
2312  * Pass asynchonous commit call down the MDS stack.
2313  *
2314  * \param env environment
2315  * \param mdt the mdt device
2316  */
2317 static void mdt_device_commit_async(const struct lu_env *env,
2318                                     struct mdt_device *mdt)
2319 {
2320         struct dt_device *dt = mdt->mdt_bottom;
2321         int rc;
2322         ENTRY;
2323
2324         rc = dt->dd_ops->dt_commit_async(env, dt);
2325         if (unlikely(rc != 0))
2326                 CWARN("%s: async commit start failed: rc = %d\n",
2327                       mdt_obd_name(mdt), rc);
2328         atomic_inc(&mdt->mdt_async_commit_count);
2329         EXIT;
2330 }
2331
2332 /**
2333  * Mark the lock as "synchonous".
2334  *
2335  * Mark the lock to deffer transaction commit to the unlock time.
2336  *
2337  * \param lock the lock to mark as "synchonous"
2338  *
2339  * \see mdt_is_lock_sync
2340  * \see mdt_save_lock
2341  */
2342 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
2343 {
2344         lock->l_ast_data = (void*)1;
2345 }
2346
2347 /**
2348  * Check whehter the lock "synchonous" or not.
2349  *
2350  * \param lock the lock to check
2351  * \retval 1 the lock is "synchonous"
2352  * \retval 0 the lock isn't "synchronous"
2353  *
2354  * \see mdt_set_lock_sync
2355  * \see mdt_save_lock
2356  */
2357 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
2358 {
2359         return lock->l_ast_data != NULL;
2360 }
2361
2362 /**
2363  * Blocking AST for mdt locks.
2364  *
2365  * Starts transaction commit if in case of COS lock conflict or
2366  * deffers such a commit to the mdt_save_lock.
2367  *
2368  * \param lock the lock which blocks a request or cancelling lock
2369  * \param desc unused
2370  * \param data unused
2371  * \param flag indicates whether this cancelling or blocking callback
2372  * \retval 0
2373  * \see ldlm_blocking_ast_nocheck
2374  */
2375 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2376                      void *data, int flag)
2377 {
2378         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2379         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
2380         int rc;
2381         ENTRY;
2382
2383         if (flag == LDLM_CB_CANCELING)
2384                 RETURN(0);
2385
2386         lock_res_and_lock(lock);
2387         if (lock->l_blocking_ast != mdt_blocking_ast) {
2388                 unlock_res_and_lock(lock);
2389                 RETURN(0);
2390         }
2391         if (lock->l_req_mode & (LCK_PW | LCK_EX) &&
2392             lock->l_blocking_lock != NULL) {
2393                 if (mdt_cos_is_enabled(mdt) &&
2394                     lock->l_client_cookie !=
2395                     lock->l_blocking_lock->l_client_cookie)
2396                         mdt_set_lock_sync(lock);
2397                 else if (mdt_slc_is_enabled(mdt) &&
2398                          ldlm_is_cos_incompat(lock->l_blocking_lock))
2399                         mdt_set_lock_sync(lock);
2400         }
2401         rc = ldlm_blocking_ast_nocheck(lock);
2402
2403         /* There is no lock conflict if l_blocking_lock == NULL,
2404          * it indicates a blocking ast sent from ldlm_lock_decref_internal
2405          * when the last reference to a local lock was released */
2406         if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) {
2407                 struct lu_env env;
2408
2409                 rc = lu_env_init(&env, LCT_LOCAL);
2410                 if (unlikely(rc != 0))
2411                         CWARN("%s: lu_env initialization failed, cannot "
2412                               "start asynchronous commit: rc = %d\n",
2413                               obd->obd_name, rc);
2414                 else
2415                         mdt_device_commit_async(&env, mdt);
2416                 lu_env_fini(&env);
2417         }
2418         RETURN(rc);
2419 }
2420
2421 /*
2422  * Blocking AST for cross-MDT lock
2423  *
2424  * Discard lock from uncommitted_slc_locks and cancel it.
2425  *
2426  * \param lock  the lock which blocks a request or cancelling lock
2427  * \param desc  unused
2428  * \param data  unused
2429  * \param flag  indicates whether this cancelling or blocking callback
2430  * \retval      0 on success
2431  * \retval      negative number on error
2432  */
2433 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
2434                             void *data, int flag)
2435 {
2436         int rc = 0;
2437         ENTRY;
2438
2439         switch (flag) {
2440         case LDLM_CB_BLOCKING: {
2441                 struct lustre_handle lockh;
2442
2443                 ldlm_lock2handle(lock, &lockh);
2444                 rc = ldlm_cli_cancel(&lockh,
2445                         ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
2446                 if (rc < 0) {
2447                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
2448                         RETURN(rc);
2449                 }
2450                 break;
2451         }
2452         case LDLM_CB_CANCELING: {
2453                 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
2454                 struct mdt_device *mdt =
2455                                 mdt_dev(obd->obd_lu_dev->ld_site->ls_top_dev);
2456
2457                 LDLM_DEBUG(lock, "Revoke remote lock\n");
2458
2459                 /* discard slc lock here so that it can be cleaned anytime,
2460                  * especially for cleanup_resource() */
2461                 tgt_discard_slc_lock(&mdt->mdt_lut, lock);
2462
2463                 /* once we cache lock, l_ast_data is set to mdt_object */
2464                 if (lock->l_ast_data != NULL) {
2465                         struct mdt_object *mo = lock->l_ast_data;
2466                         struct lu_env env;
2467
2468                         rc = lu_env_init(&env, LCT_MD_THREAD);
2469                         if (unlikely(rc != 0)) {
2470                                 CWARN("%s: lu_env initialization failed, object"
2471                                       "%p "DFID" is leaked!\n",
2472                                       obd->obd_name, mo,
2473                                       PFID(mdt_object_fid(mo)));
2474                                 RETURN(rc);
2475                         }
2476
2477                         if (lock->l_policy_data.l_inodebits.bits &
2478                             (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)) {
2479                                 rc = mo_invalidate(&env, mdt_object_child(mo));
2480                                 mo->mot_cache_attr = 0;
2481                         }
2482                         mdt_object_put(&env, mo);
2483                         lu_env_fini(&env);
2484                 }
2485                 break;
2486         }
2487         default:
2488                 LBUG();
2489         }
2490
2491         RETURN(rc);
2492 }
2493
2494 int mdt_check_resent_lock(struct mdt_thread_info *info,
2495                           struct mdt_object *mo,
2496                           struct mdt_lock_handle *lhc)
2497 {
2498         /* the lock might already be gotten in ldlm_handle_enqueue() */
2499         if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
2500                 struct ptlrpc_request *req = mdt_info_req(info);
2501                 struct ldlm_lock      *lock;
2502
2503                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2504                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
2505                 if (lock == NULL) {
2506                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
2507                          * a resend case, however, it could be already destroyed
2508                          * due to client eviction or a raced cancel RPC. */
2509                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx",
2510                                           lhc->mlh_reg_lh.cookie);
2511                         RETURN(-ESTALE);
2512                 }
2513
2514                 if (!fid_res_name_eq(mdt_object_fid(mo),
2515                                      &lock->l_resource->lr_name)) {
2516                         CWARN("%s: Although resent, but still not "
2517                               "get child lock:"DFID"\n",
2518                               info->mti_exp->exp_obd->obd_name,
2519                               PFID(mdt_object_fid(mo)));
2520                         LDLM_LOCK_PUT(lock);
2521                         RETURN(-EPROTO);
2522                 }
2523                 LDLM_LOCK_PUT(lock);
2524                 return 0;
2525         }
2526         return 1;
2527 }
2528
2529 int mdt_remote_object_lock(struct mdt_thread_info *mti, struct mdt_object *o,
2530                            const struct lu_fid *fid, struct lustre_handle *lh,
2531                            enum ldlm_mode mode, __u64 ibits, bool nonblock,
2532                            bool cache)
2533 {
2534         struct ldlm_enqueue_info *einfo = &mti->mti_einfo;
2535         union ldlm_policy_data *policy = &mti->mti_policy;
2536         struct ldlm_res_id *res_id = &mti->mti_res_id;
2537         int rc = 0;
2538         ENTRY;
2539
2540         LASSERT(mdt_object_remote(o));
2541
2542         fid_build_reg_res_name(fid, res_id);
2543
2544         memset(einfo, 0, sizeof(*einfo));
2545         einfo->ei_type = LDLM_IBITS;
2546         einfo->ei_mode = mode;
2547         einfo->ei_cb_bl = mdt_remote_blocking_ast;
2548         einfo->ei_cb_cp = ldlm_completion_ast;
2549         einfo->ei_enq_slave = 0;
2550         einfo->ei_res_id = res_id;
2551         if (nonblock)
2552                 einfo->ei_nonblock = 1;
2553         if (cache) {
2554                 /*
2555                  * if we cache lock, couple lock with mdt_object, so that object
2556                  * can be easily found in lock ASTs.
2557                  */
2558                 mdt_object_get(mti->mti_env, o);
2559                 einfo->ei_cbdata = o;
2560         }
2561
2562         memset(policy, 0, sizeof(*policy));
2563         policy->l_inodebits.bits = ibits;
2564
2565         rc = mo_object_lock(mti->mti_env, mdt_object_child(o), lh, einfo,
2566                             policy);
2567         if (rc < 0 && cache) {
2568                 mdt_object_put(mti->mti_env, o);
2569                 einfo->ei_cbdata = NULL;
2570         }
2571         RETURN(rc);
2572 }
2573
2574 static int mdt_object_local_lock(struct mdt_thread_info *info,
2575                                  struct mdt_object *o,
2576                                  struct mdt_lock_handle *lh, __u64 ibits,
2577                                  bool nonblock, bool cos_incompat)
2578 {
2579         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
2580         union ldlm_policy_data *policy = &info->mti_policy;
2581         struct ldlm_res_id *res_id = &info->mti_res_id;
2582         __u64 dlmflags = 0;
2583         int rc;
2584         ENTRY;
2585
2586         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
2587         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
2588         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
2589         LASSERT(lh->mlh_type != MDT_NUL_LOCK);
2590
2591         if (cos_incompat) {
2592                 LASSERT(lh->mlh_reg_mode == LCK_PW ||
2593                         lh->mlh_reg_mode == LCK_EX);
2594                 dlmflags |= LDLM_FL_COS_INCOMPAT;
2595         } else if (mdt_cos_is_enabled(info->mti_mdt)) {
2596                 dlmflags |= LDLM_FL_COS_ENABLED;
2597         }
2598
2599         /* Only enqueue LOOKUP lock for remote object */
2600         if (mdt_object_remote(o))
2601                 LASSERT(ibits == MDS_INODELOCK_LOOKUP);
2602
2603         if (lh->mlh_type == MDT_PDO_LOCK) {
2604                 /* check for exists after object is locked */
2605                 if (mdt_object_exists(o) == 0) {
2606                         /* Non-existent object shouldn't have PDO lock */
2607                         RETURN(-ESTALE);
2608                 } else {
2609                         /* Non-dir object shouldn't have PDO lock */
2610                         if (!S_ISDIR(lu_object_attr(&o->mot_obj)))
2611                                 RETURN(-ENOTDIR);
2612                 }
2613         }
2614
2615         memset(policy, 0, sizeof(*policy));
2616         fid_build_reg_res_name(mdt_object_fid(o), res_id);
2617
2618         dlmflags |= LDLM_FL_ATOMIC_CB;
2619         if (nonblock)
2620                 dlmflags |= LDLM_FL_BLOCK_NOWAIT;
2621
2622         /*
2623          * Take PDO lock on whole directory and build correct @res_id for lock
2624          * on part of directory.
2625          */
2626         if (lh->mlh_pdo_hash != 0) {
2627                 LASSERT(lh->mlh_type == MDT_PDO_LOCK);
2628                 mdt_lock_pdo_mode(info, o, lh);
2629                 if (lh->mlh_pdo_mode != LCK_NL) {
2630                         /*
2631                          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it
2632                          * is never going to be sent to client and we do not
2633                          * want it slowed down due to possible cancels.
2634                          */
2635                         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
2636                         rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
2637                                           policy, res_id, dlmflags,
2638                                           info->mti_exp == NULL ? NULL :
2639                                           &info->mti_exp->exp_handle.h_cookie);
2640                         if (unlikely(rc != 0))
2641                                 GOTO(out_unlock, rc);
2642                 }
2643
2644                 /*
2645                  * Finish res_id initializing by name hash marking part of
2646                  * directory which is taking modification.
2647                  */
2648                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
2649         }
2650
2651         policy->l_inodebits.bits = ibits;
2652
2653         /*
2654          * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
2655          * going to be sent to client. If it is - mdt_intent_policy() path will
2656          * fix it up and turn FL_LOCAL flag off.
2657          */
2658         rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
2659                           res_id, LDLM_FL_LOCAL_ONLY | dlmflags,
2660                           info->mti_exp == NULL ? NULL :
2661                           &info->mti_exp->exp_handle.h_cookie);
2662 out_unlock:
2663         if (rc != 0)
2664                 mdt_object_unlock(info, o, lh, 1);
2665         else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) &&
2666                    lh->mlh_pdo_hash != 0 &&
2667                    (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
2668                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
2669
2670         RETURN(rc);
2671 }
2672
2673 static int
2674 mdt_object_lock_internal(struct mdt_thread_info *info, struct mdt_object *o,
2675                          struct mdt_lock_handle *lh, __u64 ibits, bool nonblock,
2676                          bool cos_incompat)
2677 {
2678         struct mdt_lock_handle *local_lh = NULL;
2679         int rc;
2680         ENTRY;
2681
2682         if (!mdt_object_remote(o)) {
2683                 rc = mdt_object_local_lock(info, o, lh, ibits, nonblock,
2684                                            cos_incompat);
2685                 RETURN(rc);
2686         }
2687
2688         /* XXX do not support PERM/LAYOUT/XATTR lock for remote object yet */
2689         ibits &= ~(MDS_INODELOCK_PERM | MDS_INODELOCK_LAYOUT |
2690                    MDS_INODELOCK_XATTR);
2691
2692         /* Only enqueue LOOKUP lock for remote object */
2693         if (ibits & MDS_INODELOCK_LOOKUP) {
2694                 rc = mdt_object_local_lock(info, o, lh, MDS_INODELOCK_LOOKUP,
2695                                            nonblock, cos_incompat);
2696                 if (rc != ELDLM_OK)
2697                         RETURN(rc);
2698
2699                 local_lh = lh;
2700         }
2701
2702         if (ibits & MDS_INODELOCK_UPDATE) {
2703                 /* Sigh, PDO needs to enqueue 2 locks right now, but
2704                  * enqueue RPC can only request 1 lock, to avoid extra
2705                  * RPC, so it will instead enqueue EX lock for remote
2706                  * object anyway XXX*/
2707                 if (lh->mlh_type == MDT_PDO_LOCK &&
2708                     lh->mlh_pdo_hash != 0) {
2709                         CDEBUG(D_INFO, "%s: "DFID" convert PDO lock to"
2710                                "EX lock.\n", mdt_obd_name(info->mti_mdt),
2711                                PFID(mdt_object_fid(o)));
2712                         lh->mlh_pdo_hash = 0;
2713                         lh->mlh_rreg_mode = LCK_EX;
2714                         lh->mlh_type = MDT_REG_LOCK;
2715                 }
2716                 rc = mdt_remote_object_lock(info, o, mdt_object_fid(o),
2717                                             &lh->mlh_rreg_lh,
2718                                             lh->mlh_rreg_mode,
2719                                             MDS_INODELOCK_UPDATE, nonblock,
2720                                             false);
2721                 if (rc != ELDLM_OK) {
2722                         if (local_lh != NULL)
2723                                 mdt_object_unlock(info, o, local_lh, rc);
2724                         RETURN(rc);
2725                 }
2726         }
2727
2728         RETURN(0);
2729 }
2730
2731 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2732                     struct mdt_lock_handle *lh, __u64 ibits)
2733 {
2734         return mdt_object_lock_internal(info, o, lh, ibits, false, false);
2735 }
2736
2737 int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
2738                           struct mdt_lock_handle *lh, __u64 ibits,
2739                           bool cos_incompat)
2740 {
2741         LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
2742         return mdt_object_lock_internal(info, o, lh, ibits, false,
2743                                         cos_incompat);
2744 }
2745
2746 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *o,
2747                         struct mdt_lock_handle *lh, __u64 ibits)
2748 {
2749         struct mdt_lock_handle tmp = *lh;
2750         int rc;
2751
2752         rc = mdt_object_lock_internal(info, o, &tmp, ibits, true, false);
2753         if (rc == 0)
2754                 *lh = tmp;
2755
2756         return rc == 0;
2757 }
2758
2759 int mdt_reint_object_lock_try(struct mdt_thread_info *info,
2760                               struct mdt_object *o, struct mdt_lock_handle *lh,
2761                               __u64 ibits, bool cos_incompat)
2762 {
2763         struct mdt_lock_handle tmp = *lh;
2764         int rc;
2765
2766         LASSERT(lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX);
2767         rc = mdt_object_lock_internal(info, o, &tmp, ibits, true, cos_incompat);
2768         if (rc == 0)
2769                 *lh = tmp;
2770
2771         return rc == 0;
2772 }
2773
2774 /**
2775  * Save a lock within request object.
2776  *
2777  * Keep the lock referenced until whether client ACK or transaction
2778  * commit happens or release the lock immediately depending on input
2779  * parameters. If COS is ON, a write lock is converted to COS lock
2780  * before saving.
