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