2781  *
2782  * \param info thead info object
2783  * \param h lock handle
2784  * \param mode lock mode
2785  * \param decref force immediate lock releasing
2786  */
2787 static void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
2788                           enum ldlm_mode mode, int decref)
2789 {
2790         ENTRY;
2791
2792         if (lustre_handle_is_used(h)) {
2793                 if (decref || !info->mti_has_trans ||
2794                     !(mode & (LCK_PW | LCK_EX))) {
2795                         mdt_fid_unlock(h, mode);
2796                 } else {
2797                         struct mdt_device *mdt = info->mti_mdt;
2798                         struct ldlm_lock *lock = ldlm_handle2lock(h);
2799                         struct ptlrpc_request *req = mdt_info_req(info);
2800                         int cos;
2801
2802                         cos = (mdt_cos_is_enabled(mdt) ||
2803                                mdt_slc_is_enabled(mdt));
2804
2805                         LASSERTF(lock != NULL, "no lock for cookie %#llx\n",
2806                                  h->cookie);
2807
2808                         /* there is no request if mdt_object_unlock() is called
2809                          * from mdt_export_cleanup()->mdt_add_dirty_flag() */
2810                         if (likely(req != NULL)) {
2811                                 CDEBUG(D_HA, "request = %p reply state = %p"
2812                                        " transno = %lld\n", req,
2813                                        req->rq_reply_state, req->rq_transno);
2814                                 if (cos) {
2815                                         ldlm_lock_downgrade(lock, LCK_COS);
2816                                         mode = LCK_COS;
2817                                 }
2818                                 ptlrpc_save_lock(req, h, mode, cos);
2819                         } else {
2820                                 mdt_fid_unlock(h, mode);
2821                         }
2822                         if (mdt_is_lock_sync(lock)) {
2823                                 CDEBUG(D_HA, "found sync-lock,"
2824                                        " async commit started\n");
2825                                 mdt_device_commit_async(info->mti_env,
2826                                                         mdt);
2827                         }
2828                         LDLM_LOCK_PUT(lock);
2829                 }
2830                 h->cookie = 0ull;
2831         }
2832
2833         EXIT;
2834 }
2835
2836 /**
2837  * Save cross-MDT lock in uncommitted_slc_locks
2838  *
2839  * Keep the lock referenced until transaction commit happens or release the lock
2840  * immediately depending on input parameters.
2841  *
2842  * \param info thead info object
2843  * \param h lock handle
2844  * \param mode lock mode
2845  * \param decref force immediate lock releasing
2846  */
2847 static void mdt_save_remote_lock(struct mdt_thread_info *info,
2848                                  struct mdt_object *o, struct lustre_handle *h,
2849                                  enum ldlm_mode mode, int decref)
2850 {
2851         ENTRY;
2852
2853         if (lustre_handle_is_used(h)) {
2854                 struct ldlm_lock *lock = ldlm_handle2lock(h);
2855
2856                 if (o != NULL &&
2857                     (lock->l_policy_data.l_inodebits.bits &
2858                      (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)))
2859                         mo_invalidate(info->mti_env, mdt_object_child(o));
2860
2861                 if (decref || !info->mti_has_trans ||
2862                     !(mode & (LCK_PW | LCK_EX))) {
2863                         ldlm_lock_decref_and_cancel(h, mode);
2864                         LDLM_LOCK_PUT(lock);
2865                 } else {
2866                         struct ptlrpc_request *req = mdt_info_req(info);
2867
2868                         LASSERT(req != NULL);
2869                         tgt_save_slc_lock(&info->mti_mdt->mdt_lut, lock,
2870                                           req->rq_transno);
2871                         ldlm_lock_decref(h, mode);
2872                 }
2873                 h->cookie = 0ull;
2874         }
2875
2876         EXIT;
2877 }
2878
2879 /**
2880  * Unlock mdt object.
2881  *
2882  * Immeditely release the regular lock and the PDO lock or save the
2883  * lock in request and keep them referenced until client ACK or
2884  * transaction commit.
2885  *
2886  * \param info thread info object
2887  * \param o mdt object
2888  * \param lh mdt lock handle referencing regular and PDO locks
2889  * \param decref force immediate lock releasing
2890  *
2891  * XXX o is not used and may be NULL, see hsm_cdt_request_completed().
2892  */
2893 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
2894                        struct mdt_lock_handle *lh, int decref)
2895 {
2896         ENTRY;
2897
2898         mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
2899         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
2900         mdt_save_remote_lock(info, o, &lh->mlh_rreg_lh, lh->mlh_rreg_mode,
2901                              decref);
2902
2903         EXIT;
2904 }
2905
2906 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
2907                                         const struct lu_fid *f,
2908                                         struct mdt_lock_handle *lh,
2909                                         __u64 ibits)
2910 {
2911         struct mdt_object *o;
2912
2913         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
2914         if (!IS_ERR(o)) {
2915                 int rc;
2916
2917                 rc = mdt_object_lock(info, o, lh, ibits);
2918                 if (rc != 0) {
2919                         mdt_object_put(info->mti_env, o);
2920                         o = ERR_PTR(rc);
2921                 }
2922         }
2923         return o;
2924 }
2925
2926 void mdt_object_unlock_put(struct mdt_thread_info * info,
2927                            struct mdt_object * o,
2928                            struct mdt_lock_handle *lh,
2929                            int decref)
2930 {
2931         mdt_object_unlock(info, o, lh, decref);
2932         mdt_object_put(info->mti_env, o);
2933 }
2934
2935 /*
2936  * Generic code handling requests that have struct mdt_body passed in:
2937  *
2938  *  - extract mdt_body from request and save it in @info, if present;
2939  *
2940  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
2941  *  @info;
2942  *
2943  *  - if HABEO_CORPUS flag is set for this request type check whether object
2944  *  actually exists on storage (lu_object_exists()).
2945  *
2946  */
2947 static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags)
2948 {
2949         const struct mdt_body    *body;
2950         struct mdt_object        *obj;
2951         const struct lu_env      *env;
2952         struct req_capsule       *pill;
2953         int                       rc;
2954         ENTRY;
2955
2956         env = info->mti_env;
2957         pill = info->mti_pill;
2958
2959         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
2960         if (body == NULL)
2961                 RETURN(-EFAULT);
2962
2963         if (!(body->mbo_valid & OBD_MD_FLID))
2964                 RETURN(0);
2965
2966         if (!fid_is_sane(&body->mbo_fid1)) {
2967                 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
2968                 RETURN(-EINVAL);
2969         }
2970
2971         obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
2972         if (!IS_ERR(obj)) {
2973                 if ((flags & HABEO_CORPUS) && !mdt_object_exists(obj)) {
2974                         mdt_object_put(env, obj);
2975                         rc = -ENOENT;
2976                 } else {
2977                         info->mti_object = obj;
2978                         rc = 0;
2979                 }
2980         } else
2981                 rc = PTR_ERR(obj);
2982
2983         RETURN(rc);
2984 }
2985
2986 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
2987 {
2988         struct req_capsule *pill = info->mti_pill;
2989         int rc;
2990         ENTRY;
2991
2992         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
2993                 rc = mdt_body_unpack(info, flags);
2994         else
2995                 rc = 0;
2996
2997         if (rc == 0 && (flags & HABEO_REFERO)) {
2998                 /* Pack reply. */
2999                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
3000                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
3001                                              DEF_REP_MD_SIZE);
3002                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
3003                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
3004                                              RCL_SERVER, 0);
3005
3006                 rc = req_capsule_server_pack(pill);
3007         }
3008         RETURN(rc);
3009 }
3010
3011 void mdt_lock_handle_init(struct mdt_lock_handle *lh)
3012 {
3013         lh->mlh_type = MDT_NUL_LOCK;
3014         lh->mlh_reg_lh.cookie = 0ull;
3015         lh->mlh_reg_mode = LCK_MINMODE;
3016         lh->mlh_pdo_lh.cookie = 0ull;
3017         lh->mlh_pdo_mode = LCK_MINMODE;
3018         lh->mlh_rreg_lh.cookie = 0ull;
3019         lh->mlh_rreg_mode = LCK_MINMODE;
3020 }
3021
3022 void mdt_lock_handle_fini(struct mdt_lock_handle *lh)
3023 {
3024         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
3025         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
3026 }
3027
3028 /*
3029  * Initialize fields of struct mdt_thread_info. Other fields are left in
3030  * uninitialized state, because it's too expensive to zero out whole
3031  * mdt_thread_info (> 1K) on each request arrival.
3032  */
3033 void mdt_thread_info_init(struct ptlrpc_request *req,
3034                           struct mdt_thread_info *info)
3035 {
3036         int i;
3037
3038         info->mti_pill = &req->rq_pill;
3039
3040         /* lock handle */
3041         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3042                 mdt_lock_handle_init(&info->mti_lh[i]);
3043
3044         /* mdt device: it can be NULL while CONNECT */
3045         if (req->rq_export) {
3046                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
3047                 info->mti_exp = req->rq_export;
3048         } else
3049                 info->mti_mdt = NULL;
3050         info->mti_env = req->rq_svc_thread->t_env;
3051         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
3052
3053         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
3054         info->mti_big_buf = LU_BUF_NULL;
3055         info->mti_body = NULL;
3056         info->mti_object = NULL;
3057         info->mti_dlm_req = NULL;
3058         info->mti_has_trans = 0;
3059         info->mti_cross_ref = 0;
3060         info->mti_opdata = 0;
3061         info->mti_big_lmm_used = 0;
3062
3063         info->mti_spec.no_create = 0;
3064         info->mti_spec.sp_rm_entry = 0;
3065         info->mti_spec.sp_permitted = 0;
3066         info->mti_spec.sp_migrate_close = 0;
3067
3068         info->mti_spec.u.sp_ea.eadata = NULL;
3069         info->mti_spec.u.sp_ea.eadatalen = 0;
3070 }
3071
3072 void mdt_thread_info_fini(struct mdt_thread_info *info)
3073 {
3074         int i;
3075
3076         if (info->mti_object != NULL) {
3077                 mdt_object_put(info->mti_env, info->mti_object);
3078                 info->mti_object = NULL;
3079         }
3080
3081         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
3082                 mdt_lock_handle_fini(&info->mti_lh[i]);
3083         info->mti_env = NULL;
3084         info->mti_pill = NULL;
3085         info->mti_exp = NULL;
3086
3087         if (unlikely(info->mti_big_buf.lb_buf != NULL))
3088                 lu_buf_free(&info->mti_big_buf);
3089 }
3090
3091 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
3092 {
3093         struct mdt_thread_info  *mti;
3094
3095         mti = mdt_th_info(tsi->tsi_env);
3096         LASSERT(mti != NULL);
3097
3098         mdt_thread_info_init(tgt_ses_req(tsi), mti);
3099         if (tsi->tsi_corpus != NULL) {
3100                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
3101                 lu_object_get(tsi->tsi_corpus);
3102         }
3103         mti->mti_body = tsi->tsi_mdt_body;
3104         mti->mti_dlm_req = tsi->tsi_dlm_req;
3105
3106         return mti;
3107 }
3108
3109 static int mdt_tgt_connect(struct tgt_session_info *tsi)
3110 {
3111         if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
3112             cfs_fail_val ==
3113             tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id) {
3114                 set_current_state(TASK_UNINTERRUPTIBLE);
3115                 schedule_timeout(msecs_to_jiffies(3 * MSEC_PER_SEC));
3116         }
3117
3118         return tgt_connect(tsi);
3119 }
3120
3121 enum mdt_it_code {
3122         MDT_IT_OPEN,
3123         MDT_IT_OCREAT,
3124         MDT_IT_CREATE,
3125         MDT_IT_GETATTR,
3126         MDT_IT_READDIR,
3127         MDT_IT_LOOKUP,
3128         MDT_IT_UNLINK,
3129         MDT_IT_TRUNC,
3130         MDT_IT_GETXATTR,
3131         MDT_IT_LAYOUT,
3132         MDT_IT_QUOTA,
3133         MDT_IT_NR
3134 };
3135
3136 static int mdt_intent_getattr(enum mdt_it_code opcode,
3137                               struct mdt_thread_info *info,
3138                               struct ldlm_lock **,
3139                               __u64);
3140
3141 static int mdt_intent_getxattr(enum mdt_it_code opcode,
3142                                 struct mdt_thread_info *info,
3143                                 struct ldlm_lock **lockp,
3144                                 __u64 flags);
3145
3146 static int mdt_intent_layout(enum mdt_it_code opcode,
3147                              struct mdt_thread_info *info,
3148                              struct ldlm_lock **,
3149                              __u64);
3150 static int mdt_intent_reint(enum mdt_it_code opcode,
3151                             struct mdt_thread_info *info,
3152                             struct ldlm_lock **,
3153                             __u64);
3154
3155 static struct mdt_it_flavor {
3156         const struct req_format *it_fmt;
3157         __u32                    it_flags;
3158         int                    (*it_act)(enum mdt_it_code ,
3159                                          struct mdt_thread_info *,
3160                                          struct ldlm_lock **,
3161                                          __u64);
3162         long                     it_reint;
3163 } mdt_it_flavor[] = {
3164         [MDT_IT_OPEN]     = {
3165                 .it_fmt   = &RQF_LDLM_INTENT,
3166                 /*.it_flags = HABEO_REFERO,*/
3167                 .it_flags = 0,
3168                 .it_act   = mdt_intent_reint,
3169                 .it_reint = REINT_OPEN
3170         },
3171         [MDT_IT_OCREAT]   = {
3172                 .it_fmt   = &RQF_LDLM_INTENT,
3173                 /*
3174                  * OCREAT is not a MUTABOR request as if the file
3175                  * already exists.
3176                  * We do the extra check of OBD_CONNECT_RDONLY in
3177                  * mdt_reint_open() when we really need to create
3178                  * the object.
3179                  */
3180                 .it_flags = 0,
3181                 .it_act   = mdt_intent_reint,
3182                 .it_reint = REINT_OPEN
3183         },
3184         [MDT_IT_CREATE]   = {
3185                 .it_fmt   = &RQF_LDLM_INTENT,
3186                 .it_flags = MUTABOR,
3187                 .it_act   = mdt_intent_reint,
3188                 .it_reint = REINT_CREATE
3189         },
3190         [MDT_IT_GETATTR]  = {
3191                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3192                 .it_flags = HABEO_REFERO,
3193                 .it_act   = mdt_intent_getattr
3194         },
3195         [MDT_IT_READDIR]  = {
3196                 .it_fmt   = NULL,
3197                 .it_flags = 0,
3198                 .it_act   = NULL
3199         },
3200         [MDT_IT_LOOKUP]   = {
3201                 .it_fmt   = &RQF_LDLM_INTENT_GETATTR,
3202                 .it_flags = HABEO_REFERO,
3203                 .it_act   = mdt_intent_getattr
3204         },
3205         [MDT_IT_UNLINK]   = {
3206                 .it_fmt   = &RQF_LDLM_INTENT_UNLINK,
3207                 .it_flags = MUTABOR,
3208                 .it_act   = NULL,
3209                 .it_reint = REINT_UNLINK
3210         },
3211         [MDT_IT_TRUNC]    = {
3212                 .it_fmt   = NULL,
3213                 .it_flags = MUTABOR,
3214                 .it_act   = NULL
3215         },
3216         [MDT_IT_GETXATTR] = {
3217                 .it_fmt   = &RQF_LDLM_INTENT_GETXATTR,
3218                 .it_flags = HABEO_CORPUS,
3219                 .it_act   = mdt_intent_getxattr
3220         },
3221         [MDT_IT_LAYOUT] = {
3222                 .it_fmt   = &RQF_LDLM_INTENT_LAYOUT,
3223                 .it_flags = 0,
3224                 .it_act   = mdt_intent_layout
3225         }
3226 };
3227
3228 static int
3229 mdt_intent_lock_replace(struct mdt_thread_info *info,
3230                         struct ldlm_lock **lockp,
3231                         struct mdt_lock_handle *lh,
3232                         __u64 flags, int result)
3233 {
3234         struct ptlrpc_request  *req = mdt_info_req(info);
3235         struct ldlm_lock       *lock = *lockp;
3236         struct ldlm_lock       *new_lock;
3237
3238         /* If possible resent found a lock, @lh is set to its handle */
3239         new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
3240
3241         if (new_lock == NULL && (flags & LDLM_FL_INTENT_ONLY)) {
3242                 lh->mlh_reg_lh.cookie = 0;
3243                 RETURN(0);
3244         }
3245
3246         if (new_lock == NULL && (flags & LDLM_FL_RESENT)) {
3247                 /* Lock is pinned by ldlm_handle_enqueue0() as it is
3248                  * a resend case, however, it could be already destroyed
3249                  * due to client eviction or a raced cancel RPC. */
3250                 LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx\n",
3251                                   lh->mlh_reg_lh.cookie);
3252                 lh->mlh_reg_lh.cookie = 0;
3253                 RETURN(-ESTALE);
3254         }
3255
3256         LASSERTF(new_lock != NULL,
3257                  "lockh %#llx flags %#llx : rc = %d\n",
3258                  lh->mlh_reg_lh.cookie, flags, result);
3259
3260         /*
3261          * If we've already given this lock to a client once, then we should
3262          * have no readers or writers.  Otherwise, we should have one reader
3263          * _or_ writer ref (which will be zeroed below) before returning the
3264          * lock to a client.
3265          */
3266         if (new_lock->l_export == req->rq_export) {
3267                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
3268         } else {
3269                 LASSERT(new_lock->l_export == NULL);
3270                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
3271         }
3272
3273         *lockp = new_lock;
3274
3275         if (new_lock->l_export == req->rq_export) {
3276                 /*
3277                  * Already gave this to the client, which means that we
3278                  * reconstructed a reply.
3279                  */
3280                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
3281                         MSG_RESENT);
3282
3283                 LDLM_LOCK_RELEASE(new_lock);
3284                 lh->mlh_reg_lh.cookie = 0;
3285                 RETURN(ELDLM_LOCK_REPLACED);
3286         }
3287
3288         /*
3289          * Fixup the lock to be given to the client.
3290          */
3291         lock_res_and_lock(new_lock);
3292         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
3293          * possible blocking AST. */
3294         while (new_lock->l_readers > 0) {
3295                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
3296                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3297                 new_lock->l_readers--;
3298         }
3299         while (new_lock->l_writers > 0) {
3300                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
3301                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
3302                 new_lock->l_writers--;
3303         }
3304
3305         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
3306         new_lock->l_blocking_ast = lock->l_blocking_ast;
3307         new_lock->l_completion_ast = lock->l_completion_ast;
3308         new_lock->l_remote_handle = lock->l_remote_handle;
3309         new_lock->l_flags &= ~LDLM_FL_LOCAL;
3310
3311         unlock_res_and_lock(new_lock);
3312
3313         cfs_hash_add(new_lock->l_export->exp_lock_hash,
3314                      &new_lock->l_remote_handle,
3315                      &new_lock->l_exp_hash);
3316
3317         LDLM_LOCK_RELEASE(new_lock);
3318         lh->mlh_reg_lh.cookie = 0;
3319
3320         RETURN(ELDLM_LOCK_REPLACED);
3321 }
3322
3323 static void mdt_intent_fixup_resent(struct mdt_thread_info *info,
3324                                     struct ldlm_lock *new_lock,
3325                                     struct mdt_lock_handle *lh,
3326                                     __u64 flags)
3327 {
3328         struct ptlrpc_request  *req = mdt_info_req(info);
3329         struct ldlm_request    *dlmreq;
3330
3331         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
3332                 return;
3333
3334         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
3335
3336         /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
3337          * lock was found by ldlm_handle_enqueue(); if so @lh must be
3338          * initialized. */
3339         if (flags & LDLM_FL_RESENT) {
3340                 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
3341                 lh->mlh_reg_mode = new_lock->l_granted_mode;
3342
3343                 LDLM_DEBUG(new_lock, "Restoring lock cookie");
3344                 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie %#llx",
3345                           lh->mlh_reg_lh.cookie);
3346                 return;
3347         }
3348
3349         /*
3350          * If the xid matches, then we know this is a resent request, and allow
3351          * it. (It's probably an OPEN, for which we don't send a lock.
3352          */
3353         if (req_can_reconstruct(req, NULL))
3354                 return;
3355
3356         /*
3357          * This remote handle isn't enqueued, so we never received or processed
3358          * this request.  Clear MSG_RESENT, because it can be handled like any
3359          * normal request now.
3360          */
3361         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
3362
3363         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle %#llx",
3364                   dlmreq->lock_handle[0].cookie);
3365 }
3366
3367 static int mdt_intent_getxattr(enum mdt_it_code opcode,
3368                                 struct mdt_thread_info *info,
3369                                 struct ldlm_lock **lockp,
3370                                 __u64 flags)
3371 {
3372         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3373         struct ldlm_reply      *ldlm_rep = NULL;
3374         int rc;
3375         ENTRY;
3376
3377         /*
3378          * Initialize lhc->mlh_reg_lh either from a previously granted lock
3379          * (for the resend case) or a new lock. Below we will use it to
3380          * replace the original lock.
3381          */
3382         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3383         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
3384                 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
3385                 rc = mdt_object_lock(info, info->mti_object, lhc,
3386                                      MDS_INODELOCK_XATTR);
3387                 if (rc)
3388                         return rc;
3389         }
3390
3391         rc = mdt_getxattr(info);
3392
3393         if (mdt_info_req(info)->rq_repmsg != NULL)
3394                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3395
3396         if (ldlm_rep == NULL ||
3397             OBD_FAIL_CHECK(OBD_FAIL_MDS_XATTR_REP)) {
3398                 mdt_object_unlock(info,  info->mti_object, lhc, 1);
3399                 RETURN(err_serious(-EFAULT));
3400         }
3401
3402         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3403
3404         /* This is left for interop instead of adding a new interop flag.
3405          * LU-7433 */
3406 #if LUSTRE_VERSION_CODE > OBD_OCD_VERSION(3, 0, 0, 0)
3407         if (ldlm_rep->lock_policy_res2) {
3408                 mdt_object_unlock(info, info->mti_object, lhc, 1);
3409                 RETURN(ELDLM_LOCK_ABORTED);
3410         }
3411 #endif
3412
3413         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3414         RETURN(rc);
3415 }
3416
3417 static int mdt_intent_getattr(enum mdt_it_code opcode,
3418                               struct mdt_thread_info *info,
3419                               struct ldlm_lock **lockp,
3420                               __u64 flags)
3421 {
3422         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3423         __u64                   child_bits;
3424         struct ldlm_reply      *ldlm_rep;
3425         struct mdt_body        *reqbody;
3426         struct mdt_body        *repbody;
3427         int                     rc, rc2;
3428         ENTRY;
3429
3430         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
3431         LASSERT(reqbody);
3432
3433         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3434         LASSERT(repbody);
3435
3436         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
3437         repbody->mbo_eadatasize = 0;
3438         repbody->mbo_aclsize = 0;
3439
3440         switch (opcode) {
3441         case MDT_IT_LOOKUP:
3442                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
3443                 break;
3444         case MDT_IT_GETATTR:
3445                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
3446                              MDS_INODELOCK_PERM;
3447                 break;
3448         default:
3449                 CERROR("Unsupported intent (%d)\n", opcode);
3450                 GOTO(out_shrink, rc = -EINVAL);
3451         }
3452
3453         rc = mdt_init_ucred_intent_getattr(info, reqbody);
3454         if (rc)
3455                 GOTO(out_shrink, rc);
3456
3457         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3458         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
3459
3460         /* Get lock from request for possible resent case. */
3461         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3462
3463         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
3464         ldlm_rep->lock_policy_res2 = clear_serious(rc);
3465
3466         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG))
3467                 ldlm_rep->lock_policy_res2 = 0;
3468         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
3469             ldlm_rep->lock_policy_res2) {
3470                 lhc->mlh_reg_lh.cookie = 0ull;
3471                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
3472         }
3473
3474         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3475         EXIT;
3476 out_ucred:
3477         mdt_exit_ucred(info);
3478 out_shrink:
3479         mdt_client_compatibility(info);
3480         rc2 = mdt_fix_reply(info);
3481         if (rc == 0)
3482                 rc = rc2;
3483         return rc;
3484 }
3485
3486 static int mdt_intent_layout(enum mdt_it_code opcode,
3487                              struct mdt_thread_info *info,
3488                              struct ldlm_lock **lockp,
3489                              __u64 flags)
3490 {
3491         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_LAYOUT];
3492         struct layout_intent *layout;
3493         struct lu_fid *fid;
3494         struct mdt_object *obj = NULL;
3495         bool layout_change = false;
3496         int layout_size = 0;
3497         int rc = 0;
3498         ENTRY;
3499
3500         if (opcode != MDT_IT_LAYOUT) {
3501                 CERROR("%s: Unknown intent (%d)\n", mdt_obd_name(info->mti_mdt),
3502                         opcode);
3503                 RETURN(-EINVAL);
3504         }
3505
3506         layout = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
3507         if (layout == NULL)
3508                 RETURN(-EPROTO);
3509
3510         switch (layout->li_opc) {
3511         case LAYOUT_INTENT_TRUNC:
3512         case LAYOUT_INTENT_WRITE:
3513                 layout_change = true;
3514                 break;
3515         case LAYOUT_INTENT_ACCESS:
3516                 break;
3517         case LAYOUT_INTENT_READ:
3518         case LAYOUT_INTENT_GLIMPSE:
3519         case LAYOUT_INTENT_RELEASE:
3520         case LAYOUT_INTENT_RESTORE:
3521                 CERROR("%s: Unsupported layout intent opc %d\n",
3522                        mdt_obd_name(info->mti_mdt), layout->li_opc);
3523                 rc = -ENOTSUPP;
3524                 break;
3525         default:
3526                 CERROR("%s: Unknown layout intent opc %d\n",
3527                        mdt_obd_name(info->mti_mdt), layout->li_opc);
3528                 rc = -EINVAL;
3529                 break;
3530         }
3531         if (rc < 0)
3532                 RETURN(rc);
3533
3534         fid = &info->mti_tmp_fid2;
3535         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
3536
3537         /* Get lock from request for possible resent case. */
3538         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3539
3540         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
3541         if (IS_ERR(obj))
3542                 GOTO(out, rc = PTR_ERR(obj));
3543
3544         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
3545                 layout_size = mdt_attr_get_eabuf_size(info, obj);
3546                 if (layout_size < 0)
3547                         GOTO(out_obj, rc = layout_size);
3548
3549                 if (layout_size > info->mti_mdt->mdt_max_mdsize)
3550                         info->mti_mdt->mdt_max_mdsize = layout_size;
3551         }
3552
3553         /*
3554          * set reply buffer size, so that ldlm_handle_enqueue0()->
3555          * ldlm_lvbo_fill() will fill the reply buffer with lovea.
3556          */
3557         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
3558         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
3559                              layout_size);
3560         rc = req_capsule_server_pack(info->mti_pill);
3561         if (rc)
3562                 GOTO(out_obj, rc);
3563
3564
3565         if (layout_change) {
3566                 struct lu_buf *buf = &info->mti_buf;
3567
3568                 buf->lb_buf = NULL;
3569                 buf->lb_len = 0;
3570                 if (unlikely(req_is_replay(mdt_info_req(info)))) {
3571                         buf->lb_buf = req_capsule_client_get(info->mti_pill,
3572                                         &RMF_EADATA);
3573                         buf->lb_len = req_capsule_get_size(info->mti_pill,
3574                                         &RMF_EADATA, RCL_CLIENT);
3575                         /*
3576                          * If it's a replay of layout write intent RPC, the
3577                          * client has saved the extended lovea when
3578                          * it get reply then.
3579                          */
3580                         if (buf->lb_len > 0)
3581                                 mdt_fix_lov_magic(info, buf->lb_buf);
3582                 }
3583
3584                 /*
3585                  * Instantiate some layout components, if @buf contains
3586                  * lovea, then it's a replay of the layout intent write
3587                  * RPC.
3588                  */
3589                 rc = mdt_layout_change(info, obj, layout, buf);
3590                 if (rc)
3591                         GOTO(out_obj, rc);
3592         }
3593 out_obj:
3594         mdt_object_put(info->mti_env, obj);
3595
3596         if (rc == 0 && lustre_handle_is_used(&lhc->mlh_reg_lh))
3597                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3598
3599 out:
3600         lhc->mlh_reg_lh.cookie = 0;
3601
3602         return rc;
3603 }
3604
3605 static int mdt_intent_reint(enum mdt_it_code opcode,
3606                             struct mdt_thread_info *info,
3607                             struct ldlm_lock **lockp,
3608                             __u64 flags)
3609 {
3610         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
3611         struct ldlm_reply      *rep = NULL;
3612         long                    opc;
3613         int                     rc;
3614
3615         static const struct req_format *intent_fmts[REINT_MAX] = {
3616                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
3617                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
3618         };
3619
3620         ENTRY;
3621
3622         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
3623         if (opc < 0)
3624                 RETURN(opc);
3625
3626         if (mdt_it_flavor[opcode].it_reint != opc) {
3627                 CERROR("Reint code %ld doesn't match intent: %d\n",
3628                        opc, opcode);
3629                 RETURN(err_serious(-EPROTO));
3630         }
3631
3632         /* Get lock from request for possible resent case. */
3633         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
3634
3635         rc = mdt_reint_internal(info, lhc, opc);
3636
3637         /* Check whether the reply has been packed successfully. */
3638         if (mdt_info_req(info)->rq_repmsg != NULL)
3639                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
3640         if (rep == NULL)
3641                 RETURN(err_serious(-EFAULT));
3642
3643         /* MDC expects this in any case */
3644         if (rc != 0)
3645                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
3646
3647         /* the open lock or the lock for cross-ref object should be
3648          * returned to the client */
3649         if (lustre_handle_is_used(&lhc->mlh_reg_lh) &&
3650             (rc == 0 || rc == -MDT_EREMOTE_OPEN)) {
3651                 rep->lock_policy_res2 = 0;
3652                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
3653                 RETURN(rc);
3654         }
3655
3656         rep->lock_policy_res2 = clear_serious(rc);
3657
3658         if (rep->lock_policy_res2 == -ENOENT &&
3659             mdt_get_disposition(rep, DISP_LOOKUP_NEG) &&
3660             !mdt_get_disposition(rep, DISP_OPEN_CREATE))
3661                 rep->lock_policy_res2 = 0;
3662
3663         lhc->mlh_reg_lh.cookie = 0ull;
3664         if (rc == -ENOTCONN || rc == -ENODEV ||
3665             rc == -EOVERFLOW) { /**< if VBR failure then return error */
3666                 /*
3667                  * If it is the disconnect error (ENODEV & ENOCONN), the error
3668                  * will be returned by rq_status, and client at ptlrpc layer
3669                  * will detect this, then disconnect, reconnect the import
3670                  * immediately, instead of impacting the following the rpc.
3671                  */
3672                 RETURN(rc);
3673         }
3674         /*
3675          * For other cases, the error will be returned by intent, and client
3676          * will retrieve the result from intent.
3677          */
3678         RETURN(ELDLM_LOCK_ABORTED);
3679 }
3680
3681 static int mdt_intent_code(enum ldlm_intent_flags itcode)
3682 {
3683         int rc;
3684
3685         switch (itcode) {
3686         case IT_OPEN:
3687                 rc = MDT_IT_OPEN;
3688                 break;
3689         case IT_OPEN|IT_CREAT:
3690                 rc = MDT_IT_OCREAT;
3691                 break;
3692         case IT_CREAT:
3693                 rc = MDT_IT_CREATE;
3694                 break;
3695         case IT_READDIR:
3696                 rc = MDT_IT_READDIR;
3697                 break;
3698         case IT_GETATTR:
3699                 rc = MDT_IT_GETATTR;
3700                 break;
3701         case IT_LOOKUP:
3702                 rc = MDT_IT_LOOKUP;
3703                 break;
3704         case IT_UNLINK:
3705                 rc = MDT_IT_UNLINK;
3706                 break;
3707         case IT_TRUNC:
3708                 rc = MDT_IT_TRUNC;
3709                 break;
3710         case IT_GETXATTR:
3711                 rc = MDT_IT_GETXATTR;
3712                 break;
3713         case IT_LAYOUT:
3714                 rc = MDT_IT_LAYOUT;
3715                 break;
3716         case IT_QUOTA_DQACQ:
3717         case IT_QUOTA_CONN:
3718                 rc = MDT_IT_QUOTA;
3719                 break;
3720         default:
3721                 CERROR("Unknown intent opcode: 0x%08x\n", itcode);
3722                 rc = -EINVAL;
3723                 break;
3724         }
3725         return rc;
3726 }
3727
3728 static int mdt_intent_opc(enum ldlm_intent_flags itopc,
3729                           struct mdt_thread_info *info,
3730                           struct ldlm_lock **lockp, __u64 flags)
3731 {
3732         struct req_capsule      *pill = info->mti_pill;
3733         struct ptlrpc_request   *req = mdt_info_req(info);
3734         struct mdt_it_flavor    *flv;
3735         int opc;
3736         int rc;
3737         ENTRY;
3738
3739         opc = mdt_intent_code(itopc);
3740         if (opc < 0)
3741                 RETURN(-EINVAL);
3742
3743         if (opc == MDT_IT_QUOTA) {
3744                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
3745
3746                 if (qmt == NULL)
3747                         RETURN(-EOPNOTSUPP);
3748
3749                 if (mdt_rdonly(req->rq_export))
3750                         RETURN(-EROFS);
3751
3752                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
3753                 /* pass the request to quota master */
3754                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
3755                                                  mdt_info_req(info), lockp,
3756                                                  flags);
3757                 RETURN(rc);
3758         }
3759
3760         flv = &mdt_it_flavor[opc];
3761         if (flv->it_fmt != NULL)
3762                 req_capsule_extend(pill, flv->it_fmt);
3763
3764         rc = mdt_unpack_req_pack_rep(info, flv->it_flags);
3765         if (rc < 0)
3766                 RETURN(rc);
3767
3768         if (flv->it_flags & MUTABOR && mdt_rdonly(req->rq_export))
3769                 RETURN(-EROFS);
3770
3771         if (flv->it_act != NULL) {
3772                 struct ldlm_reply *rep;
3773
3774                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_INTENT_DELAY, 10);
3775
3776                 /* execute policy */
3777                 rc = flv->it_act(opc, info, lockp, flags);
3778
3779                 /* Check whether the reply has been packed successfully. */
3780                 if (req->rq_repmsg != NULL) {
3781                         rep = req_capsule_server_get(info->mti_pill,
3782                                                      &RMF_DLM_REP);
3783                         rep->lock_policy_res2 =
3784                                 ptlrpc_status_hton(rep->lock_policy_res2);
3785                 }
3786         }
3787
3788         RETURN(rc);
3789 }
3790
3791 static int mdt_intent_policy(struct ldlm_namespace *ns,
3792                              struct ldlm_lock **lockp, void *req_cookie,
3793                              enum ldlm_mode mode, __u64 flags, void *data)
3794 {
3795         struct tgt_session_info *tsi;
3796         struct mdt_thread_info  *info;
3797         struct ptlrpc_request   *req  =  req_cookie;
3798         struct ldlm_intent      *it;
3799         struct req_capsule      *pill;
3800         int rc;
3801
3802         ENTRY;
3803
3804         LASSERT(req != NULL);
3805
3806         tsi = tgt_ses_info(req->rq_svc_thread->t_env);
3807
3808         info = tsi2mdt_info(tsi);
3809         LASSERT(info != NULL);
3810         pill = info->mti_pill;
3811         LASSERT(pill->rc_req == req);
3812
3813         if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
3814                 req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
3815                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
3816                 if (it != NULL) {
3817                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
3818                         if (rc == 0)
3819                                 rc = ELDLM_OK;
3820
3821                         /* Lock without inodebits makes no sense and will oops
3822                          * later in ldlm. Let's check it now to see if we have
3823                          * ibits corrupted somewhere in mdt_intent_opc().
3824                          * The case for client miss to set ibits has been
3825                          * processed by others. */
3826                         LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\
3827                                         lr_type == LDLM_IBITS,
3828                                      info->mti_dlm_req->lock_desc.\
3829                                         l_policy_data.l_inodebits.bits != 0));
3830                 } else
3831                         rc = err_serious(-EFAULT);
3832         } else {
3833                 /* No intent was provided */
3834                 LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE);
3835                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
3836                 rc = req_capsule_server_pack(pill);
3837                 if (rc)
3838                         rc = err_serious(rc);
3839         }
3840         mdt_thread_info_fini(info);
3841         RETURN(rc);
3842 }
3843
3844 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
3845 {
3846         struct seq_server_site  *ss = mdt_seq_site(mdt);
3847
3848         if (ss->ss_node_id == 0)
3849                 return;
3850
3851         if (ss->ss_client_seq != NULL) {
3852                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3853                 ss->ss_client_seq->lcs_exp = NULL;
3854         }
3855
3856         if (ss->ss_server_fld != NULL) {
3857                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
3858                 ss->ss_server_fld->lsf_control_exp = NULL;
3859         }
3860 }
3861
3862 static void mdt_seq_fini_cli(struct mdt_device *mdt)
3863 {
3864         struct seq_server_site *ss = mdt_seq_site(mdt);
3865
3866         if (ss == NULL)
3867                 return;
3868
3869         if (ss->ss_server_seq != NULL)
3870                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
3871 }
3872
3873 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
3874 {
3875         mdt_seq_fini_cli(mdt);
3876         mdt_deregister_seq_exp(mdt);
3877
3878         return seq_site_fini(env, mdt_seq_site(mdt));
3879 }
3880
3881 /**
3882  * It will retrieve its FLDB entries from MDT0, and it only happens
3883  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
3884  * and it needs to refresh FLDB from the MDT0.
3885  **/
3886 static int mdt_register_lwp_callback(void *data)
3887 {
3888         struct lu_env           env;
3889         struct mdt_device       *mdt = data;
3890         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
3891         int                     rc;
3892         ENTRY;
3893
3894         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
3895
3896         rc = lu_env_init(&env, LCT_MD_THREAD);
3897         if (rc < 0) {
3898                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
3899                 RETURN(rc);
3900         }
3901
3902         /* Allocate new sequence now to avoid creating local transaction
3903          * in the normal transaction process */
3904         rc = seq_server_check_and_alloc_super(&env,
3905                                               mdt_seq_site(mdt)->ss_server_seq);
3906         if (rc < 0)
3907                 GOTO(out, rc);
3908
3909         if (fld->lsf_new) {
3910                 rc = fld_update_from_controller(&env, fld);
3911                 if (rc != 0) {
3912                         CERROR("%s: cannot update controller: rc = %d\n",
3913                                mdt_obd_name(mdt), rc);
3914                         GOTO(out, rc);
3915                 }
3916         }
3917 out:
3918         lu_env_fini(&env);
3919         RETURN(rc);
3920 }
3921
3922 static int mdt_register_seq_exp(struct mdt_device *mdt)
3923 {
3924         struct seq_server_site  *ss = mdt_seq_site(mdt);
3925         char                    *lwp_name = NULL;
3926         int                     rc;
3927
3928         if (ss->ss_node_id == 0)
3929                 return 0;
3930
3931         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
3932         if (lwp_name == NULL)
3933                 GOTO(out_free, rc = -ENOMEM);
3934
3935         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
3936         if (rc != 0)
3937                 GOTO(out_free, rc);
3938
3939         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
3940                                       NULL, NULL);
3941         if (rc != 0)
3942                 GOTO(out_free, rc);
3943
3944         rc = lustre_register_lwp_item(lwp_name,
3945                                       &ss->ss_server_fld->lsf_control_exp,
3946                                       mdt_register_lwp_callback, mdt);
3947         if (rc != 0) {
3948                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
3949                 ss->ss_client_seq->lcs_exp = NULL;
3950                 GOTO(out_free, rc);
3951         }
3952 out_free:
3953         if (lwp_name != NULL)
3954                 OBD_FREE(lwp_name, MAX_OBD_NAME);
3955
3956         return rc;
3957 }
3958
3959 /*
3960  * Init client sequence manager which is used by local MDS to talk to sequence
3961  * controller on remote node.
3962  */
3963 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
3964 {
3965         struct seq_server_site  *ss = mdt_seq_site(mdt);
3966         int                     rc;
3967         char                    *prefix;
3968         ENTRY;
3969
3970         /* check if this is adding the first MDC and controller is not yet
3971          * initialized. */
3972         OBD_ALLOC_PTR(ss->ss_client_seq);
3973         if (ss->ss_client_seq == NULL)
3974                 RETURN(-ENOMEM);
3975
3976         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
3977         if (prefix == NULL) {
3978                 OBD_FREE_PTR(ss->ss_client_seq);
3979                 ss->ss_client_seq = NULL;
3980                 RETURN(-ENOMEM);
3981         }
3982
3983         /* Note: seq_client_fini will be called in seq_site_fini */
3984         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
3985         rc = seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
3986                              prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
3987                                                             NULL);
3988         OBD_FREE(prefix, MAX_OBD_NAME + 5);
3989         if (rc != 0) {
3990                 OBD_FREE_PTR(ss->ss_client_seq);
3991                 ss->ss_client_seq = NULL;
3992                 RETURN(rc);
3993         }
3994
3995         rc = seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq);
3996
3997         RETURN(rc);
3998 }
3999
4000 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
4001 {
4002         struct seq_server_site  *ss;
4003         int                     rc;
4004         ENTRY;
4005
4006         ss = mdt_seq_site(mdt);
4007         /* init sequence controller server(MDT0) */
4008         if (ss->ss_node_id == 0) {
4009                 OBD_ALLOC_PTR(ss->ss_control_seq);
4010                 if (ss->ss_control_seq == NULL)
4011                         RETURN(-ENOMEM);
4012
4013                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
4014                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
4015                                      ss);
4016                 if (rc)
4017                         GOTO(out_seq_fini, rc);
4018         }
4019
4020         /* Init normal sequence server */
4021         OBD_ALLOC_PTR(ss->ss_server_seq);
4022         if (ss->ss_server_seq == NULL)
4023                 GOTO(out_seq_fini, rc = -ENOMEM);
4024
4025         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
4026                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
4027         if (rc)
4028                 GOTO(out_seq_fini, rc);
4029
4030         /* init seq client for seq server to talk to seq controller(MDT0) */
4031         rc = mdt_seq_init_cli(env, mdt);
4032         if (rc != 0)
4033                 GOTO(out_seq_fini, rc);
4034
4035         if (ss->ss_node_id != 0)
4036                 /* register controller export through lwp */
4037                 rc = mdt_register_seq_exp(mdt);
4038
4039         EXIT;
4040 out_seq_fini:
4041         if (rc)
4042                 mdt_seq_fini(env, mdt);
4043
4044         return rc;
4045 }
4046
4047 /*
4048  * FLD wrappers
4049  */
4050 static int mdt_fld_fini(const struct lu_env *env,
4051                         struct mdt_device *m)
4052 {
4053         struct seq_server_site *ss = mdt_seq_site(m);
4054         ENTRY;
4055
4056         if (ss && ss->ss_server_fld) {
4057                 fld_server_fini(env, ss->ss_server_fld);
4058                 OBD_FREE_PTR(ss->ss_server_fld);
4059                 ss->ss_server_fld = NULL;
4060         }
4061
4062         RETURN(0);
4063 }
4064
4065 static int mdt_fld_init(const struct lu_env *env,
4066                         const char *uuid,
4067                         struct mdt_device *m)
4068 {
4069         struct seq_server_site *ss;
4070         int rc;
4071         ENTRY;
4072
4073         ss = mdt_seq_site(m);
4074
4075         OBD_ALLOC_PTR(ss->ss_server_fld);
4076         if (ss->ss_server_fld == NULL)
4077                 RETURN(rc = -ENOMEM);
4078
4079         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
4080                              LU_SEQ_RANGE_MDT);
4081         if (rc) {
4082                 OBD_FREE_PTR(ss->ss_server_fld);
4083                 ss->ss_server_fld = NULL;
4084                 RETURN(rc);
4085         }
4086
4087         RETURN(0);
4088 }
4089
4090 static void mdt_stack_pre_fini(const struct lu_env *env,
4091                            struct mdt_device *m, struct lu_device *top)
4092 {
4093         struct lustre_cfg_bufs  *bufs;
4094         struct lustre_cfg       *lcfg;
4095         struct mdt_thread_info  *info;
4096         ENTRY;
4097
4098         LASSERT(top);
4099
4100         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4101         LASSERT(info != NULL);
4102
4103         bufs = &info->mti_u.bufs;
4104
4105         LASSERT(m->mdt_child_exp);
4106         LASSERT(m->mdt_child_exp->exp_obd);
4107
4108         /* process cleanup, pass mdt obd name to get obd umount flags */
4109         /* XXX: this is needed because all layers are referenced by
4110          * objects (some of them are pinned by osd, for example *
4111          * the proper solution should be a model where object used
4112          * by osd only doesn't have mdt/mdd slices -bzzz */
4113         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4114         lustre_cfg_bufs_set_string(bufs, 1, NULL);
4115         lcfg = lustre_cfg_new(LCFG_PRE_CLEANUP, bufs);
4116         if (lcfg == NULL)
4117                 RETURN_EXIT;
4118
4119         top->ld_ops->ldo_process_config(env, top, lcfg);
4120         lustre_cfg_free(lcfg);
4121         EXIT;
4122 }
4123
4124 static void mdt_stack_fini(const struct lu_env *env,
4125                            struct mdt_device *m, struct lu_device *top)
4126 {
4127         struct obd_device       *obd = mdt2obd_dev(m);
4128         struct lustre_cfg_bufs  *bufs;
4129         struct lustre_cfg       *lcfg;
4130         struct mdt_thread_info  *info;
4131         char                     flags[3] = "";
4132         ENTRY;
4133
4134         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4135         LASSERT(info != NULL);
4136
4137         lu_dev_del_linkage(top->ld_site, top);
4138
4139         lu_site_purge(env, top->ld_site, -1);
4140
4141         bufs = &info->mti_u.bufs;
4142         /* process cleanup, pass mdt obd name to get obd umount flags */
4143         /* another purpose is to let all layers to release their objects */
4144         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
4145         if (obd->obd_force)
4146                 strcat(flags, "F");
4147         if (obd->obd_fail)
4148                 strcat(flags, "A");
4149         lustre_cfg_bufs_set_string(bufs, 1, flags);
4150         lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
4151         if (lcfg == NULL)
4152                 RETURN_EXIT;
4153
4154         LASSERT(top);
4155         top->ld_ops->ldo_process_config(env, top, lcfg);
4156         lustre_cfg_free(lcfg);
4157
4158         lu_site_purge(env, top->ld_site, -1);
4159
4160         m->mdt_child = NULL;
4161         m->mdt_bottom = NULL;
4162
4163         obd_disconnect(m->mdt_child_exp);
4164         m->mdt_child_exp = NULL;
4165
4166         obd_disconnect(m->mdt_bottom_exp);
4167         m->mdt_child_exp = NULL;
4168 }
4169
4170 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
4171                                const char *next, struct obd_export **exp)
4172 {
4173         struct obd_connect_data *data = NULL;
4174         struct obd_device       *obd;
4175         int                      rc;
4176         ENTRY;
4177
4178         OBD_ALLOC_PTR(data);
4179         if (data == NULL)
4180                 GOTO(out, rc = -ENOMEM);
4181
4182         obd = class_name2obd(next);
4183         if (obd == NULL) {
4184                 CERROR("%s: can't locate next device: %s\n",
4185                        mdt_obd_name(m), next);
4186                 GOTO(out, rc = -ENOTCONN);
4187         }
4188
4189         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4190         data->ocd_version = LUSTRE_VERSION_CODE;
4191
4192         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
4193         if (rc) {
4194                 CERROR("%s: cannot connect to next dev %s (%d)\n",
4195                        mdt_obd_name(m), next, rc);
4196                 GOTO(out, rc);
4197         }
4198
4199 out:
4200         if (data)
4201                 OBD_FREE_PTR(data);
4202         RETURN(rc);
4203 }
4204
4205 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
4206                           struct lustre_cfg *cfg)
4207 {
4208         char                   *dev = lustre_cfg_string(cfg, 0);
4209         int                     rc, name_size, uuid_size;
4210         char                   *name, *uuid, *p;
4211         struct lustre_cfg_bufs *bufs;
4212         struct lustre_cfg      *lcfg;
4213         struct obd_device      *obd;
4214         struct lustre_profile  *lprof;
4215         struct lu_site         *site;
4216         ENTRY;
4217
4218         /* in 1.8 we had the only device in the stack - MDS.
4219          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
4220          * in 2.3 OSD is instantiated by obd_mount.c, so we need
4221          * to generate names and setup MDT, MDD. MDT will be using
4222          * generated name to connect to MDD. for MDD the next device
4223          * will be LOD with name taken from so called "profile" which
4224          * is generated by mount_option line
4225          *
4226          * 1.8 MGS generates config. commands like this:
4227          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
4228          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
4229          * 2.0 MGS generates config. commands like this:
4230          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
4231          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4232          *                    3:lustre-MDT0000-mdtlov  4:f
4233          *
4234          * we generate MDD name from MDT one, just replacing T with D
4235          *
4236          * after all the preparations, the logical equivalent will be
4237          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
4238          *                    3:lustre-MDT0000-mdtlov  4:f
4239          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4240          *                    3:lustre-MDD0000  4:f
4241          *
4242          *  notice we build the stack from down to top: MDD first, then MDT */
4243
4244         name_size = MAX_OBD_NAME;
4245         uuid_size = MAX_OBD_NAME;
4246
4247         OBD_ALLOC(name, name_size);
4248         OBD_ALLOC(uuid, uuid_size);
4249         if (name == NULL || uuid == NULL)
4250                 GOTO(cleanup_mem, rc = -ENOMEM);
4251
4252         OBD_ALLOC_PTR(bufs);
4253         if (!bufs)
4254                 GOTO(cleanup_mem, rc = -ENOMEM);
4255
4256         strcpy(name, dev);
4257         p = strstr(name, "-MDT");
4258         if (p == NULL)
4259                 GOTO(free_bufs, rc = -ENOMEM);
4260         p[3] = 'D';
4261
4262         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
4263
4264         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4265         if (lprof == NULL || lprof->lp_dt == NULL) {
4266                 CERROR("can't find the profile: %s\n",
4267                        lustre_cfg_string(cfg, 0));
4268                 GOTO(free_bufs, rc = -EINVAL);
4269         }
4270
4271         lustre_cfg_bufs_reset(bufs, name);
4272         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
4273         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4274         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4275
4276         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4277         if (lcfg == NULL)
4278                 GOTO(put_profile, rc = -ENOMEM);
4279
4280         rc = class_attach(lcfg);
4281         if (rc)
4282                 GOTO(lcfg_cleanup, rc);
4283
4284         obd = class_name2obd(name);
4285         if (!obd) {
4286                 CERROR("Can not find obd %s (%s in config)\n",
4287                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
4288                 GOTO(lcfg_cleanup, rc = -EINVAL);
4289         }
4290
4291         lustre_cfg_free(lcfg);
4292
4293         lustre_cfg_bufs_reset(bufs, name);
4294         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4295         lustre_cfg_bufs_set_string(bufs, 2, dev);
4296         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4297
4298         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4299         if (lcfg == NULL)
4300                 GOTO(class_detach, rc = -ENOMEM);
4301
4302         rc = class_setup(obd, lcfg);
4303         if (rc)
4304                 GOTO(class_detach, rc);
4305
4306         /* connect to MDD we just setup */
4307         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
4308         if (rc)
4309                 GOTO(class_detach, rc);
4310
4311         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
4312         LASSERT(site);
4313         LASSERT(mdt_lu_site(mdt) == NULL);
4314         mdt->mdt_lu_dev.ld_site = site;
4315         site->ls_top_dev = &mdt->mdt_lu_dev;
4316         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
4317
4318         /* now connect to bottom OSD */
4319         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
4320         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
4321         if (rc)
4322                 GOTO(class_detach, rc);
4323         mdt->mdt_bottom =
4324                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
4325
4326         rc = lu_env_refill((struct lu_env *)env);
4327         if (rc != 0)
4328                 CERROR("Failure to refill session: '%d'\n", rc);
4329
4330         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
4331
4332         EXIT;
4333 class_detach:
4334         if (rc)
4335                 class_detach(obd, lcfg);
4336 lcfg_cleanup:
4337         lustre_cfg_free(lcfg);
4338 put_profile:
4339         class_put_profile(lprof);
4340 free_bufs:
4341         OBD_FREE_PTR(bufs);
4342 cleanup_mem:
4343         if (name)
4344                 OBD_FREE(name, name_size);
4345         if (uuid)
4346                 OBD_FREE(uuid, uuid_size);
4347         RETURN(rc);
4348 }
4349
4350 /* setup quota master target on MDT0 */
4351 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
4352                           struct lustre_cfg *cfg)
4353 {
4354         struct obd_device       *obd;
4355         char                    *dev = lustre_cfg_string(cfg, 0);
4356         char                    *qmtname, *uuid, *p;
4357         struct lustre_cfg_bufs  *bufs;
4358         struct lustre_cfg       *lcfg;
4359         struct lustre_profile   *lprof;
4360         struct obd_connect_data *data;
4361         int                      rc;
4362         ENTRY;
4363
4364         LASSERT(mdt->mdt_qmt_exp == NULL);
4365         LASSERT(mdt->mdt_qmt_dev == NULL);
4366
4367         /* quota master is on MDT0 only for now */
4368         if (mdt->mdt_seq_site.ss_node_id != 0)
4369                 RETURN(0);
4370
4371         /* MGS generates config commands which look as follows:
4372          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
4373          *                    3:lustre-MDT0000-mdtlov  4:f
4374          *
4375          * We generate the QMT name from the MDT one, just replacing MD with QM
4376          * after all the preparations, the logical equivalent will be:
4377          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
4378          *                    3:lustre-MDT0000-osd  4:f */
4379         OBD_ALLOC(qmtname, MAX_OBD_NAME);
4380         OBD_ALLOC(uuid, UUID_MAX);
4381         OBD_ALLOC_PTR(bufs);
4382         OBD_ALLOC_PTR(data);
4383         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
4384                 GOTO(cleanup_mem, rc = -ENOMEM);
4385
4386         strcpy(qmtname, dev);
4387         p = strstr(qmtname, "-MDT");
4388         if (p == NULL)
4389                 GOTO(cleanup_mem, rc = -ENOMEM);
4390         /* replace MD with QM */
4391         p[1] = 'Q';
4392         p[2] = 'M';
4393
4394         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
4395
4396         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
4397         if (lprof == NULL || lprof->lp_dt == NULL) {
4398                 CERROR("can't find profile for %s\n",
4399                        lustre_cfg_string(cfg, 0));
4400                 GOTO(cleanup_mem, rc = -EINVAL);
4401         }
4402
4403         lustre_cfg_bufs_reset(bufs, qmtname);
4404         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
4405         lustre_cfg_bufs_set_string(bufs, 2, uuid);
4406         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
4407
4408         lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
4409         if (lcfg == NULL)
4410                 GOTO(put_profile, rc = -ENOMEM);
4411
4412         rc = class_attach(lcfg);
4413         if (rc)
4414                 GOTO(lcfg_cleanup, rc);
4415
4416         obd = class_name2obd(qmtname);
4417         if (!obd) {
4418                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
4419                        lustre_cfg_string(cfg, 0));
4420                 GOTO(lcfg_cleanup, rc = -EINVAL);
4421         }
4422
4423         lustre_cfg_free(lcfg);
4424
4425         lustre_cfg_bufs_reset(bufs, qmtname);
4426         lustre_cfg_bufs_set_string(bufs, 1, uuid);
4427         lustre_cfg_bufs_set_string(bufs, 2, dev);
4428
4429         /* for quota, the next device should be the OSD device */
4430         lustre_cfg_bufs_set_string(bufs, 3,
4431                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
4432
4433         lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
4434         if (lcfg == NULL)
4435                 GOTO(class_detach, rc = -ENOMEM);
4436
4437         rc = class_setup(obd, lcfg);
4438         if (rc)
4439                 GOTO(class_detach, rc);
4440
4441         mdt->mdt_qmt_dev = obd->obd_lu_dev;
4442
4443         /* configure local quota objects */
4444         rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
4445                                                    &mdt->mdt_lu_dev,
4446                                                    mdt->mdt_qmt_dev);
4447         if (rc)
4448                 GOTO(class_cleanup, rc);
4449
4450         /* connect to quota master target */
4451         data->ocd_connect_flags = OBD_CONNECT_VERSION;
4452         data->ocd_version = LUSTRE_VERSION_CODE;
4453         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
4454                          data, NULL);
4455         if (rc) {
4456                 CERROR("cannot connect to quota master device %s (%d)\n",
4457                        qmtname, rc);
4458                 GOTO(class_cleanup, rc);
4459         }
4460
4461         EXIT;
4462 class_cleanup:
4463         if (rc) {
4464                 class_manual_cleanup(obd);
4465                 mdt->mdt_qmt_dev = NULL;
4466         }
4467 class_detach:
4468         if (rc)
4469                 class_detach(obd, lcfg);
4470 lcfg_cleanup:
4471         lustre_cfg_free(lcfg);
4472 put_profile:
4473         class_put_profile(lprof);
4474 cleanup_mem:
4475         if (bufs)
4476                 OBD_FREE_PTR(bufs);
4477         if (qmtname)
4478                 OBD_FREE(qmtname, MAX_OBD_NAME);
4479         if (uuid)
4480                 OBD_FREE(uuid, UUID_MAX);
4481         if (data)
4482                 OBD_FREE_PTR(data);
4483         return rc;
4484 }
4485
4486 /* Shutdown quota master target associated with mdt */
4487 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
4488 {
4489         ENTRY;
4490
4491         if (mdt->mdt_qmt_exp == NULL)
4492                 RETURN_EXIT;
4493         LASSERT(mdt->mdt_qmt_dev != NULL);
4494
4495         /* the qmt automatically shuts down when the mdt disconnects */
4496         obd_disconnect(mdt->mdt_qmt_exp);
4497         mdt->mdt_qmt_exp = NULL;
4498         mdt->mdt_qmt_dev = NULL;
4499         EXIT;
4500 }
4501
4502 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
4503  * for now. This will be removed along with converting rest of MDT code
4504  * to use tgt_session_info */
4505 static int mdt_tgt_getxattr(struct tgt_session_info *tsi)
4506 {
4507         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
4508         int                      rc;
4509
4510         rc = mdt_getxattr(info);
4511
4512         mdt_thread_info_fini(info);
4513         return rc;
4514 }
4515
4516 static struct tgt_handler mdt_tgt_handlers[] = {
4517 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4518                 0,                      MDS_CONNECT,    mdt_tgt_connect,
4519                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
4520 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4521                 0,                      MDS_DISCONNECT, tgt_disconnect,
4522                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
4523 TGT_RPC_HANDLER(MDS_FIRST_OPC,
4524                 HABEO_REFERO,           MDS_SET_INFO,   mdt_set_info,
4525                 &RQF_OBD_SET_INFO, LUSTRE_MDS_VERSION),
4526 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
4527 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_GET_ROOT,   mdt_get_root),
4528 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETATTR,    mdt_getattr),
4529 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_GETATTR_NAME,
4530                                                         mdt_getattr_name),
4531 TGT_MDT_HDL(HABEO_CORPUS,               MDS_GETXATTR,   mdt_tgt_getxattr),
4532 TGT_MDT_HDL(0           | HABEO_REFERO, MDS_STATFS,     mdt_statfs),
4533 TGT_MDT_HDL(0           | MUTABOR,      MDS_REINT,      mdt_reint),
4534 TGT_MDT_HDL(HABEO_CORPUS,               MDS_CLOSE,      mdt_close),
4535 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_READPAGE,   mdt_readpage),
4536 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_SYNC,       mdt_sync),
4537 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
4538 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_PROGRESS,
4539                                                         mdt_hsm_progress),
4540 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_REGISTER,
4541                                                         mdt_hsm_ct_register),
4542 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_CT_UNREGISTER,
4543                                                         mdt_hsm_ct_unregister),
4544 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_STATE_GET,
4545                                                         mdt_hsm_state_get),
4546 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO | MUTABOR, MDS_HSM_STATE_SET,
4547                                                         mdt_hsm_state_set),
4548 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_ACTION, mdt_hsm_action),
4549 TGT_MDT_HDL(HABEO_CORPUS| HABEO_REFERO, MDS_HSM_REQUEST,
4550                                                         mdt_hsm_request),
4551 TGT_MDT_HDL(HABEO_CLAVIS | HABEO_CORPUS | HABEO_REFERO | MUTABOR,
4552             MDS_SWAP_LAYOUTS,
4553             mdt_swap_layouts),
4554 };
4555
4556 static struct tgt_handler mdt_sec_ctx_ops[] = {
4557 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
4558 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT,mdt_sec_ctx_handle),
4559 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
4560 };
4561
4562 static struct tgt_handler mdt_quota_ops[] = {
4563 TGT_QUOTA_HDL(HABEO_REFERO,             QUOTA_DQACQ,      mdt_quota_dqacq),
4564 };
4565
4566 static struct tgt_opc_slice mdt_common_slice[] = {
4567         {
4568                 .tos_opc_start  = MDS_FIRST_OPC,
4569                 .tos_opc_end    = MDS_LAST_OPC,
4570                 .tos_hs         = mdt_tgt_handlers
4571         },
4572         {
4573                 .tos_opc_start  = OBD_FIRST_OPC,
4574                 .tos_opc_end    = OBD_LAST_OPC,
4575                 .tos_hs         = tgt_obd_handlers
4576         },
4577         {
4578                 .tos_opc_start  = LDLM_FIRST_OPC,
4579                 .tos_opc_end    = LDLM_LAST_OPC,
4580                 .tos_hs         = tgt_dlm_handlers
4581         },
4582         {
4583                 .tos_opc_start  = SEC_FIRST_OPC,
4584                 .tos_opc_end    = SEC_LAST_OPC,
4585                 .tos_hs         = mdt_sec_ctx_ops
4586         },
4587         {
4588                 .tos_opc_start  = OUT_UPDATE_FIRST_OPC,
4589                 .tos_opc_end    = OUT_UPDATE_LAST_OPC,
4590                 .tos_hs         = tgt_out_handlers
4591         },
4592         {
4593                 .tos_opc_start  = FLD_FIRST_OPC,
4594                 .tos_opc_end    = FLD_LAST_OPC,
4595                 .tos_hs         = fld_handlers
4596         },
4597         {
4598                 .tos_opc_start  = SEQ_FIRST_OPC,
4599                 .tos_opc_end    = SEQ_LAST_OPC,
4600                 .tos_hs         = seq_handlers
4601         },
4602         {
4603                 .tos_opc_start  = QUOTA_DQACQ,
4604                 .tos_opc_end    = QUOTA_LAST_OPC,
4605                 .tos_hs         = mdt_quota_ops
4606         },
4607         {
4608                 .tos_opc_start  = LLOG_FIRST_OPC,
4609                 .tos_opc_end    = LLOG_LAST_OPC,
4610                 .tos_hs         = tgt_llog_handlers
4611         },
4612         {
4613                 .tos_opc_start  = LFSCK_FIRST_OPC,
4614                 .tos_opc_end    = LFSCK_LAST_OPC,
4615                 .tos_hs         = tgt_lfsck_handlers
4616         },
4617
4618         {
4619                 .tos_hs         = NULL
4620         }
4621 };
4622
4623 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
4624 {
4625         struct md_device *next = m->mdt_child;
4626         struct lu_device *d = &m->mdt_lu_dev;
4627         struct obd_device *obd = mdt2obd_dev(m);
4628         struct lfsck_stop stop;
4629
4630         ENTRY;
4631         stop.ls_status = LS_PAUSED;
4632         stop.ls_flags = 0;
4633         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
4634
4635         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
4636         ping_evictor_stop();
4637
4638         if (m->mdt_opts.mo_coordinator)
4639                 mdt_hsm_cdt_stop(m);
4640
4641         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
4642         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
4643
4644         if (m->mdt_namespace != NULL)
4645                 ldlm_namespace_free_prior(m->mdt_namespace, NULL,
4646                                           d->ld_obd->obd_force);
4647
4648         obd_exports_barrier(obd);
4649         obd_zombie_barrier();
4650
4651         mdt_quota_fini(env, m);
4652
4653         cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
4654
4655         /* Calling the cleanup functions in the same order as in the mdt_init0
4656          * error path
4657          */
4658         mdt_procfs_fini(m);
4659
4660         target_recovery_fini(obd);
4661         upcall_cache_cleanup(m->mdt_identity_cache);
4662         m->mdt_identity_cache = NULL;
4663
4664         mdt_fs_cleanup(env, m);
4665
4666         tgt_fini(env, &m->mdt_lut);
4667
4668         mdt_hsm_cdt_fini(m);
4669
4670         if (m->mdt_namespace != NULL) {
4671                 ldlm_namespace_free_post(m->mdt_namespace);
4672                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
4673         }
4674
4675         if (m->mdt_md_root != NULL) {
4676                 mdt_object_put(env, m->mdt_md_root);
4677                 m->mdt_md_root = NULL;
4678         }
4679
4680         mdt_seq_fini(env, m);
4681
4682         mdt_fld_fini(env, m);
4683
4684         /*
4685          * Finish the stack
4686          */
4687         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4688
4689         LASSERT(atomic_read(&d->ld_ref) == 0);
4690
4691         server_put_mount(mdt_obd_name(m), true);
4692
4693         EXIT;
4694 }
4695
4696 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
4697
4698 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
4699                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
4700 {
4701         struct mdt_thread_info    *info;
4702         struct obd_device         *obd;
4703         const char                *dev = lustre_cfg_string(cfg, 0);
4704         const char                *num = lustre_cfg_string(cfg, 2);
4705         struct lustre_mount_info  *lmi = NULL;
4706         struct lustre_sb_info     *lsi;
4707         struct lu_site            *s;
4708         struct seq_server_site    *ss_site;
4709         const char                *identity_upcall = "NONE";
4710         struct md_device          *next;
4711         int                        rc;
4712         long                       node_id;
4713         mntopt_t                   mntopts;
4714         ENTRY;
4715
4716         lu_device_init(&m->mdt_lu_dev, ldt);
4717         /*
4718          * Environment (env) might be missing mdt_thread_key values at that
4719          * point, if device is allocated when mdt_thread_key is in QUIESCENT
4720          * mode.
4721          *
4722          * Usually device allocation path doesn't use module key values, but
4723          * mdt has to do a lot of work here, so allocate key value.
4724          */
4725         rc = lu_env_refill((struct lu_env *)env);
4726         if (rc != 0)
4727                 RETURN(rc);
4728
4729         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
4730         LASSERT(info != NULL);
4731
4732         obd = class_name2obd(dev);
4733         LASSERT(obd != NULL);
4734
4735         m->mdt_max_mdsize = MAX_MD_SIZE; /* 4 stripes */
4736         m->mdt_opts.mo_evict_tgt_nids = 1;
4737         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
4738
4739         /* default is coordinator off, it is started through conf_param
4740          * or /proc */
4741         m->mdt_opts.mo_coordinator = 0;
4742
4743         lmi = server_get_mount(dev);
4744         if (lmi == NULL) {
4745                 CERROR("Cannot get mount info for %s!\n", dev);
4746                 RETURN(-EFAULT);
4747         } else {
4748                 lsi = s2lsi(lmi->lmi_sb);
4749                 /* CMD is supported only in IAM mode */
4750                 LASSERT(num);
4751                 node_id = simple_strtol(num, NULL, 10);
4752                 obd->u.obt.obt_magic = OBT_MAGIC;
4753                 if (lsi->lsi_lmd != NULL &&
4754                     lsi->lsi_lmd->lmd_flags & LMD_FLG_SKIP_LFSCK)
4755                         m->mdt_skip_lfsck = 1;
4756         }
4757
4758         m->mdt_squash.rsi_uid = 0;
4759         m->mdt_squash.rsi_gid = 0;
4760         INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
4761         init_rwsem(&m->mdt_squash.rsi_sem);
4762         spin_lock_init(&m->mdt_lock);
4763         m->mdt_osfs_age = cfs_time_shift_64(-1000);
4764         m->mdt_enable_remote_dir = 0;
4765         m->mdt_enable_remote_dir_gid = 0;
4766
4767         atomic_set(&m->mdt_mds_mds_conns, 0);
4768         atomic_set(&m->mdt_async_commit_count, 0);
4769
4770         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
4771         m->mdt_lu_dev.ld_obd = obd;
4772         /* Set this lu_device to obd for error handling purposes. */
4773         obd->obd_lu_dev = &m->mdt_lu_dev;
4774
4775         /* init the stack */
4776         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
4777         if (rc) {
4778                 CERROR("%s: Can't init device stack, rc %d\n",
4779                        mdt_obd_name(m), rc);
4780                 GOTO(err_lmi, rc);
4781         }
4782
4783         s = mdt_lu_site(m);
4784         ss_site = mdt_seq_site(m);
4785         s->ld_seq_site = ss_site;
4786         ss_site->ss_lu = s;
4787
4788         /* set server index */
4789         ss_site->ss_node_id = node_id;
4790
4791         /* failover is the default
4792          * FIXME: we do not failout mds0/mgs, which may cause some problems.
4793          * assumed whose ss_node_id == 0 XXX
4794          * */
4795         obd->obd_replayable = 1;
4796         /* No connection accepted until configurations will finish */
4797         obd->obd_no_conn = 1;
4798
4799         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
4800                 char *str = lustre_cfg_string(cfg, 4);
4801                 if (strchr(str, 'n')) {
4802                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
4803                         obd->obd_replayable = 0;
4804                 }
4805         }
4806
4807         rc = mdt_fld_init(env, mdt_obd_name(m), m);
4808         if (rc)
4809                 GOTO(err_fini_stack, rc);
4810
4811         rc = mdt_seq_init(env, m);
4812         if (rc)
4813                 GOTO(err_fini_fld, rc);
4814
4815         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
4816                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
4817         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
4818                                               LDLM_NAMESPACE_SERVER,
4819                                               LDLM_NAMESPACE_GREEDY,
4820                                               LDLM_NS_TYPE_MDT);
4821         if (m->mdt_namespace == NULL)
4822                 GOTO(err_fini_seq, rc = -ENOMEM);
4823
4824         m->mdt_namespace->ns_lvbp = m;
4825         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
4826
4827         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
4828         /* set obd_namespace for compatibility with old code */
4829         obd->obd_namespace = m->mdt_namespace;
4830
4831         rc = mdt_hsm_cdt_init(m);
4832         if (rc != 0) {
4833                 CERROR("%s: error initializing coordinator, rc %d\n",
4834                        mdt_obd_name(m), rc);
4835                 GOTO(err_free_ns, rc);
4836         }
4837
4838         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
4839                       OBD_FAIL_MDS_ALL_REQUEST_NET,
4840                       OBD_FAIL_MDS_ALL_REPLY_NET);
4841         if (rc)
4842                 GOTO(err_free_hsm, rc);
4843
4844         rc = mdt_fs_setup(env, m, obd, lsi);
4845         if (rc)
4846                 GOTO(err_tgt, rc);
4847
4848         tgt_adapt_sptlrpc_conf(&m->mdt_lut);
4849
4850         next = m->mdt_child;
4851         rc = next->md_ops->mdo_iocontrol(env, next, OBD_IOC_GET_MNTOPT, 0,
4852                                          &mntopts);
4853         if (rc)
4854                 GOTO(err_fs_cleanup, rc);
4855
4856         if (mntopts & MNTOPT_USERXATTR)
4857                 m->mdt_opts.mo_user_xattr = 1;
4858         else
4859                 m->mdt_opts.mo_user_xattr = 0;
4860
4861         rc = next->md_ops->mdo_maxeasize_get(env, next, &m->mdt_max_ea_size);
4862         if (rc)
4863                 GOTO(err_fs_cleanup, rc);
4864
4865         if (mntopts & MNTOPT_ACL)
4866                 m->mdt_opts.mo_acl = 1;
4867         else
4868                 m->mdt_opts.mo_acl = 0;
4869
4870         /* XXX: to support suppgid for ACL, we enable identity_upcall
4871          * by default, otherwise, maybe got unexpected -EACCESS. */
4872         if (m->mdt_opts.mo_acl)
4873                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
4874
4875         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
4876                                                 identity_upcall,
4877                                                 &mdt_identity_upcall_cache_ops);
4878         if (IS_ERR(m->mdt_identity_cache)) {
4879                 rc = PTR_ERR(m->mdt_identity_cache);
4880                 m->mdt_identity_cache = NULL;
4881                 GOTO(err_fs_cleanup, rc);
4882         }
4883
4884         rc = mdt_procfs_init(m, dev);
4885         if (rc) {
4886                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
4887                 GOTO(err_recovery, rc);
4888         }
4889
4890         rc = mdt_quota_init(env, m, cfg);
4891         if (rc)
4892                 GOTO(err_procfs, rc);
4893
4894         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
4895         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
4896                            "mdt_ldlm_client", m->mdt_ldlm_client);
4897
4898         ping_evictor_start();
4899
4900         /* recovery will be started upon mdt_prepare()
4901          * when the whole stack is complete and ready
4902          * to serve the requests */
4903
4904         /* Reduce the initial timeout on an MDS because it doesn't need such
4905          * a long timeout as an OST does. Adaptive timeouts will adjust this
4906          * value appropriately. */
4907         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
4908                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
4909
4910         RETURN(0);
4911 err_procfs:
4912         mdt_procfs_fini(m);
4913 err_recovery:
4914         target_recovery_fini(obd);
4915         upcall_cache_cleanup(m->mdt_identity_cache);
4916         m->mdt_identity_cache = NULL;
4917 err_fs_cleanup:
4918         mdt_fs_cleanup(env, m);
4919 err_tgt:
4920         tgt_fini(env, &m->mdt_lut);
4921 err_free_hsm:
4922         mdt_hsm_cdt_fini(m);
4923 err_free_ns:
4924         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
4925         obd->obd_namespace = m->mdt_namespace = NULL;
4926 err_fini_seq:
4927         mdt_seq_fini(env, m);
4928 err_fini_fld:
4929         mdt_fld_fini(env, m);
4930 err_fini_stack:
4931         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
4932 err_lmi:
4933         if (lmi)
4934                 server_put_mount(dev, true);
4935         return(rc);
4936 }
4937
4938 /* For interoperability, the left element is old parameter, the right one
4939  * is the new version of the parameter, if some parameter is deprecated,
4940  * the new version should be set as NULL. */
4941 static struct cfg_interop_param mdt_interop_param[] = {
4942         { "mdt.group_upcall",   NULL },
4943         { "mdt.quota_type",     NULL },
4944         { "mdd.quota_type",     NULL },
4945         { "mdt.som",            NULL },
4946         { "mdt.rootsquash",     "mdt.root_squash" },
4947         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
4948         { NULL }
4949 };
4950
4951 /* used by MGS to process specific configurations */
4952 static int mdt_process_config(const struct lu_env *env,
4953                               struct lu_device *d, struct lustre_cfg *cfg)
4954 {
4955         struct mdt_device *m = mdt_dev(d);
4956         struct md_device *md_next = m->mdt_child;
4957         struct lu_device *next = md2lu_dev(md_next);
4958         int rc;
4959         ENTRY;
4960
4961         switch (cfg->lcfg_command) {
4962         case LCFG_PARAM: {
4963                 struct obd_device          *obd = d->ld_obd;
4964
4965                 /* For interoperability */
4966                 struct cfg_interop_param   *ptr = NULL;
4967                 struct lustre_cfg          *old_cfg = NULL;
4968                 char                       *param = NULL;
4969
4970                 param = lustre_cfg_string(cfg, 1);
4971                 if (param == NULL) {
4972                         CERROR("param is empty\n");
4973                         rc = -EINVAL;
4974                         break;
4975                 }
4976
4977                 ptr = class_find_old_param(param, mdt_interop_param);
4978                 if (ptr != NULL) {
4979                         if (ptr->new_param == NULL) {
4980                                 rc = 0;
4981                                 CWARN("For interoperability, skip this %s."
4982                                       " It is obsolete.\n", ptr->old_param);
4983                                 break;
4984                         }
4985
4986                         CWARN("Found old param %s, changed it to %s.\n",
4987                               ptr->old_param, ptr->new_param);
4988
4989                         old_cfg = cfg;
4990                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
4991                         if (IS_ERR(cfg)) {
4992                                 rc = PTR_ERR(cfg);
4993                                 break;
4994                         }
4995                 }
4996
4997                 rc = class_process_proc_param(PARAM_MDT, obd->obd_vars,
4998                                               cfg, obd);
4999                 if (rc > 0 || rc == -ENOSYS) {
5000                         /* is it an HSM var ? */
5001                         rc = class_process_proc_param(PARAM_HSM,
5002                                                       hsm_cdt_get_proc_vars(),
5003                                                       cfg, obd);
5004                         if (rc > 0 || rc == -ENOSYS)
5005                                 /* we don't understand; pass it on */
5006                                 rc = next->ld_ops->ldo_process_config(env, next,
5007                                                                       cfg);
5008                 }
5009
5010                 if (old_cfg != NULL)
5011                         lustre_cfg_free(cfg);
5012
5013                 break;
5014         }
5015         default:
5016                 /* others are passed further */
5017                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
5018                 break;
5019         }
5020         RETURN(rc);
5021 }
5022
5023 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
5024                                           const struct lu_object_header *hdr,
5025                                           struct lu_device *d)
5026 {
5027         struct mdt_object *mo;
5028
5029         ENTRY;
5030
5031         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
5032         if (mo != NULL) {
5033                 struct lu_object *o;
5034                 struct lu_object_header *h;
5035
5036                 o = &mo->mot_obj;
5037                 h = &mo->mot_header;
5038                 lu_object_header_init(h);
5039                 lu_object_init(o, h, d);
5040                 lu_object_add_top(h, o);
5041                 o->lo_ops = &mdt_obj_ops;
5042                 spin_lock_init(&mo->mot_write_lock);
5043                 mutex_init(&mo->mot_lov_mutex);
5044                 init_rwsem(&mo->mot_open_sem);
5045                 atomic_set(&mo->mot_open_count, 0);
5046                 RETURN(o);
5047         }
5048         RETURN(NULL);
5049 }
5050
5051 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
5052                            const struct lu_object_conf *unused)
5053 {
5054         struct mdt_device *d = mdt_dev(o->lo_dev);
5055         struct lu_device  *under;
5056         struct lu_object  *below;
5057         int                rc = 0;
5058         ENTRY;
5059
5060         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
5061                PFID(lu_object_fid(o)));
5062
5063         under = &d->mdt_child->md_lu_dev;
5064         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
5065         if (below != NULL) {
5066                 lu_object_add(o, below);
5067         } else
5068                 rc = -ENOMEM;
5069
5070         RETURN(rc);
5071 }
5072
5073 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
5074 {
5075         struct mdt_object *mo = mdt_obj(o);
5076         struct lu_object_header *h;
5077         ENTRY;
5078
5079         h = o->lo_header;
5080         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
5081                PFID(lu_object_fid(o)));
5082
5083         LASSERT(atomic_read(&mo->mot_open_count) == 0);
5084         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
5085
5086         lu_object_fini(o);
5087         lu_object_header_fini(h);
5088         OBD_SLAB_FREE_PTR(mo, mdt_object_kmem);
5089
5090         EXIT;
5091 }
5092
5093 static int mdt_object_print(const struct lu_env *env, void *cookie,
5094                             lu_printer_t p, const struct lu_object *o)
5095 {
5096         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
5097
5098         return (*p)(env, cookie,
5099                     LUSTRE_MDT_NAME"-object@%p(%s %s, writecount=%d)",
5100                     mdto, mdto->mot_lov_created ? "lov_created" : "",
5101                     mdto->mot_cache_attr ? "cache_attr" : "",
5102                     mdto->mot_write_count);
5103 }
5104
5105 static int mdt_prepare(const struct lu_env *env,
5106                 struct lu_device *pdev,
5107                 struct lu_device *cdev)
5108 {
5109         struct mdt_device *mdt = mdt_dev(cdev);
5110         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
5111         struct obd_device *obd = cdev->ld_obd;
5112         int rc;
5113
5114         ENTRY;
5115
5116         LASSERT(obd);
5117
5118         rc = next->ld_ops->ldo_prepare(env, cdev, next);
5119         if (rc)
5120                 RETURN(rc);
5121
5122         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
5123         if (rc)
5124                 RETURN(rc);
5125
5126         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
5127         if (rc)
5128                 RETURN(rc);
5129
5130         rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
5131         /* The LFSCK instance is registered just now, so it must be there when
5132          * register the namespace to such instance. */
5133         LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
5134
5135         if (mdt->mdt_seq_site.ss_node_id == 0) {
5136                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
5137                                                          &mdt->mdt_md_root_fid);
5138                 if (rc)
5139                         RETURN(rc);
5140         }
5141
5142         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
5143
5144         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
5145         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
5146         LASSERT(obd->obd_no_conn);
5147         spin_lock(&obd->obd_dev_lock);
5148         obd->obd_no_conn = 0;
5149         spin_unlock(&obd->obd_dev_lock);
5150
5151         if (obd->obd_recovering == 0)
5152                 mdt_postrecov(env, mdt);
5153
5154         RETURN(rc);
5155 }
5156
5157 const struct lu_device_operations mdt_lu_ops = {
5158         .ldo_object_alloc   = mdt_object_alloc,
5159         .ldo_process_config = mdt_process_config,
5160         .ldo_prepare        = mdt_prepare,
5161 };
5162
5163 static const struct lu_object_operations mdt_obj_ops = {
5164         .loo_object_init    = mdt_object_init,
5165         .loo_object_free    = mdt_object_free,
5166         .loo_object_print   = mdt_object_print
5167 };
5168
5169 static int mdt_obd_set_info_async(const struct lu_env *env,
5170                                   struct obd_export *exp,
5171                                   __u32 keylen, void *key,
5172                                   __u32 vallen, void *val,
5173                                   struct ptlrpc_request_set *set)
5174 {
5175         int rc;
5176
5177         ENTRY;
5178
5179         if (KEY_IS(KEY_SPTLRPC_CONF)) {
5180                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp));
5181                 RETURN(rc);
5182         }
5183
5184         RETURN(0);
5185 }
5186
5187 /**
5188  * Match client and server connection feature flags.
5189  *
5190  * Compute the compatibility flags for a connection request based on
5191  * features mutually supported by client and server.
5192  *
5193  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
5194  * must not be updated here, otherwise a partially initialized value may
5195  * be exposed. After the connection request is successfully processed,
5196  * the top-level MDT connect request handler atomically updates the export
5197  * connect flags from the obd_connect_data::ocd_connect_flags field of the
5198  * reply. \see mdt_connect().
5199  *
5200  * Before 2.7.50 clients will send a struct obd_connect_data_v1 rather than a
5201  * full struct obd_connect_data. So care must be taken when accessing fields
5202  * that are not present in struct obd_connect_data_v1. See LU-16.
5203  *
5204  * \param exp   the obd_export associated with this client/target pair
5205  * \param mdt   the target device for the connection
5206  * \param data  stores data for this connect request
5207  *
5208  * \retval 0       success
5209  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
5210  * \retval -EBADE  client and server feature requirements are incompatible
5211  */
5212 static int mdt_connect_internal(struct obd_export *exp,
5213                                 struct mdt_device *mdt,
5214                                 struct obd_connect_data *data)
5215 {
5216         LASSERT(data != NULL);
5217
5218         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
5219
5220         if (mdt->mdt_bottom->dd_rdonly &&
5221             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5222             !(data->ocd_connect_flags & OBD_CONNECT_RDONLY))
5223                 RETURN(-EACCES);
5224
5225         if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
5226                 data->ocd_connect_flags2 &= MDT_CONNECT_SUPPORTED2;
5227
5228         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
5229
5230         if (!(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5231             !(data->ocd_connect_flags & OBD_CONNECT_IBITS)) {
5232                 CWARN("%s: client %s does not support ibits lock, either "
5233                       "very old or an invalid client: flags %#llx\n",
5234                       mdt_obd_name(mdt), exp->exp_client_uuid.uuid,
5235                       data->ocd_connect_flags);
5236                 return -EBADE;
5237         }
5238
5239         if (!mdt->mdt_opts.mo_acl)
5240                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
5241
5242         if (!mdt->mdt_opts.mo_user_xattr)
5243                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
5244
5245         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE) {
5246                 data->ocd_brw_size = min(data->ocd_brw_size,
5247                                          (__u32)MD_MAX_BRW_SIZE);
5248                 if (data->ocd_brw_size == 0) {
5249                         CERROR("%s: cli %s/%p ocd_connect_flags: %#llx"
5250                                " ocd_version: %x ocd_grant: %d "
5251                                "ocd_index: %u ocd_brw_size is "
5252                                "unexpectedly zero, network data "
5253                                "corruption? Refusing connection of this"
5254                                " client\n",
5255                                mdt_obd_name(mdt),
5256                                exp->exp_client_uuid.uuid,
5257                                exp, data->ocd_connect_flags, data->ocd_version,
5258                                data->ocd_grant, data->ocd_index);
5259                         return -EPROTO;
5260                 }
5261         }
5262
5263         /* NB: Disregard the rule against updating
5264          * exp_connect_data.ocd_connect_flags in this case, since
5265          * tgt_client_new() needs to know if this is a lightweight
5266          * connection, and it is safe to expose this flag before
5267          * connection processing completes. */
5268         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
5269                 spin_lock(&exp->exp_lock);
5270                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
5271                 spin_unlock(&exp->exp_lock);
5272         }
5273
5274         data->ocd_version = LUSTRE_VERSION_CODE;
5275
5276         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
5277                 CWARN("%s: MDS requires FID support, but client not\n",
5278                       mdt_obd_name(mdt));
5279                 return -EBADE;
5280         }
5281
5282         if (OCD_HAS_FLAG(data, PINGLESS)) {
5283                 if (ptlrpc_pinger_suppress_pings()) {
5284                         spin_lock(&exp->exp_obd->obd_dev_lock);
5285                         list_del_init(&exp->exp_obd_chain_timed);
5286                         spin_unlock(&exp->exp_obd->obd_dev_lock);
5287                 } else {
5288                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
5289                 }
5290         }
5291
5292         data->ocd_max_easize = mdt->mdt_max_ea_size;
5293
5294         /* NB: Disregard the rule against updating
5295          * exp_connect_data.ocd_connect_flags in this case, since
5296          * tgt_client_new() needs to know if this is client supports
5297          * multiple modify RPCs, and it is safe to expose this flag before
5298          * connection processing completes. */
5299         if (data->ocd_connect_flags & OBD_CONNECT_MULTIMODRPCS) {
5300                 data->ocd_maxmodrpcs = max_mod_rpcs_per_client;
5301                 spin_lock(&exp->exp_lock);
5302                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_MULTIMODRPCS;
5303                 spin_unlock(&exp->exp_lock);
5304         }
5305
5306         return 0;
5307 }
5308
5309 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
5310                                    struct mdt_thread_info *info,
5311                                    struct mdt_file_data *mfd)
5312 {
5313         struct lu_context ses;
5314         int rc;
5315         ENTRY;
5316
5317         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
5318         if (rc)
5319                 RETURN(rc);
5320
5321         env->le_ses = &ses;
5322         lu_context_enter(&ses);
5323
5324         mdt_ucred(info)->uc_valid = UCRED_OLD;
5325         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
5326
5327         lu_context_exit(&ses);
5328         lu_context_fini(&ses);
5329         env->le_ses = NULL;
5330
5331         RETURN(rc);
5332 }
5333
5334 static int mdt_export_cleanup(struct obd_export *exp)
5335 {
5336         struct list_head         closing_list;
5337         struct mdt_export_data  *med = &exp->exp_mdt_data;
5338         struct obd_device       *obd = exp->exp_obd;
5339         struct mdt_device       *mdt;
5340         struct mdt_thread_info  *info;
5341         struct lu_env            env;
5342         struct mdt_file_data    *mfd, *n;
5343         int rc = 0;
5344         ENTRY;
5345
5346         INIT_LIST_HEAD(&closing_list);
5347         spin_lock(&med->med_open_lock);
5348         while (!list_empty(&med->med_open_head)) {
5349                 struct list_head *tmp = med->med_open_head.next;
5350                 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
5351
5352                 /* Remove mfd handle so it can't be found again.
5353                  * We are consuming the mfd_list reference here. */
5354                 class_handle_unhash(&mfd->mfd_handle);
5355                 list_move_tail(&mfd->mfd_list, &closing_list);
5356         }
5357         spin_unlock(&med->med_open_lock);
5358         mdt = mdt_dev(obd->obd_lu_dev);
5359         LASSERT(mdt != NULL);
5360
5361         rc = lu_env_init(&env, LCT_MD_THREAD);
5362         if (rc)
5363                 RETURN(rc);
5364
5365         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
5366         LASSERT(info != NULL);
5367         memset(info, 0, sizeof *info);
5368         info->mti_env = &env;
5369         info->mti_mdt = mdt;
5370         info->mti_exp = exp;
5371
5372         if (!list_empty(&closing_list)) {
5373                 struct md_attr *ma = &info->mti_attr;
5374
5375                 /* Close any open files (which may also cause orphan
5376                  * unlinking). */
5377                 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
5378                         list_del_init(&mfd->mfd_list);
5379                         ma->ma_need = ma->ma_valid = 0;
5380
5381                         /* This file is being closed due to an eviction, it
5382                          * could have been modified and now dirty regarding to
5383                          * HSM archive, check this!
5384                          * The logic here is to mark a file dirty if there's a
5385                          * chance it was dirtied before the client was evicted,
5386                          * so that we don't have to wait for a release attempt
5387                          * before finding out the file was actually dirty and
5388                          * fail the release. Aggressively marking it dirty here
5389                          * will cause the policy engine to attempt to
5390                          * re-archive it; when rearchiving, we can compare the
5391                          * current version to the HSM data_version and make the
5392                          * archive request into a noop if it's not actually
5393                          * dirty.
5394                          */
5395                         if (mfd->mfd_mode & FMODE_WRITE)
5396                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
5397
5398                         /* Don't unlink orphan on failover umount, LU-184 */
5399                         if (exp->exp_flags & OBD_OPT_FAILOVER) {
5400                                 ma->ma_valid = MA_FLAGS;
5401                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
5402                         }
5403                         mdt_mfd_close(info, mfd);
5404                 }
5405         }
5406         info->mti_mdt = NULL;
5407         /* cleanup client slot early */
5408         /* Do not erase record for recoverable client. */
5409         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
5410                 tgt_client_del(&env, exp);
5411         lu_env_fini(&env);
5412
5413         RETURN(rc);
5414 }
5415
5416 static inline void mdt_enable_slc(struct mdt_device *mdt)
5417 {
5418         if (mdt->mdt_lut.lut_sync_lock_cancel == NEVER_SYNC_ON_CANCEL)
5419                 mdt->mdt_lut.lut_sync_lock_cancel = BLOCKING_SYNC_ON_CANCEL;
5420 }
5421
5422 static inline void mdt_disable_slc(struct mdt_device *mdt)
5423 {
5424         if (mdt->mdt_lut.lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL)
5425                 mdt->mdt_lut.lut_sync_lock_cancel = NEVER_SYNC_ON_CANCEL;
5426 }
5427
5428 static int mdt_obd_disconnect(struct obd_export *exp)
5429 {
5430         int rc;
5431         ENTRY;
5432
5433         LASSERT(exp);
5434         class_export_get(exp);
5435
5436         if ((exp_connect_flags(exp) & OBD_CONNECT_MDS_MDS) &&
5437             !(exp_connect_flags(exp) & OBD_CONNECT_LIGHTWEIGHT)) {
5438                 struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
5439
5440                 if (atomic_dec_and_test(&mdt->mdt_mds_mds_conns))
5441                         mdt_disable_slc(mdt);
5442         }
5443
5444         rc = server_disconnect_export(exp);
5445         if (rc != 0)
5446                 CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
5447
5448         rc = mdt_export_cleanup(exp);
5449         nodemap_del_member(exp);
5450         class_export_put(exp);
5451         RETURN(rc);
5452 }
5453
5454 /* mds_connect copy */
5455 static int mdt_obd_connect(const struct lu_env *env,
5456                            struct obd_export **exp, struct obd_device *obd,
5457                            struct obd_uuid *cluuid,
5458                            struct obd_connect_data *data,
5459                            void *localdata)
5460 {
5461         struct obd_export       *lexp;
5462         struct lustre_handle    conn = { 0 };
5463         struct mdt_device       *mdt;
5464         int                      rc;
5465         lnet_nid_t              *client_nid = localdata;
5466         ENTRY;
5467
5468         LASSERT(env != NULL);
5469         LASSERT(data != NULL);
5470
5471         if (!exp || !obd || !cluuid)
5472                 RETURN(-EINVAL);
5473
5474         mdt = mdt_dev(obd->obd_lu_dev);
5475
5476         if ((data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
5477             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
5478                 atomic_inc(&mdt->mdt_mds_mds_conns);
5479                 mdt_enable_slc(mdt);
5480         }
5481
5482         /*
5483          * first, check whether the stack is ready to handle requests
5484          * XXX: probably not very appropriate method is used now
5485          *      at some point we should find a better one
5486          */
5487         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) &&
5488             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) &&
5489             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
5490                 rc = obd_get_info(env, mdt->mdt_child_exp,
5491                                   sizeof(KEY_OSP_CONNECTED),
5492                                   KEY_OSP_CONNECTED, NULL, NULL);
5493                 if (rc)
5494                         RETURN(-EAGAIN);
5495                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
5496         }
5497
5498         rc = class_connect(&conn, obd, cluuid);
5499         if (rc)
5500                 RETURN(rc);
5501
5502         lexp = class_conn2export(&conn);
5503         LASSERT(lexp != NULL);
5504
5505         rc = nodemap_add_member(*client_nid, lexp);
5506         if (rc != 0 && rc != -EEXIST)
5507                 GOTO(out, rc);
5508
5509         rc = mdt_connect_internal(lexp, mdt, data);
5510         if (rc == 0) {
5511                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
5512
5513                 LASSERT(lcd);
5514                 memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
5515                 rc = tgt_client_new(env, lexp);
5516                 if (rc == 0)
5517                         mdt_export_stats_init(obd, lexp, localdata);
5518         }
5519 out:
5520         if (rc != 0) {
5521                 class_disconnect(lexp);
5522                 nodemap_del_member(lexp);
5523                 *exp = NULL;
5524         } else {
5525                 *exp = lexp;
5526                 /* Because we do not want this export to be evicted by pinger,
5527                  * let's not add this export to the timed chain list. */
5528                 if (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) {
5529                         spin_lock(&lexp->exp_obd->obd_dev_lock);
5530                         list_del_init(&lexp->exp_obd_chain_timed);
5531                         spin_unlock(&lexp->exp_obd->obd_dev_lock);
5532                 }
5533         }
5534
5535         RETURN(rc);
5536 }
5537
5538 static int mdt_obd_reconnect(const struct lu_env *env,
5539                              struct obd_export *exp, struct obd_device *obd,
5540                              struct obd_uuid *cluuid,
5541                              struct obd_connect_data *data,
5542                              void *localdata)
5543 {
5544         lnet_nid_t             *client_nid = localdata;
5545         int                     rc;
5546         ENTRY;
5547
5548         if (exp == NULL || obd == NULL || cluuid == NULL)
5549                 RETURN(-EINVAL);
5550
5551         rc = nodemap_add_member(*client_nid, exp);
5552         if (rc != 0 && rc != -EEXIST)
5553                 RETURN(rc);
5554
5555         rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
5556         if (rc == 0)
5557                 mdt_export_stats_init(obd, exp, localdata);
5558         else
5559                 nodemap_del_member(exp);
5560
5561         RETURN(rc);
5562 }
5563
5564 /* FIXME: Can we avoid using these two interfaces? */
5565 static int mdt_init_export(struct obd_export *exp)
5566 {
5567         struct mdt_export_data *med = &exp->exp_mdt_data;
5568         int                     rc;
5569         ENTRY;
5570
5571         INIT_LIST_HEAD(&med->med_open_head);
5572         spin_lock_init(&med->med_open_lock);
5573         spin_lock(&exp->exp_lock);
5574         exp->exp_connecting = 1;
5575         spin_unlock(&exp->exp_lock);
5576
5577         /* self-export doesn't need client data and ldlm initialization */
5578         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5579                                      &exp->exp_client_uuid)))
5580                 RETURN(0);
5581
5582         rc = tgt_client_alloc(exp);
5583         if (rc)
5584                 GOTO(err, rc);
5585
5586         rc = ldlm_init_export(exp);
5587         if (rc)
5588                 GOTO(err_free, rc);
5589
5590         RETURN(rc);
5591
5592 err_free:
5593         tgt_client_free(exp);
5594 err:
5595         CERROR("%s: Failed to initialize export: rc = %d\n",
5596                exp->exp_obd->obd_name, rc);
5597         return rc;
5598 }
5599
5600 static int mdt_destroy_export(struct obd_export *exp)
5601 {
5602         ENTRY;
5603
5604         target_destroy_export(exp);
5605         /* destroy can be called from failed obd_setup, so
5606          * checking uuid is safer than obd_self_export */
5607         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
5608                                      &exp->exp_client_uuid)))
5609                 RETURN(0);
5610
5611         ldlm_destroy_export(exp);
5612         tgt_client_free(exp);
5613
5614         LASSERT(list_empty(&exp->exp_outstanding_replies));
5615         LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
5616
5617         RETURN(0);
5618 }
5619
5620 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
5621                    struct linkea_data *ldata)
5622 {
5623         int rc;
5624
5625         LASSERT(ldata->ld_buf->lb_buf != NULL);
5626
5627         if (!mdt_object_exists(mdt_obj))
5628                 return -ENODATA;
5629
5630         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5631                           ldata->ld_buf, XATTR_NAME_LINK);
5632         if (rc == -ERANGE) {
5633                 /* Buf was too small, figure out what we need. */
5634                 lu_buf_free(ldata->ld_buf);
5635                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5636                                   ldata->ld_buf, XATTR_NAME_LINK);
5637                 if (rc < 0)
5638                         return rc;
5639                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
5640                 if (ldata->ld_buf->lb_buf == NULL)
5641                         return -ENOMEM;
5642                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5643                                   ldata->ld_buf, XATTR_NAME_LINK);
5644         }
5645         if (rc < 0)
5646                 return rc;
5647
5648         return linkea_init_with_rec(ldata);
5649 }
5650
5651 /**
5652  * Given an MDT object, try to look up the full path to the object.
5653  * Part of the MDT layer implementation of lfs fid2path.
5654  *
5655  * \param[in]     info  Per-thread common data shared by MDT level handlers.
5656  * \param[in]     obj   Object to do path lookup of
5657  * \param[in,out] fp    User-provided struct to store path information
5658  * \param[in]     root_fid Root FID of current path should reach
5659  *
5660  * \retval 0 Lookup successful, path information stored in fp
5661  * \retval -EAGAIN Lookup failed, usually because object is being moved
5662  * \retval negative errno if there was a problem
5663  */
5664 static int mdt_path_current(struct mdt_thread_info *info,
5665                             struct mdt_object *obj,
5666                             struct getinfo_fid2path *fp,
5667                             struct lu_fid *root_fid)
5668 {
5669         struct mdt_device       *mdt = info->mti_mdt;
5670         struct mdt_object       *mdt_obj;
5671         struct link_ea_header   *leh;
5672         struct link_ea_entry    *lee;
5673         struct lu_name          *tmpname = &info->mti_name;
5674         struct lu_fid           *tmpfid = &info->mti_tmp_fid1;
5675         struct lu_buf           *buf = &info->mti_big_buf;
5676         char                    *ptr;
5677         int                     reclen;
5678         struct linkea_data      ldata = { NULL };
5679         int                     rc = 0;
5680         bool                    first = true;
5681         ENTRY;
5682
5683         /* temp buffer for path element, the buffer will be finally freed
5684          * in mdt_thread_info_fini */
5685         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
5686         if (buf->lb_buf == NULL)
5687                 RETURN(-ENOMEM);
5688
5689         ldata.ld_buf = buf;
5690         ptr = fp->gf_u.gf_path + fp->gf_pathlen - 1;
5691         *ptr = 0;
5692         --ptr;
5693         *tmpfid = fp->gf_fid = *mdt_object_fid(obj);
5694
5695         while (!lu_fid_eq(root_fid, &fp->gf_fid)) {
5696                 struct lu_buf           lmv_buf;
5697
5698                 if (!lu_fid_eq(root_fid, &mdt->mdt_md_root_fid) &&
5699                     lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid))
5700                         GOTO(out, rc = -ENOENT);
5701
5702                 if (lu_fid_eq(mdt_object_fid(obj), tmpfid)) {
5703                         mdt_obj = obj;
5704                         mdt_object_get(info->mti_env, mdt_obj);
5705                 } else {
5706                         mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
5707                         if (IS_ERR(mdt_obj))
5708                                 GOTO(out, rc = PTR_ERR(mdt_obj));
5709                 }
5710
5711                 if (!mdt_object_exists(mdt_obj)) {
5712                         mdt_object_put(info->mti_env, mdt_obj);
5713                         GOTO(out, rc = -ENOENT);
5714                 }
5715
5716                 if (mdt_object_remote(mdt_obj)) {
5717                         mdt_object_put(info->mti_env, mdt_obj);
5718                         GOTO(remote_out, rc = -EREMOTE);
5719                 }
5720
5721                 rc = mdt_links_read(info, mdt_obj, &ldata);
5722                 if (rc != 0) {
5723                         mdt_object_put(info->mti_env, mdt_obj);
5724                         GOTO(out, rc);
5725                 }
5726
5727                 leh = buf->lb_buf;
5728                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
5729                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
5730                 /* If set, use link #linkno for path lookup, otherwise use
5731                    link #0.  Only do this for the final path element. */
5732                 if (first && fp->gf_linkno < leh->leh_reccount) {
5733                         int count;
5734                         for (count = 0; count < fp->gf_linkno; count++) {
5735                                 lee = (struct link_ea_entry *)
5736                                      ((char *)lee + reclen);
5737                                 linkea_entry_unpack(lee, &reclen, tmpname,
5738                                                     tmpfid);
5739                         }
5740                         if (fp->gf_linkno < leh->leh_reccount - 1)
5741                                 /* indicate to user there are more links */
5742                                 fp->gf_linkno++;
5743                 }
5744
5745                 lmv_buf.lb_buf = info->mti_xattr_buf;
5746                 lmv_buf.lb_len = sizeof(info->mti_xattr_buf);
5747                 /* Check if it is slave stripes */
5748                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
5749                                   &lmv_buf, XATTR_NAME_LMV);
5750                 mdt_object_put(info->mti_env, mdt_obj);
5751                 if (rc > 0) {
5752                         union lmv_mds_md *lmm = lmv_buf.lb_buf;
5753
5754                         /* For slave stripes, get its master */
5755                         if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE) {
5756                                 fp->gf_fid = *tmpfid;
5757                                 continue;
5758                         }
5759                 } else if (rc < 0 && rc != -ENODATA) {
5760                         GOTO(out, rc);
5761                 }
5762
5763                 rc = 0;
5764
5765                 /* Pack the name in the end of the buffer */
5766                 ptr -= tmpname->ln_namelen;
5767                 if (ptr - 1 <= fp->gf_u.gf_path)
5768                         GOTO(out, rc = -EOVERFLOW);
5769                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
5770                 *(--ptr) = '/';
5771
5772                 /* keep the last resolved fid to the client, so the
5773                  * client will build the left path on another MDT for
5774                  * remote object */
5775                 fp->gf_fid = *tmpfid;
5776
5777                 first = false;
5778         }
5779
5780 remote_out:
5781         ptr++; /* skip leading / */
5782         memmove(fp->gf_u.gf_path, ptr,
5783                 fp->gf_u.gf_path + fp->gf_pathlen - ptr);
5784
5785 out:
5786         RETURN(rc);
5787 }
5788
5789 /**
5790  * Given an MDT object, use mdt_path_current to get the path.
5791  * Essentially a wrapper to retry mdt_path_current a set number of times
5792  * if -EAGAIN is returned (usually because an object is being moved).
5793  *
5794  * Part of the MDT layer implementation of lfs fid2path.
5795  *
5796  * \param[in]     info  Per-thread common data shared by mdt level handlers.
5797  * \param[in]     obj   Object to do path lookup of
5798  * \param[in,out] fp    User-provided struct for arguments and to store path
5799  *                      information
5800  *
5801  * \retval 0 Lookup successful, path information stored in fp
5802  * \retval negative errno if there was a problem
5803  */
5804 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
5805                     struct getinfo_fid2path *fp, struct lu_fid *root_fid)
5806 {
5807         struct mdt_device       *mdt = info->mti_mdt;
5808         int                     tries = 3;
5809         int                     rc = -EAGAIN;
5810         ENTRY;
5811
5812         if (fp->gf_pathlen < 3)
5813                 RETURN(-EOVERFLOW);
5814
5815         if (root_fid == NULL)
5816                 root_fid = &mdt->mdt_md_root_fid;
5817
5818         if (lu_fid_eq(root_fid, mdt_object_fid(obj))) {
5819                 fp->gf_u.gf_path[0] = '\0';
5820                 RETURN(0);
5821         }
5822
5823         /* Retry multiple times in case file is being moved */
5824         while (tries-- && rc == -EAGAIN)
5825                 rc = mdt_path_current(info, obj, fp, root_fid);
5826
5827         RETURN(rc);
5828 }
5829
5830 /**
5831  * Get the full path of the provided FID, as of changelog record recno.
5832  *
5833  * This checks sanity and looks up object for user provided FID
5834  * before calling the actual path lookup code.
5835  *
5836  * Part of the MDT layer implementation of lfs fid2path.
5837  *
5838  * \param[in]     info  Per-thread common data shared by mdt level handlers.
5839  * \param[in,out] fp    User-provided struct for arguments and to store path
5840  *                      information
5841  *
5842  * \retval 0 Lookup successful, path information and recno stored in fp
5843  * \retval -ENOENT, object does not exist
5844  * \retval negative errno if there was a problem
5845  */
5846 static int mdt_fid2path(struct mdt_thread_info *info,
5847                         struct lu_fid *root_fid,
5848                         struct getinfo_fid2path *fp)
5849 {
5850         struct mdt_device *mdt = info->mti_mdt;
5851         struct mdt_object *obj;
5852         int    rc;
5853         ENTRY;
5854
5855         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
5856                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
5857
5858         if (!fid_is_sane(&fp->gf_fid))
5859                 RETURN(-EINVAL);
5860
5861         if (!fid_is_namespace_visible(&fp->gf_fid)) {
5862                 CDEBUG(D_INFO, "%s: "DFID" is invalid, f_seq should be >= %#llx"
5863                        ", or f_oid != 0, or f_ver == 0\n", mdt_obd_name(mdt),
5864                        PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
5865                 RETURN(-EINVAL);
5866         }
5867
5868         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
5869         if (IS_ERR(obj)) {
5870                 rc = PTR_ERR(obj);
5871                 CDEBUG(D_IOCTL, "cannot find "DFID": rc = %d\n",
5872                        PFID(&fp->gf_fid), rc);
5873                 RETURN(rc);
5874         }
5875
5876         if (mdt_object_remote(obj))
5877                 rc = -EREMOTE;
5878         else if (!mdt_object_exists(obj))
5879                 rc = -ENOENT;
5880         else
5881                 rc = 0;
5882
5883         if (rc < 0) {
5884                 mdt_object_put(info->mti_env, obj);
5885                 CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n",
5886                        PFID(&fp->gf_fid), rc);
5887                 RETURN(rc);
5888         }
5889
5890         rc = mdt_path(info, obj, fp, root_fid);
5891
5892         CDEBUG(D_INFO, "fid "DFID", path %s recno %#llx linkno %u\n",
5893                PFID(&fp->gf_fid), fp->gf_u.gf_path,
5894                fp->gf_recno, fp->gf_linkno);
5895
5896         mdt_object_put(info->mti_env, obj);
5897
5898         RETURN(rc);
5899 }
5900
5901 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, int keylen,
5902                             void *val, int vallen)
5903 {
5904         struct getinfo_fid2path *fpout, *fpin;
5905         struct lu_fid *root_fid = NULL;
5906         int rc = 0;
5907
5908         fpin = key + cfs_size_round(sizeof(KEY_FID2PATH));
5909         fpout = val;
5910
5911         if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5912                 lustre_swab_fid2path(fpin);
5913
5914         memcpy(fpout, fpin, sizeof(*fpin));
5915         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
5916                 RETURN(-EINVAL);
5917
5918         if (keylen >= cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*fpin) +
5919                       sizeof(struct lu_fid)) {
5920                 /* client sent its root FID, which is normally fileset FID */
5921                 root_fid = fpin->gf_u.gf_root_fid;
5922                 if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
5923                         lustre_swab_lu_fid(root_fid);
5924
5925                 if (root_fid != NULL && !fid_is_sane(root_fid))
5926                         RETURN(-EINVAL);
5927         }
5928
5929         rc = mdt_fid2path(info, root_fid, fpout);
5930         RETURN(rc);
5931 }
5932
5933 int mdt_get_info(struct tgt_session_info *tsi)
5934 {
5935         char    *key;
5936         int      keylen;
5937         __u32   *vallen;
5938         void    *valout;
5939         int      rc;
5940
5941         ENTRY;
5942
5943         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
5944         if (key == NULL) {
5945                 CDEBUG(D_IOCTL, "No GETINFO key\n");
5946                 RETURN(err_serious(-EFAULT));
5947         }
5948         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
5949                                       RCL_CLIENT);
5950
5951         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
5952         if (vallen == NULL) {
5953                 CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n",
5954                                 tgt_name(tsi->tsi_tgt));
5955                 RETURN(err_serious(-EFAULT));
5956         }
5957
5958         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
5959                              *vallen);
5960         rc = req_capsule_server_pack(tsi->tsi_pill);
5961         if (rc)
5962                 RETURN(err_serious(rc));
5963
5964         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
5965         if (valout == NULL) {
5966                 CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n",
5967                                 tgt_name(tsi->tsi_tgt));
5968                 RETURN(err_serious(-EFAULT));
5969         }
5970
5971         if (KEY_IS(KEY_FID2PATH)) {
5972                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
5973
5974                 rc = mdt_rpc_fid2path(info, key, keylen, valout, *vallen);
5975                 mdt_thread_info_fini(info);
5976         } else {
5977                 rc = -EINVAL;
5978         }
5979         RETURN(rc);
5980 }
5981
5982 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
5983 {
5984         struct obd_ioctl_data *data = karg;
5985         struct lu_fid *fid;
5986         __u64 version;
5987         struct mdt_object *obj;
5988         struct mdt_lock_handle  *lh;
5989         int rc;
5990         ENTRY;
5991
5992         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
5993             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
5994                 RETURN(-EINVAL);
5995
5996         fid = (struct lu_fid *)data->ioc_inlbuf1;
5997
5998         if (!fid_is_sane(fid))
5999                 RETURN(-EINVAL);
6000
6001         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
6002
6003         lh = &mti->mti_lh[MDT_LH_PARENT];
6004         mdt_lock_reg_init(lh, LCK_CR);
6005
6006         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
6007         if (IS_ERR(obj))
6008                 RETURN(PTR_ERR(obj));
6009
6010         if (mdt_object_remote(obj)) {
6011                 rc = -EREMOTE;
6012                 /**
6013                  * before calling version get the correct MDS should be
6014                  * fid, this is error to find remote object here
6015                  */
6016                 CERROR("nonlocal object "DFID"\n", PFID(fid));
6017         } else if (!mdt_object_exists(obj)) {
6018                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
6019                 rc = -ENOENT;
6020         } else {
6021                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
6022                *(__u64 *)data->ioc_inlbuf2 = version;
6023                 rc = 0;
6024         }
6025         mdt_object_unlock_put(mti, obj, lh, 1);
6026         RETURN(rc);
6027 }
6028
6029 /* ioctls on obd dev */
6030 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
6031                          void *karg, void __user *uarg)
6032 {
6033         struct lu_env      env;
6034         struct obd_device *obd = exp->exp_obd;
6035         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
6036         struct dt_device  *dt = mdt->mdt_bottom;
6037         int rc;
6038
6039         ENTRY;
6040         CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd);
6041         rc = lu_env_init(&env, LCT_MD_THREAD);
6042         if (rc)
6043                 RETURN(rc);
6044
6045         switch (cmd) {
6046         case OBD_IOC_SYNC:
6047                 rc = mdt_device_sync(&env, mdt);
6048                 break;
6049         case OBD_IOC_SET_READONLY:
6050                 rc = dt_sync(&env, dt);
6051                 if (rc == 0)
6052                         rc = dt_ro(&env, dt);
6053                 break;
6054         case OBD_IOC_ABORT_RECOVERY:
6055                 CERROR("%s: Aborting recovery for device\n", mdt_obd_name(mdt));
6056                 obd->obd_abort_recovery = 1;
6057                 target_stop_recovery_thread(obd);
6058                 rc = 0;
6059                 break;
6060         case OBD_IOC_CHANGELOG_REG:
6061         case OBD_IOC_CHANGELOG_DEREG:
6062         case OBD_IOC_CHANGELOG_CLEAR:
6063                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(&env,
6064                                                            mdt->mdt_child,
6065                                                            cmd, len, karg);
6066                 break;
6067         case OBD_IOC_START_LFSCK: {
6068                 struct md_device *next = mdt->mdt_child;
6069                 struct obd_ioctl_data *data = karg;
6070                 struct lfsck_start_param lsp;
6071
6072                 if (unlikely(data == NULL)) {
6073                         rc = -EINVAL;
6074                         break;
6075                 }
6076
6077                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
6078                 lsp.lsp_index_valid = 0;
6079                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
6080                 break;
6081         }
6082         case OBD_IOC_STOP_LFSCK: {
6083                 struct md_device        *next = mdt->mdt_child;
6084                 struct obd_ioctl_data   *data = karg;
6085                 struct lfsck_stop        stop;
6086
6087                 stop.ls_status = LS_STOPPED;
6088                 /* Old lfsck utils may pass NULL @stop. */
6089                 if (data->ioc_inlbuf1 == NULL)
6090                         stop.ls_flags = 0;
6091                 else
6092                         stop.ls_flags =
6093                         ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
6094
6095                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
6096                 break;
6097         }
6098         case OBD_IOC_QUERY_LFSCK: {
6099                 struct md_device        *next = mdt->mdt_child;
6100                 struct obd_ioctl_data   *data = karg;
6101
6102                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0,
6103                                                  data->ioc_inlbuf1);
6104                 break;
6105         }
6106         case OBD_IOC_GET_OBJ_VERSION: {
6107                 struct mdt_thread_info *mti;
6108                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6109                 memset(mti, 0, sizeof *mti);
6110                 mti->mti_env = &env;
6111                 mti->mti_mdt = mdt;
6112                 mti->mti_exp = exp;
6113
6114                 rc = mdt_ioc_version_get(mti, karg);
6115                 break;
6116         }
6117         case OBD_IOC_CATLOGLIST: {
6118                 struct mdt_thread_info *mti;
6119
6120                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
6121                 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
6122                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
6123                                        &mti->mti_tmp_fid1);
6124                 break;
6125          }
6126         default:
6127                 rc = -EOPNOTSUPP;
6128                 CERROR("%s: Not supported cmd = %d, rc = %d\n",
6129                         mdt_obd_name(mdt), cmd, rc);
6130         }
6131
6132         lu_env_fini(&env);
6133         RETURN(rc);
6134 }
6135
6136 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
6137 {
6138         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
6139         int rc;
6140         ENTRY;
6141
6142         if (!mdt->mdt_skip_lfsck && !mdt->mdt_bottom->dd_rdonly) {
6143                 struct lfsck_start_param lsp;
6144
6145                 lsp.lsp_start = NULL;
6146                 lsp.lsp_index_valid = 0;
6147                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
6148                                                            OBD_IOC_START_LFSCK,
6149                                                            0, &lsp);
6150                 if (rc != 0 && rc != -EALREADY)
6151                         CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
6152                               mdt_obd_name(mdt), rc);
6153         }
6154
6155         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
6156         RETURN(rc);
6157 }
6158
6159 static int mdt_obd_postrecov(struct obd_device *obd)
6160 {
6161         struct lu_env env;
6162         int rc;
6163
6164         rc = lu_env_init(&env, LCT_MD_THREAD);
6165         if (rc)
6166                 RETURN(rc);
6167         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
6168         lu_env_fini(&env);
6169         return rc;
6170 }
6171
6172 static struct obd_ops mdt_obd_device_ops = {
6173         .o_owner          = THIS_MODULE,
6174         .o_set_info_async = mdt_obd_set_info_async,
6175         .o_connect        = mdt_obd_connect,
6176         .o_reconnect      = mdt_obd_reconnect,
6177         .o_disconnect     = mdt_obd_disconnect,
6178         .o_init_export    = mdt_init_export,
6179         .o_destroy_export = mdt_destroy_export,
6180         .o_iocontrol      = mdt_iocontrol,
6181         .o_postrecov      = mdt_obd_postrecov,
6182 };
6183
6184 static struct lu_device* mdt_device_fini(const struct lu_env *env,
6185                                          struct lu_device *d)
6186 {
6187         struct mdt_device *m = mdt_dev(d);
6188         ENTRY;
6189
6190         mdt_fini(env, m);
6191         RETURN(NULL);
6192 }
6193
6194 static struct lu_device *mdt_device_free(const struct lu_env *env,
6195                                          struct lu_device *d)
6196 {
6197         struct mdt_device *m = mdt_dev(d);
6198         ENTRY;
6199
6200         lu_device_fini(&m->mdt_lu_dev);
6201         OBD_FREE_PTR(m);
6202
6203         RETURN(NULL);
6204 }
6205
6206 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
6207                                           struct lu_device_type *t,
6208                                           struct lustre_cfg *cfg)
6209 {
6210         struct lu_device  *l;
6211         struct mdt_device *m;
6212
6213         OBD_ALLOC_PTR(m);
6214         if (m != NULL) {
6215                 int rc;
6216
6217                 l = &m->mdt_lu_dev;
6218                 rc = mdt_init0(env, m, t, cfg);
6219                 if (rc != 0) {
6220                         mdt_device_free(env, l);
6221                         l = ERR_PTR(rc);
6222                         return l;
6223                 }
6224         } else
6225                 l = ERR_PTR(-ENOMEM);
6226         return l;
6227 }
6228
6229 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
6230 LU_KEY_INIT(mdt, struct mdt_thread_info);
6231
6232 static void mdt_key_fini(const struct lu_context *ctx,
6233                          struct lu_context_key *key, void* data)
6234 {
6235         struct mdt_thread_info *info = data;
6236
6237         if (info->mti_big_lmm) {
6238                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
6239                 info->mti_big_lmm = NULL;
6240                 info->mti_big_lmmsize = 0;
6241         }
6242         OBD_FREE_PTR(info);
6243 }
6244
6245 /* context key: mdt_thread_key */
6246 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
6247
6248 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
6249 {
6250         return lu_ucred(info->mti_env);
6251 }
6252
6253 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
6254 {
6255         return lu_ucred_check(info->mti_env);
6256 }
6257
6258 /**
6259  * Enable/disable COS (Commit On Sharing).
6260  *
6261  * Set/Clear the COS flag in mdt options.
6262  *
6263  * \param mdt mdt device
6264  * \param val 0 disables COS, other values enable COS
6265  */
6266 void mdt_enable_cos(struct mdt_device *mdt, int val)
6267 {
6268         struct lu_env env;
6269         int rc;
6270
6271         mdt->mdt_opts.mo_cos = !!val;
6272         rc = lu_env_init(&env, LCT_LOCAL);
6273         if (unlikely(rc != 0)) {
6274                 CWARN("%s: lu_env initialization failed, cannot "
6275                       "sync: rc = %d\n", mdt_obd_name(mdt), rc);
6276                 return;
6277         }
6278         mdt_device_sync(&env, mdt);
6279         lu_env_fini(&env);
6280 }
6281
6282 /**
6283  * Check COS (Commit On Sharing) status.
6284  *
6285  * Return COS flag status.
6286  *
6287  * \param mdt mdt device
6288  */
6289 int mdt_cos_is_enabled(struct mdt_device *mdt)
6290 {
6291         return mdt->mdt_opts.mo_cos != 0;
6292 }
6293
6294 static struct lu_device_type_operations mdt_device_type_ops = {
6295         .ldto_device_alloc = mdt_device_alloc,
6296         .ldto_device_free  = mdt_device_free,
6297         .ldto_device_fini  = mdt_device_fini
6298 };
6299
6300 static struct lu_device_type mdt_device_type = {
6301         .ldt_tags     = LU_DEVICE_MD,
6302         .ldt_name     = LUSTRE_MDT_NAME,
6303         .ldt_ops      = &mdt_device_type_ops,
6304         .ldt_ctx_tags = LCT_MD_THREAD
6305 };
6306
6307 static int __init mdt_init(void)
6308 {
6309         int rc;
6310
6311         CLASSERT(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") ==
6312                  FID_NOBRACE_LEN + 1);
6313         CLASSERT(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") ==
6314                  FID_LEN + 1);
6315         rc = lu_kmem_init(mdt_caches);
6316         if (rc)
6317                 return rc;
6318
6319         rc = mds_mod_init();
6320         if (rc)
6321                 GOTO(lu_fini, rc);
6322
6323         rc = class_register_type(&mdt_obd_device_ops, NULL, true, NULL,
6324                                  LUSTRE_MDT_NAME, &mdt_device_type);
6325         if (rc)
6326                 GOTO(mds_fini, rc);
6327 lu_fini:
6328         if (rc)
6329                 lu_kmem_fini(mdt_caches);
6330 mds_fini:
6331         if (rc)
6332                 mds_mod_exit();
6333         return rc;
6334 }
6335
6336 static void __exit mdt_exit(void)
6337 {
6338         class_unregister_type(LUSTRE_MDT_NAME);
6339         mds_mod_exit();
6340         lu_kmem_fini(mdt_caches);
6341 }
6342
6343 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
6344 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
6345 MODULE_VERSION(LUSTRE_VERSION_STRING);
6346 MODULE_LICENSE("GPL");
6347
6348 module_init(mdt_init);
6349 module_exit(mdt_exit);